CN114842363A - Identification method and system for key power equipment in digital twin platform area - Google Patents
Identification method and system for key power equipment in digital twin platform area Download PDFInfo
- Publication number
- CN114842363A CN114842363A CN202210776349.9A CN202210776349A CN114842363A CN 114842363 A CN114842363 A CN 114842363A CN 202210776349 A CN202210776349 A CN 202210776349A CN 114842363 A CN114842363 A CN 114842363A
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- module
- power equipment
- convolutional neural
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 74
- 238000003708 edge detection Methods 0.000 claims abstract description 18
- 238000007781 pre-processing Methods 0.000 claims abstract description 16
- 230000006872 improvement Effects 0.000 claims abstract description 9
- 239000010410 layer Substances 0.000 claims description 59
- 230000007246 mechanism Effects 0.000 claims description 45
- 238000000605 extraction Methods 0.000 claims description 31
- 230000004927 fusion Effects 0.000 claims description 24
- 238000013528 artificial neural network Methods 0.000 claims description 14
- 239000000284 extract Substances 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 7
- 230000009977 dual effect Effects 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 2
- 239000002355 dual-layer Substances 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 16
- 238000004364 calculation method Methods 0.000 description 13
- 239000012212 insulator Substances 0.000 description 10
- 238000011176 pooling Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 9
- 230000007613 environmental effect Effects 0.000 description 6
- 238000001914 filtration Methods 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 5
- 238000001514 detection method Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000003062 neural network model Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000005611 electricity Effects 0.000 description 3
- 238000003709 image segmentation Methods 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000011229 interlayer Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning 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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method and a system for identifying key electric power equipment in a digital twin platform area, wherein the method comprises the following steps: acquiring an aerial image of a digital twin platform area of the low-voltage distribution network, and preprocessing the aerial image to obtain a preprocessed aerial image; extracting a foreground image containing key power equipment from the preprocessed aerial image by using an improved edge detection operator, wherein the key power equipment at least comprises a power line; and inputting the foreground image into a trained lightweight convolutional neural network model, and identifying key power equipment, wherein the lightweight convolutional neural network model is obtained by carrying out lightweight improvement on a backbone network of a YOLOv4 model. The method can still give consideration to speed and precision for positioning and extracting the power line under the complex background, and has higher practicability and expandability.
Description
Technical Field
The invention relates to the technical field of digital twinning of a power distribution network, in particular to a method and a system for identifying key power equipment in a digital twinning station area.
Background
The digital twin aims to construct holographic mapping of a complex physical entity from a real space to a virtual digital space, and real-time states and dynamic characteristics of a physical system are described and simulated through mutual linkage of virtual and actual information. In the context of the construction of new power systems and digital power grids, the digital twin system of a power system is one of the most complex systems concerning the national economy, while the digital twin of a power distribution system is one of the most critical ones. In the digital twin of a power distribution system, the realization of comprehensive and accurate digital representation of the running state of a power distribution network is an important foundation.
In an electric power system, a low-voltage distribution network is positioned at the tail end of an electric power network and is the last kilometer for connecting the electric power network with a user, along with the high-speed development of economy, the domestic electricity consumption is greatly increased, and the construction and transformation speed of the low-voltage distribution network is far less than the increase speed of a load. Due to the complex network structure and the huge number of lines, the perfection degree of basic data in all aspects is far lower than that of a high-voltage transmission network, and the relation data from the transformer of the transformer area to the specific loads of all users are in a large blank and chaotic non-transparent state for a long time.
The low-voltage distribution line has important significance for line inspection maintenance, digital twinning and public safety involved in electricity of the low-voltage distribution network, but the environment of the low-voltage distribution network is complex and has a lot of shelters, so that the low-voltage distribution line is difficult to detect quickly and accurately.
Disclosure of Invention
The invention aims to provide a method and a system for identifying key power equipment in a digital twin distribution area, which aim to solve the technical problem that speed and precision are difficult to be considered when a low-voltage distribution line is detected in the prior art.
The purpose of the invention can be realized by the following technical scheme:
a method for identifying key power equipment in a digital twin region comprises the following steps:
acquiring an aerial image of a digital twin platform area of the low-voltage distribution network, and preprocessing the aerial image to obtain a preprocessed aerial image;
extracting a foreground image containing key power equipment from the preprocessed aerial image by using an improved edge detection operator, wherein the key power equipment at least comprises a power line;
and inputting the foreground image into a trained lightweight convolutional neural network model, and identifying the key power equipment, wherein the lightweight convolutional neural network model is obtained by carrying out lightweight improvement on a backbone network of a YOLOv4 model.
Optionally, the extracting, from the preprocessed aerial image, a foreground image containing a key power device by using an improved edge detection operator includes:
and carrying out rapid Gabor transformation on the preprocessed aerial image by utilizing an improved Gabor operator to extract Gabort characteristics, distinguishing a foreground image and a background image of the preprocessed aerial image according to the Gabort characteristics, and reserving the foreground image.
Optionally, the lightweight convolutional neural network model comprises:
a backbone network, a double-layer feature extraction module and a feature pyramid fusion module of a double attention mechanism are added;
the foreground image feature extraction module extracts a low-order geometric feature and a high-order semantic feature from the foreground image feature extraction module, the feature pyramid fusion module performs feature fusion on the low-order geometric feature and the high-order semantic feature to obtain image features of different scales, the dual attention mechanism comprises a channel attention mechanism and a space attention mechanism, the low-order geometric feature is used for representing geometric detail information of the foreground image, and the high-order semantic feature is used for representing semantic information of the foreground image.
Optionally, the backbone network incorporating the dual attention mechanism includes:
the system comprises a three-layer 2D convolutional neural network module and a three-layer residual error convolutional module;
the two previous layers of 2D convolutional neural network modules are connected in series, then are connected in series with the three layers of residual convolutional modules, and finally are connected in series with the third layer of 2D convolutional neural network modules, and a double attention mechanism is added between every two adjacent residual convolutional modules.
Optionally, the 2D convolutional neural network module includes:
a layer of normal two-dimensional convolution Conv2D, a layer of normalized Batchnormalization and a layer of activation function LeakyReLU connected in series.
Optionally, the residual convolution module is formed by connecting two layers of 2D convolution neural network modules in series and then connecting the two layers of 2D convolution neural network modules in parallel with the skip connection structure.
Optionally, a double attention mechanism is added between two adjacent residual convolution modules, including:
adding a channel attention mechanism between two adjacent 2D convolution neural network modules of the residual convolution module, wherein the channel attention mechanism is used for adjusting the weight of each channel of the residual convolution module;
and a spatial attention mechanism is connected in series behind the channel attention mechanism and is used for adjusting the spatial feature map output by the 2D convolutional neural network module.
Optionally, the extracting the low-order geometric features and the high-order semantic features from the backbone network by the two-layer feature extraction module includes:
the double-layer feature extraction module extracts the low-order geometric features from a first layer of residual convolution module of the backbone network, and extracts the high-order semantic features from a third layer of residual convolution module of the backbone network.
Optionally, the feature pyramid fusion module performs feature fusion on the low-order geometric features and the high-order semantic features to obtain image features of different scales, where the feature pyramid fusion module includes:
and the feature pyramid fusion module performs up-sampling on the high-order semantic features, aligns the up-sampled features and then splices the aligned features with the low-order geometric features to obtain image features with different scales.
The invention also provides a system for identifying the key power equipment in the digital twin station area, which comprises the following steps:
the image acquisition and preprocessing module is used for acquiring an aerial image of a digital twin platform area of the low-voltage distribution network and preprocessing the aerial image to obtain a preprocessed aerial image;
the foreground image acquisition module is used for extracting a foreground image containing key power equipment from the preprocessed aerial image by using an improved edge detection operator, wherein the key power equipment at least comprises a power line;
and the key power equipment identification module is used for inputting the foreground image into a trained light-weight convolutional neural network model to identify the key power equipment, wherein the light-weight convolutional neural network model is obtained by carrying out light-weight improvement on a backbone network of a YOLOv4 model.
The invention provides a method and a system for identifying key power equipment in a digital twin station area, wherein the method comprises the following steps: acquiring an aerial image of a digital twin platform area of the low-voltage distribution network, and preprocessing the aerial image to obtain a preprocessed aerial image; extracting a foreground image containing key power equipment from the preprocessed aerial image by using an improved edge detection operator, wherein the key power equipment at least comprises a power line; and inputting the foreground image into a trained lightweight convolutional neural network model, and identifying the key power equipment, wherein the lightweight convolutional neural network model is obtained by carrying out lightweight improvement on a backbone network of a YOLOv4 model.
Therefore, the invention has the beneficial effects that:
aiming at the problems of complex background and serious shielding of the low-voltage distribution network overhead line, the foreground image containing the power line can be quickly and accurately extracted from the high-resolution unmanned aerial vehicle aerial image by utilizing the improved edge detection operator; the lightweight convolutional neural network model obtained by improving the YOLO convolutional neural network model is more suitable for power line identification, the lightweight convolutional neural network model only needs to identify the image characteristics of the foreground image, the extraction precision of the power line can be improved, the calculation load of the convolutional neural network is greatly reduced, and the method is very suitable for the algorithm environment with limited computing resources, such as an unmanned aerial vehicle embedded platform. The method can still give consideration to speed and precision for positioning and extracting the power line under the complex background, and has higher practicability and expandability.
Drawings
FIG. 1 is an initial view of a high resolution aerial image of the present invention;
FIG. 2 is a schematic flow chart of the identification method of the present invention;
FIG. 3 is an aerial image of the present invention after graying and Gaussian filtering preprocessing;
FIG. 4 is a diagram of 4 Gabor features extracted using different Gabor operators according to the present invention;
FIG. 5 is an edge feature map after feature fusion of 4 Gabor feature maps according to the present invention;
FIG. 6 is a foreground image segmented from a pre-processed aerial image in accordance with the present invention;
FIG. 7 is a diagram of the final identified result of the lightweight convolutional neural network model of the present invention;
FIG. 8 is a schematic illustration of the process of pre-processing an aerial image according to the present invention;
FIG. 9 is a schematic diagram of the structure of the lightweight convolutional neural network model of the present invention;
FIG. 10 is a schematic diagram of the dual-attention machine mechanism of the present invention;
FIG. 11 is a schematic structural diagram of a two-layer feature extraction module and a feature pyramid fusion module according to the present invention;
FIG. 12 is a block diagram of the residual convolution module of the present invention;
FIG. 13 is a schematic structural diagram of an identification system according to the present invention.
Detailed Description
The embodiment of the invention aims to provide a method and a system for identifying key power equipment in a digital twin distribution area, so as to solve the technical problem that speed and precision are difficult to be considered when a low-voltage distribution line is detected in the prior art.
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
The relation data between the transformer of the transformer area and each user load in the low-voltage distribution network mainly comprises three aspects of a user variable relation, a circuit topological structure and line trend, wherein the line trend information is almost in a completely blank state, and the relation data has important significance for line inspection maintenance, digital twinning and electricity-related public safety of the low-voltage distribution network. The method has great potential safety hazard, is very unfavorable for daily maintenance and repair of the line and power restoration after the fault, and urgently needs to utilize an informatization means to complement and perfect basic data such as the line trend of the low-voltage distribution network, the line topology and the corresponding relation. Simultaneously, along with the rapid development of unmanned aerial vehicle technique, unmanned aerial vehicle's stability, duration and load capacity have obtained great promotion, and unmanned aerial vehicle patrols the line and also applies gradually to the low voltage distribution network power line and patrols and examines.
The low-voltage distribution line is divided into 2 types of buried cables and overhead lines, and the geographic trend information of the overhead lines comprises power key equipment information extraction and geographic information extraction. The intelligent extraction of the key equipment information of the electric power is the basis for perfecting the geographical trend of the low-voltage distribution overhead line, and comprises the identification and calibration of targets such as a desk-top or box-type transformer, a power line, a telegraph pole or a mechanical supporting point, various insulators, a C-shaped wire clamp and the like in a low-voltage distribution channel.
The environment of low voltage distribution network is complicated and shelter from the thing many, when adopting unmanned aerial vehicle to patrol and examine, generally can not be to low voltage distribution lines near shooting but keep considerable safe distance for guaranteeing safety, under this condition, can clearly effectively demonstrate transmission line and insulator and generally can use very high resolution ratio in order to guarantee during unmanned aerial vehicle takes photo by plane, and resolution ratio can reach 8000 x 6000 and more than, and the size of a picture is more than 10M.
As shown in FIG. 1, in the aerial image with high resolution, the proportion of the power line and the insulator in the image is only 1/50-1/100. As is well known, the image recognition method based on deep learning is very sensitive to resolution, and for such high-resolution images, if the image resolution is directly reduced to a small fixed value such as 608 × 608 like other existing image recognition methods based on deep learning, the characteristics of power lines and insulators are inevitably lost, so that the characteristics become ambiguous and cannot be recognized; if the high-resolution image is directly input into the convolutional neural network for identification without any processing, the computational power consumption is very slow, parameter explosion is very likely to occur, and the benefit is very low, because most of the original image is meaningless environmental background. However, the conventional method insensitive to resolution, such as edge detection, can only identify features such as power lines and insulators and electric facilities with simple outlines, and is difficult to adapt to complex and variable environmental noise in aerial images.
In order to solve the problems, the invention provides a method for identifying key electric power equipment in a digital twin distribution area, which is used for carrying out light-weight high-precision identification on key electric equipment in a low-voltage distribution network based on Gabor-YOLO of a high-resolution image.
Referring to fig. 2, the following is an embodiment of a method for identifying a key power device in a digital twin zone according to the present invention, including the following steps:
s100: acquiring an aerial image of a digital twin platform area of the low-voltage distribution network, and preprocessing the aerial image to obtain a preprocessed aerial image;
s200: extracting a foreground image containing key power equipment from the preprocessed aerial image by using an improved edge detection operator, wherein the key power equipment at least comprises a power line;
s300: and inputting the foreground image into a trained lightweight convolutional neural network model, and identifying the key power equipment, wherein the lightweight convolutional neural network model is obtained by carrying out lightweight improvement on a backbone network of a YOLOv4 model.
In this embodiment, the image of taking photo by plane that unmanned aerial vehicle gathered is RGB color model, carries out the preliminary treatment to initial image of taking photo by plane, mainly includes: and carrying out graying processing and Gaussian filtering. The gray image is obtained by carrying out gray processing on the target image through gray computing, and the gray image can overcome the defects of large communication channel space occupied by an RGB color model, low transmission efficiency and low speed, and reduce the data volume. The Gaussian filter is used for carrying out linear smooth filtering on the gray level image, a large amount of noise in the image in the shooting and transmission process of the unmanned aerial vehicle can be filtered out, the Gaussian noise can be eliminated, and the subsequent influence of the Gaussian noise can be eliminated as far as possible.
From the unmanned aerial vehicle image of taking photo by plane on the one hand, the background of low voltage distribution network transmission line is complicated changeable and shelters from seriously, and on the other hand power line part all has longer, narrower linear structure. The straight line is a special edge, and the edge can be regarded as a pixel point set with severe image spectral feature change, so that the power line extraction problem is often converted into an edge detection process in the prior art. In a low-voltage distribution network, a large number of background pixel points can have great influence on the performance of a traditional algorithm, and it is not appropriate to directly adopt a traditional edge detection method to extract a power line. However, the edge detection operator, such as a Gabor operator, has the characteristics of simple structure and high operation speed, and can be used for performing edge detection on an image first and establishing a linear pixel point candidate pool, so that the accuracy of subsequent power line extraction can be improved, and the calculation amount of convolutional neural network feature extraction can be greatly reduced.
In this embodiment, 8 groups of Gabor filters with different wavelengths and bandwidths perform filtering; the 8 groups of Gabor filters are connected in parallel, each group of Gabor filters only responding to the corresponding frequency and bandwidth
While the energy of other textures is suppressed. Wherein x and y are the horizontal and vertical coordinates of the pixel points in the gray level image,
is the angle of direction of the Gabort wavelet; the parameters including the angle of orientation
Width parameter of sum Gaussian function
And the center frequency of the filter bandwidth
These thresholds and the subsequent neural network of the present inventionThe final recognition results of the models are closely related, and therefore, the present invention sets these thresholds as the differentiable parameters of the networked models that the present invention refreshes.
In practical operation, the input of the algorithm is a discrete value, and discretization processing is performed on the result of the Gabor operator, in this embodiment, a spectrum energy function is adopted
As the response of the input image to the Gabor filter, 8 groups of Gabor filters are designed according to the difference of the edge and the texture of the distribution channel of the low-voltage distribution network and the environmental background, and each Gabor filter only responds to the frequency and the bandwidth corresponding to the Gabor filter
And the energy of other textures is suppressed, so that the texture characteristics of the low-voltage distribution channel are analyzed and extracted, and most environmental backgrounds are quickly filtered. It can be understood that the power distribution channel of the low-voltage distribution network is a partial image containing power lines, and a foreground image containing the power lines can be extracted by using a Gabor filter.
In one embodiment, the distribution channel is distinguished from useless environmental background by a Gabor filter, and back propagation is carried out according to the final recognition effect of the neural network model to adjust the differential derivative of the threshold value so as to realize the iteration of the threshold value. Referring to fig. 4, in order to ensure the effect, the algorithm of the present embodiment adopts four sets of thresholds to extract Gabor features respectively to obtain 4 Gabor feature maps, and obtains a final result after comparison.
In the embodiment, the power line foreground region (namely, the partial image containing the power line) can be quickly and accurately extracted from the high-resolution unmanned aerial vehicle aerial image by using the Gabor operator, the foreground region accounts for less than 30% of the original image, and the subsequent convolutional neural network model only needs to extract the image characteristics of the foreground region and only needs to perform image identification on the foreground image containing the power line, so that the calculation load of the convolutional neural network is greatly reduced, and the identification accuracy is improved.
In step S300, the foreground image is feature-extracted by using a lightweight convolutional neural network model, in a preferred embodiment, the foreground image is feature-extracted by using a Darknet53-tiny lightweight convolutional neural network model, the low-order contour features and the high-order semantic features of the image are sequentially extracted by using a Darknet53-tiny to the input image in a convolutional calculation manner, and subsequent algorithms such as feature fusion are developed based on the features extracted by using a Darknet53-tiny neural network.
It should be noted that the Darknet53-tiny lightweight convolutional neural network model is improved based on the YOLOv4 backbone network after lightweight, and many convolutional blocks and branches in the YOLOv4 backbone network are deleted. The light weight of the Darknet53-tiny light weight convolution neural network model is mainly reflected in the light weight of the Backbone network Backbone, the Backbone network of the whole Darknet53-tiny light weight convolution neural network model is formed by only stacking six layers of convolution modules (a three-layer 2D convolution neural network module and a three-layer residual convolution module), the size of a model file of the whole algorithm is about 20M, and the size of a model file of a conventional convolution neural network is more than 300M.
Referring to fig. 9, the Darknet53-tiny lightweight convolutional neural network model includes: a backbone network with a double attention mechanism, a double-layer feature extraction module and a feature pyramid fusion module are added.
The main network added with the double attention mechanism comprises: a triple-layer 2D convolutional neural network module (darknencov 2D _ Block) and a triple-layer residual convolutional module (Resblock _ CBAM _ body); the concrete connection mode is as follows: the first two layers of DarknetConv2D _ Block modules are connected in series, then connected in series with the three layers of Resblock _ CBAM _ body modules, and finally connected in series with the third layer of DarknetConv2D _ Block modules.
Specifically, the DarknetConv2D _ Block module is composed of: a module is formed by a Conv2D layer (normal two-dimensional convolution) + a BatchNormalization layer (normalization) + a leakage relu layer (activation function) in series.
Specifically, the Resblock _ CBAM _ body module is formed by connecting two layers of 2D convolutional neural networks in series and then connecting the two layers of 2D convolutional neural networks in parallel with a skip connection structure on the basis of a DarknetConv2D _ Block, and the specific structure is shown in fig. 12 in the specification. CBAM represents a dual attention mechanism, which is added between the two layers of 2D convolutional neural networks of each Resblock _ CBAM _ body module, as shown in fig. 11.
The Darknet53-tiny lightweight convolutional neural network model takes a residual convolutional neural network as a key structure. The convolutional neural network is a feedforward neural network which comprises convolutional calculation and has a deep structure, and the convolutional kernel parameter sharing in an implicit layer and the sparsity of interlayer connection enable the convolutional neural network to calculate complex features with smaller calculation amount; the residual convolution module is a convolution neural network formed on the basis of jump-connected residual blocks; the residual convolutional neural network comprises a forward-connected 2D convolutional neural network and a residual edge connected in parallel with the forward connection, and a skip connection structure formed by the two.
The Darknet53-tiny convolution neural network model in the embodiment is a main network taking Darknet53-tiny as the whole algorithm; the Darknet53-tiny sequentially extracts the low-order contour feature and the high-order semantic feature of the image in a convolution calculation mode on the input image.
On one hand, aiming at the characteristics that the power line features are weak and high-order semantic features are not needed, the embodiment abandons the adoption of a large and deep convolutional neural network and uses a light-weight shallow neural network; on the other hand, the shallow neural network is improved, the image characteristics of the power line are enhanced by adopting a double attention mechanism, and a plurality of differentiated characteristic extraction branches are designed to adjust the receptive field.
The dual attention mechanism in this embodiment includes: a channel attention mechanism and a spatial attention mechanism connected in series. A channel attention mechanism is introduced between the 2D convolutional neural networks of two adjacent residual convolutional modules of the backbone network to adjust the weight of each channel of the residual convolutional modules, and a space attention mechanism is connected in series behind the channel attention mechanism to adjust the space characteristic diagram output by each layer of the 2D convolutional neural networks.
Specifically, the channel attention mechanism is to perform weighted calculation on each channel in the convolutional neural network, give a high weight to a meaningful channel of interest and suppress a weight of a meaningless background channel, and achieve the purpose by connecting the channels in series in a global average pooling mode and a maximum pooling mode, wherein a calculation formula of the channel attention mechanism is as follows (1):
in the formula (I), the compound is shown in the specification,
representing a sigmoid function;
the shared network in the representation module is composed of a hidden layer and a plurality of layers of perceptrons. Wherein the activation size of the hidden layer is set to
In multilayer sensors
And
are respectively arranged as
In the present invention, r is set to 16.
And
respectively representing the operation of the module for performing average pooling and maximum pooling on the feature map space information,
and
global average pooling and maximum averaging, respectively, representing channel attention mechanismsAnd (4) performing pooling operation.
Specifically, the spatial attention mechanism is to perform weighted calculation and screening on a spatial feature graph output by each Resblock _ CBAM _ body module, improve the weight of meaningful features such as a power distribution channel and the like, and transplant the spatial feature weight of corresponding features of an environment background, and the purpose is achieved through point-to-point convolution operation and pooling operation, wherein a calculation formula of the spatial attention mechanism is shown as formula (2):
in the formula:
represents a convolution calculation of size 7 × 7;
and
the global average pooling and maximum average pooling operations of the spatial attention mechanism are shown separately.
In this embodiment, a channel attention mechanism is added between two adjacent residual convolution modules in the Backbone network backhaul to calculate and screen the characteristics of each channel, so that a significant characteristic channel in a characteristic diagram can be further highlighted and a background characteristic channel can be suppressed. The defect that image position information cannot be well acquired in a channel attention mechanism is considered, a space attention mechanism is introduced to focus on a space area, the space attention mechanism calculates and screens characteristic graphs corresponding to channels, the area characteristics of low-voltage power distribution channels are further highlighted, and the problems that a target to be detected in an aerial image of a low-voltage power distribution network is few in pixels and weak in characteristics and is easily influenced by a complex background and partial shielding are solved.
It should be noted that, the three layers of residual convolution modules in the Backbone network Backbone are connected in series, so that a channel attention mechanism and a space attention mechanism are added between two adjacent residual convolution modules.
In this embodiment, the double-layer feature extraction module in the lightweight convolutional neural network model extracts a low-order geometric feature from the first-layer residual convolution module of the backbone network to represent geometric detail information of the image, and extracts a high-order semantic feature from the third-layer residual convolution module of the backbone network to represent semantic information of the image.
In this embodiment, the feature pyramid fusion module in the lightweight convolutional neural network model fuses the low-order geometric features and the high-order semantic features obtained by the double-layer feature extraction module, and specifically, the method is to perform upsampling on the high-order semantic features, align the upsampled features, and then splice the aligned features with the low-order geometric features to obtain image features of different scales. It is worth to be noted that the feature pyramid fusion module performs multi-scale fusion on the features, where the multi-scale means that the sizes of the receptive fields of the low-order geometric features and the high-order semantic features are different, and the data dimensions of the two are also different.
In the embodiment, when the convolutional neural network is used for extracting the characteristics of the power line, the insulator and the like in a layer-by-layer abstract mode, the receptive field of a high-level network is large, the semantic information representation capability is strong, but the representation capability of the geometric information of the characteristic diagram is weak, and the space lacks of geometric characteristic details; and the receptive field of the lower network is small, the representation capability of the geometric detail information is strong, but the representation capability of the semantic information is weak. High-order semantic information can help to accurately detect or segment generalized targets such as power distribution channels from targets, and low-order geometric detail information can accurately detect or segment tiny targets such as power lines.
Aiming at the characteristics of a long and narrow line type structure of a power line in an image, weak insulator characteristics and the like, the embodiment adopts a double-layer characteristic extraction module and a characteristic pyramid fusion module, extracts low-order geometric characteristics and high-order semantic characteristics of a foreground image by using the double-layer characteristic extraction module, and fuses the low-order geometric characteristics and the high-order semantic characteristics of different scales by using the characteristic pyramid fusion module, so that the optimal detection effect can be obtained.
According to the identification method of the key power equipment in the digital twin station area, aiming at the problems of complex background and serious shielding of the overhead line of the low-voltage distribution network, the improved edge detection operator can be used for quickly and accurately extracting the foreground image containing the power line from the aerial image of the unmanned aerial vehicle with high resolution; the lightweight convolutional neural network model obtained by improving the YOLO convolutional neural network model is more suitable for power line identification, the lightweight convolutional neural network model only needs to identify the image characteristics of the foreground image, the extraction precision of the power line can be improved, the calculation load of the convolutional neural network is greatly reduced, and the method is very suitable for the algorithm environment with limited computing resources, such as an unmanned aerial vehicle embedded platform. The method can still give consideration to speed and precision for positioning and extracting the power line under the complex background, and has higher practicability and expandability.
The power line extraction experiment is carried out on the overhead line of the low-voltage distribution network in a village in a city, and the algorithm provided by the invention has higher detection precision and better real-time property under the conditions that the complex background in a high-resolution image and a target have partial shielding.
According to fig. 1, 3-7, in the process of extracting key power equipment of a low-voltage overhead low-voltage distribution network, fig. 1 is an aerial image initial diagram shot by an unmanned aerial vehicle, fig. 3 is a preprocessing diagram after image graying and gaussian filtering, fig. 4 is a characteristic diagram extracted by a Gabor operator, fig. 5 is an edge characteristic diagram after fusion of characteristics of different Gabor operators, fig. 6 is a foreground diagram segmented by a foreground extractor, and fig. 7 is a final result diagram of a lightweight convolutional neural network model. As can be seen from fig. 6, the foreground map accounts for 27.4% of the original map, which greatly contributes to improving the accuracy and speed of the subsequent convolutional neural network model, and as can be seen from fig. 7, the improved lightweight convolutional neural network model accurately completes extraction of the key power equipment, which only takes 25 ms.
Table 1 compares the results of identifying power lines for different algorithms as follows:
TABLE 1
As can be seen from the comparison of the algorithm in Table 1, mAP (mean Average precision) is the Average accuracy, and the highest value of mAP is 93.4%. Thanks to the rapidity of Gabor transformation, the recognition speed of the algorithm can reach 37 frames/s, which is second to YOLOV3-tiny, so that the real-time performance of the algorithm can be fully ensured, and the unmanned aerial vehicle can be conveniently carried. The detection speed of YOLOV3-tiny is obviously higher than that of other materials, but the mAP value is the lowest, and a large amount of false positives occur, especially power lines and insulators. The increase of YOLOV3 compared with YOLOV3-tiny deepens the network structure, the recognition speed is greatly reduced, but the improvement of the mAP value is limited, which also proves the view that the power line extraction is not sensitive to high-order semantic features. Fasterncn, like the algorithm herein, is a two-stage identification algorithm with a top value of mAP significantly higher than YOLOV3 algorithm, but with a detection speed much lower than other algorithms, so as to be significantly unsuitable for drones.
The algorithm can accurately identify the power distribution channel and extract the power line in the power distribution channel, and has the highest detection precision. The yolov3-tiny algorithm has the condition of missing reports, a part of power lines and insulators are not detected, and meanwhile, the miscellaneous lines, pipelines and the like outside the power distribution channel are mistakenly identified as the power lines. Compared with yolov3-tiny, the yolov3 algorithm adds abundant high-order semantic features, but false alarm and false alarm conditions exist, a part of power lines and insulators are not detected, and miscellaneous lines, road surface white identification lines and the like outside a power distribution channel are mistakenly identified as the power lines.
The invention is specially carried on the unmanned aerial vehicle of patrolling and examining to the overhead line of low voltage distribution, it is the pioneer at home and abroad, position and extract the electric power apparatus under the complicated background and can give consideration to speed and precision, have higher practicability.
In subsequent work, the algorithm and the geographic information acquired in real time can be fused and then transplanted into embedded equipment for carrying by the unmanned aerial vehicle, so that the comprehensive automatic identification of the line trend based on the aerial photography of the unmanned aerial vehicle is realized, the basic data of the low-voltage distribution network is effectively perfected, and the work burden of basic personnel is relieved.
The invention provides a method for identifying key electric power equipment in a digital twin region aiming at the problems of complex background and serious shielding of an overhead line of a low-voltage distribution network, which is an electric power equipment identification method based on a Gabor-YOLO algorithm.
Referring to fig. 13, the present invention further provides an embodiment of a system for identifying a key power device in a digital twin platform area, including:
the image acquisition and preprocessing module 11 is used for acquiring an aerial image of a digital twin platform area of the low-voltage distribution network and preprocessing the aerial image to obtain a preprocessed aerial image;
a foreground image obtaining module 22, configured to extract a foreground image including a key power device from the preprocessed aerial image by using an improved edge detection operator, where the key power device at least includes a power line;
and the key power equipment identification module 33 is configured to input the foreground image into a trained lightweight convolutional neural network model, and identify the key power equipment, where the lightweight convolutional neural network model is obtained by performing lightweight improvement on a backbone network of a YOLOv4 model.
In order to better illustrate the advantages of the algorithm in the invention, the following takes the unmanned aerial vehicle to inspect the low-voltage distribution network as an example to illustrate the function of the unmanned aerial vehicle:
1. and eliminating the environmental background of the aerial image, and extracting the foreground part of the image.
(1) The image acquisition and preprocessing module quickly performs preprocessing such as graying, Gaussian filtering and the like on the received aerial image;
(2) the foreground image acquisition module utilizes an improved edge detection operator, namely a Gabor operator, to perform rapid Gabor conversion on the preprocessed image, correspondingly distinguishes a foreground part and a background part of the image according to characteristics, reserves the foreground image and removes the background part.
2. And identifying the key electrical equipment from the foreground image by using a key electrical equipment identification module.
(1) And carrying out feature extraction on the foreground image by using a backbone network in the trained lightweight convolutional neural network model to obtain rich low-order geometric features and high-order semantic features.
(2) And transforming and fusing image features such as low-order geometric features, high-order semantic features and the like by using a double-layer feature extraction module (a feature extraction branch) and a feature pyramid fusion module in the lightweight convolutional neural network model, and identifying and obtaining the position and the type of the key electrical equipment in the aerial image.
The image Gabor method can quickly and accurately extract the foreground area of the power line (namely the partial image containing the power line) from the aerial image of the high-resolution unmanned aerial vehicle, the foreground area accounts for less than 30% of the original image, and the subsequent convolutional neural network only needs to extract the image characteristics of the foreground area for image recognition, so that the computational burden of the convolutional neural network is greatly reduced, and the accuracy is improved. The deep convolutional network can rapidly extract and identify the characteristics of the electrical equipment of the power distribution channel on the basis of Gabor image segmentation, the calculated amount in the algorithm process is greatly reduced compared with that of a conventional convolutional neural network, and the deep convolutional network is very suitable for the algorithm environment with limited computing resources, such as an unmanned aerial vehicle embedded platform.
(1) The biggest innovation point of the method is that high-precision high-resolution image recognition is realized in an algorithm environment with limited computing resources, and the method is very suitable for the urgent needs of current mainstream unmanned aerial vehicle aerial photography, which is not available in other patents.
(2) The function of the Gabor image segmentation algorithm of the invention is to extract foreground region information and input the foreground region information into a convolutional neural network instead of segmenting out specific objects, which is different from other patents.
(3) The lightweight convolutional neural network can only process low-resolution images in other patents, and cannot process high-resolution large images, and in the invention, the lightweight convolutional neural network can process large images of more than 10M under the transformation of a Gabor image segmentation algorithm, so that the high-resolution images can be accurately identified.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for identifying key power equipment in a digital twin region is characterized by comprising the following steps:
acquiring an aerial image of a digital twin platform area of the low-voltage distribution network, and preprocessing the aerial image to obtain a preprocessed aerial image;
extracting a foreground image containing key power equipment from the preprocessed aerial image by using an improved edge detection operator, wherein the key power equipment at least comprises a power line;
and inputting the foreground image into a trained lightweight convolutional neural network model, and identifying the key power equipment, wherein the lightweight convolutional neural network model is obtained by carrying out lightweight improvement on a backbone network of a YOLOv4 model.
2. The method for identifying key power devices in a digital twin region according to claim 1, wherein extracting a foreground image containing key power devices from the preprocessed aerial image by using an improved edge detection operator comprises:
and carrying out rapid Gabor transformation on the preprocessed aerial image by utilizing an improved Gabor operator to extract Gabort characteristics, distinguishing a foreground image and a background image of the preprocessed aerial image according to the Gabort characteristics, and reserving the foreground image.
3. The method of identifying digital twin zone critical power equipment of claim 1, wherein the lightweight convolutional neural network model comprises:
a backbone network, a double-layer feature extraction module and a feature pyramid fusion module of a double attention mechanism are added;
the foreground image feature extraction module extracts a low-order geometric feature and a high-order semantic feature from the foreground image feature extraction module, the feature pyramid fusion module performs feature fusion on the low-order geometric feature and the high-order semantic feature to obtain image features of different scales, the dual attention mechanism comprises a channel attention mechanism and a space attention mechanism, the low-order geometric feature is used for representing geometric detail information of the foreground image, and the high-order semantic feature is used for representing semantic information of the foreground image.
4. The method for identifying digital twin zone key power equipment as claimed in claim 3, wherein said backbone network incorporating a dual attention mechanism comprises:
the system comprises a three-layer 2D convolutional neural network module and a three-layer residual error convolutional module;
the two previous layers of 2D convolutional neural network modules are connected in series, then are connected in series with the three layers of residual convolutional modules, and finally are connected in series with the third layer of 2D convolutional neural network modules, and a double attention mechanism is added between every two adjacent residual convolutional modules.
5. The method for identifying digital twin zone critical power equipment as claimed in claim 4, wherein said 2D convolutional neural network module comprises:
a layer of normal two-dimensional convolution Conv2D, a layer of normalized Batchnormalization and a layer of activation function LeakyReLU connected in series.
6. The method for identifying the key power equipment of the digital twin platform area according to claim 4, wherein the residual convolution module is formed by connecting two layers of 2D convolution neural network modules in series and then connecting the two layers of 2D convolution neural network modules in parallel with a skipping connection structure.
7. The method for identifying the key power equipment of the digital twin region as claimed in claim 6, wherein a double attention mechanism is added between two adjacent residual convolution modules, and the double attention mechanism comprises:
adding a channel attention mechanism between two adjacent 2D convolution neural network modules of the residual convolution module, wherein the channel attention mechanism is used for adjusting the weight of each channel of the residual convolution module;
and a spatial attention mechanism is connected in series behind the channel attention mechanism and is used for adjusting the spatial feature map output by the 2D convolutional neural network module.
8. The method for identifying digital twin zone key power devices as claimed in claim 3, wherein said dual-layer feature extraction module extracting said lower-order geometric features and said higher-order semantic features from said backbone network comprises:
the double-layer feature extraction module extracts the low-order geometric features from a first layer of residual convolution module of the backbone network, and extracts the high-order semantic features from a third layer of residual convolution module of the backbone network.
9. The method for identifying the key power equipment in the digital twin platform area according to claim 3, wherein the feature pyramid fusion module performs feature fusion on the low-order geometric features and the high-order semantic features to obtain image features of different scales, and comprises:
and the feature pyramid fusion module performs up-sampling on the high-order semantic features, aligns the up-sampled features and then splices the aligned features with the low-order geometric features to obtain image features with different scales.
10. A system for identifying a digital twin zone critical power device, comprising:
the image acquisition and preprocessing module is used for acquiring an aerial image of a digital twin platform area of the low-voltage distribution network and preprocessing the aerial image to obtain a preprocessed aerial image;
the foreground image acquisition module is used for extracting a foreground image containing key power equipment from the preprocessed aerial image by using an improved edge detection operator, wherein the key power equipment at least comprises a power line;
and the key power equipment identification module is used for inputting the foreground image into a trained light-weight convolutional neural network model to identify the key power equipment, wherein the light-weight convolutional neural network model is obtained by carrying out light-weight improvement on a backbone network of a YOLOv4 model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210776349.9A CN114842363B (en) | 2022-07-04 | 2022-07-04 | Identification method and system for key power equipment in digital twin platform area |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210776349.9A CN114842363B (en) | 2022-07-04 | 2022-07-04 | Identification method and system for key power equipment in digital twin platform area |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114842363A true CN114842363A (en) | 2022-08-02 |
| CN114842363B CN114842363B (en) | 2022-10-18 |
Family
ID=82573414
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210776349.9A Active CN114842363B (en) | 2022-07-04 | 2022-07-04 | Identification method and system for key power equipment in digital twin platform area |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114842363B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115562191A (en) * | 2022-09-26 | 2023-01-03 | 北京能科瑞元数字技术有限公司 | Productivity intermediate station intelligent conjecture analysis method based on industrial digital twin |
| CN116542203A (en) * | 2023-06-27 | 2023-08-04 | 贵州轻工职业技术学院 | Circuit simulation method and device for realizing wearable equipment based on embedded and digital mode |
| CN117574259A (en) * | 2023-10-12 | 2024-02-20 | 南京工业大学 | Attention twin intelligent migration interpretability diagnosis method suitable for high-end equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106251344A (en) * | 2016-07-26 | 2016-12-21 | 北京理工大学 | A kind of multiple dimensioned infrared target self-adapting detecting method of view-based access control model receptive field |
| WO2021149857A1 (en) * | 2020-01-22 | 2021-07-29 | 고려대학교 세종산학협력단 | Accurate animal detection method and apparatus using yolo-based light-weight bounding box detection and image processing |
| CN113313032A (en) * | 2021-05-31 | 2021-08-27 | 广东电网有限责任公司 | Detection algorithm and related device for external force damage abnormal target of power transmission line |
| CN113537119A (en) * | 2021-07-28 | 2021-10-22 | 国网河南省电力公司电力科学研究院 | Transmission line connecting part detection method based on improved Yolov4-tiny |
| CN113888514A (en) * | 2021-10-09 | 2022-01-04 | 中国南方电网有限责任公司超高压输电公司广州局 | Method and device for detecting defects of ground wire, edge computing equipment and storage medium |
| CN114283469A (en) * | 2021-12-14 | 2022-04-05 | 贵州大学 | Lightweight target detection method and system based on improved YOLOv4-tiny |
| CN114463257A (en) * | 2021-12-23 | 2022-05-10 | 国网湖南省电力有限公司 | Power equipment infrared image detection method and system based on deep learning |
-
2022
- 2022-07-04 CN CN202210776349.9A patent/CN114842363B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106251344A (en) * | 2016-07-26 | 2016-12-21 | 北京理工大学 | A kind of multiple dimensioned infrared target self-adapting detecting method of view-based access control model receptive field |
| WO2021149857A1 (en) * | 2020-01-22 | 2021-07-29 | 고려대학교 세종산학협력단 | Accurate animal detection method and apparatus using yolo-based light-weight bounding box detection and image processing |
| CN113313032A (en) * | 2021-05-31 | 2021-08-27 | 广东电网有限责任公司 | Detection algorithm and related device for external force damage abnormal target of power transmission line |
| CN113537119A (en) * | 2021-07-28 | 2021-10-22 | 国网河南省电力公司电力科学研究院 | Transmission line connecting part detection method based on improved Yolov4-tiny |
| CN113888514A (en) * | 2021-10-09 | 2022-01-04 | 中国南方电网有限责任公司超高压输电公司广州局 | Method and device for detecting defects of ground wire, edge computing equipment and storage medium |
| CN114283469A (en) * | 2021-12-14 | 2022-04-05 | 贵州大学 | Lightweight target detection method and system based on improved YOLOv4-tiny |
| CN114463257A (en) * | 2021-12-23 | 2022-05-10 | 国网湖南省电力有限公司 | Power equipment infrared image detection method and system based on deep learning |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115562191A (en) * | 2022-09-26 | 2023-01-03 | 北京能科瑞元数字技术有限公司 | Productivity intermediate station intelligent conjecture analysis method based on industrial digital twin |
| CN115562191B (en) * | 2022-09-26 | 2024-02-27 | 北京能科瑞元数字技术有限公司 | Industrial digital twin-based intelligent presumption analysis method for productivity center |
| CN116542203A (en) * | 2023-06-27 | 2023-08-04 | 贵州轻工职业技术学院 | Circuit simulation method and device for realizing wearable equipment based on embedded and digital mode |
| CN116542203B (en) * | 2023-06-27 | 2023-09-01 | 贵州轻工职业技术学院 | Circuit simulation method and device for realizing wearable equipment based on embedded and digital mode |
| CN117574259A (en) * | 2023-10-12 | 2024-02-20 | 南京工业大学 | Attention twin intelligent migration interpretability diagnosis method suitable for high-end equipment |
| CN117574259B (en) * | 2023-10-12 | 2024-05-07 | 南京工业大学 | Attention twin intelligent migration interpretability diagnosis method suitable for high-end equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114842363B (en) | 2022-10-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114842363B (en) | Identification method and system for key power equipment in digital twin platform area | |
| CN112651978B (en) | Sublingual microcirculation image segmentation method and device, electronic equipment and storage medium | |
| CN108734143A (en) | A kind of transmission line of electricity online test method based on binocular vision of crusing robot | |
| Lu et al. | Cascaded multi-task road extraction network for road surface, centerline, and edge extraction | |
| CN106295613A (en) | A kind of unmanned plane target localization method and system | |
| CN112489050A (en) | Semi-supervised instance segmentation algorithm based on feature migration | |
| Chen et al. | Surface defect detection of electric power equipment in substation based on improved YOLOv4 algorithm | |
| CN113066089B (en) | Real-time image semantic segmentation method based on attention guide mechanism | |
| CN115223017B (en) | Multi-scale feature fusion bridge detection method based on depth separable convolution | |
| CN113312499A (en) | Power safety early warning method and system based on knowledge graph | |
| CN112419333A (en) | Remote sensing image self-adaptive feature selection segmentation method and system | |
| CN114821508B (en) | Road three-dimensional target detection method based on implicit context learning | |
| CN117315509A (en) | Yolov 7-based unmanned aerial vehicle aerial image small target detection method | |
| CN115424016A (en) | High-voltage transmission line identification method | |
| CN113870160B (en) | Point cloud data processing method based on transformer neural network | |
| CN114998610A (en) | Target detection method, device, equipment and storage medium | |
| CN118172334A (en) | Electric network wiring diagram electric element cascade detection method based on transducer and residual convolution | |
| CN113012107B (en) | Power grid defect detection method and system | |
| CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
| Zhang et al. | A context-aware road extraction method for remote sensing imagery based on transformer network | |
| CN111583171B (en) | Insulator defect detection method integrating foreground compact characteristic and multi-environment information | |
| Zhou et al. | Class-aware edge-assisted lightweight semantic segmentation network for power transmission line inspection | |
| CN117315370A (en) | Floating object detection method, device, storage medium and equipment based on remote sensing image | |
| Hao et al. | Research on image semantic segmentation based on FCN-VGG and pyramid pooling module | |
| CN116912774A (en) | Infrared image target identification method, electronic device and storage medium of power transmission and transformation equipment based on edge calculation |
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 |