CN116563841B - Detection method and detection device for power distribution network equipment identification plate and electronic equipment - Google Patents
Detection method and detection device for power distribution network equipment identification plate and electronic equipment Download PDFInfo
- Publication number
- CN116563841B CN116563841B CN202310830073.2A CN202310830073A CN116563841B CN 116563841 B CN116563841 B CN 116563841B CN 202310830073 A CN202310830073 A CN 202310830073A CN 116563841 B CN116563841 B CN 116563841B
- Authority
- CN
- China
- Prior art keywords
- image
- signboard
- distribution network
- power distribution
- network equipment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000009826 distribution Methods 0.000 title claims abstract description 391
- 238000001514 detection method Methods 0.000 title claims abstract description 149
- 238000000034 method Methods 0.000 claims abstract description 80
- 238000005259 measurement Methods 0.000 claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 claims description 203
- 230000006870 function Effects 0.000 claims description 93
- 238000012549 training Methods 0.000 claims description 37
- 238000010801 machine learning Methods 0.000 claims description 14
- 238000011524 similarity measure Methods 0.000 claims 1
- 238000007689 inspection Methods 0.000 abstract description 30
- 230000008569 process Effects 0.000 description 23
- 238000005457 optimization Methods 0.000 description 22
- 238000004590 computer program Methods 0.000 description 12
- 238000010586 diagram Methods 0.000 description 10
- 238000002372 labelling Methods 0.000 description 10
- 238000003062 neural network model Methods 0.000 description 10
- 238000003860 storage Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 9
- 230000008901 benefit Effects 0.000 description 8
- 238000005286 illumination Methods 0.000 description 8
- 238000010606 normalization Methods 0.000 description 8
- 230000005540 biological transmission Effects 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 230000003321 amplification Effects 0.000 description 4
- 238000004140 cleaning Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000011478 gradient descent method Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 238000003199 nucleic acid amplification method Methods 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 4
- 238000012935 Averaging Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 238000011895 specific detection Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The application provides a detection method, a detection device and electronic equipment for a power distribution network equipment signboard. The method comprises the following steps: acquiring an image of power distribution network equipment to obtain the image of the power distribution network equipment; determining the category of the power distribution network equipment, and determining a signboard reference category and a signboard reference image corresponding to the category of the power distribution network equipment in the association database according to the category of the power distribution network equipment; determining the category of the identification plate in the power distribution network equipment image and the position information of the identification plate image, comparing the category of the identification plate with the reference category of the identification plate, and determining whether the category of the identification plate is correct or not; obtaining a signboard image according to the position information of the signboard, carrying out similarity measurement on the signboard image and a signboard reference image, determining whether the signboard image is clear or not, and replacing the signboard image under the condition that the signboard image is not clear. The application solves the problem of low efficiency of manual inspection of the equipment identification plate of the power distribution network, and achieves the purpose of high-efficiency inspection of the equipment identification plate.
Description
Technical Field
The application relates to the technical field of power inspection, in particular to a detection method, a detection device, an electronic device and electronic equipment for a power distribution network equipment signboard.
Background
With the rapid development of the power industry, distribution network equipment needs to be widely deployed in various areas in large quantities. In order to ensure the electricity utilization safety, the life and property safety of people and avoid the occurrence of environmental personal safety hazards of power distribution network equipment, the electric power inspection technology aiming at the power distribution network equipment is endlessly layered, wherein the detection technology of the environmental personal safety hazards, which is provided with unclear and inaccurate power distribution network equipment identification labels, is an important ring.
The traditional inspection mode is mainly based on manual work, and the following problems often occur in the manual inspection process: when checking the power distribution network equipment identification label, various detection and recording equipment is required to be worn, so that the inspection cost is high and the detection efficiency is low; some distribution network equipment is in a special geographic position or the equipment is difficult to detect, and when special inspection equipment such as an unmanned plane and a telescope is not available, the identification label is difficult to detect effectively; when severe extreme weather conditions such as strong wind, severe cold, high temperature and the like are met, the difficulty of manual inspection is increased, so that the inspection efficiency is greatly reduced; manual inspection is affected by personal ability of inspectors, and problems such as incapability of accurately or timely distinguishing potential safety hazards and the like can be caused.
Therefore, how to ensure efficient and accurate execution of the power inspection task is a problem to be solved.
Disclosure of Invention
The application mainly aims to provide a detection method, a detection device, an electronic device and electronic equipment for a power distribution network equipment signboard, and aims to at least solve the problem that in the prior art, the efficiency of manually inspecting the power distribution network equipment signboard is low.
In order to achieve the above object, according to one aspect of the present application, there is provided a method for detecting a power distribution network device signboard, including: acquiring an image of power distribution network equipment to obtain an image of the power distribution network equipment, wherein the image of the power distribution network equipment comprises a signboard image; analyzing the power distribution network equipment image by using a residual neural network, determining the type of the power distribution network equipment, and determining a signboard reference type and a signboard reference image corresponding to the type of the power distribution network equipment in a correlation database according to the type of the power distribution network equipment, wherein the residual neural network is trained by using a plurality of groups of first data through machine learning, and each group of data in the plurality of groups of first data comprises: the association database comprises standard images corresponding to the power distribution network equipment in all categories and the identification plate reference image corresponding to each category; analyzing the power distribution network equipment image by using a YOLOX target detection neural network, determining the category of a signboard in the power distribution network equipment image and the position information of the signboard image, comparing the category of the signboard with the reference category of the signboard, and determining whether the category of the signboard is correct, wherein the YOLOX target detection neural network is trained by using a plurality of sets of second data through machine learning, and each set of data in the plurality of sets of second data comprises: the signboard images of all the categories and the categories corresponding to the signboards of all the categories; and acquiring the signboard image according to the position information of the signboard, carrying out similarity measurement on the signboard image and the signboard reference image, determining whether the signboard image is clear or not, and replacing the signboard image under the condition that the signboard image is not clear.
Optionally, analyzing the power distribution network device image by using a residual neural network, determining a category of the power distribution network device, including: inputting the power distribution network equipment image into the residual neural network, and obtaining an output result of the residual neural network to obtain the corresponding prediction probability of the power distribution network equipment of each category; and taking the category of the power distribution network equipment with the maximum prediction probability as the category of the power distribution network equipment.
Optionally, analyzing the power distribution network device image by using a YOLOX target detection neural network, and determining the category of the signboard in the power distribution network device image and the position information of the signboard image, including: inputting the power distribution network equipment image into the YOLOX target detection neural network; by passing throughDetermining the position information of the identification plate image in the distribution network equipment image and passing +.>Determining the category of the identification plate, wherein X represents the abscissa of the identification plate image on the power distribution network equipment image, Y represents the ordinate of the identification plate image on the power distribution network equipment image, grid x Abscissa, grid representing pixel point positions of characteristic images of power distribution network equipment y An ordinate, offset, representing the pixel position of the characteristic image of the distribution network device x An offset representing an abscissa offset of the signboard image y Representing the ordinate offset and stride of the signboard image x Representing the abscissa ratio of the signboard image to the distribution network equipment image, and stride y Representing the ordinate ratio of the signboard image mapped to the power distribution network equipment image, W representing the width of the signboard reference image, H representing the height of the signboard reference image, W representing the width of the signboard image, H representing the height of the signboard image, R b Representing the predictive probability of the sign for each category, O b k An output vector representing a category of the kth signboard corresponding to the YOLOX target detection neural network, O b i And an output vector representing the category of the ith signboard corresponding to the YOLOX target detection neural network.
Optionally, aPerforming similarity measurement on the signboard image and the signboard reference image to determine whether the signboard image is clear, including: according to the formulaNormalizing the signboard image and the signboard reference image, wherein RI (x, y) represents the signboard reference image, PI (x, y) represents the signboard image, RI '(x, y) represents the normalized signboard reference image, PI' (x, y) represents the normalized signboard image, RI max Representing the maximum value of pixel values of all pixel points in the reference image of the signboard, RI min Minimum value of pixel values representing all pixel points in the reference image of the signboard, PI max Maximum value of pixel values representing all pixel points in the signboard image, PI min Representing the minimum value of the pixel values of all pixel points in the signboard image; according to the formula->Calculating the similarity score of the signboard image and the signboard reference image, determining that the signboard image is clear when the similarity score is larger than a preset threshold value, and determining that the signboard image is not clear when the similarity score is smaller than the preset threshold value, wherein score represents the similarity score.
Optionally, the detection method further comprises: using cross entropy loss functionAs a loss function of an initial residual neural network, determining that the initial residual neural network training is completed under the condition that the cross entropy loss function converges to obtain the residual neural network, wherein L is as follows cla s Representing the cross entropy loss function, y ic N represents the number of the distribution network equipment, M represents the category number of the distribution network equipment and p is a sign function ic Representing the probability that the ith power distribution network device belongs to the c-th category. Optionally, the detectingThe method further comprises the steps of: use of the focal loss function +.>As a loss function of the initial YOLOX target detection neural network, an intermediate YOLOX target detection neural network is obtained with convergence of the focal loss function, wherein L box b Representing the focal loss function, p t Representing the proximity degree of the predicted category and the real category of the signboard, wherein gamma represents the adjustable factor of the YOLOX target detection neural network; using the cross-ratio loss function->Obtaining the YOLOX target detection neural network as a loss function of the intermediate YOLOX target detection neural network under the condition that the cross-ratio loss function is converged, wherein L box c And representing the cross ratio loss function, wherein A represents a predicted area of the signboard image, and B represents a real area of the signboard image.
Optionally, the detection method further comprises: according to the formulaCalculating an average value of the precision of the prediction result of the residual neural network to obtain an average precision value, and determining that the precision of the prediction result of the residual neural network meets the requirement under the condition that the average precision value is within a first preset range, wherein the prediction result of the residual neural network represents the type of the power distribution network equipment predicted by the residual neural network, mAP represents the average precision value, P (R) represents a PR curve function, R represents a recall maximum value, and the PR curve function is used for representing the relationship between the precision of the type of the power distribution network equipment predicted by the residual neural network and the recall; according to the formula- >Calculating the overlapping degree of the predicted result of the YOLOX target detection neural network, and determining the overlapping degree in the case that the overlapping degree is within a second preset rangeThe accuracy of the predicted result of the YOLOX target detection neural network satisfies a requirement, wherein the predicted result of the YOLOX target detection neural network represents the category of the signboard predicted by the YOLOX target detection neural network and the position information of the signboard image, ioU represents the overlapping degree, the overlapping degree represents the accuracy of the predicted result of the YOLOX target detection neural network, a represents the predicted region of the signboard image, and B represents the real region of the signboard image.
According to another aspect of the present application, there is provided a detection apparatus for a power distribution network device signboard, including: the power distribution network equipment image acquisition unit is used for acquiring the image of the power distribution network equipment to obtain the power distribution network equipment image, wherein the power distribution network equipment image comprises a signboard image; the first determining unit is configured to analyze the power distribution network device image by using a residual neural network, determine a category of the power distribution network device, and determine a reference category of a signboard and a reference image of the signboard, which correspond to the category of the power distribution network device in an association database according to the category of the power distribution network device, where the residual neural network is trained by using a plurality of sets of first data through machine learning, and each set of data in the plurality of sets of first data includes: the association database comprises standard images corresponding to the power distribution network equipment in all categories and the identification plate reference image corresponding to each category; a second determining unit, configured to analyze the power distribution network device image by using a YOLOX target detection neural network, determine a category of a signboard in the power distribution network device image and position information of the signboard image, compare the category of the signboard with the reference category of the signboard, and determine whether the category of the signboard is correct, where the YOLOX target detection neural network is trained by using a plurality of sets of second data through machine learning, and each set of data in the plurality of sets of second data includes: the signboard images of all the categories and the categories corresponding to the signboards of all the categories; the replacing unit is used for acquiring the signboard image according to the position information of the signboard, carrying out similarity measurement on the signboard image and the signboard reference image, determining whether the signboard image is clear or not, and replacing the signboard image under the condition that the signboard image is not clear.
According to a further aspect of the present application there is provided an electronic device comprising a memory and a processor, the memory having stored therein a computer program, the processor being arranged to perform any one of the detection methods by means of the computer program.
According to still another aspect of the present application, there is provided an electronic apparatus including: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising instructions for performing any one of the detection methods.
By applying the technical scheme, firstly, a power distribution network equipment image is obtained, the power distribution network equipment image is analyzed through a residual neural network, the category of the power distribution network equipment is determined, and the category of the power distribution network equipment is searched in a correlation database according to the category of the power distribution network equipment, so that a signboard reference category and a signboard reference image corresponding to the category of the power distribution network equipment are obtained; and then, analyzing the power distribution network equipment image through a YOLOX target detection neural network, determining the position information of the signboard image and the category of the signboard, comparing the category of the signboard with the signboard reference category, determining whether the signboard category is correct, acquiring the signboard image according to the position information of the signboard image on the power distribution network equipment, carrying out similarity measurement on the signboard image and the signboard reference image, and determining whether the signboard image is clear. Compared with the prior art, the method and the device for manually inspecting the identification plate on the power distribution network equipment generally adopt a manual inspection mode to inspect and judge the identification plate on the power distribution network equipment, not only have low efficiency, but also can solve the problem of inaccurate manual identification, the method and the device can be used for analyzing the image of the power distribution network equipment by combining a neural network, comparing the image with a reference type and the reference image, and calculating the value of similarity measurement, so that whether the identification plate type and the identification plate image are correct and clear is determined, and the problem of inaccurate judgment result possibly existing in manual inspection is avoided while the inspection efficiency is improved, thereby solving the problem of low efficiency of manually inspecting the identification plate of the power distribution network equipment, and achieving the purpose of efficiently inspecting the equipment identification plate.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:
fig. 1 shows a hardware block diagram of a mobile terminal for executing a detection method of a power distribution network equipment signboard according to an embodiment of the present application;
fig. 2 is a schematic flow chart of a detection method of a power distribution network equipment signboard according to an embodiment of the present application;
fig. 3 is a schematic flow chart of a detection method of a specific power distribution network equipment signboard according to an embodiment of the present application;
fig. 4 is a schematic flow chart of judging whether a signboard is clear or not in a specific detection method of a power distribution network equipment signboard according to an embodiment of the present application;
fig. 5 shows a block diagram of a detection device for a power distribution network equipment signboard according to an embodiment of the present application.
Wherein the above figures include the following reference numerals:
102. a processor; 104. a memory; 106. a transmission device; 108. and an input/output device.
Detailed Description
It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other. The application will be described in detail below with reference to the drawings in connection with embodiments.
In order that those skilled in the art will better understand the present application, a technical solution in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present application without making any inventive effort, shall fall within the scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate in order to describe the embodiments of the application herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
As described in the background art, in the prior art, the efficiency of manually inspecting the power distribution network equipment identification plate is low, and in order to solve the problem of low efficiency of manually inspecting the power distribution network equipment identification plate, the embodiment of the application provides a detection method, a detection device, an electronic device and electronic equipment for the power distribution network equipment identification plate.
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application.
The method embodiments provided in the embodiments of the present application may be performed in a mobile terminal, a computer terminal or similar computing device. Taking the operation on a mobile terminal as an example, fig. 1 is a block diagram of a hardware structure of a mobile terminal of a detection method of a power distribution network equipment signboard according to an embodiment of the present application. As shown in fig. 1, a mobile terminal may include one or more (only one is shown in fig. 1) processors 102 (the processor 102 may include, but is not limited to, a microprocessor MCU or a processing device such as a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 for communication functions and an input-output device 108. It will be appreciated by those skilled in the art that the structure shown in fig. 1 is merely illustrative and not limiting of the structure of the mobile terminal described above. For example, the mobile terminal may also include more or fewer components than shown in fig. 1, or have a different configuration than shown in fig. 1.
The memory 104 may be used to store a computer program, for example, a software program of application software and a module, such as a computer program corresponding to a method for detecting a power distribution network equipment identification board in an embodiment of the present invention, and the processor 102 executes the computer program stored in the memory 104, thereby performing various functional applications and data processing, that is, implementing the method described above. Memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely located relative to the processor 102, which may be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or transmit data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, simply referred to as NIC) that can connect to other network devices through a base station to communicate with the internet. In one example, the transmission device 106 may be a Radio Frequency (RF) module, which is configured to communicate with the internet wirelessly.
In the present embodiment, a method for detecting a power distribution network device signboard operating on a mobile terminal, a computer terminal or the like is provided, and it is to be noted that the steps shown in the flowchart of the drawings may be executed in a computer system such as a set of computer executable instructions, and that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown or described herein.
Fig. 2 is a flowchart of a method for detecting a power distribution network equipment signboard according to an embodiment of the present application. As shown in fig. 2, the method comprises the steps of:
step S201, acquiring an image of power distribution network equipment to obtain the image of the power distribution network equipment, wherein the image of the power distribution network equipment comprises a signboard image;
specifically, the image of distribution network equipment can acquire through multiple image acquisition equipment, and image acquisition equipment includes but is not limited to handheld image acquisition equipment, unmanned aerial vehicle machine carries camera, surveillance camera head, to intelligent inspection technique, mainly adopts unmanned aerial vehicle machine to carry camera, surveillance camera head to acquire distribution network equipment image, and wherein, unmanned aerial vehicle machine carries the camera to acquire distribution network equipment image has following advantage: the method has the advantages that the power distribution network equipment images can be shot at multiple angles, and the richness of the data set is improved in the construction of the data set and the training stage of the neural network model, so that the robustness of the neural network is enhanced; in the prediction stage of the neural network, the multi-angle images can be used for simultaneously predicting, so that the accuracy of the algorithm is improved. The monitoring camera acquires the image with the following advantages: the method has the advantages that the image of the power distribution network equipment can be shot in real time, the image of the power distribution network equipment under various environments and illumination can be obtained, the data set can be enriched, the algorithm accuracy can be improved, the power distribution network equipment can be monitored in real time, and the safety of the power distribution network is guaranteed. Because the image of the signboard is generally attached to the power distribution network device, the image of the signboard is included in the image of the power distribution network device captured by the image capturing device, such as an unmanned aerial vehicle camera, that is, the image of the power distribution network device includes the image of the signboard.
Step S202, analyzing the power distribution network equipment image by using a residual neural network, determining the type of the power distribution network equipment, and determining a signboard reference type and a signboard reference image corresponding to the type of the power distribution network equipment in a correlation database according to the type of the power distribution network equipment, wherein the residual neural network is trained by using a plurality of groups of first data through machine learning, and each group of data in the plurality of groups of first data comprises: the association database comprises standard images corresponding to the power distribution network equipment in all categories and the identification plate reference image corresponding to each category;
in a specific implementation process, the steps construct two image data sets and an associated database. Firstly, data cleaning is carried out on acquired power distribution network equipment images, images with poor imaging effects are selected and removed, the images are prevented from affecting the optimization of the whole neural network parameters, then classification labeling and target detection labeling (labeling of the types of identification plates in the power distribution network equipment images) are carried out on the processed power distribution network equipment images, the power distribution network equipment images and the power distribution network equipment images after classification labeling are constructed into a power distribution network equipment classification data set, the power distribution network equipment images and the power distribution network equipment images after target detection labeling are constructed into identification plate image data sets, finally the power distribution network equipment images and the identification plate images are constructed into associated databases, and namely the power distribution network equipment image data sets, the identification plate image data sets and the associated databases of the power distribution network equipment images and the identification plate images are respectively established. The power distribution network equipment image data set is used for training an initial residual neural network model, and the residual neural network after training is completed analyzes the acquired power distribution network equipment image to determine the type of the power distribution network equipment in the power distribution network equipment image. Because the association database contains standard images of all types of power distribution network equipment and identification plate reference images of each type of power distribution network equipment, the identification plate reference images and the identification plate reference images corresponding to the power distribution network equipment in the power distribution network equipment images are obtained by searching in the association database according to the determined types in the power distribution network equipment images. The association database is represented by S, the reference category of the signboard is represented by RC, and the reference image of the signboard is represented by RI, and the following relationship is satisfied Cls represents the category of the power distribution network equipment, and the identification plate reference category RC and the identification plate reference image RI are in one-to-one correspondence and are unique.
Step S203, analyzing the power distribution network device image by using a YOLOX target detection neural network, determining a category of a signboard in the power distribution network device image and position information of the signboard image, comparing the category of the signboard with the reference category of the signboard, and determining whether the category of the signboard is correct, wherein the YOLOX target detection neural network is trained by using a plurality of sets of second data through machine learning, and each set of data in the plurality of sets of second data includes: all kinds of the signboard images and all kinds of the classes corresponding to the signboard;
specifically, the signboard image data set is used for training an initial YOLOX target detection neural network, and the YOLOX target detection neural network after training is completed analyzes the acquired power distribution network equipment image to determine the position of the signboard image and the type of the signboard in the power distribution network equipment image. The identification board reference class and the identification board reference image are determined, the class of the identification board in the determined power distribution network equipment image is compared with the identification board reference class, whether the class of the identification board is the same as the identification board reference class or not is determined by using the identification board reference class as the reference, if the class of the identification board is the same as the identification board reference class, the identification board attached to the power distribution network equipment is correct, the next judgment is continued, and if the class of the identification board attached to the power distribution network equipment is incorrect, the identification board is required to be corrected. The types of power distribution network equipment are divided into transformers, high-voltage cabinets, low-voltage cabinets, bus bridges, high-voltage cables and the like, the identification boards of the equipment in each type are obviously different, and the identification board images usually comprise warning sentences, warning symbols, warning icons and the like so as to play a role in classifying and warning.
Step S204, obtaining the signboard image according to the position information of the signboard, carrying out similarity measurement on the signboard image and the signboard reference image, determining whether the signboard image is clear or not, and replacing the signboard image when the signboard image is not clear.
Specifically, after the identification board is determined to be correct in category, the identification board image can be obtained according to the position information of the identification board in the power distribution network equipment image, and the identification board image and the identification board reference image are subjected to similarity measurement, namely, one measurement for evaluating the similarity degree between the identification board image and the identification board reference image is larger as the similarity measurement is closer, otherwise, the similarity measurement is smaller as the similarity measurement is farther. And determining whether the signboard image is clear or not according to the similarity measurement value, and if not, indicating that the signboard image needs to be replaced.
Through the embodiment, firstly, a power distribution network equipment image is obtained, the power distribution network equipment image is analyzed through a residual neural network, the category of the power distribution network equipment is determined, and searching is carried out in a correlation database according to the category of the power distribution network equipment, so that a signboard reference category and a signboard reference image corresponding to the category of the power distribution network equipment are obtained; and then, analyzing the power distribution network equipment image through a YOLOX target detection neural network, determining the position information of the signboard image and the category of the signboard, comparing the category of the signboard with the signboard reference category, determining whether the signboard category is correct, acquiring the signboard image according to the position information of the signboard image on the power distribution network equipment, carrying out similarity measurement on the signboard image and the signboard reference image, and determining whether the signboard image is clear. Compared with the prior art, the method and the device for manually inspecting the identification plate on the power distribution network equipment generally adopt a manual inspection mode to inspect and judge the identification plate on the power distribution network equipment, not only have low efficiency, but also can solve the problem of inaccurate manual identification, the method and the device can be used for analyzing the image of the power distribution network equipment by combining a neural network, comparing the image with a reference type and the reference image, and calculating the value of similarity measurement, so that whether the identification plate type and the identification plate image are correct and clear is determined, and the problem of inaccurate judgment result possibly existing in manual inspection is avoided while the inspection efficiency is improved, thereby solving the problem of low efficiency of manually inspecting the identification plate of the power distribution network equipment, and achieving the purpose of efficiently inspecting the equipment identification plate.
In a specific implementation process, the step S202 may be implemented by the following steps: inputting the power distribution network equipment image into the residual neural network, and obtaining an output result of the residual neural network to obtain the corresponding prediction probability of the power distribution network equipment of each category; and taking the type of the power distribution network equipment with the highest prediction probability as the type of the power distribution network equipment. The method takes the category with the largest prediction probability output by the residual neural network as the category of the power distribution network equipment, so that the category of the power distribution network equipment can be accurately obtained.
In some optional embodiments, the residual neural network outputs a prediction probability, the result of the prediction probability is controlled to be 0-1, and in order to ensure the accuracy of the prediction of the neural network model, the category of the power distribution network equipment corresponding to the maximum value of the prediction probability is selected as the final prediction category.
In order to determine the position of the signboard image in the power distribution network device image, the above step S203 of the present application may be implemented by: inputting the power distribution network equipment image into the YOLOX target detection neural network; by passing throughDetermining the position information of the signboard image in the power distribution network equipment image and passing +. >Determining the category of the identification plate, wherein X represents the abscissa of the identification plate image on the power distribution network equipment image, Y represents the ordinate of the identification plate image on the power distribution network equipment image, grid x Abscissa, grid representing pixel point position of characteristic image of power distribution network equipment y Ordinate, offset representing pixel point position of characteristic image of power distribution network equipment x An offset representing an abscissa offset of the signboard image y Representing the ordinate offset and stride of the signboard image x Representing the abscissa ratio and stride of the image of the signboard mapped to the image of the power distribution network equipment y Representing a ratio of the ordinate of the signboard image to the power distribution network equipment image, W representing a width of the signboard reference image, H representing a height of the signboard reference image, W representing a width of the signboard image, H representing a height of the signboard image, R b Representing the above-mentioned identification for each categoryPredictive probability of cards, O b k An output vector O representing the type of the kth signboard corresponding to the YOLOX target detection neural network b i An output vector indicating the type of the ith signboard corresponding to the YOLOX target detection neural network. According to the method, the position of the signboard image is determined by analyzing the power distribution network equipment image, so that the signboard image can be accurately obtained according to the position information and used for confirming the follow-up signboard.
In the specific implementation process, the image after preprocessing (preprocessing operations such as data cleaning and the like, removing images with poor imaging effect caused by, for example, overexposure of a camera, underexposure of the camera, virtual focus of the camera, dynamic blurring, internal parameter calibration errors and the like, removing deformation and incomplete images of power distribution network equipment and identification labels caused by shooting angles) is normalized:g' represents the normalized signboard image, G represents the signboard image before normalization, mean G Pixel mean, var, representing statistics of an identification signage image dataset G Representing the variance of the statistics of the identification signage image dataset, x representing the abscissa of the target image, and y representing the ordinate of the target image. Taking the normalized power distribution network equipment image as input of a YOLOX target detection neural network, and performing post-processing on the obtained output result, wherein the input result is obtained byDetermining the position information of the signboard image in the power distribution network equipment image, including the abscissa, the ordinate, the length and the width of the signboard image in the power distribution network equipment image, and passing + ->Calculating to obtain the prediction probability of the power distribution network equipment of each category, and taking the category with the highest probability as the category of the power distribution network equipment, stride x Representing the abscissa ratio and stride of the image of the signboard mapped to the image of the power distribution network equipment y Representing the signboardThe image is mapped to the ordinate proportion of the power distribution network equipment image, and the sizes of the signboard image and the power distribution network equipment image are usually different, so that the signboard image is usually smaller than the power distribution network equipment image, and the signboard image needs to be amplified by a certain multiple under the condition that the signboard image and the power distribution network equipment image are of the same size, the amplification multiple on the abscissa is the abscissa proportion, and the amplification multiple on the ordinate is the ordinate proportion.
The above step S204 may also be implemented in other manners, for example: according to the formulaNormalizing the signboard image and the signboard reference image, wherein RI (x, y) represents the signboard reference image, PI (x, y) represents the signboard image, RI '(x, y) represents the normalized signboard reference image, PI' (x, y) represents the normalized signboard image, and RI max Representing the maximum value of pixel values of all pixel points in the reference image of the signboard, RI min Minimum value of pixel values representing all pixel points in the reference image of the signboard, PI max Maximum value of pixel values representing all pixel points in the signboard image, PI min A minimum value representing the pixel values of all the pixel points in the signboard image; according to the formulaCalculating a similarity score of the signboard image and the signboard reference image, determining that the signboard image is clear when the similarity score is greater than a preset threshold value, and determining that the signboard image is not clear when the similarity score is less than the preset threshold value, wherein score represents the similarity score. The method determines whether the signboard images are clear or not through the similarity measurement, so that whether the signboard images are clear or not can be objectively determined through a numerical calculation method. The selection of the reference image RI of the signboard satisfies the following conditions:PC is a prediction category, RC identification plate reference category, S is a correlation database, cls is a category of power distribution network equipment, and the identification plate reference category RC and the identification plate reference image RI are in one-to-one correspondence and are unique.
Specifically, comparing the predicted identification plate category with a corresponding identification plate reference category RC in an associated database, if the corresponding identification plate reference category RC in the associated database is equal to or contains the predicted category PC, the condition that the identification plate category of the power distribution network equipment is inaccurate does not exist, and selecting an identification plate reference image RI in the associated database corresponding to the predicted category PC to judge whether the identification plate is unclear; otherwise, the condition that the identification label is inaccurate exists, and the hidden danger is judged. Then, the predicted position information of the signboard is utilized to obtain a signboard image PI in the image, similarity measurement is carried out on the signboard image PI in the association database corresponding to the selected prediction type PC, and the utilization is utilized Calculating the similarity score, if the similarity score is smaller than a preset threshold value of 0.7, judging that hidden danger exists under the condition that the power distribution network equipment identification plate is unclear, otherwise, judging that the power distribution network equipment identification plate is accurate and clear. When the similarity is measured, the identification plate reference image RI corresponding to the prediction type of the power distribution network equipment in the associated database is required to be adjusted to be the size of the predicted identification plate image PI, and normalization operation is carried out on the identification plate reference image RI and the predicted identification plate image PI, so that the influence of different illumination on the matching score is eliminated. The resizing adopts a bilinear interpolation method:wherein->，X represents the abscissa of the image after interpolation, y represents the ordinate of the image after interpolation, x 1 Represents the abscissa, y of the reference image RI of the signboard 1 Representing a reference image of a signboardOrdinate, x of RI 2 Representing the abscissa, y, of the predicted signboard image PI 1 Representing the ordinate of the predicted signboard image PI. Then, normalization operation is carried out on the reference image RI of the signboard and the predicted image PI of the signboard, so that the influence of factors such as illumination and the like in the image acquisition process is eliminated, and the user is in a position of +.>X represents the abscissa in the image, y represents the ordinate in the image, RI (x, y) can represent the pixel value of each pixel point in the reference image of the signboard, and if factors such as illumination and the like are different, the pixel values are also different, and RI max Maximum value of pixel values representing all pixel points in reference image of signboard, RI min Representing the minimum value of the pixel values of all the pixel points, so that the standard image of the signboard can be normalized through the formula and the maximum value and the minimum value of the pixel values, and in the same way, PI (x, y) can represent the pixel value of each pixel point in the image of the signboard, and PI max Maximum value of pixel values representing all pixel points in the signboard image, PI min The minimum value of the pixel values of all the pixel points in the signboard image is represented, so that the signboard image can be normalized through the formula and the maximum value and the minimum value of the signboard image.
In order to train to obtain the residual neural network model, in some embodiments, the above method may further be implemented by the following steps: using cross entropy loss functionAs a loss function of the initial residual neural network, determining that the initial residual neural network training is completed under the condition that the cross entropy loss function converges to obtain the residual neural network, wherein L is cla s Representing the cross entropy loss function, y ic N represents the number of the distribution network devices, M represents the category number of the distribution network devices and p is a sign function ic Representing the probability that the ith aforementioned distribution network device belongs to the c-th category. The method uses the loss function as the loss function of the residual neural networkIn this way, whether the residual neural network is trained can be determined through the convergence of the cross entropy loss function.
In some alternative embodiments, the neural network model is trained, the network parameters are updated by using a random gradient descent method by using the difference between the minimized predicted value and the actual value as a neural network optimization criterion, and the loss function is designed for training the power distribution network equipment image classification network and the identification label target detection network. Using cross entropy loss functionsAs a classification loss function of the equipment image classification neural network, a small batch random gradient descent method is used as a parameter optimization method: />Wherein->，x t+1 Representing the learnable parameters, x, of the residual neural network after optimization t The learning parameter before the residual neural network optimization is represented, lr represents the learning rate, li represents the loss function, m represents the total number of samples, and i represents the ith sample. When the cross entropy loss function converges, the initial residual neural network training is completed.
In some embodiments, the above method may be further implemented by: using a focal loss function As a loss function of the initial YOLOX target detection neural network, an intermediate YOLOX target detection neural network is obtained in the case where the above-mentioned focal loss function converges, wherein L box b Representing the focal loss function, p t Indicating the proximity of the predicted category and the real category of the signboard, wherein gamma indicates the tunable factor of the YOLOX target detection neural network; using the cross-ratio loss function->Obtaining the YOLOX target detection neural network as a loss function of the intermediate YOLOX target detection neural network, wherein L is box c And a step of representing the cross ratio loss function, wherein A represents a predicted area of the signboard image, and B represents a real area of the signboard image. The method uses a focal loss function and an intersection ratio loss function as the loss functions of the initial YOLOX target detection neural network respectively, so that whether the YOLOX target detection neural network is trained can be determined by whether the loss functions are converged.
Specifically, in the process of training the YOLOX target detection neural network, a focal loss is used as a classification loss function of the signboard image target detection network, so that the influence of unbalance of different signboard image data amounts and identification complexity on network stability is reduced; and adopting the cross ratio loss function as a loss function of the positioning of the signboard, namely, determining the loss function of the position information. Then, an adaptive optimization method is used as a parameter optimization method: ，/>，，/>，/>，x t+1 、x t 、x t-1 The learnable parameters of the YOLOX target detection neural network are represented, f represents the objective function after optimization, lr represents the learning rate, and m represents the step size of each optimization t Index moving average representing gradient, m 0 Initialized to 0, beta 1 Representing the exponential decay rate, controlling the weight distribution of momentum and current gradient, typically takes a value close to 1, where 0.9, v may be defaulted t Finger representing the square of a gradientNumber moving average, v 0 Initialized to 0, beta 2 For the exponential decay rate, the current impact of the square of the gradient before control, similar to the RMSProp algorithm, is weighted averaged, here defaulting to 0.999, +.>Representation of the mean value m of the gradients t Result after deviation correction, ++>Mean value v of representation pair gradient t The result after deviation correction is carried out in order to reduce the influence of the initial training phase bias 0 on the initial training phase due to the initialization of 0, +.>Representing a very small non-zero number, avoiding network optimization failure caused by a divisor of 0 in the optimization process, which is defaulted to 10 -8 . When the focal loss function and the cross-ratio loss function converge, the initial YOLOX target detection neural network training is indicated to be completed.
In order to determine the accuracy of the prediction results of the residual neural network and the YOLOX target detection neural network, in some embodiments, the above method may also be implemented by: according to the formula Calculating an average value of the precision of the prediction result of the residual neural network to obtain an average precision value, and determining that the precision of the prediction result of the residual neural network meets the requirement when the average precision value is within a first preset range, wherein the prediction result of the residual neural network represents the type of the power distribution network equipment predicted by the residual neural network, mAP represents the average precision value, P (R) represents a PR curve function, R represents a recall maximum value, and the PR curve function is used for representing the relationship between the precision of the type of the power distribution network equipment predicted by the residual neural network and the recall; according to the formula->And calculating a degree of overlap of the predicted result of the YOLOX target detection neural network, wherein the predicted result of the YOLOX target detection neural network indicates the type of the signboard predicted by the YOLOX target detection neural network and the position information of the signboard image, ioU indicates the degree of overlap, the degree of overlap indicates the degree of overlap of the predicted result of the YOLOX target detection neural network, a indicates the predicted area of the signboard image, and B indicates the real area of the signboard image, and the degree of overlap is within a second preset range. The method determines the accuracy of the prediction results of the residual neural network and the YOLOX target detection neural network by calculating the values of the maps and IoU.
In some alternative embodiments, the prediction results of the residual neural network and the YOLOX target detection neural network after the training are evaluated, the mAP is adopted to evaluate the classification effect of the residual neural network on the power distribution network equipment, and the mAP and IoU are adopted to evaluate the classification and position location effect of the YOLOX target detection neural network on the identification board. Calculating mAP by adopting a mode of averaging after calculating the average Precision AP of each category, constructing PR curves of Precision and Recall ratio (Recall) for the average Precision AP of each category of power distribution network equipment, and calculating the areas between the PR curves and coordinate axes to obtain the average Precision AP of each category; ioU the area of each predicted signboard image and the area of the real signboard image are calculated by the area-to-area ratio calculation method, and the common area and the total area are calculated respectively and then divided by the total area to obtain the final result. In the training process, when the mAP and IoU evaluation indexes tend to be stable, the prediction accuracy of the corresponding neural network is higher, the trained network parameters are stored as weight files, the repeated training is avoided, meanwhile, the calculation resources are saved, and the network migration and deployment are convenient to use for the prediction of the subsequent neural network. When new power distribution network equipment image data is imported, network parameters can be dynamically optimized, real-time updating of the network is guaranteed, and training of the network parameters is not needed again.
In order to enable those skilled in the art to more clearly understand the technical scheme of the present application, the implementation process of the detection method of the power distribution network equipment identification board of the present application will be described in detail below with reference to specific embodiments.
The embodiment relates to a specific detection method of a power distribution network equipment signboard, as shown in fig. 3 and fig. 4, comprising the following steps:
step S1: acquiring power distribution network equipment images by utilizing various image sensors;
step S2: constructing a distribution network equipment image dataset (comprising distribution network equipment images and categories of the distribution network equipment), constructing a distribution network equipment and identification tag (identification tag) association database (comprising standard images corresponding to all the distribution network equipment categories and identification tag reference images corresponding to each category), and constructing an identification tag image (identification tag image) dataset (comprising identification tag images of all the categories and categories corresponding to all the identification tags);
step S3: training a ResNet classification neural network (residual neural network) using the device identification signage image dataset, and training a Yolox target detection neural network using the identification signage image dataset;
Step S4: classifying devices in an image by using a ResNet network (residual neural network) to obtain the types of power distribution network devices, classifying and positioning identification tags in the image by using a YOLOX network, and determining the types of identification tags in the power distribution network device image and the position information of the identification tag image;
step S5: acquiring the category and the image (the reference category of the identification tag and the reference image of the identification tag) of the association database by using the predicted equipment category (the category of the power distribution network equipment), and judging whether the identification tag is inaccurate or not by comparing the predicted identification tag category with the identification tag category of the association database; obtaining the identification label image and the association database identification label image to carry out similarity measurement by utilizing the predicted identification label positioning information, and judging whether the identification label is unclear;
step S6: the process of determining whether the identification tag is clear is as shown in fig. 4, classifying the power distribution network equipment by using the residual neural network, determining the category of the image of the power distribution network equipment, acquiring the identification tag category C (identification tag reference category) and the image I (identification tag reference image) corresponding to the power distribution network equipment in the association database, comparing the identification tag prediction category cls with the identification tag category C, and judging cls=c? If cls=c is yes, continuing to perform the next step of judgment, and if cls=c is no, determining that the mark and the sign (signboard) are inaccurate;
Step S7: analyzing the power distribution network equipment image by utilizing the YOLOX target detection neural network to obtain a label prediction target box (position information of a label image), acquiring a label target image img (label image) in the image, calculating a similarity score of the label target image img (label image) and a corresponding label image I (label reference image) in an association database, and judging that score is more than or equal to Threshold? In the case of yes, the identification and the sign (signboard) are determined to be accurate and clear, and in the case of no, the identification and the sign (signboard) are determined to be unclear and the signboard is replaced.
The embodiment of the application also provides a detection device for the power distribution network equipment identification plate, and the detection device for the power distribution network equipment identification plate can be used for executing the detection method for the power distribution network equipment identification plate. The device is used for realizing the above embodiments and preferred embodiments, and is not described in detail. As used below, the term "module" may be a combination of software and/or hardware that implements a predetermined function. While the means described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.
The detection device of the power distribution network equipment identification plate provided by the embodiment of the application is described below.
Fig. 5 is a schematic diagram of a detection device for a power distribution network equipment signboard according to an embodiment of the present application. As shown in fig. 5, the apparatus includes:
an obtaining unit 10, configured to obtain an image of a power distribution network device, and obtain an image of the power distribution network device, where the image of the power distribution network device includes a signboard image;
specifically, the image of distribution network equipment can acquire through multiple image acquisition equipment, and image acquisition equipment includes but is not limited to handheld image acquisition equipment, unmanned aerial vehicle machine carries camera, surveillance camera head, to intelligent inspection technique, mainly adopts unmanned aerial vehicle machine to carry camera, surveillance camera head to acquire distribution network equipment image, and wherein, unmanned aerial vehicle machine carries the camera to acquire distribution network equipment image has following advantage: the method has the advantages that the power distribution network equipment images can be shot at multiple angles, and the richness of the data set is improved in the construction of the data set and the training stage of the neural network model, so that the robustness of the neural network is enhanced; in the prediction stage of the neural network, the multi-angle images can be used for simultaneously predicting, so that the accuracy of the algorithm is improved. The monitoring camera acquires the image with the following advantages: the method has the advantages that the image of the power distribution network equipment can be shot in real time, the image of the power distribution network equipment under various environments and illumination can be obtained, the data set can be enriched, the algorithm accuracy can be improved, the power distribution network equipment can be monitored in real time, and the safety of the power distribution network is guaranteed. Because the image of the signboard is generally attached to the power distribution network device, the image of the signboard is included in the image of the power distribution network device captured by the image capturing device, such as an unmanned aerial vehicle camera, that is, the image of the power distribution network device includes the image of the signboard.
A first determining unit 20, configured to analyze the power distribution network device image by using a residual neural network, determine a category of the power distribution network device, and determine a reference category of a signboard and a reference image of a signboard in an association database corresponding to the category of the power distribution network device according to the category of the power distribution network device, where the residual neural network is trained by using a plurality of sets of first data through machine learning, and each set of data in the plurality of sets of first data includes: the association database comprises standard images corresponding to the power distribution network equipment in all categories and the identification plate reference image corresponding to each category;
in a specific implementation process, the steps construct two image data sets and an associated database. Firstly, data cleaning is carried out on acquired power distribution network equipment images, images with poor imaging effects are selected and removed, the images are prevented from affecting the optimization of the whole neural network parameters, then classification labeling and target detection labeling (labeling of the types of identification plates in the power distribution network equipment images) are carried out on the processed power distribution network equipment images, the power distribution network equipment images and the power distribution network equipment images after classification labeling are constructed into a power distribution network equipment classification data set, the power distribution network equipment images and the power distribution network equipment images after target detection labeling are constructed into identification plate image data sets, finally the power distribution network equipment images and the identification plate images are constructed into associated databases, and namely the power distribution network equipment image data sets, the identification plate image data sets and the associated databases of the power distribution network equipment images and the identification plate images are respectively established. The power distribution network equipment image data set is used for training an initial residual neural network model, and the residual neural network after training is completed analyzes the acquired power distribution network equipment image to determine the type of the power distribution network equipment in the power distribution network equipment image. Because the association database contains standard images of all types of power distribution network equipment and identification plate reference images of each type of power distribution network equipment, the identification plate reference images and the identification plate reference images corresponding to the power distribution network equipment in the power distribution network equipment images are obtained by searching in the association database according to the determined types in the power distribution network equipment images. The association database is represented by S, the reference category of the signboard is represented by RC, and the reference image of the signboard is represented by RI, and the following relationship is satisfied Cls represents the category of the power distribution network equipment, and the identification plate reference category RC and the identification plate reference image RI are in one-to-one correspondence and are unique.
A second determining unit 30, configured to analyze the power distribution network device image using a YOLOX target detection neural network, determine a category of a signboard in the power distribution network device image and position information of the signboard image, compare the category of the signboard with the reference category of the signboard, and determine whether the category of the signboard is correct, where the YOLOX target detection neural network is trained by machine learning using a plurality of sets of second data, and each set of data in the plurality of sets of second data includes: all kinds of the signboard images and all kinds of the classes corresponding to the signboard;
specifically, the signboard image data set is used for training an initial YOLOX target detection neural network, and the YOLOX target detection neural network after training is completed analyzes the acquired power distribution network equipment image to determine the position of the signboard image and the type of the signboard in the power distribution network equipment image. The identification board reference class and the identification board reference image are determined, the class of the identification board in the determined power distribution network equipment image is compared with the identification board reference class, whether the class of the identification board is the same as the identification board reference class or not is determined by using the identification board reference class as the reference, if the class of the identification board is the same as the identification board reference class, the identification board attached to the power distribution network equipment is correct, the next judgment is continued, and if the class of the identification board attached to the power distribution network equipment is incorrect, the identification board is required to be corrected. The types of power distribution network equipment are divided into transformers, high-voltage cabinets, low-voltage cabinets, bus bridges, high-voltage cables and the like, the identification boards of the equipment in each type are obviously different, and the identification board images usually comprise warning sentences, warning symbols, warning icons and the like so as to play a role in classifying and warning.
And a replacing unit 40 for acquiring the signboard image based on the position information of the signboard, performing similarity measurement on the signboard image and the signboard reference image, and determining whether the signboard image is clear or not, and replacing the signboard image when the signboard image is not clear.
Specifically, after the identification board is determined to be correct in category, the identification board image can be obtained according to the position information of the identification board in the power distribution network equipment image, and the identification board image and the identification board reference image are subjected to similarity measurement, namely, one measurement for evaluating the similarity degree between the identification board image and the identification board reference image is larger as the similarity measurement is closer, otherwise, the similarity measurement is smaller as the similarity measurement is farther. And determining whether the signboard image is clear or not according to the similarity measurement value, and if not, indicating that the signboard image needs to be replaced.
Through the embodiment, firstly, a power distribution network equipment image is obtained, the power distribution network equipment image is analyzed through a residual neural network, the category of the power distribution network equipment is determined, and searching is carried out in a correlation database according to the category of the power distribution network equipment, so that a signboard reference category and a signboard reference image corresponding to the category of the power distribution network equipment are obtained; and then, analyzing the power distribution network equipment image through a YOLOX target detection neural network, determining the position information of the signboard image and the category of the signboard, comparing the category of the signboard with the signboard reference category, determining whether the signboard category is correct, acquiring the signboard image according to the position information of the signboard image on the power distribution network equipment, carrying out similarity measurement on the signboard image and the signboard reference image, and determining whether the signboard image is clear. Compared with the prior art, the method and the device for manually inspecting the identification plate on the power distribution network equipment generally adopt a manual inspection mode to inspect and judge the identification plate on the power distribution network equipment, not only have low efficiency, but also can solve the problem of inaccurate manual identification, the method and the device can be used for analyzing the image of the power distribution network equipment by combining a neural network, comparing the image with a reference type and the reference image, and calculating the value of similarity measurement, so that whether the identification plate type and the identification plate image are correct and clear is determined, and the problem of inaccurate judgment result possibly existing in manual inspection is avoided while the inspection efficiency is improved, thereby solving the problem of low efficiency of manually inspecting the identification plate of the power distribution network equipment, and achieving the purpose of efficiently inspecting the equipment identification plate.
In the specific implementation process, the second determining unit comprises a first obtaining module and an executing module, wherein the first obtaining module is used for inputting the power distribution network equipment image into the residual neural network and obtaining an output result of the residual neural network to obtain the corresponding prediction probability of the power distribution network equipment of each category; and the execution module is used for taking the type of the power distribution network equipment with the highest prediction probability as the type of the power distribution network equipment. The device takes the category with the largest prediction probability output by the residual neural network as the category of the power distribution network equipment, so that the category of the power distribution network equipment can be accurately obtained.
In some optional embodiments, the residual neural network outputs a prediction probability, the result of the prediction probability is controlled to be 0-1, and in order to ensure the accuracy of the prediction of the neural network model, the category of the power distribution network equipment corresponding to the maximum value of the prediction probability is selected as the final prediction category.
In order to determine the position of the identification plate image in the power distribution network equipment image, the second determining unit comprises an input module and a first determining module, wherein the input module is used for inputting the power distribution network equipment image into the YOLOX target detection neural network; the first determining module is used for passing Determining the position information of the signboard image in the power distribution network equipment image and passing +.>Determining the category of the identification plate, wherein X represents the abscissa of the identification plate image on the power distribution network equipment image, Y represents the ordinate of the identification plate image on the power distribution network equipment image, grid x Abscissa, grid representing pixel point position of characteristic image of power distribution network equipment y Ordinate, offset representing pixel point position of characteristic image of power distribution network equipment x An offset representing an abscissa offset of the signboard image y Representing the ordinate offset and stride of the signboard image x Representing the abscissa ratio and stride of the image of the signboard mapped to the image of the power distribution network equipment y Representing the ordinate ratio of the signboard image to the distribution network equipment image, W representing the width of the signboard reference image, H representing the height of the signboard reference image, W representing the width of the signboard image, H representing the height of the signboard image,R b representing the predictive probability of the above-mentioned signboard for each category, O b k An output vector O representing the type of the kth signboard corresponding to the YOLOX target detection neural network b i An output vector indicating the type of the ith signboard corresponding to the YOLOX target detection neural network. The device determines the position of the signboard image by analyzing the power distribution network equipment image, so that the signboard image can be accurately obtained according to the position information for the confirmation of the follow-up signboard.
In the specific implementation process, the image after preprocessing (preprocessing operations such as data cleaning and the like, removing images with poor imaging effect caused by, for example, overexposure of a camera, underexposure of the camera, virtual focus of the camera, dynamic blurring, internal parameter calibration errors and the like, removing deformation and incomplete images of power distribution network equipment and identification labels caused by shooting angles) is normalized:g' represents the normalized signboard image, G represents the signboard image before normalization, mean G Pixel mean, var, representing statistics of an identification signage image dataset G Representing the variance of the statistics of the identification signage image dataset, x representing the abscissa of the target image, and y representing the ordinate of the target image. Taking the normalized power distribution network equipment image as input of a YOLOX target detection neural network, and performing post-processing on the obtained output result, wherein the input result is obtained by Determining the position information of the signboard image in the power distribution network equipment image, including the abscissa, the ordinate, the length and the width of the signboard image in the power distribution network equipment image, and passing + ->Calculating to obtain the prediction probability of the power distribution network equipment of each category, and taking the category with the highest probability as the category of the power distribution network equipment, stride x Abscissa representing mapping of the signboard image to the distribution network equipment imageProportion, stride y The ordinate ratio of the sign board image mapped to the power distribution network equipment image is represented, and the sign board image and the power distribution network equipment image are usually different in size, so that the sign board image is required to be amplified by a certain multiple under the condition that the sign board image and the power distribution network equipment image are the same in size, the amplification factor on the abscissa is the abscissa ratio, and the amplification factor on the ordinate is the ordinate ratio.
The replacement unit comprises a normalization module and a second determination module, wherein the normalization module is used for determining the replacement unit according to a formulaNormalizing the signboard image and the signboard reference image, wherein RI (x, y) represents the signboard reference image, PI (x, y) represents the signboard image, RI '(x, y) represents the normalized signboard reference image, PI' (x, y) represents the normalized signboard image, and RI max Representing the maximum value of pixel values of all pixel points in the reference image of the signboard, RI min Minimum value of pixel values representing all pixel points in the reference image of the signboard, PI max Maximum value of pixel values representing all pixel points in the signboard image, PI min A minimum value representing the pixel values of all the pixel points in the signboard image; the second determination module is used for determining the formula +.>Calculating a similarity score of the signboard image and the signboard reference image, determining that the signboard image is clear when the similarity score is greater than a preset threshold value, and determining that the signboard image is not clear when the similarity score is less than the preset threshold value, wherein score represents the similarity score. The device determines whether the signboard images are clear through the similarity measurement, so that whether the signboard images are clear can be objectively determined through a numerical calculation device. The selection of the reference image RI of the signboard satisfies the following conditions: />PC is a prediction category, RC identification plate reference category, S is a correlation database, cls is a category of power distribution network equipment, and the identification plate reference category RC and the identification plate reference image RI are in one-to-one correspondence and are unique.
Specifically, comparing the predicted identification plate category with a corresponding identification plate reference category RC in an associated database, if the corresponding identification plate reference category RC in the associated database is equal to or contains the predicted category PC, the condition that the identification plate category of the power distribution network equipment is inaccurate does not exist, and selecting an identification plate reference image RI in the associated database corresponding to the predicted category PC to judge whether the identification plate is unclear; otherwise, the condition that the identification label is inaccurate exists, and the hidden danger is judged. Then, the predicted position information of the signboard is utilized to obtain a signboard image PI in the image, similarity measurement is carried out on the signboard image PI in the association database corresponding to the selected prediction type PC, and the utilization is utilizedCalculating the similarity score, if the similarity score is smaller than a preset threshold value of 0.7, judging that hidden danger exists under the condition that the power distribution network equipment identification plate is unclear, otherwise, judging that the power distribution network equipment identification plate is accurate and clear. When the similarity is measured, the identification plate reference image RI corresponding to the prediction type of the power distribution network equipment in the associated database is required to be adjusted to be the size of the predicted identification plate image PI, and normalization operation is carried out on the identification plate reference image RI and the predicted identification plate image PI, so that the influence of different illumination on the matching score is eliminated. The device for adjusting the size adopts bilinear interpolation: Wherein->，X represents the abscissa of the image after interpolation, y represents the image after interpolationOrdinate, x 1 Represents the abscissa, y of the reference image RI of the signboard 1 Representing the ordinate, x of the reference image RI of the sign 2 Representing the abscissa, y, of the predicted signboard image PI 1 Representing the ordinate of the predicted signboard image PI. Then, normalization operation is carried out on the reference image RI of the signboard and the predicted image PI of the signboard, so that the influence of factors such as illumination and the like in the image acquisition process is eliminated, and the user is in a position of +.>X represents the abscissa in the image, y represents the ordinate in the image, RI (x, y) can represent the pixel value of each pixel point in the reference image of the signboard, and if factors such as illumination and the like are different, the pixel values are also different, and RI max Maximum value of pixel values representing all pixel points in reference image of signboard, RI min Representing the minimum value of the pixel values of all the pixel points, so that the standard image of the signboard can be normalized through the formula and the maximum value and the minimum value of the pixel values, and in the same way, PI (x, y) can represent the pixel value of each pixel point in the image of the signboard, and PI max Maximum value of pixel values representing all pixel points in the signboard image, PI min The minimum value of the pixel values of all the pixel points in the signboard image is represented, so that the signboard image can be normalized through the formula and the maximum value and the minimum value of the signboard image.
For training to obtain a residual neural network model, in some embodiments, the apparatus further comprises a third determination unit for using a cross entropy loss functionAs a loss function of the initial residual neural network, determining that the initial residual neural network training is completed under the condition that the cross entropy loss function converges to obtain the residual neural network, wherein L is cla s Representing the cross entropy loss function, y ic N represents the number of the distribution network devices, M represents the category number of the distribution network devices and p is a sign function ic Representing the ith aforementioned distribution networkProbability that the device belongs to category c. The device uses the loss function as the loss function of the residual neural network, so that whether the residual neural network is trained can be determined through convergence of the cross entropy loss function.
In some alternative embodiments, the neural network model is trained, the network parameters are updated by using a random gradient descent method by using the difference between the minimized predicted value and the actual value as a neural network optimization criterion, and the loss function is designed for training the power distribution network equipment image classification network and the identification label target detection network. Using cross entropy loss functions As a classification loss function of the equipment image classification neural network, a small batch random gradient descent method is used as a parameter optimization device: />Wherein->，x t+1 Representing the learnable parameters, x, of the residual neural network after optimization t The learning parameter before the residual neural network optimization is represented, lr represents the learning rate, li represents the loss function, m represents the total number of samples, and i represents the ith sample. When the cross entropy loss function converges, the initial residual neural network training is completed.
In some embodiments, the apparatus further comprises a fourth determining unit and a fifth determining unit, wherein the fourth determining unit is configured to use a focal loss functionAs a loss function of the initial YOLOX target detection neural network, an intermediate YOLOX target detection neural network is obtained in the case where the above-mentioned focal loss function converges, wherein L box b Representing the focal loss function, p t Indicating the proximity of the predicted category to the actual category of the signboard, and γ indicating the YOLOX target detection neural networkAdjusting factors; a fifth determining unit for using the cross-ratio loss function +.>Obtaining the YOLOX target detection neural network as a loss function of the intermediate YOLOX target detection neural network, wherein L is box c And a step of representing the cross ratio loss function, wherein A represents a predicted area of the signboard image, and B represents a real area of the signboard image. The device uses the focal loss function and the cross-ratio loss function as the loss functions of the initial YOLOX target detection neural network respectively, so that whether the YOLOX target detection neural network is trained can be determined by whether the loss functions are converged.
Specifically, in the process of training the YOLOX target detection neural network, a focal loss is used as a classification loss function of the signboard image target detection network, so that the influence of unbalance of different signboard image data amounts and identification complexity on network stability is reduced; and adopting the cross ratio loss function as a loss function of the positioning of the signboard, namely, determining the loss function of the position information. Then, using an adaptive optimization method as a parameter optimization device:，/>，，/>，/>，x t+1 、x t 、x t-1 the learnable parameters of the YOLOX target detection neural network are represented, f represents the objective function after optimization, lr represents the learning rate, and m represents the step size of each optimization t Index moving average representing gradient, m 0 Initialization ofIs 0, beta 1 Representing the exponential decay rate, controlling the weight distribution of momentum and current gradient, typically takes a value close to 1, where 0.9, v may be defaulted t Exponential moving average, v, representing the square of the gradient 0 Initialized to 0, beta 2 For the exponential decay rate, the current impact of the square of the gradient before control, similar to the RMSProp algorithm, is weighted averaged, here defaulting to 0.999, +.>Representation of the mean value m of the gradients t Result after deviation correction, ++>Mean value v of representation pair gradient t The result after deviation correction is carried out in order to reduce the influence of the initial training phase bias 0 on the initial training phase due to the initialization of 0, +.>Representing a very small non-zero number, avoiding network optimization failure caused by a divisor of 0 in the optimization process, which is defaulted to 10 -8 . When the focal loss function and the cross-ratio loss function converge, the initial YOLOX target detection neural network training is indicated to be completed.
In order to determine the accuracy of the prediction results of the residual neural network and the YOLOX target detection neural network, in some embodiments, the apparatus further comprises a sixth determination unit and a seventh determination unit, wherein the sixth determination unit is configured toCalculating an average value of the accuracy of the prediction results of the residual neural network to obtain an average accuracy value, and determining that the accuracy of the prediction results of the residual neural network meets the requirement when the average accuracy value is within a first preset range, wherein the prediction results of the residual neural network represent the type of the power distribution network equipment predicted by the residual neural network, mAP represents the average accuracy value, P (R) represents a PR curve function, R represents a recall maximum value, and the PR curve The function is used for expressing the relation between the accuracy rate and recall rate of the category of the power distribution network equipment predicted by the residual neural network; the seventh determination unit is used for determining according to the formula +.>And calculating a degree of overlap of the predicted result of the YOLOX target detection neural network, wherein the predicted result of the YOLOX target detection neural network indicates the type of the signboard predicted by the YOLOX target detection neural network and the position information of the signboard image, ioU indicates the degree of overlap, the degree of overlap indicates the degree of overlap of the predicted result of the YOLOX target detection neural network, a indicates the predicted area of the signboard image, and B indicates the real area of the signboard image, and the degree of overlap is within a second preset range. The apparatus determines the accuracy of the prediction results of the residual neural network and the YOLOX target detection neural network by calculating the values of maps and IoU.
In some alternative embodiments, the prediction results of the residual neural network and the YOLOX target detection neural network after the training are evaluated, the mAP is adopted to evaluate the classification effect of the residual neural network on the power distribution network equipment, and the mAP and IoU are adopted to evaluate the classification and position location effect of the YOLOX target detection neural network on the identification board. Calculating mAP by adopting a mode of averaging after calculating the average Precision AP of each category, constructing PR curves of Precision and Recall ratio (Recall) for the average Precision AP of each category of power distribution network equipment, and calculating the areas between the PR curves and coordinate axes to obtain the average Precision AP of each category; ioU the area of each predicted signboard image and the area of the real signboard image are calculated by the area-to-area ratio calculation method, and the common area and the total area are calculated respectively and then divided by the total area to obtain the final result. In the training process, when the mAP and IoU evaluation indexes tend to be stable, the prediction accuracy of the corresponding neural network is higher, the trained network parameters are stored as weight files, the repeated training is avoided, meanwhile, the calculation resources are saved, and the network migration and deployment are convenient to use for the prediction of the subsequent neural network. When new power distribution network equipment image data is imported, network parameters can be dynamically optimized, real-time updating of the network is guaranteed, and training of the network parameters is not needed again.
The detection device of the power distribution network equipment identification plate comprises a processor and a memory, wherein the acquisition unit, the first determination unit, the second determination unit, the replacement unit and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to realize corresponding functions. The modules are all located in the same processor; alternatively, the above modules may be located in different processors in any combination.
The processor includes a kernel, and the kernel fetches the corresponding program unit from the memory. The kernel can be provided with one or more than one, and the aim of high-efficiency inspection of the equipment identification plate is achieved by adjusting kernel parameters.
The memory may include volatile memory, random Access Memory (RAM), and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM), among other forms in computer readable media, the memory including at least one memory chip.
The embodiment of the invention provides equipment, which comprises a processor, a memory and a program stored in the memory and capable of running on the processor, wherein the processor realizes at least the following steps when executing the program:
Step S201, acquiring an image of power distribution network equipment to obtain the image of the power distribution network equipment, wherein the image of the power distribution network equipment comprises a signboard image;
step S202, analyzing the power distribution network equipment image by using a residual neural network, determining the type of the power distribution network equipment, and determining a signboard reference type and a signboard reference image corresponding to the type of the power distribution network equipment in a correlation database according to the type of the power distribution network equipment, wherein the residual neural network is trained by using a plurality of groups of first data through machine learning, and each group of data in the plurality of groups of first data comprises: the association database comprises standard images corresponding to the power distribution network equipment in all categories and the identification plate reference image corresponding to each category;
step S203, analyzing the power distribution network device image by using a YOLOX target detection neural network, determining a category of a signboard in the power distribution network device image and position information of the signboard image, comparing the category of the signboard with the reference category of the signboard, and determining whether the category of the signboard is correct, wherein the YOLOX target detection neural network is trained by using a plurality of sets of second data through machine learning, and each set of data in the plurality of sets of second data includes: all kinds of the signboard images and all kinds of the classes corresponding to the signboard;
Step S204, obtaining the signboard image according to the position information of the signboard, carrying out similarity measurement on the signboard image and the signboard reference image, determining whether the signboard image is clear or not, and replacing the signboard image when the signboard image is not clear.
The device herein may be a server, PC, PAD, cell phone, etc.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented in a general purpose computing device, they may be concentrated on a single computing device, or distributed across a network of computing devices, they may be implemented in program code executable by computing devices, so that they may be stored in a storage device for execution by computing devices, and in some cases, the steps shown or described may be performed in a different order than that shown or described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple modules or steps of them may be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, etc., such as Read Only Memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises an element.
From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:
1) According to the detection method of the power distribution network equipment identification plate, the power distribution network equipment image is obtained, the power distribution network equipment image is analyzed through the residual neural network, the type of the power distribution network equipment is determined, and the identification plate reference type and the identification plate reference image corresponding to the type of the power distribution network equipment are obtained by searching in the association database according to the type of the power distribution network equipment; and then, analyzing the power distribution network equipment image through a YOLOX target detection neural network, determining the position information of the signboard image and the category of the signboard, comparing the category of the signboard with the signboard reference category, determining whether the signboard category is correct, acquiring the signboard image according to the position information of the signboard image on the power distribution network equipment, carrying out similarity measurement on the signboard image and the signboard reference image, and determining whether the signboard image is clear. Compared with the prior art, the method and the device for manually inspecting the identification plate on the power distribution network equipment generally adopt a manual inspection mode to inspect and judge the identification plate on the power distribution network equipment, not only have low efficiency, but also can solve the problem of inaccurate manual identification, the method and the device can be used for analyzing the image of the power distribution network equipment by combining a neural network, comparing the image with a reference type and the reference image, and calculating the value of similarity measurement, so that whether the identification plate type and the identification plate image are correct and clear is determined, and the problem of inaccurate judgment result possibly existing in manual inspection is avoided while the inspection efficiency is improved, thereby solving the problem of low efficiency of manually inspecting the identification plate of the power distribution network equipment, and achieving the purpose of efficiently inspecting the equipment identification plate.
2) In the detection device for the identification plate of the power distribution network equipment, the power distribution network equipment image is obtained, the power distribution network equipment image is analyzed through the residual neural network, the type of the power distribution network equipment is determined, and the identification plate reference type and the identification plate reference image corresponding to the type of the power distribution network equipment are obtained by searching in the association database according to the type of the power distribution network equipment; and then, analyzing the power distribution network equipment image through a YOLOX target detection neural network, determining the position information of the signboard image and the category of the signboard, comparing the category of the signboard with the signboard reference category, determining whether the signboard category is correct, acquiring the signboard image according to the position information of the signboard image on the power distribution network equipment, carrying out similarity measurement on the signboard image and the signboard reference image, and determining whether the signboard image is clear. Compared with the prior art, the method and the device for manually inspecting the identification plate on the power distribution network equipment generally adopt a manual inspection mode to inspect and judge the identification plate on the power distribution network equipment, not only have low efficiency, but also can solve the problem of inaccurate manual identification, the method and the device can be used for analyzing the image of the power distribution network equipment by combining a neural network, comparing the image with a reference type and the reference image, and calculating the value of similarity measurement, so that whether the identification plate type and the identification plate image are correct and clear is determined, and the problem of inaccurate judgment result possibly existing in manual inspection is avoided while the inspection efficiency is improved, thereby solving the problem of low efficiency of manually inspecting the identification plate of the power distribution network equipment, and achieving the purpose of efficiently inspecting the equipment identification plate.
The above description is only of the preferred embodiments of the present application and is not intended to limit the present application, but various modifications and variations can be made to the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims (9)
1. The detection method of the power distribution network equipment identification plate is characterized by comprising the following steps of:
acquiring an image of power distribution network equipment to obtain an image of the power distribution network equipment, wherein the image of the power distribution network equipment comprises a signboard image;
analyzing the power distribution network equipment image by using a residual neural network, determining the type of the power distribution network equipment, and determining a signboard reference type and a signboard reference image corresponding to the type of the power distribution network equipment in a correlation database according to the type of the power distribution network equipment, wherein the residual neural network is trained by using a plurality of groups of first data through machine learning, and each group of data in the plurality of groups of first data comprises: the association database comprises standard images corresponding to the power distribution network equipment in all categories and the identification plate reference image corresponding to each category;
Analyzing the power distribution network equipment image by using a YOLOX target detection neural network, determining the category of a signboard in the power distribution network equipment image and the position information of the signboard image, comparing the category of the signboard with the reference category of the signboard, and determining whether the category of the signboard is correct, wherein the YOLOX target detection neural network is trained by using a plurality of sets of second data through machine learning, and each set of data in the plurality of sets of second data comprises: the signboard images of all the categories and the categories corresponding to the signboards of all the categories;
and acquiring the signboard image according to the position information of the signboard, carrying out similarity measurement on the signboard image and the signboard reference image, determining whether the signboard image is clear or not, and replacing the signboard image under the condition that the signboard image is not clear.
2. The method of claim 1, wherein analyzing the power distribution network device image using a residual neural network to determine the category of the power distribution network device comprises:
inputting the power distribution network equipment image into the residual neural network, and obtaining an output result of the residual neural network to obtain the corresponding prediction probability of the power distribution network equipment of each category;
And taking the category of the power distribution network equipment with the maximum prediction probability as the category of the power distribution network equipment.
3. The detection method according to claim 1, wherein analyzing the distribution network device image using a YOLOX target detection neural network, determining the category of the signboard in the distribution network device image and the positional information of the signboard image, comprises:
inputting the power distribution network equipment image into the YOLOX target detection neural network;
by passing throughDetermining the position information of the identification plate image in the distribution network equipment image and passing +.>Determining the category of the identification plate, wherein X represents the abscissa of the identification plate image on the power distribution network equipment image, Y represents the ordinate of the identification plate image on the power distribution network equipment image, grid x Abscissa, grid representing pixel point positions of characteristic images of power distribution network equipment y An ordinate, offset, representing the pixel position of the characteristic image of the distribution network device x An offset representing an abscissa offset of the signboard image y Representing the ordinate offset and stride of the signboard image x Representing the abscissa ratio of the signboard image to the distribution network equipment image, and stride y Representing the ordinate ratio of the signboard image mapped to the power distribution network equipment image, W representing the width of the signboard reference image, H representing the height of the signboard reference image, W representing the width of the signboard image, H representing the height of the signboard image, R b Representing the predictive probability of the sign for each category, O b k An output vector representing a category of the kth signboard corresponding to the YOLOX target detection neural network, O b i And an output vector representing the category of the ith signboard corresponding to the YOLOX target detection neural network.
4. The method of detecting according to claim 1, wherein performing a similarity measure on the signboard image and the signboard reference image to determine whether the signboard image is clear includes:
according to the formulaNormalizing the signboard image and the signboard reference image, whichWherein RI (x, y) represents the signboard reference image, PI (x, y) represents the signboard image, RI '(x, y) represents the normalized signboard reference image, PI' (x, y) represents the normalized signboard image, and RI max Representing the maximum value of pixel values of all pixel points in the reference image of the signboard, RI min Minimum value of pixel values representing all pixel points in the reference image of the signboard, PI max Maximum value of pixel values representing all pixel points in the signboard image, PI min Representing the minimum value of the pixel values of all pixel points in the signboard image;
according to the formulaCalculating the similarity score of the signboard image and the signboard reference image, determining that the signboard image is clear when the similarity score is larger than a preset threshold value, and determining that the signboard image is not clear when the similarity score is smaller than the preset threshold value, wherein score represents the similarity score.
5. The method of detection according to claim 1, wherein the method of detection further comprises:
using cross entropy loss functionAs a loss function of an initial residual neural network, determining that the initial residual neural network training is completed under the condition that the cross entropy loss function converges to obtain the residual neural network, wherein L is as follows cla s Representing the cross entropy loss function, y ic N represents the number of the distribution network equipment, M represents the category number of the distribution network equipment and p is a sign function ic Representing the probability that the ith power distribution network device belongs to the c-th category.
6. The method of detection according to claim 1, wherein the method of detection further comprises:
using a focal loss functionAs a loss function of the initial YOLOX target detection neural network, an intermediate YOLOX target detection neural network is obtained with convergence of the focal loss function, wherein L box b Representing the focal loss function, p t Representing the proximity degree of the predicted category and the real category of the signboard, wherein gamma represents the adjustable factor of the YOLOX target detection neural network;
using cross-ratio loss functionsObtaining the YOLOX target detection neural network as a loss function of the intermediate YOLOX target detection neural network under the condition that the cross-ratio loss function is converged, wherein L box c And representing the cross ratio loss function, wherein A represents a predicted area of the signboard image, and B represents a real area of the signboard image.
7. The method of detection according to claim 1, wherein the method of detection further comprises:
according to the formulaCalculating an average value of the precision of the prediction result of the residual neural network to obtain an average precision value, and determining that the precision of the prediction result of the residual neural network meets the requirement under the condition that the average precision value is within a first preset range, wherein the prediction result of the residual neural network represents the type of the power distribution network equipment predicted by the residual neural network, mAP represents the average precision value, P (R) represents a PR curve function, R represents a recall maximum value, and the PR curve function is used for representing the relationship between the precision of the type of the power distribution network equipment predicted by the residual neural network and the recall;
According toFormula (VI)Calculating the overlapping degree of the predicted result of the YOLOX target detection neural network, and determining that the precision of the predicted result of the YOLOX target detection neural network meets the requirement when the overlapping degree is within a second preset range, wherein the predicted result of the YOLOX target detection neural network represents the category of the signboard predicted by the YOLOX target detection neural network and the position information of the signboard image, ioU represents the overlapping degree, the overlapping degree represents the precision of the predicted result of the YOLOX target detection neural network, A represents the predicted area of the signboard image, and B represents the real area of the signboard image.
8. Detection device of distribution network equipment signboard, characterized by includes:
the power distribution network equipment image acquisition unit is used for acquiring the image of the power distribution network equipment to obtain the power distribution network equipment image, wherein the power distribution network equipment image comprises a signboard image;
the first determining unit is configured to analyze the power distribution network device image by using a residual neural network, determine a category of the power distribution network device, and determine a reference category of a signboard and a reference image of the signboard, which correspond to the category of the power distribution network device in an association database according to the category of the power distribution network device, where the residual neural network is trained by using a plurality of sets of first data through machine learning, and each set of data in the plurality of sets of first data includes: the association database comprises standard images corresponding to the power distribution network equipment in all categories and the identification plate reference image corresponding to each category;
A second determining unit, configured to analyze the power distribution network device image by using a YOLOX target detection neural network, determine a category of a signboard in the power distribution network device image and position information of the signboard image, compare the category of the signboard with the reference category of the signboard, and determine whether the category of the signboard is correct, where the YOLOX target detection neural network is trained by using a plurality of sets of second data through machine learning, and each set of data in the plurality of sets of second data includes: the signboard images of all the categories and the categories corresponding to the signboards of all the categories;
the replacing unit is used for acquiring the signboard image according to the position information of the signboard, carrying out similarity measurement on the signboard image and the signboard reference image, determining whether the signboard image is clear or not, and replacing the signboard image under the condition that the signboard image is not clear.
9. An electronic device, comprising: one or more processors, memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising instructions for performing the detection method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310830073.2A CN116563841B (en) | 2023-07-07 | 2023-07-07 | Detection method and detection device for power distribution network equipment identification plate and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310830073.2A CN116563841B (en) | 2023-07-07 | 2023-07-07 | Detection method and detection device for power distribution network equipment identification plate and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116563841A CN116563841A (en) | 2023-08-08 |
| CN116563841B true CN116563841B (en) | 2023-11-10 |
Family
ID=87488295
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310830073.2A Active CN116563841B (en) | 2023-07-07 | 2023-07-07 | Detection method and detection device for power distribution network equipment identification plate and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116563841B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117392561A (en) * | 2023-10-07 | 2024-01-12 | 中国公路工程咨询集团有限公司 | Remote sensing unmanned aerial vehicle image processing method and system for intelligent traffic construction data acquisition |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796210A (en) * | 2018-08-03 | 2020-02-14 | 珠海格力电器股份有限公司 | Method and device for identifying label information |
| CN111292327A (en) * | 2020-05-08 | 2020-06-16 | 成都派沃特科技股份有限公司 | Machine room inspection method, device, equipment and storage medium |
| CN113936278A (en) * | 2021-09-30 | 2022-01-14 | 杭州飞步科技有限公司 | Method, device and equipment for determining abrasion of positioning identifier |
| CN115294373A (en) * | 2022-08-11 | 2022-11-04 | 南方电网调峰调频发电有限公司检修试验分公司 | Equipment parameter information updating method and updating system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10515275B2 (en) * | 2017-11-17 | 2019-12-24 | Adobe Inc. | Intelligent digital image scene detection |
| US20220284199A1 (en) * | 2021-03-02 | 2022-09-08 | Michael Arno | System and method for determining the condition of an article |
-
2023
- 2023-07-07 CN CN202310830073.2A patent/CN116563841B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796210A (en) * | 2018-08-03 | 2020-02-14 | 珠海格力电器股份有限公司 | Method and device for identifying label information |
| CN111292327A (en) * | 2020-05-08 | 2020-06-16 | 成都派沃特科技股份有限公司 | Machine room inspection method, device, equipment and storage medium |
| CN113936278A (en) * | 2021-09-30 | 2022-01-14 | 杭州飞步科技有限公司 | Method, device and equipment for determining abrasion of positioning identifier |
| CN115294373A (en) * | 2022-08-11 | 2022-11-04 | 南方电网调峰调频发电有限公司检修试验分公司 | Equipment parameter information updating method and updating system |
Non-Patent Citations (2)
| Title |
|---|
| Methods of image recognition of overhead power line insulators and ice types based on deep weakly-supervised and transfer and ice types based on deep weakly-supervised and transfer;Yanpeng Hao 等;《IET Generation, Transmission & Distribution》;第16卷(第11期);第2140-2153页 * |
| 基于YOLO-V3算法的电气输电设备识别研究;耿子进;《基于YOLO-V3算法的电气输电设备识别研究》(第2期);第C042-1722页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116563841A (en) | 2023-08-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113705478B (en) | Mangrove single wood target detection method based on improved YOLOv5 | |
| CN111398176B (en) | Water body water color abnormity remote sensing identification method and device based on pixel scale characteristics | |
| CN116563841B (en) | Detection method and detection device for power distribution network equipment identification plate and electronic equipment | |
| CN111209958B (en) | Substation equipment detection method and device based on deep learning | |
| CN110175519B (en) | Method and device for identifying separation and combination identification instrument of transformer substation and storage medium | |
| CN112613569A (en) | Image recognition method, and training method and device of image classification model | |
| CN115660262B (en) | Engineering intelligent quality inspection method, system and medium based on database application | |
| CN111524394A (en) | Method, device and system for improving accuracy of comprehensive track monitoring data of apron | |
| CN114037907A (en) | Detection method and device for power transmission line, computer equipment and storage medium | |
| Wang et al. | Pointer meter recognition in UAV inspection of overhead transmission lines | |
| CN110320544B (en) | Method, device, equipment and storage medium for identifying position of terminal equipment | |
| CN114227717A (en) | Intelligent inspection method, device, equipment and storage medium based on inspection robot | |
| CN116805387B (en) | Model training method, quality inspection method and related equipment based on knowledge distillation | |
| CN113516177A (en) | Wheat lodging region identification method based on spectral texture features and support vector machine | |
| CN110909804B (en) | Method, device, server and storage medium for detecting abnormal data of base station | |
| CN116403071B (en) | Method and device for detecting few-sample concrete defects based on feature reconstruction | |
| CN116823793A (en) | Device defect detection method, device, electronic device and readable storage medium | |
| CN116645612A (en) | Forest resource asset determination method and system | |
| CN111738259A (en) | Tower state detection method and device | |
| CN115372877B (en) | Lightning arrester leakage ammeter inspection method of transformer substation based on unmanned aerial vehicle | |
| CN115932661A (en) | Method for analyzing and judging noise compliance of transformer | |
| CN116756560A (en) | Dust aerosol recognition method, system, model training method, medium and equipment | |
| CN111931721B (en) | Method and device for detecting color and number of annual inspection label and electronic equipment | |
| CN113989632A (en) | Bridge detection method and device for remote sensing image, electronic equipment and storage medium | |
| CN115035481A (en) | Image object distance fusion method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |