CN116338392A - Method, device and equipment for identifying lightning discharge defects of glass insulator - Google Patents

Method, device and equipment for identifying lightning discharge defects of glass insulator Download PDF

Info

Publication number
CN116338392A
CN116338392A CN202310327200.7A CN202310327200A CN116338392A CN 116338392 A CN116338392 A CN 116338392A CN 202310327200 A CN202310327200 A CN 202310327200A CN 116338392 A CN116338392 A CN 116338392A
Authority
CN
China
Prior art keywords
defect
yolov4
detection model
feature
glass insulator
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.)
Pending
Application number
CN202310327200.7A
Other languages
Chinese (zh)
Inventor
何勇
黄诗文
陆林
原瀚杰
孙仝
郑耀华
刘永浩
谭海傲
陈亮
董丽梦
尤德柱
罗建斌
黄城
姜天航
邝建东
潘绮彤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Power Grid Co Ltd
Zhaoqing Power Supply Bureau of Guangdong Power Grid Co Ltd
Original Assignee
Guangdong Power Grid Co Ltd
Zhaoqing Power Supply Bureau of Guangdong Power Grid Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangdong Power Grid Co Ltd, Zhaoqing Power Supply Bureau of Guangdong Power Grid Co Ltd filed Critical Guangdong Power Grid Co Ltd
Priority to CN202310327200.7A priority Critical patent/CN116338392A/en
Publication of CN116338392A publication Critical patent/CN116338392A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
    • G01R31/1245—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of line insulators or spacers, e.g. ceramic overhead line cap insulators; of insulators in HV bushings
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • 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
    • G06V10/77—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
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • 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
    • 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 relates to a method, a device and equipment for identifying lightning strike discharge defects of a glass insulator, wherein the method comprises the steps of constructing an RCNN detection model and a YOLOv4 detection model; acquiring a patrol image and patrol route information of a glass insulator on a power transmission line, and detecting the patrol image by adopting a YOLOv4 detection model to obtain a detection recognition result; if the detection and identification result shows that the glass insulator has the lightning flashover defect, detecting the inspection image according to the inspection route information by adopting an RCNN detection model to obtain the position and type of the lightning flashover defect. The RCNN detection model and the YOLOv4 detection model constructed by the glass insulator lightning stroke discharge defect identification method are matched with each other to realize lightning stroke flashover defect identification, defect type identification and defect position determination of the glass insulator, so that the defect identification efficiency of lightning stroke discharge of the glass insulator of the power transmission line is improved, and the working force is reduced.

Description

Method, device and equipment for identifying lightning discharge defects of glass insulator
Technical Field
The application relates to the technical field of insulators, in particular to a method, a device and equipment for identifying lightning strike discharge defects of a glass insulator.
Background
The insulator plays an important role in electric insulation and mechanical support in the power transmission line, the good insulator has an important significance for safe and stable operation of the power transmission line, and compared with other types of insulators, the glass insulator has the advantages of high mechanical strength, convenience in detection, zero-value self-explosion, lower maintenance cost and the like, so that the glass insulator is widely applied in the power transmission line. However, due to the difference of the operation conditions, the glass insulators are easy to be subjected to lightning strike discharge of different degrees, so that the service life of the insulator equipment is shortened, degradation self-explosion of different degrees is easy to occur, and when the number of the self-explosion insulators does not meet the requirement of the operation regulations of overhead power transmission lines, the tripping accident of the power transmission lines is caused, so that the safe and stable operation of a power grid and equipment is seriously threatened. Therefore, effective countermeasure is adopted to avoid tripping accident of the transmission line caused by insulator self-explosion.
The existing method for detecting insulators on a power transmission line by adopting manual field test is low in efficiency, tens of thousands of towers are arranged in a power grid, the annual inspection task is heavy, and the manual field test method is time-consuming and labor-consuming.
Disclosure of Invention
The embodiment of the application provides a method, a device and equipment for identifying lightning strike discharge defects of a glass insulator, which are used for solving the technical problems that the detection efficiency is low, time and labor are wasted by adopting a manual mode in the existing inspection of the insulator operation process on a transmission line.
In order to achieve the above purpose, the embodiment of the present application provides the following technical solutions:
a method for identifying lightning strike discharge defects of a glass insulator comprises the following steps:
constructing an RCNN detection model and a YOLOv4 detection model, wherein the YOLOv4 detection model is constructed based on an adaptive feature fusion improved YOLOv4 network;
acquiring a patrol image and patrol route information of a glass insulator on a power transmission line, and detecting the patrol image by adopting the YOLOv4 detection model to obtain a detection recognition result;
and if the detection and identification result shows that the glass insulator has the lightning flashover defect, detecting the inspection image according to the inspection route information and by adopting the RCNN detection model to obtain the position and type of the lightning flashover defect.
Preferably, constructing the YOLOv4 detection model includes:
acquiring first defect image data with lightning flashover defects, wherein the first defect image data comprises N pieces of first defect images;
and combining the self-adaptive feature fusion and the PAN structure of the YOLOv4 network to perform training processing on the first defect image data to obtain a YOLOv4 detection model.
Preferably, combining the adaptive feature fusion and the PAN structure of the YOLOv4 network with the training process of the first defect image data includes:
processing the first defect image data through a PAN structure of a YOLOv4 network to obtain at least three feature graphs with different scales;
performing equal scaling treatment on each feature map by adopting self-adaptive feature fusion to obtain feature scaled maps with the same scale, and calculating the weight of the corresponding scale of the feature scaled maps;
and carrying out fusion calculation on weights corresponding to all the feature zoom diagrams by adopting self-adaptive feature fusion to obtain spatial weights of mapping fusion of the features of each scale.
Preferably, the YOLOv4 detection model calculates the weight of the scale corresponding to the feature zoom map by using a weight calculation formula, and the weight calculation formula is as follows:
Figure BDA0004153739720000031
Figure BDA0004153739720000032
Figure BDA0004153739720000033
in the formula, alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the transverse and longitudinal coordinates of the spatial position of the feature map, and lambda is the convolution kernel calculation control coefficient in the YOLOv4 network.
Preferably, the YOLOv4 detection model calculates the weights of all the feature zoom diagrams by adopting a fusion calculation formula to obtain the spatial weights of mapping fusion of the features of each scale; the fusion calculation formula is as follows:
Figure BDA0004153739720000034
Figure BDA0004153739720000035
Figure BDA0004153739720000036
wherein alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the horizontal and vertical coordinates of the spatial position of the feature map,
Figure BDA0004153739720000037
the vector of the graph at spatial position (i, j) is scaled for the a-th feature.
Preferably, the training process of the first defect image data by combining the adaptive feature fusion and the PAN structure of the YOLOv4 network includes: and splicing and reconstructing data output by the YOLOv4 network convolution layer by adopting a decoupling head to obtain the size of the feature map.
Preferably, constructing the RCNN detection model includes:
acquiring second defect image data marked with defect types and position information, wherein the second defect image data comprises N pieces of second defect images;
and training the second defect image data by adopting an RPN network of three cascade detection heads to obtain an RCNN detection model capable of identifying the defect position and the defect category.
The application also provides a glass insulator lightning stroke discharge defect recognition device, include: the system comprises a model construction module, a defect identification module and a position type identification module;
the model construction module is used for constructing an RCNN detection model and a YOLOv4 detection model, and the YOLOv4 detection model is constructed based on an adaptive feature fusion improved YOLOv4 network;
the defect identification module is used for acquiring a patrol image and patrol route information of the glass insulator on the power transmission line, and detecting the patrol image by adopting the YOLOv4 detection model to obtain a detection and identification result;
the position type recognition module is used for detecting the position and the type of the lightning flashover defect of the glass insulator according to the detection recognition result, detecting the inspection image according to the inspection route information and by adopting the RCNN detection model.
Preferably, the model building module comprises a first model building module and a second model building module;
the first model construction module is used for acquiring first defect image data with lightning flashover defects, and performing training processing on the first defect image data by combining self-adaptive feature fusion and a PAN structure of a YOLOv4 network to obtain a YOLOv4 detection model;
the model construction module is used for acquiring second defect image data marked with defect types and position information, and training the second defect image data by adopting an RPN network of three cascade detection heads to obtain an RCNN detection model capable of identifying the defect positions and the defect types;
the first model building module is further used for processing the first defect image data through a PAN structure of a YOLOv4 network to obtain at least three feature diagrams with different scales; performing equal scaling treatment on each feature map by adopting self-adaptive feature fusion to obtain feature scaled maps with the same scale, and calculating the weight of the corresponding scale of the feature scaled maps by adopting a weight calculation formula; carrying out fusion calculation on weights corresponding to all the feature zoom diagrams through self-adaptive feature fusion by adopting a fusion calculation formula to obtain spatial weights of mapping fusion of the features of each scale;
the weight calculation formula is as follows:
Figure BDA0004153739720000051
Figure BDA0004153739720000052
Figure BDA0004153739720000053
the fusion calculation formula is as follows:
Figure BDA0004153739720000054
Figure BDA0004153739720000055
Figure BDA0004153739720000056
wherein, alpha, beta and gamma are respectively the numerical values of weights, l is the size of the scale,i. j is the numerical value of the abscissa of the spatial position of the feature map, lambda is the convolution kernel calculation control coefficient in the YOLOv4 network,
Figure BDA0004153739720000057
the vector of the graph at spatial position (i, j) is scaled for the a-th feature.
The application also provides a terminal device, which comprises a processor and a memory;
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is used for executing the method for identifying the lightning strike discharge defects of the glass insulator according to the instructions in the program codes.
From the above technical solutions, the embodiments of the present application have the following advantages: the method comprises the steps of constructing an RCNN detection model and a YOLOv4 detection model; acquiring a patrol image and patrol route information of a glass insulator on a power transmission line, and detecting the patrol image by adopting a YOLOv4 detection model to obtain a detection recognition result; if the detection and identification result shows that the glass insulator has the lightning flashover defect, detecting the inspection image according to the inspection route information by adopting an RCNN detection model to obtain the position and type of the lightning flashover defect. The RCNN detection model and the YOLOv4 detection model constructed by the glass insulator lightning stroke discharge defect identification method are matched with each other to realize lightning stroke flashover defect identification, defect type identification and defect position determination of the glass insulator, so that the defect identification efficiency of lightning stroke discharge of the glass insulator of the power transmission line is improved, and the working force is reduced. The technical problems that the detection efficiency is low, time and labor are wasted are solved by adopting a manual mode in the existing inspection of the insulator operation process on the transmission line.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained according to these drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of steps of a method for identifying lightning strike discharge defects of a glass insulator according to an embodiment of the present application;
FIG. 2 is a frame flow chart of a YOLOv4 detection model in a method for identifying lightning strike discharge defects of a glass insulator according to an embodiment of the present application;
FIG. 3 is a frame diagram of a YOLOv4 network of a YOLOv4 detection model in a glass insulator lightning discharge defect identification method according to an embodiment of the present application;
fig. 4 is a frame diagram of a glass insulator lightning discharge defect recognition device according to an embodiment of the present application.
Detailed Description
In order to make the objects, features and advantages of the present invention more obvious and understandable, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the embodiments described below are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
The embodiment of the application provides a method, a device and equipment for identifying lightning strike discharge defects of a glass insulator, which are used for solving the technical problems that the detection efficiency is low, time and labor are wasted by adopting a manual mode in the existing inspection of the operation process of the insulator on a transmission line. The method, the device and the equipment for identifying the lightning stroke discharge defects of the glass insulator are used for detecting the lightning stroke flashover front end of the glass insulator.
Embodiment one:
fig. 1 is a flowchart of steps of a method for identifying lightning strike discharge defects of a glass insulator according to an embodiment of the present application.
As shown in fig. 1, an embodiment of the present application provides a method for identifying a lightning strike discharge defect of a glass insulator, including the following steps:
s1, constructing an RCNN detection model and a YOLOv4 detection model, wherein the YOLOv4 detection model is constructed based on self-adaptive feature fusion improved YOLOv4 network.
In step S1, the purpose is to construct a model for identifying lightning strike discharge defects of the glass insulator. In this embodiment, the YOLOv4 detection model may be set on an edge computing box for power transmission line inspection, and the RCNN detection model may be set on a cloud high-performance computing cluster for power transmission line inspection. The cloud high-performance computing cluster is large in memory and high in computing capacity, and can meet the requirements of real-time detection. The edge computing box with the inspection is smaller in memory, the floating point computing capability is relatively weak, the cradle head needs to be adjusted in real time according to the detection result, and the detection model needs to have the capability of real-time detection, so that the detection model which is small in size and can have real-time performance needs to be deployed, and the computing box with the YOLOv4 detection model deployed at the edge end is constructed.
S2, acquiring a patrol image and patrol route information of the glass insulator on the transmission line, and detecting the patrol image by adopting a YOLOv4 detection model to obtain a detection and identification result.
In the step S2, firstly, a patrol image for identifying the lightning discharge defect of the glass insulator and route information of the patrol image are obtained; and secondly, detecting the inspection image by using a YOLOv4 detection model to obtain a detection and identification result.
In the embodiment of the application, the line power transmission line can be inspected by the unmanned aerial vehicle according to the inspection navigation, and the inspection image is acquired by adopting the camera equipment of the unmanned aerial vehicle.
The unmanned aerial vehicle receives the navigation point coordinates of the camera shooting acquisition images in real time according to the routing inspection route as routing inspection route information. The image pickup apparatus may be a camera. According to the glass insulator lightning stroke discharge defect identification method, the acquired inspection image can be transmitted to the RCNN detection model of the cloud through the wireless communication module of the camera equipment.
S3, if the detection and identification result shows that the glass insulator has the lightning flashover defect, detecting the inspection image according to the inspection route information by adopting an RCNN detection model to obtain the position and type of the lightning flashover defect.
In the step S3, if the lightning flashover defect exists in the glass insulator according to the detection and identification result in the step S2, the type and the position of the lightning flashover defect in the inspection image are further identified again by adopting an RCNN detection model. According to the lightning stroke discharge defect identification method for the glass insulator, the RCNN detection model and the YOLOv4 detection model are matched with each other to realize lightning stroke flashover defect identification and position determination of the glass insulator.
In the embodiment of the application, the method for identifying the lightning strike discharge defects of the glass insulator also correlates the detection and identification result to the coordinate position on the inspection route information and generates a visual report so that power staff can analyze and process the power transmission line.
The application provides a method for identifying lightning strike discharge defects of a glass insulator, which comprises the steps of constructing an RCNN detection model and a YOLOv4 detection model; acquiring a patrol image and patrol route information of a glass insulator on a power transmission line, and detecting the patrol image by adopting a YOLOv4 detection model to obtain a detection recognition result; if the detection and identification result shows that the glass insulator has the lightning flashover defect, detecting the inspection image according to the inspection route information by adopting an RCNN detection model to obtain the position and type of the lightning flashover defect. The RCNN detection model and the YOLOv4 detection model constructed by the glass insulator lightning stroke discharge defect identification method are matched with each other to realize lightning stroke flashover defect identification, defect type identification and defect position determination of the glass insulator, so that the defect identification efficiency of lightning stroke discharge of the glass insulator of the power transmission line is improved, and the working force is reduced. The technical problems that the detection efficiency is low, time and labor are wasted are solved by adopting a manual mode in the existing inspection of the insulator operation process on the transmission line.
In one embodiment of the present application, constructing the YOLOv4 detection model includes:
acquiring first defect image data with lightning flashover defects, wherein the first defect image data comprises N pieces of first defect images;
combining the self-adaptive feature fusion and a PAN structure of the YOLOv4 network to perform training processing on the first defect image data to obtain a YOLOv4 detection model;
combining the adaptive feature fusion and the PAN structure of the YOLOv4 network with the training processing of the first defect image data comprises the following steps:
processing the first defect image data through a PAN structure of the YOLOv4 network to obtain at least three feature graphs with different scales;
performing equal scaling treatment on each feature map by adopting self-adaptive feature fusion to obtain feature scaled maps with the same scale, and calculating the weight of the corresponding scale of the feature scaled maps;
performing fusion calculation on weights corresponding to all feature zoom diagrams by adopting self-adaptive feature fusion to obtain spatial weights of mapping fusion of the features of each scale;
the YOLOv4 detection model adopts a weight calculation formula to calculate the weight of the corresponding scale of the feature zoom map, and the weight calculation formula is as follows:
Figure BDA0004153739720000091
Figure BDA0004153739720000092
Figure BDA0004153739720000093
wherein alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the transverse and longitudinal coordinates of the spatial position of the feature map, and lambda is the convolution kernel calculation control coefficient in the YOLOv4 network;
the YOLOv4 detection model calculates the weights of all the feature zoom diagrams by adopting a fusion calculation formula to obtain the spatial weights of the mapping fusion of the feature of each scale; the fusion calculation formula is as follows:
Figure BDA0004153739720000094
Figure BDA0004153739720000095
Figure BDA0004153739720000096
wherein alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the horizontal and vertical coordinates of the spatial position of the feature map,
Figure BDA0004153739720000097
the vector of the graph at spatial position (i, j) is scaled for the a-th feature.
In the method for identifying the lightning strike discharge defects of the glass insulator, in the process of constructing a YOLOv4 detection model, three characteristic diagrams with different scales are generated by processing first defect image data by adopting a PAN structure of a YOLOv4 network, and a is E [1,2,3]The characteristic diagram is marked as X 1 、X 2 And X 3 . If the feature map with the a-th scale is subjected to constant scaling processing, obtaining a feature scaling map X with the same scale 1→l 、X 2→l 、X 3→l And scaling feature graphs with different scales to feature scaling graphs with the same shape and size, and then carrying out fusion calculation on the weights of all the feature scaling graphs by adopting self-adaptive feature fusion to obtain the spatial weights of feature mapping fusion of each scale. According to the glass insulator lightning discharge defect identification method, the problem of low accuracy of detection and identification results caused by characteristic influence among different layers of a YOLOv4 network is avoided through a YOLOv4 detection model constructed based on self-adaptive characteristic fusion and a PAN structure of the YOLOv4 network.
Fig. 2 is a frame flow chart of a YOLOv4 detection model in the method for identifying lightning discharge defects of a glass insulator according to an embodiment of the present application.
As shown in fig. 2, in the embodiment of the present application, a YOLOv4 detection model of an adaptive feature fusion ASFF is added, and it is seen that the adaptive feature fusion ASFF fuses three feature graphs of different scales in a PAN structure of a YOLOv4 network into three feature graphs of corresponding scales, and then learns a fusion weight to learn the contribution of different feature scales to a predicted feature graph, instead of simply adopting a cascading manner to perform multi-level feature fusion in the existing YOLOv4 model. The space weight of each scale feature map fusion is adaptively learned by the PAN structure ASFF of the YOLOv4 network, so that the detection precision of objects with different scales is effectively improved, and the method is more suitable for positioning a plurality of defect targets with different scales from the inspection images.
Fig. 3 is a frame diagram of a YOLOv4 network of a YOLOv4 detection model in the method for identifying lightning discharge defects of a glass insulator according to an embodiment of the present application.
In one embodiment of the present application, combining the adaptive feature fusion with the PAN structure of the YOLOv4 network to perform training processing on the first defect image data includes: and splicing and reconstructing data output by the YOLOv4 network convolution layer by adopting a decoupling head to obtain the size of the feature map.
It should be noted that, as shown in fig. 3, the method for identifying lightning strike discharge defects of the glass insulator can greatly improve the detection capability, accelerate the convergence speed of the network and also improve the speed of positioning various power elements with different scales in the inspection process by adopting the YOLOv4 detection model of the decoupling head. In this embodiment, if the feature size of the input first defect image data is (13, 13, 512), the type prediction output of the convolution layer in the YOLOv4 detection model for the first defect image data is cls_out, the judgment output of the convolution layer in the YOLOv4 detection model for the foreground or background of the first defect image data is obj_out, the coordinate information output of the convolution layer in the YOLOv4 detection model for the first defect image data is reg_out, and the feature information of the dimension data is obtained by performing splicing and fusion on the information output by the 3 convolution layers in the YOLOv4 detection model; and reconstructing the spliced characteristic information to obtain the dimension data of the feature map. The splicing refers to that the alignment of the data of the first two dimensions of information output by a convolution layer in a YOLOv4 detection model is kept unchanged, and the data of the third dimension is stacked to obtain the characteristic information of the three dimensions after the splicing. The reconstruction refers to that the spliced characteristic information (13, 7) corresponds to the length, the width and the channel number respectively, the 2-dimensional matrix corresponding to the length, the width (13, 13) is straightened into a vector (13 x 13=169) of 1 dimension, the channel number is kept unchanged, and the size of the reconstructed characteristic diagram is (169,7). The dimension data of the feature map can be used for loss function calculation and network parameter update of the subsequent YOLOV4 detection model.
In one embodiment of the present application, constructing the RCNN detection model includes:
acquiring second defect image data marked with defect types and position information, wherein the second defect image data comprises N second defect images;
and training the second defect image data by adopting an RPN network of three cascade detection heads to obtain an RCNN detection model capable of identifying the defect position and the defect category.
It should be noted that, the RCNN detection model adopts the RPN network to perform the coarse detection in the first stage, so as to detect the defect position in the second defect image data as much as possible, and improve the recall rate of the RCNN detection model; and then the RPN network of the three cascade detection heads is utilized to refine the detection result step by step in the second stage, the false alarm result in the first stage is screened, the accuracy of the RCNN detection model is improved, and the RCNN detection model can efficiently perform defect identification and position location of lightning stroke discharge of the glass insulator.
In the embodiment of the application, in the step 1, in the construction of the RCNN detection model and the YOLOv4 detection model by the glass insulator lightning discharge defect identification method, the method can be realized by three steps of constructing a data set, training the model and testing the model,
constructing a data set: image data under multiple scenes are collected through inspection of the power transmission line by the unmanned aerial vehicle, a large number of images with abnormal brightness, large noise, blurring and the like are cleaned to serve as data of a data set in a manual sorting mode, and high-quality image training data are beneficial to improving robustness of a constructed model. And marking the electric power element and the defect target in the image by using a marking tool, wherein the marked information comprises category names and position information Xmin (X coordinate at the upper left corner of the marking frame), ymin (Y coordinate at the upper left corner of the marking frame), xmax (X coordinate at the lower right corner of the marking frame) and Ymax (Y coordinate at the lower right corner of the marking frame), and the marked information is stored into an xml tag file in a VOC data format. And dividing the image data in the data set and the corresponding annotation files into a training set and a testing set according to the quantity ratio of 4:1, and completing the construction of the data set.
Model training: and training a training set training model in the data set, wherein the model training adopts an adaptive moment estimation optimizer as a training optimization strategy of the network, and the total training is 100 rounds. The learning rate is dynamically adjusted from 0.001, and after each round of updating is completed, the learning rate is multiplied by 0.9. And observing the change of LOSS values LOSS in the YOLOv4 network and the RPN network in the training process, and when the LOSS value is not reduced for 5 continuous rounds, considering that the model is converged, stopping training, and obtaining the model with the best convergence effect (the lowest final LOSS value).
Model test: and testing the trained model by using a test set in the data set. The precision, recall, mAP (mean Average Precision), FPS (Frames Per Second) are used as performance indicators for model testing. The definition formula of precision and recall rate is as follows
Figure BDA0004153739720000121
Figure BDA0004153739720000122
Wherein x is TP，A For the number of correctly identified targets in class a, x FP，A Is the target number, x, of the error identified as positive in class A FN，A Representing the number of incorrectly identified positive examples in class A; p (P) A A target duty cycle that is correctly identified for all targets identified as category a; r is R A Is the target duty ratio of the correctly identified targets with the category A in the picture. Confidence is that a certain object belongs to a categoryAnd (A) probability, sequencing the detected targets of each category in the test set from high to low according to confidence, drawing an accuracy recall rate curve of the category, taking the area under the accuracy recall rate curve as the AP of the category, and then averaging the APs of all the categories to obtain mAP, wherein the mAP reflects the average detection accuracy of the model on each category. The FPS refers to the model per second processing, reflecting the detection speed of the target. FLPs refer to the computational effort of a model, and are used to measure the complexity of the model. At R A 、P A The values of mAP are more than 85%, the value of YOLOv4 detection model is more than 0.8, the FPS value of RCNN detection model is more than 10, and the FPS value of RCNN detection model is more than 1.5, which means that the RCNN detection model and the YOLOv4 detection model are optimal models.
In the embodiment of the application, the lightning stroke discharge defect identification method for the glass insulator carries out lightning stroke discharge defect identification on the images of the inspection glass insulator through the optimal RCNN detection model and the YOLOv4 detection model, and has good defect identification effect and high efficiency. The detection precision of objects with different scales is effectively improved through the YOLOv4 detection model with the self-adaptive feature fusion ASFF, and the method is more suitable for positioning various power elements with different scales from the inspection image.
Embodiment two:
fig. 4 is a frame flow chart of a glass insulator lightning strike discharge defect recognition device according to an embodiment of the application.
As shown in fig. 4, an embodiment of the present application provides a device for identifying a lightning strike discharge defect of a glass insulator, including: a model construction module 10, a defect identification module 20, and a location type identification module 30;
the model construction module 10 is used for constructing an RCNN detection model and a Yolov4 detection model, wherein the Yolov4 detection model is constructed based on an adaptive feature fusion improved Yolov4 network;
the defect recognition module 20 is used for acquiring the inspection image and the inspection route information of the glass insulator on the power transmission line, and detecting the inspection image by adopting a YOLOv4 detection model to obtain a detection recognition result;
the position type recognition module 30 is configured to detect, according to the detection and recognition result, that the glass insulator has a lightning flashover defect, and according to the inspection route information and by adopting the RCNN detection model, the inspection image, so as to obtain the position and type of the lightning flashover defect.
In the present embodiment, the model building module 10 includes a first model building module and a second model building module;
the first model construction module is used for acquiring first defect image data with lightning flashover defects, and combining the self-adaptive feature fusion and a PAN structure of the YOLOv4 network to perform training treatment on the first defect image data to obtain a YOLOv4 detection model;
the model construction module is used for acquiring second defect image data marked with defect types and position information, and training the second defect image data by adopting an RPN network of three cascade detection heads to obtain an RCNN detection model capable of identifying the defect positions and the defect types;
the first model building module is further used for processing the first defect image data through a PAN structure of the YOLOv4 network to obtain at least three feature graphs with different scales; performing equal scaling treatment on each feature map by adopting self-adaptive feature fusion to obtain feature scaled maps with the same scale, and calculating the weight of the corresponding scale of the feature scaled maps by adopting a weight calculation formula; carrying out fusion calculation on weights corresponding to all feature zoom diagrams through self-adaptive feature fusion by adopting a fusion calculation formula to obtain spatial weights of mapping fusion of the features of each scale;
the weight calculation formula is:
Figure BDA0004153739720000141
Figure BDA0004153739720000142
Figure BDA0004153739720000143
the fusion calculation formula is as follows:
Figure BDA0004153739720000144
Figure BDA0004153739720000145
Figure BDA0004153739720000146
wherein alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the transverse and longitudinal coordinates of the spatial position of the feature map, lambda is the convolution kernel calculation control coefficient in the YOLOv4 network,
Figure BDA0004153739720000147
the vector of the graph at spatial position (i, j) is scaled for the a-th feature.
It should be noted that, the module in the second device corresponds to the steps in the method in the first embodiment, and the content of the method for identifying lightning strike discharge defects of the glass insulator is described in detail in the first embodiment, and the content of the module in the second device is not described in detail in the second embodiment.
Embodiment III:
the embodiment of the application provides terminal equipment, which comprises a processor and a memory;
a memory for storing program code and transmitting the program code to the processor;
and the processor is used for executing the method for identifying the lightning strike discharge defect of the glass insulator according to the instructions in the program codes.
It should be noted that the processor is configured to execute the steps in the embodiment of the method for identifying a lightning strike discharge defect of a glass insulator according to the instructions in the program code. In the alternative, the processor, when executing the computer program, performs the functions of the modules/units in the system/apparatus embodiments described above.
For example, a computer program may be split into one or more modules/units, which are stored in a memory and executed by a processor to complete the present application. One or more of the modules/units may be a series of computer program instruction segments capable of performing specific functions for describing the execution of the computer program in the terminal device.
The terminal device may be a computing device such as a desktop computer, a notebook computer, a palm computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor, a memory. It will be appreciated by those skilled in the art that the terminal device is not limited and may include more or less components than those illustrated, or may be combined with certain components, or different components, e.g., the terminal device may also include input and output devices, network access devices, buses, etc.
The processor may be a central processing unit (Central Processing Unit, CPU), other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory may be an internal storage unit of the terminal device, such as a hard disk or a memory of the terminal device. The memory may also be an external storage device of the terminal device, such as a plug-in hard disk provided on the terminal device, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), or the like. Further, the memory may also include both an internal storage unit of the terminal device and an external storage device. The memory is used for storing computer programs and other programs and data required by the terminal device. The memory may also be used to temporarily store data that has been output or is to be output.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
In the several embodiments provided in the present application, it should be understood that the disclosed systems, apparatuses, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other forms.
The units described as separate units may or may not be physically separate, and units shown 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 may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above embodiments are merely for illustrating the technical solution of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.

Claims (10)

1. The method for identifying the lightning strike discharge defect of the glass insulator is characterized by comprising the following steps of:
constructing an RCNN detection model and a YOLOv4 detection model, wherein the YOLOv4 detection model is constructed based on an adaptive feature fusion improved YOLOv4 network;
acquiring a patrol image and patrol route information of a glass insulator on a power transmission line, and detecting the patrol image by adopting the YOLOv4 detection model to obtain a detection recognition result;
and if the detection and identification result shows that the glass insulator has the lightning flashover defect, detecting the inspection image according to the inspection route information and by adopting the RCNN detection model to obtain the position and type of the lightning flashover defect.
2. The method for identifying lightning strike discharge defects of glass insulators according to claim 1, wherein constructing a YOLOv4 detection model comprises:
acquiring first defect image data with lightning flashover defects, wherein the first defect image data comprises N pieces of first defect images;
and combining the self-adaptive feature fusion and the PAN structure of the YOLOv4 network to perform training processing on the first defect image data to obtain a YOLOv4 detection model.
3. The method for identifying a lightning strike discharge defect of a glass insulator according to claim 2, wherein combining the adaptive feature fusion and the PAN structure of the YOLOv4 network with the training process of the first defect image data comprises:
processing the first defect image data through a PAN structure of a YOLOv4 network to obtain at least three feature graphs with different scales;
performing equal scaling treatment on each feature map by adopting self-adaptive feature fusion to obtain feature scaled maps with the same scale, and calculating the weight of the corresponding scale of the feature scaled maps;
and carrying out fusion calculation on weights corresponding to all the feature zoom diagrams by adopting self-adaptive feature fusion to obtain spatial weights of mapping fusion of the features of each scale.
4. The method for identifying a lightning discharge defect of a glass insulator according to claim 3, wherein the YOLOv4 detection model calculates the weight of the scale corresponding to the characteristic scaling map by using a weight calculation formula, and the weight calculation formula is:
Figure FDA0004153739700000021
Figure FDA0004153739700000022
Figure FDA0004153739700000023
in the formula, alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the transverse and longitudinal coordinates of the spatial position of the feature map, and lambda is the convolution kernel calculation control coefficient in the YOLOv4 network.
5. The method for identifying the lightning discharge defects of the glass insulator according to claim 3, wherein the YOLOv4 detection model calculates the weights of all the feature zoom diagrams by adopting a fusion calculation formula to obtain the spatial weights of the mapping fusion of the features of each scale; the fusion calculation formula is as follows:
Figure FDA0004153739700000024
Figure FDA0004153739700000025
Figure FDA0004153739700000026
wherein alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the horizontal and vertical coordinates of the spatial position of the feature map,
Figure FDA0004153739700000027
the vector of the graph at spatial position (i, j) is scaled for the a-th feature.
6. The method for identifying the lightning strike discharge defect of the glass insulator according to claim 2, wherein the process of combining the adaptive feature fusion and the PAN structure of the YOLOv4 network to train the first defect image data comprises the following steps: and splicing and reconstructing data output by the YOLOv4 network convolution layer by adopting a decoupling head to obtain the size of the feature map.
7. The method for identifying lightning strike discharge defects of glass insulators according to claim 1, wherein constructing an RCNN detection model comprises:
acquiring second defect image data marked with defect types and position information, wherein the second defect image data comprises N pieces of second defect images;
and training the second defect image data by adopting an RPN network of three cascade detection heads to obtain an RCNN detection model capable of identifying the defect position and the defect category.
8. The utility model provides a glass insulator lightning stroke discharge defect recognition device which characterized in that includes: the system comprises a model construction module, a defect identification module and a position type identification module;
the model construction module is used for constructing an RCNN detection model and a YOLOv4 detection model, and the YOLOv4 detection model is constructed based on an adaptive feature fusion improved YOLOv4 network;
the defect identification module is used for acquiring a patrol image and patrol route information of the glass insulator on the power transmission line, and detecting the patrol image by adopting the YOLOv4 detection model to obtain a detection and identification result;
the position type recognition module is used for detecting the position and the type of the lightning flashover defect of the glass insulator according to the detection recognition result, detecting the inspection image according to the inspection route information and by adopting the RCNN detection model.
9. The glass insulator lightning strike discharge defect identification device of claim 8, wherein the model building module comprises a first model building module and a second model building module;
the first model construction module is used for acquiring first defect image data with lightning flashover defects, and performing training processing on the first defect image data by combining self-adaptive feature fusion and a PAN structure of a YOLOv4 network to obtain a YOLOv4 detection model;
the model construction module is used for acquiring second defect image data marked with defect types and position information, and training the second defect image data by adopting an RPN network of three cascade detection heads to obtain an RCNN detection model capable of identifying the defect positions and the defect types;
the first model building module is further used for processing the first defect image data through a PAN structure of a YOLOv4 network to obtain at least three feature diagrams with different scales; performing equal scaling treatment on each feature map by adopting self-adaptive feature fusion to obtain feature scaled maps with the same scale, and calculating the weight of the corresponding scale of the feature scaled maps by adopting a weight calculation formula; carrying out fusion calculation on weights corresponding to all the feature zoom diagrams through self-adaptive feature fusion by adopting a fusion calculation formula to obtain spatial weights of mapping fusion of the features of each scale;
the weight calculation formula is as follows:
Figure FDA0004153739700000041
Figure FDA0004153739700000042
Figure FDA0004153739700000043
the fusion calculation formula is as follows:
Figure FDA0004153739700000044
Figure FDA0004153739700000045
Figure FDA0004153739700000046
wherein alpha, beta and gamma are respectively the numerical values of weights, l is the size of a scale, i and j are respectively the numerical values of the transverse and longitudinal coordinates of the spatial position of the feature map, lambda is the convolution kernel calculation control coefficient in the YOLOv4 network,
Figure FDA0004153739700000047
the vector of the graph at spatial position (i, j) is scaled for the a-th feature.
10. A terminal device comprising a processor and a memory;
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the glass insulator lightning strike discharge defect identification method according to any one of claims 1-7 according to instructions in the program code.
CN202310327200.7A 2023-03-29 2023-03-29 Method, device and equipment for identifying lightning discharge defects of glass insulator Pending CN116338392A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310327200.7A CN116338392A (en) 2023-03-29 2023-03-29 Method, device and equipment for identifying lightning discharge defects of glass insulator

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310327200.7A CN116338392A (en) 2023-03-29 2023-03-29 Method, device and equipment for identifying lightning discharge defects of glass insulator

Publications (1)

Publication Number Publication Date
CN116338392A true CN116338392A (en) 2023-06-27

Family

ID=86875916

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310327200.7A Pending CN116338392A (en) 2023-03-29 2023-03-29 Method, device and equipment for identifying lightning discharge defects of glass insulator

Country Status (1)

Country Link
CN (1) CN116338392A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116797604A (en) * 2023-08-28 2023-09-22 中江立江电子有限公司 Glass insulator defect identification method, device, equipment and medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116797604A (en) * 2023-08-28 2023-09-22 中江立江电子有限公司 Glass insulator defect identification method, device, equipment and medium
CN116797604B (en) * 2023-08-28 2023-12-26 中江立江电子有限公司 Glass insulator defect identification method, device, equipment and medium

Similar Documents

Publication Publication Date Title
CN109086811B (en) Multi-label image classification method and device and electronic equipment
CN114332697A (en) Method, system, equipment and medium for detecting faults of multiple types of targets in power transmission line
CN112766110A (en) Training method of object defect recognition model, object defect recognition method and device
CN113313047B (en) Lane line detection method and system based on lane structure prior
CN116503399B (en) Insulator pollution flashover detection method based on YOLO-AFPS
CN116338392A (en) Method, device and equipment for identifying lightning discharge defects of glass insulator
CN113192646A (en) Target detection model construction method and different target distance monitoring method and device
CN110958467A (en) Video quality prediction method and device and electronic equipment
CN116229052A (en) Method for detecting state change of substation equipment based on twin network
CN115170988A (en) Power grid line fault identification method and system based on deep learning
CN113012107B (en) Power grid defect detection method and system
CN116805387B (en) Model training method, quality inspection method and related equipment based on knowledge distillation
CN113592839A (en) Distribution network line typical defect diagnosis method and system based on improved fast RCNN
CN111104965A (en) Vehicle target identification method and device
CN111667450A (en) Ship quantity counting method and device and electronic equipment
CN115931895A (en) Method and device for checking defects of power transmission line
CN113284103B (en) Substation equipment defect online detection method based on space transformation fast R-CNN model
CN115222691A (en) Image defect detection method, system and related device
CN113780278A (en) Method and device for identifying license plate content, electronic equipment and storage medium
CN113989632A (en) Bridge detection method and device for remote sensing image, electronic equipment and storage medium
CN113269678A (en) Fault point positioning method for contact network transmission line
CN112733681A (en) Image processing method, model training method, device, equipment and storage medium
CN117349734B (en) Water meter equipment identification method and device, electronic equipment and storage medium
CN116229419B (en) Pedestrian detection method and device
CN113808049B (en) Train line image distortion correction method and electronic equipment

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