CN113222949B - X-ray image automatic detection method for plugging position of power equipment conductor - Google Patents

X-ray image automatic detection method for plugging position of power equipment conductor Download PDF

Info

Publication number
CN113222949B
CN113222949B CN202110548257.0A CN202110548257A CN113222949B CN 113222949 B CN113222949 B CN 113222949B CN 202110548257 A CN202110548257 A CN 202110548257A CN 113222949 B CN113222949 B CN 113222949B
Authority
CN
China
Prior art keywords
image
conductor
ray
detection
loss
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110548257.0A
Other languages
Chinese (zh)
Other versions
CN113222949A (en
Inventor
周静波
刘荣海
郭新良
陈国坤
代克顺
杨迎春
许宏伟
郑欣
焦宗寒
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.)
Electric Power Research Institute of Yunnan Power Grid Co Ltd
Original Assignee
Electric Power Research Institute of Yunnan 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 Electric Power Research Institute of Yunnan Power Grid Co Ltd filed Critical Electric Power Research Institute of Yunnan Power Grid Co Ltd
Priority to CN202110548257.0A priority Critical patent/CN113222949B/en
Publication of CN113222949A publication Critical patent/CN113222949A/en
Application granted granted Critical
Publication of CN113222949B publication Critical patent/CN113222949B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N23/00—Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00
    • G01N23/02—Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material
    • G01N23/04—Investigating or analysing materials by the use of wave or particle radiation, e.g. X-rays or neutrons, not covered by groups G01N3/00 – G01N17/00, G01N21/00 or G01N22/00 by transmitting the radiation through the material and forming images of the material
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • G06T5/70—
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2223/00—Investigating materials by wave or particle radiation
    • G01N2223/10—Different kinds of radiation or particles
    • G01N2223/101—Different kinds of radiation or particles electromagnetic radiation
    • G01N2223/1016—X-ray
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2223/00—Investigating materials by wave or particle radiation
    • G01N2223/60—Specific applications or type of materials
    • G01N2223/646—Specific applications or type of materials flaws, defects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10116—X-ray image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Abstract

The application provides an automatic X-ray image detection method for a conductor plugging position of power equipment, which comprises the steps of X-ray imaging, image denoising and feature enhancement, image size format standardization, unsupervised data enhancement-GAN method, data annotation, RetinaNet target detection technology and detection result acquisition. The method has the advantages of high detection speed, high precision and high reliability, solves the problems of low X-ray image diagnosis efficiency, poor reliability and low intelligent level of X-ray automatic detection equipment in assembling and action positions of power closed equipment such as GIS, circuit breakers, wires and cables, and can quickly and accurately detect whether the plugging positions of conductors and contacts in the images are normal or not by using an intelligent recognition method for deep learning on the basis of realizing X-ray digital imaging of GIS, circuit breakers, power transmission cable crimping positions and the like, realize automatic detection of relative positions in the power equipment and improve the diagnosis efficiency and reliability of X-ray detection images.

Description

X-ray image automatic detection method for plugging position of power equipment conductor
Technical Field
The application relates to the field of image detection, in particular to an X-ray image automatic detection method for a conductor plugging position of power equipment.
Background
With the development of science and technology, smart grid construction is receiving much attention at present, and state detection and performance evaluation of power equipment develop towards refinement and intellectualization. Gas insulated metal totally-enclosed switch GIS equipment, tank circuit breaker, switch and transmission cable etc. closed equipment is the important pivot equipment of electric power system, undertakes the important task of transformer and transmission of electricity. For GIS equipment, its inner structure is complicated, difficult dismantlement, connecting pieces such as conductor, contact are sealed in metal casing, whether the effective contact depth of grafting position satisfies the requirement can't be monitored with the naked eye, other detection means also are difficult visual qualitative and quantitative analysis, if conductor grafting depth is not enough and the virtual contact, lead to the conductor high temperature, thereby make conductor surface oxidation aggravate contact resistance increase with higher speed, so circulate, lead to drawing the arc at last, even power equipment explodes and leads to the conflagration to cause the loss on a large scale. For the compression joint position of an overhead line and the compression joint positions of an aluminum compression joint pipe and a steel anchor, the bearing range of the connection of a lead is determined, the bearing capacity of the line is reduced due to the error of the compression joint position, and the line breaking and power failure accidents occur.
In order to be able to know exactly where the conductor is inserted in order to ensure a safe and stable operation of the device. The X-ray digital imaging technology is the most direct method for judging the internal condition of GIS equipment at present, and can detect the insertion depth of a conductor through perspective imaging and calculation. Compared with other judgment methods, the result of the method can provide powerful basis for the equipment to continue to operate safely when the position of the disconnecting link is at the critical value. Like the X-ray detection images in the medical industry need diagnosis by professional doctors, the complexity of the structure and the material of the power equipment has higher requirements on the technical level of detection and diagnosis personnel, the contradiction between the shortage of the professional technicians and the increasing work load of the X-ray detection is more and more prominent, and because the conditions of visual fatigue and the like are generated due to insufficient technical level of the workers or high labor intensity of the manual X-ray detection, errors can occur in the judgment of the inserting position of the internal connecting part of the equipment, so that unnecessary accidents are caused.
Therefore, an automatic detection method of the X-ray image of the electric equipment is provided, and the efficiency and the reliability of image diagnosis are improved. The application provides an automatic X-ray image detection method for the plugging position of a power equipment connecting piece based on a Retina Net target detection technology, which can accurately and quickly identify the plugging depth or relative position of a conductor, a contact and the connecting piece, improve the diagnosis efficiency and accuracy of an X-ray obtained detection result, realize intelligent diagnosis and digital application of an X-ray image of the power equipment and ensure safe and stable operation of the power equipment.
Disclosure of Invention
The application provides an automatic X-ray image detection method for a conductor plugging position of power equipment, which aims to solve the problems that the complexity of the structure and the material of the power equipment has higher requirements on the technical level of detection and diagnosis personnel, the shortage of professional technicians and the increasing contradiction of the X-ray detection workload are more and more prominent, visual fatigue and the like are generated due to insufficient technical level of the workers or high labor intensity of manual X-ray detection, errors can possibly occur in the judgment of the plugging position of an internal connecting piece of the equipment, and unnecessary accidents are caused.
The application provides an X-ray image automatic detection method of a conductor plugging position of power equipment, which comprises the following steps:
x-ray imaging, namely acquiring an X-ray picture of the insertion or assembly position of a GIS device conductor by using an X-ray digital imaging technology;
denoising and enhancing the characteristics of the image, denoising the original image of the inserting position of the X-ray conductor by a filtering algorithm in a linear denoising method, and clearly showing the specific position and characteristics of the spring;
standardizing the size and format of the image, and unifying the size and format of the denoised image according to the requirements of the convolutional neural network on the size and format of the input image;
an unsupervised data enhancement-GAN method for generating a training data image by performing X-ray sample image enhancement on a standard image based on a generated countermeasure network GAN method;
marking data, namely marking a training data image through a marking tool;
the RetinaNet target detection technology is used for automatically detecting the position of an X-ray conductor inserting frame of a training data image based on the RetinaNet target detection technology;
and obtaining a detection result, dividing the marked training data image into a training set, a verification set and a test set according to a proportion, sending the training set into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining the features to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain a specific inserting position of the conductor and giving a corresponding category.
Optionally, the image denoising and feature enhancing step includes:
filtering an original image at the inserting position of an X-ray conductor by a 3 multiplied by 3 boundary enhanced convolution kernel, a sharpening convolution kernel and a smooth convolution kernel, performing convolution traversal on pixels of the original image, performing convolution operation on a pixel matrix which is intercepted and has the same size as the convolution kernel, and performing nonlinear operation on a gray value of an input image to enable the gray value of the output image and the gray value of the input image to be in an exponential relationship, wherein the formula is as follows:
Figure BDA0003074427770000021
the index in the formula is gamma, where V in The value range of (a) is 0 to 1, so normalization operation is required and then an index is taken.
Optionally, the step of standardizing the image size format includes:
the size of the denoised image is set to be 256 multiplied by 256, and the format is unified into a jpg format.
Optionally, the step of the unsupervised data enhancement-GAN method includes:
adopting a GAN method based on a generated countermeasure network to enhance an X-ray sample image, wherein the generated countermeasure network consists of a generator G and a discriminator D, the generator continuously trains random noise input into the generator to enable the random noise to be fitted with the spatial distribution of an original image data set, samples similar to the original image are generated from the beginning to the end and the discriminator is confused, the output of the discriminator for a forged sample input into the discriminator is close to 0, and the output for a real sample is close to 1; the input of a generation model G of the countermeasure network is a random vector z in a two-dimensional Gaussian model, the output of the generation model is a synthesized image G (z) of a forged conductor inserting position, a real image x of the conductor inserting position in an original image data set is obtained through indexing, the synthesized image G (z) of the conductor inserting position and the real image x are synchronously transmitted to a discrimination model D, the discrimination model gives a discrimination result, so that the quality of a picture generated by the generation model is improved, and an optimization objective function is as follows:
Figure BDA0003074427770000031
optionally, the step of data annotation includes:
after data expansion, image data are marked by a marking tool, the image data inserted into one spring position are marked as 'one _ abrormal', the image data inserted into two spring positions are marked as 'two _ abrormal', the image data inserted into three spring positions are marked as 'three _ normal', the image and the label are placed in the same folder, the inserted label is stored in one json file, and the file content comprises a folder name, an image name, a file path, an image size, a category name of an object and frame coordinates.
Optionally, the steps of the RetinaNet target detection technology include:
the depth network model built by the method for automatically detecting the X-ray image of the conductor plugging position of the power equipment based on the RetinaNet target detection technology comprises a backbone network ResNet + FCN and 2 sub-networks with specific tasks, and P is obtained by extracting input images through ResNet characteristics 3 ～P 7 A feature map pyramid, wherein subscript l represents the number of layers of the feature pyramid, and 256 channels are obtained in each layer of the obtained feature pyramid; with the reduction of spatial resolution caused by deep convolution and the loss of spatial information, high-level semantic information is detected, a layer with high resolution and rich semantics is constructed, but as up-and-down sampling is carried out continuously and the target position is changed, transverse connection, namely, concatenate operation, is constructed between the reconstructed layer and the corresponding feature map;
the loss function of RetinaNet is divided into 3 terms, the first term is the frame regression loss, the second term is the confidence coefficient loss, the third term is the classification loss, wherein:
A. frame regression loss:
Figure BDA0003074427770000032
CIOU loss is an improved IOU loss calculation method, wherein L CIOU Represents the bounding Box regression loss, IOU (b) t ,b p ) Representing the real box b t And a prediction block b p Cross-over ratio between R CIOU Is a penalty term which, in the first term,
Figure BDA0003074427770000033
representing the center point of a real box
Figure BDA0003074427770000034
And the center point of the prediction frame
Figure BDA0003074427770000035
C represents the real box b t And a prediction block b p The diagonal length of the smallest enclosing frame; in the second item, alpha is a positive number, v is used for measuring the consistency of the length-width ratio, the function of alpha v is to control the width and height of the prediction frame to be close to the width and height of the real frame, and w is t 、h t Width and height, w, of the real box p 、h p Representing the width and height of the prediction box;
B. confidence loss:
Figure BDA0003074427770000041
the confidence loss measures whether a target exists in a prediction frame, and two-classification cross entropy loss is adopted;
in the above formula, the first and second carbon atoms are,
Figure BDA0003074427770000042
labels and predicted values representing confidence scores, respectively;
Figure BDA0003074427770000043
and
Figure BDA0003074427770000044
the value of (A) is divided into 2 cases:
②
Figure BDA0003074427770000045
indicating that there is a target in grid i;
②
Figure BDA0003074427770000046
indicating that there is no target in grid i;
C. class loss:
Figure BDA0003074427770000047
because the inserting position of the conductor in the X-ray image is small in the whole picture, and the candidate area containing the conductor is less than the candidate area not containing the conductor, the problem is solved by using a classification Loss function Focal local to replace the original cross entropy Loss (CE), and if the model predicts that a target exists in a certain grid, the model predicts that the target exists in the grid
Figure BDA0003074427770000048
Calculated according to the above formula, wherein
Figure BDA0003074427770000049
And
Figure BDA00030744277700000410
and the classification probabilities of the positive samples and the negative samples are basically consistent in the initial training stage, so that easy example cannot be inhibited, the bias of the last layer of convolution is slightly changed, the last layer of convolution is initialized to a special value b-log ((1-pi)/pi), and pi is 0.01, so that the classification probability of the positive samples in the initial training stage is improved.
Optionally, the step of obtaining the detection result includes:
dividing the marked conductor splicing position image into a training set, a verification set and a test set according to the proportion of 7:2:1, wherein the training set is sent into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining in an FPN (feature pyramid network) to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain the specific conductor splicing position and give a corresponding class, wherein the anchor point ratio of each layer in the FPN is {1:2,1:1,2:1 }; in the training process, the optimizer selects an adam (adaptive motion estimation) algorithm, the learning rate is 0.001 before 10000 iterations, then 0.0001, the total iteration number is 20000, a network outputs a plurality of candidate frames, and the candidate frames are screened according to the NMS algorithm and then output as final detection frames; in order to evaluate the performance of the model, an image of the plugging position of the conductor for testing is input, and the specific plugging position of the conductor and the specific type of the defect in the image are judged through the trained image detection model.
According to the technical scheme, the method for automatically detecting the X-ray image of the insertion position of the conductor of the power equipment comprises the steps of X-ray imaging, wherein an X-ray picture of the insertion or assembly position of the conductor of the GIS equipment is obtained by utilizing an X-ray digital imaging technology; denoising and enhancing the characteristics of the image, denoising the original image of the inserting position of the X-ray conductor by a filtering algorithm in a linear denoising method, and clearly showing the specific position and characteristics of the spring; standardizing the size and format of the image, and unifying the size and format of the denoised image according to the requirements of the convolutional neural network on the size and format of the input image; an unsupervised data enhancement-GAN method for generating a training data image by performing X-ray sample image enhancement on a standard image based on a generated countermeasure network GAN method; marking data, namely marking a training data image through a marking tool; the RetinaNet target detection technology is used for automatically detecting the position of an X-ray conductor inserting frame of a training data image based on the RetinaNet target detection technology; and obtaining a detection result, dividing the marked training data image into a training set, a verification set and a test set according to a proportion, sending the training set into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining the features to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain a specific inserting position of the conductor and giving a corresponding category.
The X-ray image automatic detection method for the conductor plugging position of the power equipment has the advantages of being high in detection speed, high in precision and high in reliability, the problems that X-ray image diagnosis efficiency is low, reliability is poor and the intelligent level of X-ray automatic detection equipment is low in assembling and action positions of power closed equipment such as GIS, a circuit breaker, a wire and a cable are low are solved, on the basis of achieving X-ray digital imaging of GIS, the circuit breaker, a crimping position of a power transmission cable and the like, whether the plugging position of the conductor and a contact in the image is normal or not is rapidly and accurately detected by means of an intelligent recognition method for deep learning, automatic detection of the internal relative position of the power equipment is achieved, diagnosis efficiency and reliability of an X-ray detection image are improved, and the intelligent level of automatic detection equipment such as a robot is improved.
Drawings
In order to more clearly explain the technical solution of the present application, the drawings needed to be passed through in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
Fig. 1 is a schematic overall flow chart of an automatic X-ray image detection method for a conductor plugging position of an electrical device according to the present disclosure;
fig. 2 is a schematic flow chart illustrating image preprocessing in an automatic X-ray image detection method for the plugging position of a conductor of an electrical device according to the present disclosure;
fig. 3 is a schematic diagram illustrating a process of enhancing data of a conductor plugging position based on a generation countermeasure network in an automatic X-ray image detection method of a conductor plugging position of an electrical device according to the present application;
fig. 4 is a schematic diagram of a conductor plugging position detection method based on a RetinaNet target detection method in an X-ray image automatic detection method for a conductor plugging position of an electrical device according to the present application;
fig. 5 is a schematic diagram of a test sample of a conductor plugging position in an X-ray image automatic detection method for a conductor plugging position of an electrical device according to the present application.
Detailed Description
Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following examples do not represent all embodiments consistent with the present application. But merely as exemplifications of systems and methods consistent with certain aspects of the application, as recited in the claims.
In view of the reason that the number of X-ray image samples of the power equipment is small, the method provides that the countermeasure network GAN is generated based on the depth to expand the defect samples, so that the problem of insufficient sample size is solved, and the depth network model RetinaNet is responsible for predicting the accurate position of conductor plugging and judging whether plugging is normal. Firstly, acquiring an X-ray picture by utilizing an X-ray digital imaging technology at the insertion or assembly position of a GIS device conductor, and then preprocessing the shot original picture; carrying out data expansion on the preprocessed picture by using a traditional data enhancement mode, and further expanding the data by using a generated countermeasure network; and marking the inserting positions of the conductors in the expanded image, training a detection network by using marked data, and evaluating the performance of the trained model by using new test data.
Referring to fig. 1, an overall flow chart of an automatic X-ray image detection method for a conductor plugging position of an electrical device provided by the present application is shown, including:
x-ray imaging, namely acquiring an X-ray picture of the insertion or assembly position of a GIS device conductor by using an X-ray digital imaging technology;
denoising and enhancing the characteristics of the image, denoising the original image of the inserting position of the X-ray conductor by a filtering algorithm in a linear denoising method, and clearly showing the specific position and characteristics of the spring;
standardizing the size and format of the image, and unifying the size and format of the denoised image according to the requirements of the convolutional neural network on the size and format of the input image;
an unsupervised data enhancement-GAN method for generating a training data image by performing X-ray sample image enhancement on a standard image based on a generated countermeasure network GAN method;
marking data, namely marking a training data image through a marking tool;
the RetinaNet target detection technology is used for automatically detecting the position of an X-ray conductor inserting frame of a training data image based on the RetinaNet target detection technology;
and obtaining a detection result, dividing the marked training data image into a training set, a verification set and a test set according to a proportion, sending the training set into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining the features to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain a specific inserting position of the conductor and giving a corresponding category.
In practical application, the image denoising and feature enhancing step includes:
filtering an original image at the inserting position of an X-ray conductor by a 3 multiplied by 3 boundary enhanced convolution kernel, a sharpening convolution kernel and a smooth convolution kernel, performing convolution traversal on pixels of the original image, performing convolution operation on a pixel matrix which is intercepted and has the same size as the convolution kernel, and performing nonlinear operation on a gray value of an input image to enable the gray value of the output image and the gray value of the input image to be in an exponential relationship, wherein the formula is as follows:
Figure BDA0003074427770000071
the index in the formula is gamma, where V in The value range of (a) is 0 to 1, so normalization operation is required and then an index is taken.
Because the characteristics of the spring in the conductor are influenced by surrounding noise, the shielding case and the spring in the shielding case cannot be clearly distinguished, and the detection result is influenced and obtained, the original image needs to be denoised by a filtering algorithm in a linear denoising method, and the specific position and the characteristics of the spring can be clearly shown.
Referring to fig. 2, in practical application, the step of standardizing the image size format includes:
the size of the denoised image is set to be 256 multiplied by 256, and the format is unified into a jpg format.
Since the convolutional neural network has certain requirements on the size and format of the input image, the image size and format of the conductor plug position need to be unified. After the preprocessing, the size and the image quality of the image data set at the plugging position of the X-ray conductor already meet the requirements of model training, but the requirement of deep learning on the number of images in the data set is large, and the number of the image data sets collected at present is only less than 50, is far away from the data amount required by the model training, and is not enough to meet the requirements of the model training. Therefore, the data are subjected to various transformations in aspects of shooting angle, shooting distance, contrast transformation, rotation, translation, clipping, overturning, brightness transformation and the like, and data enhancement of the traditional method is carried out to form more training data. Because the amount of data generated by the traditional method is limited, the method based on the generation countermeasure network GAN is adopted to enhance the X-ray sample image so as to generate more images of conductor plugging positions which are close to the real data distribution.
Referring to fig. 3, in an actual application, a schematic diagram of a conductor plugging position data enhancement process based on a generation countermeasure network in an X-ray image automatic detection method for a conductor plugging position of an electrical device provided by the present application is shown, where the unsupervised data enhancement-GAN method includes the steps of:
adopting a GAN method based on a generated countermeasure network to enhance an X-ray sample image, wherein the generated countermeasure network consists of a generator G and a discriminator D, the generator continuously trains random noise input into the generator to enable the random noise to be fitted with the spatial distribution of an original image data set, samples similar to the original image are generated from the beginning to the end and the discriminator is confused, the output of the discriminator for a forged sample input into the discriminator is close to 0, and the output for a real sample is close to 1; the input of a generation model G of the countermeasure network is a random vector z in a two-dimensional Gaussian model, the output of the generation model is a synthesized image G (z) of a forged conductor inserting position, a real image x of the conductor inserting position in an original image data set is obtained through indexing, the synthesized image G (z) of the conductor inserting position and the real image x are synchronously transmitted to a discrimination model D, the discrimination model gives a discrimination result, so that the quality of a picture generated by the generation model is improved, and an optimization objective function is as follows:
Figure BDA0003074427770000072
the discriminator essentially belongs to a classifier, and the main task of the discriminator is to distinguish real samples from fake samples.
In practical applications, the step of data annotation includes:
after data expansion, image data are marked by a marking tool, the image data inserted into one spring position are marked as 'one _ abrormal', the image data inserted into two spring positions are marked as 'two _ abrormal', the image data inserted into three spring positions are marked as 'three _ normal', the image and the label are placed in the same folder, the inserted label is stored in one json file, and the file content comprises a folder name, an image name, a file path, an image size, a category name of an object and frame coordinates.
Referring to fig. 4, in an X-ray image automatic detection method for a conductor plugging position of an electrical device according to the present invention, a schematic diagram of a conductor plugging position detection method based on a RetinaNet target detection method is shown, and in practical applications, the steps of the RetinaNet target detection technique include:
the depth network model built by the method for automatically detecting the X-ray image of the conductor plugging position of the power equipment based on the RetinaNet target detection technology comprises a backbone network ResNet + FCN and 2 sub-networks with specific tasks, and P is obtained by extracting input images through ResNet characteristics 3 ～P 7 A feature map pyramid, wherein subscript l represents the number of layers of the feature pyramid, and 256 channels are obtained in each layer of the obtained feature pyramid; with the reduction of spatial resolution caused by deep convolution and the loss of spatial information, high-level semantic information is detected, a layer with high resolution and rich semantics is constructed, but as up-and-down sampling is carried out continuously and the target position is changed, transverse connection, namely, concatenate operation, is constructed between the reconstructed layer and the corresponding feature map;
the loss function of RetinaNet is divided into 3 terms, the first term is the frame regression loss, the second term is the confidence coefficient loss, the third term is the classification loss, wherein:
A. frame regression loss:
Figure BDA0003074427770000081
CIOU lossLosers are an improved IOU loss calculation method, where L CIOU Represents the bounding Box regression loss, IOU (b) t ,b p ) Representing the real box b t And a prediction block b p Cross-over ratio between R CIOU Is a penalty term which, in the first term,
Figure BDA0003074427770000082
representing the center point of a real box
Figure BDA0003074427770000083
And the center point of the prediction frame
Figure BDA0003074427770000084
C represents the real box b t And a prediction block b p The diagonal length of the smallest enclosing frame; in the second item, alpha is a positive number, v is used for measuring the consistency of the length-width ratio, the function of alpha v is to control the width and height of the prediction frame to be close to the width and height of the real frame, and w is t 、h t Width and height, w, of the real box p 、h p Representing the width and height of the prediction box;
B. confidence loss:
Figure BDA0003074427770000091
the confidence loss measures whether a target exists in a prediction frame, and two-classification cross entropy loss is adopted;
in the above formula, the first and second carbon atoms are,
Figure BDA0003074427770000092
labels and predicted values representing confidence scores, respectively;
Figure BDA0003074427770000093
and
Figure BDA0003074427770000094
the value of (A) is divided into 2 cases:
③
Figure BDA0003074427770000095
indicating that there is a target in grid i;
②
Figure BDA0003074427770000096
indicating that there is no target in grid i;
C. class loss:
Figure BDA0003074427770000097
because the inserting position of the conductor in the X-ray image is small in the whole picture, and the candidate area containing the conductor is less than the candidate area not containing the conductor, the problem is solved by using a classification Loss function Focal local to replace the original cross entropy Loss (CE), and if the model predicts that a target exists in a certain grid, the model predicts that the target exists in the grid
Figure BDA0003074427770000098
Calculated according to the above formula, wherein
Figure BDA0003074427770000099
And
Figure BDA00030744277700000910
and the classification probabilities of the positive samples and the negative samples are basically consistent in the initial training stage, so that easy example cannot be inhibited, the bias of the last layer of convolution is slightly changed, the last layer of convolution is initialized to a special value b-log ((1-pi)/pi), and pi is 0.01, so that the classification probability of the positive samples in the initial training stage is improved.
One of the most important loops in the deep learning target detection method is the feature extraction network, which is generally called backbone network in the target detection framework. Many classical convolutional neural networks, such as VGG, inclusion, DenseNet, ResNet, and other network structures, are available for selection in different application fields. In addition, many path aggregation networks such as FPN, PAN, BiFPN, ASFF, etc. tend to assist in enhancing the feature extraction capability of convolutional neural networks.
The detection network mainly comprises: the method comprises the steps of extracting features by using a depth residual error network, fusing high-level semantic information and low-level detail information by using a feature pyramid network, classifying and coordinate regression of extracted large conductor splicing positions shot in a short distance and small conductor splicing position features shot in a long distance according to the conductor splicing position features of each scale, and therefore accurate conductor insertion positions can be accurately detected and specific types of defects can be given.
Referring to fig. 5, in practical application, a schematic diagram of a test sample of a conductor plugging position in an X-ray image automatic detection method for a conductor plugging position of an electrical device provided by the present application includes:
dividing the marked conductor splicing position image into a training set, a verification set and a test set according to the proportion of 7:2:1, wherein the training set is sent into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining in an FPN (feature pyramid network) to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain the specific conductor splicing position and give a corresponding class, wherein the anchor point ratio of each layer in the FPN is {1:2,1:1,2:1 }; in the training process, the optimizer selects an adam (adaptive motion estimation) algorithm, the learning rate is 0.001 before 10000 iterations, then 0.0001, the total iteration number is 20000, a network outputs a plurality of candidate frames, and the candidate frames are screened according to the NMS algorithm and then output as final detection frames; in order to evaluate the performance of the model, an image of the plugging position of the conductor for testing is input, and the specific plugging position of the conductor and the specific type of the defect in the image are judged through the trained image detection model.
The X-ray image automatic detection method for the conductor plugging position of the power equipment provided by the application breaks the defect of the traditional power system inspection mode, a large number of professionals are not needed to find the defect position in a picture, the influence of the conductor plugging position detection caused by the type imbalance caused by the defect sample shortage of the conductor plugging position in the actual situation is considered, and the Focal local is provided to replace the original cross entropy Loss. The target detection algorithm puts more attention on samples with difficulty and wrong classification in the training process, and the method is superior to the traditional detection method in both detection real-time property and detection precision and has the following advantages:
(1) the detection efficiency is high. The method adopts a target detection algorithm based on regression to directly output and obtain a detection result by adopting a single convolutional neural network, so that end-to-end detection is realized, and the detection efficiency is improved.
(2) The recall rate is high. In the method, the FPN characteristic pyramid network is added in the multi-scale characteristic extraction, and the low-level detail information and the high-level semantic information are fused, so that the recall rate of the small target object is improved.
(3) The accuracy is high. Considering that the target proportion is usually smaller than the background proportion in the picture, so the proportion of the negative samples in the samples is far larger than that of the positive samples, which results in too large loss value and submerges the loss of the positive samples, the Focal loss is proposed to reduce the contribution of the negative samples to the loss value and improve the detection accuracy of the target.
The method for automatically detecting the X-ray image of the insertion position of the conductor of the power equipment comprises the following steps of X-ray imaging, wherein an X-ray picture of the insertion or assembly position of the conductor of the GIS equipment is obtained by utilizing an X-ray digital imaging technology; denoising and enhancing the characteristics of the image, denoising the original image of the inserting position of the X-ray conductor by a filtering algorithm in a linear denoising method, and clearly showing the specific position and characteristics of the spring; standardizing the size and format of the image, and unifying the size and format of the denoised image according to the requirements of the convolutional neural network on the size and format of the input image; an unsupervised data enhancement-GAN method for generating a training data image by performing X-ray sample image enhancement on a standard image based on a generated countermeasure network GAN method; marking data, namely marking a training data image through a marking tool; the RetinaNet target detection technology is used for automatically detecting the position of an X-ray conductor inserting frame of a training data image based on the RetinaNet target detection technology; and obtaining a detection result, dividing the marked training data image into a training set, a verification set and a test set according to a proportion, sending the training set into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining the features to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain a specific inserting position of the conductor and giving a corresponding category. The method has the advantages of high detection speed, high precision and high reliability, solves the problems of low X-ray image diagnosis efficiency, poor reliability and low intelligent level of X-ray automatic detection equipment in assembling and action positions of power closed equipment such as GIS, a circuit breaker, a wire and a cable, and on the basis of realizing X-ray digital imaging of GIS, a circuit breaker, a crimping position of a power transmission cable and the like, quickly and accurately detects whether the plugging positions of a conductor and a contact in an image are normal by using an intelligent recognition method for deep learning, realizes automatic detection of relative positions in the power equipment, improves the diagnosis efficiency and reliability of X-ray detection images, and improves the intelligent level of automatic detection equipment such as robots.
While there have been shown and described what are at present considered the fundamental principles and essential features of the application, and advantages thereof, it will be apparent to those skilled in the art that the application is not limited to the details of the foregoing exemplary embodiments, but is capable of other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.
The embodiments provided in the present application are only a few examples of the general concept of the present application, and do not limit the scope of the present application. Any other embodiments extended according to the scheme of the present application without inventive efforts will be within the scope of protection of the present application for a person skilled in the art.

Claims (7)

1. An X-ray image automatic detection method for a conductor plugging position of electrical equipment is characterized by comprising the following steps:
x-ray imaging, namely acquiring an X-ray picture of the insertion or assembly position of a GIS device conductor by using an X-ray digital imaging technology;
denoising and enhancing the characteristics of the image, denoising the original image of the inserting position of the X-ray conductor by a filtering algorithm in a linear denoising method, and clearly showing the specific position and characteristics of the spring;
standardizing the size and format of the image, and unifying the size and format of the denoised image according to the requirements of the convolutional neural network on the size and format of the input image;
an unsupervised data enhancement-GAN method for generating a training data image by performing X-ray sample image enhancement on a standard image based on a generated countermeasure network GAN method;
marking data, namely marking a training data image through a marking tool;
the RetinaNet target detection technology is used for automatically detecting the position of an X-ray conductor inserting frame of a training data image based on the RetinaNet target detection technology;
and obtaining a detection result, dividing the marked training data image into a training set, a verification set and a test set according to a proportion, sending the training set into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining the features to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain a specific inserting position of the conductor and giving a corresponding category.
2. The method for automatically detecting the X-ray image of the plugging position of the conductor of the power equipment as claimed in claim 1, wherein the image denoising and feature enhancing steps comprise:
filtering an original image at the inserting position of an X-ray conductor by a 3 multiplied by 3 boundary enhanced convolution kernel, a sharpening convolution kernel and a smooth convolution kernel, performing convolution traversal on pixels of the original image, performing convolution operation on a pixel matrix which is intercepted and has the same size as the convolution kernel, and performing nonlinear operation on a gray value of an input image to enable the gray value of the output image and the gray value of the input image to be in an exponential relationship, wherein the formula is as follows:
Figure FDA0003074427760000011
the index in the formula is gamma, wherein V in The value range of (a) is 0 to 1, so normalization operation is required and then an index is taken.
3. The method according to claim 1, wherein the step of standardizing the image size format comprises:
the size of the denoised image is set to be 256 multiplied by 256, and the format is unified into a jpg format.
4. The method according to claim 1, wherein the step of the unsupervised data enhancement-GAN method comprises:
adopting a GAN method based on a generated countermeasure network to enhance an X-ray sample image, wherein the generated countermeasure network consists of a generator G and a discriminator D, the generator continuously trains random noise input into the generator to enable the random noise to be fitted with the spatial distribution of an original image data set, samples similar to the original image are generated from the beginning to the end and the discriminator is confused, the output of the discriminator for a forged sample input into the discriminator is close to 0, and the output for a real sample is close to 1; the input of a generation model G of the countermeasure network is a random vector z in a two-dimensional Gaussian model, the output of the generation model is a synthesized image G (z) of a forged conductor inserting position, a real image x of the conductor inserting position in an original image data set is obtained through indexing, the synthesized image G (z) of the conductor inserting position and the real image x are synchronously transmitted to a discrimination model D, the discrimination model gives a discrimination result, so that the quality of a picture generated by the generation model is improved, and an optimization objective function is as follows:
Figure FDA0003074427760000021
5. the method for automatically detecting the plugging position of the conductor of the power equipment according to the claim 1, wherein the step of data labeling comprises the following steps:
after data expansion, image data are marked by a marking tool, the image data inserted into one spring position are marked as 'one _ abrormal', the image data inserted into two spring positions are marked as 'two _ abrormal', the image data inserted into three spring positions are marked as 'three _ normal', the image and the label are placed in the same folder, the inserted label is stored in one json file, and the file content comprises a folder name, an image name, a file path, an image size, a category name of an object and frame coordinates.
6. The method for automatically detecting the plugging position of the conductor of the power equipment according to the X-ray image of claim 1, wherein the RetinaNet target detection technology comprises the following steps:
the depth network model built by the method for automatically detecting the X-ray image of the conductor plugging position of the power equipment based on the RetinaNet target detection technology comprises a backbone network ResNet + FCN and 2 sub-networks with specific tasks, and P is obtained by extracting input images through ResNet characteristics 3 ～P 7 A feature map pyramid, wherein subscript l represents the number of layers of the feature pyramid, and 256 channels are obtained in each layer of the obtained feature pyramid; with the reduction of spatial resolution caused by deep convolution and the loss of spatial information, high-level semantic information is detected, a layer with high resolution and rich semantics is constructed, but as up-and-down sampling is carried out continuously and the target position is changed, transverse connection, namely, concatenate operation, is constructed between the reconstructed layer and the corresponding feature map;
the loss function of RetinaNet is divided into 3 terms, the first term is the frame regression loss, the second term is the confidence coefficient loss, the third term is the classification loss, wherein:
A. frame regression loss:
Figure FDA0003074427760000022
CIOU loss is an improved IOU loss calculation method, wherein L CIOU Represents the bounding Box regression loss, IOU (b) t ,b p ) Representing the real box b t And a prediction block b p Cross-over ratio between R CIOU Is a penalty term which, in the first term,
Figure FDA0003074427760000023
representing the center point of a real box
Figure FDA0003074427760000024
And the center point of the prediction frame
Figure FDA0003074427760000025
C represents the real box b t And a prediction block b p The diagonal length of the smallest enclosing frame; in the second item, alpha is a positive number, v is used for measuring the consistency of the length-width ratio, the function of alpha v is to control the width and height of the prediction frame to be close to the width and height of the real frame, and w is t 、h t Width and height, w, of the real box p 、h p Representing the width and height of the prediction box;
B. confidence loss:
Figure FDA0003074427760000031
the confidence loss measures whether a target exists in a prediction frame, and two-classification cross entropy loss is adopted;
in the above formula, the first and second carbon atoms are,
Figure FDA0003074427760000032
labels and predicted values representing confidence scores, respectively;
Figure FDA0003074427760000033
and
Figure FDA0003074427760000034
the value of (2) is divided into 2 cases:
①
Figure FDA0003074427760000035
indicating that there is a target in grid i;
②
Figure FDA0003074427760000036
indicating that there is no target in grid i;
C. class loss:
Figure FDA0003074427760000037
because the inserting position of the conductor in the X-ray image is small in the whole picture, and the candidate area containing the conductor is less than the candidate area not containing the conductor, the problem is solved by using a classification Loss function Focal local to replace the original cross entropy Loss (CE), and if the model predicts that a target exists in a certain grid, the model predicts that the target exists in the grid
Figure FDA0003074427760000038
Calculated according to the above formula, wherein
Figure FDA0003074427760000039
And
Figure FDA00030744277600000310
respectively representing the class label and the prediction probability of the current target, and carrying out the final layer of convolution on the convolutional layer because the classification probabilities of the positive sample and the negative sample are basically consistent in the initial training stage and the easy example cannot be inhibitedThe bias of (2) is slightly changed, and the bias is initialized to a special value b-log ((1-pi)/pi), wherein pi is 0.01, so that the classification probability of the positive samples in the initial training stage is improved.
7. The method according to claim 1, wherein the step of obtaining the detection result comprises:
dividing the marked conductor splicing position image into a training set, a verification set and a test set according to the proportion of 7:2:1, wherein the training set is sent into a detection network model for training, preliminarily extracting features through a deep network ResNet, recombining in an FPN (feature pyramid network) to generate feature maps with different scales, sending the feature maps into a classification and regression sub-network to obtain the specific conductor splicing position and give a corresponding class, wherein the anchor point ratio of each layer in the FPN is {1:2,1:1,2:1 }; in the training process, the optimizer selects an adam (adaptive motion estimation) algorithm, the learning rate is 0.001 before 10000 iterations, then 0.0001, the total iteration number is 20000, a network outputs a plurality of candidate frames, and the candidate frames are screened according to the NMS algorithm and then output as final detection frames; in order to evaluate the performance of the model, an image of the plugging position of the conductor for testing is input, and the specific plugging position of the conductor and the specific type of the defect in the image are judged through the trained image detection model.
CN202110548257.0A 2021-05-19 2021-05-19 X-ray image automatic detection method for plugging position of power equipment conductor Active CN113222949B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110548257.0A CN113222949B (en) 2021-05-19 2021-05-19 X-ray image automatic detection method for plugging position of power equipment conductor

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110548257.0A CN113222949B (en) 2021-05-19 2021-05-19 X-ray image automatic detection method for plugging position of power equipment conductor

Publications (2)

Publication Number Publication Date
CN113222949A CN113222949A (en) 2021-08-06
CN113222949B true CN113222949B (en) 2022-08-02

Family

ID=77093498

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110548257.0A Active CN113222949B (en) 2021-05-19 2021-05-19 X-ray image automatic detection method for plugging position of power equipment conductor

Country Status (1)

Country Link
CN (1) CN113222949B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114332084B (en) * 2022-03-11 2022-09-16 齐鲁工业大学 PCB surface defect detection method based on deep learning
CN114677568B (en) * 2022-05-30 2022-08-23 山东极视角科技有限公司 Linear target detection method, module and system based on neural network
CN115115895A (en) * 2022-07-28 2022-09-27 吉林大学 Explosive mobile phone X-ray image classification method based on attention mechanism

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105891215A (en) * 2016-03-31 2016-08-24 浙江工业大学 Welding visual detection method and device based on convolutional neural network
US10304191B1 (en) * 2016-10-11 2019-05-28 Zoox, Inc. Three dimensional bounding box estimation from two dimensional images
CN110705374A (en) * 2019-09-11 2020-01-17 武汉敏捷数云信息技术有限公司 Transmission line defect identification method based on improved RetinaNet network
CN112711981A (en) * 2020-12-02 2021-04-27 国网陕西省电力公司检修公司 High-precision transmission line defect identification method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105891215A (en) * 2016-03-31 2016-08-24 浙江工业大学 Welding visual detection method and device based on convolutional neural network
US10304191B1 (en) * 2016-10-11 2019-05-28 Zoox, Inc. Three dimensional bounding box estimation from two dimensional images
CN110705374A (en) * 2019-09-11 2020-01-17 武汉敏捷数云信息技术有限公司 Transmission line defect identification method based on improved RetinaNet network
CN112711981A (en) * 2020-12-02 2021-04-27 国网陕西省电力公司检修公司 High-precision transmission line defect identification method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Yona Falinie A.Gaus等."Evaluation of a Dual Convolutional Neural Network Architecture for Object-wise Anomaly Detection in Cluttered X-ray Security Imagery".《2019 International Joint Conference on Neural Networks(IJCNN)》.2019, *
刘荣海等."X射线影像识别技术在GIS缺陷诊断中的作用".《高压电器》.2019, *
虞骅等."基于深度学习的电力设备状态检测的研究".《信息通信技术》.2021, *

Also Published As

Publication number Publication date
CN113222949A (en) 2021-08-06

Similar Documents

Publication Publication Date Title
CN113222949B (en) X-ray image automatic detection method for plugging position of power equipment conductor
Wang et al. Automatic fault diagnosis of infrared insulator images based on image instance segmentation and temperature analysis
CN112200178B (en) Transformer substation insulator infrared image detection method based on artificial intelligence
CN112184654A (en) High-voltage line insulator defect detection method based on generation countermeasure network
CN108664997A (en) High iron catenary equipotential line defective mode detection method based on cascade Faster R-CNN
CN111027631B (en) X-ray image classification and identification method for judging crimping defects of high-voltage strain clamp
CN112069930A (en) Vibration signal processing method and device for improving GIS equipment fault diagnosis accuracy
CN113837991A (en) Image processing method, device, equipment and storage medium
CN111738156A (en) Intelligent inspection management method and system for state of high-voltage switchgear
CN112508019A (en) GIS isolation/grounding switch state detection method and system based on image recognition
CN113344852A (en) Target detection method and device for power scene general-purpose article and storage medium
CN116883801A (en) YOLOv8 target detection method based on attention mechanism and multi-scale feature fusion
CN115239646A (en) Defect detection method and device for power transmission line, electronic equipment and storage medium
CN113723189A (en) Intelligent power equipment fault diagnosis method based on single-order infrared image target detection
CN115546664A (en) Cascaded network-based insulator self-explosion detection method and system
CN116503318A (en) Aerial insulator multi-defect detection method, system and equipment integrating CAT-BiFPN and attention mechanism
CN111507398A (en) Transformer substation metal instrument corrosion identification method based on target detection
CN111401437A (en) Deep learning-based power transmission channel hidden danger early warning grade analysis method
CN116580285B (en) Railway insulator night target identification and detection method
CN110992306B (en) Method and device for segmenting deteriorated insulator in infrared image based on deep learning
CN112734637A (en) Thermal infrared image processing method and system for monitoring temperature of lead
CN109919936B (en) Method, device and equipment for analyzing running state of composite insulator
CN111027569A (en) Degraded insulator identification and diagnosis method and device based on infrared image
KR20210037199A (en) Apparatus for dividing, tagging an image and for detecting defect of facilities using the same
CN114638989A (en) Fault classification visualization method based on target detection and fine-grained identification

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