CN112215055A - Method for detecting conductor defects in power transmission line based on single classification method - Google Patents

Method for detecting conductor defects in power transmission line based on single classification method Download PDF

Info

Publication number
CN112215055A
CN112215055A CN202010822113.5A CN202010822113A CN112215055A CN 112215055 A CN112215055 A CN 112215055A CN 202010822113 A CN202010822113 A CN 202010822113A CN 112215055 A CN112215055 A CN 112215055A
Authority
CN
China
Prior art keywords
training
generator
pictures
classifier
classification
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
CN202010822113.5A
Other languages
Chinese (zh)
Inventor
及浩然
杨阳
侯春萍
华中华
王霄聪
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.)
Tianjin University
Original Assignee
Tianjin University
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 Tianjin University filed Critical Tianjin University
Priority to CN202010822113.5A priority Critical patent/CN112215055A/en
Publication of CN112215055A publication Critical patent/CN112215055A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a method for detecting a wire defect in a power transmission line based on a single classification method, which comprises the following steps: constructing a training set and a testing set for detecting the defects of the wires in the power transmission line; building a network structure; and training a detection model, setting parameters according to the characteristics of the conductor defect picture, and realizing single-classification training by using a method based on generation of a countermeasure network. The training process is divided into two major parts in total: the first part is a classifier C in a training network, and the second part is used for carrying out countermeasure training of a generator and a discriminator; and classifying the pictures to be tested by using the test model. After the training is finished, the to-be-detected pictures are sent to the classifier for classification only by the classifier in the network structure, a result without defects is obtained, and the pictures are stored into the corresponding folders according to the classification result.

Description

Method for detecting conductor defects in power transmission line based on single classification method
Technical Field
The invention belongs to the field of remote sensing images, and relates to a method for identifying a wire defect in an unmanned aerial vehicle inspection aerial image based on a deep learning technology.
Background
The transmission conductor is an important component in an electric power system, and due to the special geographical distribution position, the transmission conductor can bear the internal pressure of electric power load and mechanical load, and can also be subjected to the invasion of severe environments such as wind, snow, wind, sand, electric lightning stroke and the like for a long time, so that faults such as wire jumping, conductor strand breakage, wire breakage and the like are easy to occur, and malignant electric power accidents are often caused if the faults are not processed in time. In recent years, the deterioration of environmental climate aggravates the complexity of the operation environment of the power transmission line, and the failure rate of the overhead power transmission line is in a growing trend, so that the failure reasons of the power transmission line are comprehensively analyzed, failure diagnosis and precautionary measures are actively explored, and the method has important significance on the safe and reliable operation of a power grid [1 ].
At present, the detection of the line mainly depends on manual inspection, but the detection efficiency is low, the working strength is high, the detection speed is low, and new methods are urgently needed to solve the problems. A great deal of research of scholars at home and abroad shows that the conventional power transmission line detection methods mainly comprise manual detection, an infrared imaging method, an ultraviolet imaging method, an ultrasonic detection method, a magnetic flux leakage detection method, an eddy current detection method and the like, and all the methods have advantages and disadvantages and are limited in application range. Therefore, on-line monitoring technologies such as robot inspection, helicopter inspection, unmanned aerial vehicle inspection and the like are developed, and the technologies such as machine vision, image processing, deep learning and the like are combined by means of a flexible image acquisition mode, so that the system is widely applied to a power system. In the aspect of wire identification, wire extraction is mainly realized by reducing background influence, such as: the image is segmented through a total variation model denoising and simulated annealing algorithm, and the target extraction of the wire is realized by adopting an improved Freeman chain code; the automatic detection of the power line is achieved by Radon transformation, line segment clustering and a Kalman filter, but the calculated amount is large and the requirement on real-time inspection is difficult to meet. In the aspect of conductor detection, the method for detecting the broken strand image of the power transmission conductor by using the optimized Gabor filter is used for establishing an optimized design model of the improved genetic algorithm Gabor filter for solving and segmenting, so that the fault information of the broken strand position of the power transmission conductor can be well extracted, and the method is only suitable for the image with a simple background. [2-4]
Aiming at the problems, the biggest pain point of the wire detection problem is that the difficulty of acquiring a data set is high, and the methods need to use a large amount of positive and negative data sets for training. Therefore, the model is trained by using the single classification method and only using the data set of the easily obtained normal wire, so that the cost and the error caused by incomplete and irregular data sets are reduced.
[1] Li Xiaohui, 2018, analysis of damage conditions of freezing disasters of Hubei power grid transmission line equipment [ J ]. Hubei power, 2018, 42 (3): 1-4.
[2] The method for detecting the strand breaking defect of the airplane line patrol conducting wire based on the image processing technology [ J ]. Heilongjiang electric power 2017,39 (6): 522-526.
[3] Application of unmanned aerial vehicle in power line inspection [ J ]. china power, 2013,46 (3): 35-38.
[4] Power line strand breakage and foreign object defect detection method based on unmanned aerial vehicle images [ J ] computer application, 2015,35 (8): 2404-2408.
Disclosure of Invention
The method based on the single classification thought in the deep learning determines the geographic position of the defect of the lead in the picture shot by the unmanned aerial vehicle, finds out the specific area with the problem of the lead, can reduce part of workload, and can position, find and solve the problem more quickly. The technical scheme is as follows:
a method for detecting the defects of a lead in a power transmission line based on a single classification method comprises the following steps:
1) constructing a training set and a testing set for detecting the defects of the wires in the power transmission line;
2) and (5) building a network structure.
Selecting a scheme for generating a countermeasure network to realize by using a single classification idea, wherein a generator uses a denoising autoencoder of a convolutional neural network structure to restore a pre-denoised image; the optimization process is a minimum and maximum game problem, and the optimization target is to achieve Nash equilibrium, namely whether a false sample generated by a generator is true or false is not recognized by a discrimination model.
3) And training a detection model.
And setting parameters according to the characteristics of the conductor defect picture, and realizing single-classification training by using a method based on generation of a countermeasure network.
The training process is divided into two major parts in total: the first part is to train the classifier C in the network, and the second part is to carry out the countermeasure training of the generator and the discriminator.
Firstly, fixing the rest network structures except the classifier, and training by using the reconstructed original image and the special sample generated by the generator to enable the classifier to have certain classification capability. In this step, the original image and the generated image are marked as they are.
After the first training step is finished, the classifier C is fixed, and the generator and the identifier in the network are subjected to countermeasure training. Including three generators and two discriminators. The three generators respectively have the functions of generating pictures from random Gaussian noise, generating pictures from noise samples and restoring original pictures from original pictures which are subjected to noise addition in advance. Two discriminators being discriminators D of the images respectivelyvAnd a discriminator D of the feature spacel. Wherein DvJudging between the picture generated by sampling noise and the original picture; dlThe discrimination is made between the mapping of the feature space and the sampling noise itself by the noisy picture.
In this process, training D is performed firstvAnd DlThe purpose of the training is to reduce the distribution distance between the generated picture and the original picture;
and finally, carrying out an optimization process of the generator, and training the generator by taking the original picture as an input.
4) And classifying the pictures to be tested by using the test model. After the training is finished, the to-be-detected pictures are sent to the classifier for classification only by the classifier in the network structure, a result without defects is obtained, and the pictures are stored into the corresponding folders according to the classification result.
Drawings
FIG. 1 is a detailed structure of a single-class network
Accuracy of classification of FIG. 2
Detailed Description
In order to make the technical solution of the present invention clearer, the following describes the embodiments of the present invention with reference to the accompanying drawings.
1. And (5) building a network structure.
Using the single classification concept, a solution is chosen to generate a countermeasure network, typically based on a game model, where the generator must compete with its opponent discriminator. The generator typically generates false samples from a standard gaussian distribution and the discriminant model attempts to distinguish between the generator-generated samples (false samples) and the samples extracted from the training data (true samples). The generative countermeasure network (GAN) is a generative model in which the generator attempts to learn the characteristic distribution of real data samples and generates new data samples. The discriminator is essentially a two-classifier that discriminates whether the input is real data or a generated sample.
In the scheme, the generator recovers the pre-noisy image by using a denoising autoencoder of a convolutional neural network structure. The optimization process is a minimum and maximum game problem, and the optimization target is to achieve Nash equilibrium, namely whether a false sample generated by a generator is true or false is not recognized by a discrimination model.
The generative countermeasure network is formed by a generator G which is responsible for distributing the samples a priori over p and a discriminator Dz(z)Is mapped to resemble the true data distribution pdata(x)Generated data distribution pG(z)(ii) a Discriminator D can be seen as a function mapping data to discriminant probabilities D (x) → (0,1), the data trained for input being from the true data distribution pdata(x)Or generating a data distribution pG(z). G and D respectively and continuously improve the generation capability and the discrimination capability of the G and the D in the process of mutual confrontation training until pG(z)Enough to perfectly match pdata(x)The discriminator will give a probability value of 0.5 for all inputs, at which point both also reach a dynamic nash equilibrium. Such a training task can be viewed as a minuscule game of loss functions V (D, G). The mathematical expression is as follows:
Figure BDA0002634825830000031
wherein D is discriminator, G is generator, x is original image, and z is noise.
2. Construction of data sets
And (5) correspondingly placing the obtained data sets in a specified folder and marking the categories respectively for training.
3. And training a detection model.
Parameters are set according to the characteristics of the lead defect picture, so that the training effect is more ideal. In consideration of the characteristic of less data sets of the lead defect pictures, in order to improve the robustness and the practical use difficulty of the system, the method based on generation of the countermeasure network is used for realizing the training of single classification.
The training process is divided into two major parts in total: the first part is to train the classifier C in the network, and the second part is to carry out the countermeasure training of the generator and the discriminator.
Firstly, fixing the rest network structures except the classifier, and training by using the reconstructed original image and the special sample generated by the generator to enable the classifier to have certain classification capability. In this step, the original image and the generated image are marked as they are.
After the first training step is finished, the classifier C is fixed, and the generator and the identifier in the network are subjected to countermeasure training. Including three generators and two discriminators (fig. 1). The three generators respectively have the functions of generating pictures from random Gaussian noise, generating pictures from noise samples and restoring original pictures from original pictures which are subjected to noise addition in advance. The two discriminators being discriminators of the images (D)v) And a discriminator of the feature space (D)l). Wherein DvJudging between the picture generated by sampling noise and the original picture; dlThe discrimination is made between the mapping of the feature space and the sampling noise itself by the noisy picture.
In this process, training D is performed firstvAnd DlThe training aims to reduce the distribution distance between the generated picture and the original pictureMinimization of the loss function: llatent+lvisual：
Figure BDA0002634825830000032
Figure BDA0002634825830000033
Where En is the self-encoder, De is the decoder, n is random gaussian noise, and s is the noise sample. And then optimizing a noise sampling method, wherein a specific means is to train and improve the sampling effect by using a classifier C for the image recovered from the sampling noise.
And finally, carrying out an optimization process of the generator, and training the generator by taking the original picture as an input. Except for D before usevAnd DlAnd the generator produces the confrontational training. Also used was MSE Loss:
lmse＝||x-De(l1)||2 (4)
wherein x is the original picture, l1For Gaussian distribution, the purpose of this term is to reduce the distance between the generated picture and the original picture at the generator angle, and the final overall loss function is 10 × lMSE+lvisual+llatent. The difference between the distribution in the generated picture and the original picture is reduced, and the effect of countermeasures is achieved.
5. The results of the processing of the experimental data are presented below:
the experimental results are as follows: in the early stage of training, the accuracy is at the peak, the accuracy is rapidly increased in the early stage, the accuracy is about 0.35 in the late stage, the analysis is caused by less training data sets, and in the real training, the data sets are increased or an early stopping strategy is adopted, so that a better effect can be achieved. The specific AUC effect is shown in fig. 2.

Claims (1)

1. A method for detecting the defects of a lead in a power transmission line based on a single classification method comprises the following steps:
1) constructing a training set and a testing set for detecting the defects of the wires in the power transmission line;
2) building a network structure;
selecting a scheme for generating a countermeasure network to realize by using a single classification idea, wherein a generator uses a denoising autoencoder of a convolutional neural network structure to restore a pre-denoised image; the optimization process is a minimum and maximum game problem, and the optimization target is to achieve Nash equilibrium, namely whether a false sample generated by a generator is true or false is not recognized by a discrimination model.
3) Training a detection model;
setting parameters according to the characteristics of the lead defect picture, and realizing single-classification training by using a method based on generation of a countermeasure network;
the training process is divided into two major parts in total: the first part is a classifier C in a training network, and the second part is used for carrying out countermeasure training of a generator and a discriminator;
firstly, fixing the rest network structures except the classifier, and training by using the reconstructed original image and a special sample generated by a generator to enable the classifier to have certain classification capability; in the step, marking the original image and the generated image according to the real condition;
after the first training step is finished, fixing the classifier C, and carrying out countermeasure training on a generator and a discriminator in the network; three generators and two discriminators are included; the three generators respectively have the functions of generating pictures from random Gaussian noise, generating pictures from noise samples and restoring original pictures from original pictures which are subjected to noise addition in advance; two discriminators being discriminators D of the images respectivelyvAnd a discriminator D of the feature spacel(ii) a Wherein DvJudging between the picture generated by sampling noise and the original picture; dlJudging whether the mapping of the noise-added picture in the feature space and the sampling noise are judged;
in this process, training D is performed firstvAnd DlThe purpose of the training is to reduce the distribution distance between the generated picture and the original picture;
finally, performing an optimization process of the generator, and training the generator by taking the original picture as input;
4) classifying the pictures to be tested by using the test model; after the training is finished, the to-be-detected pictures are sent to the classifier for classification only by the classifier in the network structure, a result without defects is obtained, and the pictures are stored into the corresponding folders according to the classification result.
CN202010822113.5A 2020-08-15 2020-08-15 Method for detecting conductor defects in power transmission line based on single classification method Pending CN112215055A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010822113.5A CN112215055A (en) 2020-08-15 2020-08-15 Method for detecting conductor defects in power transmission line based on single classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010822113.5A CN112215055A (en) 2020-08-15 2020-08-15 Method for detecting conductor defects in power transmission line based on single classification method

Publications (1)

Publication Number Publication Date
CN112215055A true CN112215055A (en) 2021-01-12

Family

ID=74058995

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010822113.5A Pending CN112215055A (en) 2020-08-15 2020-08-15 Method for detecting conductor defects in power transmission line based on single classification method

Country Status (1)

Country Link
CN (1) CN112215055A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109544666A (en) * 2018-10-26 2019-03-29 中国科学院计算技术研究所 A kind of full automatic model deformation transmission method and system
CN110097543A (en) * 2019-04-25 2019-08-06 东北大学 Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network
CN111340791A (en) * 2020-03-02 2020-06-26 浙江浙能技术研究院有限公司 Photovoltaic module unsupervised defect detection method based on GAN improved algorithm

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109544666A (en) * 2018-10-26 2019-03-29 中国科学院计算技术研究所 A kind of full automatic model deformation transmission method and system
CN110097543A (en) * 2019-04-25 2019-08-06 东北大学 Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network
CN111340791A (en) * 2020-03-02 2020-06-26 浙江浙能技术研究院有限公司 Photovoltaic module unsupervised defect detection method based on GAN improved algorithm

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
PRAMUDITHA PERERA ET AL.: "OCGAN: One-class Novelty Detection Using GANs with Constrained Latent Representations", 《IEEE XPLORE》 *

Similar Documents

Publication Publication Date Title
CN109118479B (en) Capsule network-based insulator defect identification and positioning device and method
CN112199993B (en) Method for identifying transformer substation insulator infrared image detection model in any direction based on artificial intelligence
CN112200178B (en) Transformer substation insulator infrared image detection method based on artificial intelligence
CN111428748A (en) Infrared image insulator recognition and detection method based on HOG characteristics and SVM
CN111832398B (en) Unmanned aerial vehicle image distribution line pole tower ground wire broken strand image detection method
CN109300118B (en) High-voltage power line unmanned aerial vehicle inspection method based on RGB image
CN109523543A (en) A kind of wire strand breakage detection method based on Edge Distance
CN111862065A (en) Power transmission line diagnosis method and system based on multitask deep convolutional neural network
CN116681979A (en) Power equipment target detection method under complex environment
Wanguo et al. Typical defect detection technology of transmission line based on deep learning
Wang et al. Railway insulator detection based on adaptive cascaded convolutional neural network
CN106355187A (en) Application of visual information to electrical equipment monitoring
CN111931577A (en) Intelligent inspection method for specific foreign matters of power grid line
CN112215055A (en) Method for detecting conductor defects in power transmission line based on single classification method
Chen et al. Attention mechanism-based CNN for surface damage detection of wind turbine blades
CN112508946B (en) Cable tunnel anomaly detection method based on antagonistic neural network
CN112183191A (en) Method for detecting bird nest in power transmission line based on countermeasure training
CN112183186A (en) Insulator defect detection method based on generation countermeasure network
CN113989209A (en) Power line foreign matter detection method based on fast R-CNN
CN113506290A (en) Method and device for detecting defects of line insulator
Ye et al. Research on detection method of tower corrosion based on hough transform
CN117437188B (en) Insulator defect detection system for smart power grid
CN117235583B (en) Monitoring method and system for GIS breaker actuating mechanism
Park et al. Displacement Detection of Wooden Cultural Properties Using Unsupervised Learning
CN111797925B (en) Visual image classification method and device for power system

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20210112