CN116805398A - Transmission line damper defect detection method - Google Patents

Transmission line damper defect detection method Download PDF

Info

Publication number
CN116805398A
CN116805398A CN202310454766.6A CN202310454766A CN116805398A CN 116805398 A CN116805398 A CN 116805398A CN 202310454766 A CN202310454766 A CN 202310454766A CN 116805398 A CN116805398 A CN 116805398A
Authority
CN
China
Prior art keywords
transmission line
network
defect detection
damper
defect
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
CN202310454766.6A
Other languages
Chinese (zh)
Inventor
吴玉香
陈恩泽
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.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
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 South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202310454766.6A priority Critical patent/CN116805398A/en
Publication of CN116805398A publication Critical patent/CN116805398A/en
Pending legal-status Critical Current

Links

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
    • G06V20/17—Terrestrial scenes taken from planes or by drones
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Remote Sensing (AREA)
  • Image Processing (AREA)

Abstract

The application discloses a method for detecting the defect of a vibration damper of a power transmission line, which comprises the following steps: acquiring a transmission line damper image through unmanned aerial vehicle equipment, preprocessing the transmission line damper image, and establishing a sample data set; performing cluster analysis on the sample data set by using a K-means clustering algorithm, and predicting the size of an anchor frame; constructing a defect detection network; the defect detection network comprises an input preprocessing network, a feature extraction network, a feature fusion network and a detector; training a defect detection network by utilizing the sample data set and the predicted anchor frame size to obtain a defect detection model; and carrying out defect detection on the transmission line damper by adopting a defect detection model to obtain a detection result. The method solves the problem of low accuracy of detection of the vibration damper defects of the power transmission line, can effectively counteract noise interference and accurately position the defects, and has the characteristics of high accuracy, low time delay and strong robustness.

Description

Transmission line damper defect detection method
Technical Field
The application belongs to the technical field of vibration damper defect detection, and particularly relates to a method for detecting vibration damper defects of a power transmission line.
Background
The defect detection of the vibration damper of the power transmission line is always one of research hotspots in the field of detection of the running state of power equipment. In the operation of power equipment, due to the reasons of external environment, material aging and the like, the shock-proof hammer of the power transmission line has defects of different degrees; if the defects cannot be found and treated in time, the defects can have adverse effects on the operation safety of the power equipment, and even accidents can be caused when the defects are serious, so that casualties and property loss are caused. At present, the traditional method for detecting the shock-proof hammer of the power transmission line mainly adopts the modes of manual inspection and manual diagnosis, has the problems of low detection efficiency, high labor cost and the like, is difficult to detect comprehensively in a large scale, and cannot meet the operation requirement of modern power equipment.
In recent years, with the rise of image detection technology, a new opportunity is brought to power equipment inspection, wherein a target detection algorithm is applied to a plurality of actual scenes by virtue of the characteristics of high automation degree, high processing speed, high accuracy and the like, but the detection capability of various target detection algorithms proposed by the prior art for the shock-proof hammer defects is very limited, the problems of large model parameter quantity, difficult deployment and the like exist, and the target detection algorithm cannot be effectively applied to actual inspection.
Disclosure of Invention
The application mainly aims to overcome the defects and the shortcomings of the prior art, and provides a method for detecting the defects of the vibration damper of a power transmission line, which is used for effectively identifying and detecting the defects of the vibration damper under the condition of complex background.
In order to achieve the above purpose, the present application adopts the following technical scheme:
the method for detecting the shock-proof hammer defect of the power transmission line is characterized by comprising the following steps of:
step 1, acquiring a transmission line damper image through unmanned aerial vehicle equipment, preprocessing the transmission line damper image, and establishing a sample data set;
step 2, carrying out cluster analysis on the sample data set by using a K-means clustering algorithm, and predicting the size of an anchor frame;
step 3, constructing a defect detection network; the defect detection network comprises an input preprocessing network, a feature extraction network, a feature fusion network and a detector;
step 4, training a defect detection network by utilizing the sample data set and the predicted anchor frame size to obtain a defect detection model;
and 5, performing defect detection on the transmission line damper by adopting a defect detection model to obtain a detection result.
As a preferable technical solution, the step S1 includes the following steps:
step 1.1, shooting and collecting vibration-proof hammers of a power transmission line under various complex backgrounds through unmanned aerial vehicle equipment; the background includes sky, forest, field, sand, lake, mine and building; the shooting modes comprise close-range shooting, long-range shooting and multi-angle shooting;
step 1.2, cleaning the collected damper image, and removing a blurred and low-quality image;
step 1.3, classifying the removed damper images according to categories; the categories include normal vibration damper, falling of vibration damper, collision of vibration damper, displacement of vibration damper and rust of vibration damper; setting a damper image of a normal class of damper as a comparison sample;
step 1.4, marking the anti-vibration hammer images of each category by using a target frame, and amplifying the marked anti-vibration hammer images to establish a sample data set; the amplification includes small angle rotation, brightness change, flipping, cropping, sliding window slicing, copy-paste, and oversampling.
As a preferred technical solution, the step 3 of constructing the defect detection network includes:
step 3.1, constructing an input preprocessing network: the response to the defect target is improved, the influence of a complex background is restrained, and a processed image is obtained;
step 3.2, designing a feature extraction network: extracting features of the processed images to obtain feature images with different sizes;
step 3.3, constructing a feature fusion network: fusing the obtained feature graphs with different sizes, and outputting a fused feature graph;
step 3.4, building a detector: performing defect detection according to the fusion feature map, and marking in the damper image by using a prediction frame;
step 3.5, defining an optimized regression loss function: calculating loss when the defect detection network is trained, and iteratively updating parameters of the defect detection network through a back propagation algorithm.
As a preferable technical solution, the step 3.1 specifically includes:
the input preprocessing network consists of a Conv block, a Gusholle module, concat operation and up-sampling operation;
the Conv block is composed of a convolution layer, a batch normalization layer and a SiLU activation function;
the GShuffle module equally divides the input channels into two groups: a group of identity mapping is carried out; the other group is operated by a Ghost module; then splicing the two groups of output channels, and rearranging the channels; and finally outputting through Conv blocks.
As a preferable technical solution, the step 3.2 specifically includes:
the feature extraction network consists of a Conv block and an OSA_n module;
each Conv block in the OSA_n module is divided into two paths downwards, one path is connected to the Conv block of the next layer, the other path carries out identity mapping, and finally output channels of the two paths are combined and output;
where n in the OSA_n module represents that n OSA operations are performed.
As a preferable technical solution, the step 3.3 specifically includes:
the feature fusion module adopts a bidirectional fusion module of a target detection algorithm PAFPN from bottom to top and from top to bottom;
the fusion module is composed of Conv blocks, a maximum pooling operation and a Concat operation.
As a preferable technical solution, the step 3.4 specifically includes:
the detector sets three scales according to the model parameter and the target size, three anchor frames are set for each scale to carry out regression analysis, and nine anchor frames are set in total.
As a preferable technical solution, the step 3.5 specifically includes:
the optimized regression loss function includes a BIOU location loss function, a confidence loss function, and a classification loss function, expressed as:
wherein, K, S 2 The method comprises the steps of respectively detecting a terminal, traversing all grid cells and traversing all prediction frames; lambda (lambda) * Weights for the corresponding items;indicating whether the kth detection end, the ith grid unit and the jth prediction frame are positive samples, if so, 1, and if not, 0; alpha k The weight of the detection end under each scale is balanced; l (L) local Positioning the loss for the BIOU; l (L) conf Is a confidence loss; l (L) cls Is a classification loss;
the BIOU positioning loss L local The calculation is carried out through the cross ratio, and the formula is as follows:
wherein B is a prediction frame, B gt For the target frame, IOU is the intersection ratio of the predicted frame and the target frame, b is the center point of the predicted frame, b gt As a center point of the target frame, ρ () represents the euclidean distance between the prediction frame and the center point of the target frame, and c is the diagonal length of the minimum frame covering the prediction frame and the target frame; w (w) gt ,h gt The width and the height of the target frame are respectively; w and h are the width and height of the prediction frame respectively;
the confidence constraint L conf Obtained by calculating the binary cross entropy, expressed as:
wherein P is p Confidence scores for the prediction frames; p (P) gt IOU values for the corresponding target frames;is a binary cross entropy function; omega conf A positive sample weight for a preset confidence level;
the classification loss L cls Obtained by calculating the binary cross entropy, expressed as:
wherein C is p A category score for the prediction box; c (C) gt A one-hot representation for the target box class;is a binary cross entropy function; omega cls Is the positive sample weight of the classification.
Compared with the prior art, the application has the following advantages and beneficial effects:
1. according to the application, the defect detection network is constructed to detect the defect of the transmission line damper, and the input pretreatment network, the feature extraction network, the feature fusion network and the detector are designed, so that the defect feature can be effectively extracted under the condition of complex background, and the generalization capability and the detection precision of the model are improved.
2. The application calculates a loss value using an optimized regression loss function that includes a positioning loss, a confidence loss, and a classification loss; the positioning loss function calculates errors of a prediction frame and a target frame based on BIOU, is more sensitive to the difference of length and width, and is consistent with an actual regression process in any condition, so that the model can better return the errors along the minimum gradient direction in the training process, and the weight value in the forward calculation is corrected to realize autonomous parameter adjustment, so that the robustness of the application is obviously improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of a method for detecting a vibration damper defect of a power transmission line according to an embodiment of the present application.
Fig. 2 is a schematic diagram of a defect detection network according to an embodiment of the present application.
Fig. 3 is a schematic structural diagram of a gushubble module according to an embodiment of the present application.
Fig. 4 is a schematic structural diagram of an osa_n module according to an embodiment of the present application.
Fig. 5 is a schematic path diagram of a bidirectional fusion module of a target detection algorithm PAFPN in an embodiment of the present application.
Fig. 6 is a schematic structural diagram of a fusion module according to an embodiment of the application.
Fig. 7 (a) is an original image of the damper collected, and fig. 7 (b) is a view of the damper image defect detection result.
Detailed Description
In order to enable those skilled in the art to better understand the present application, the following description will make clear and complete descriptions of the technical solutions according to the embodiments of the present application with reference to the accompanying drawings. It will be apparent that the described embodiments are only some, but not all, embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Reference in the specification to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the described embodiments of the application may be combined with other embodiments.
As shown in fig. 1, the method for detecting the vibration damper defect of the power transmission line according to the embodiment includes the following steps:
step 1, acquiring a transmission line damper image through unmanned aerial vehicle equipment, preprocessing the transmission line damper image, and establishing a sample data set;
specifically, step 1 of creating a sample dataset includes:
step 1.1, shooting and acquiring images of transmission line vibration hammers under various complex backgrounds through unmanned aerial vehicle equipment; wherein the background includes sky, forest, field, sand, lake, mine, building, etc.; shooting modes comprise close-range shooting, long-range shooting, multi-angle shooting and the like, and the shot images are ensured to contain targets with different pixels and different angles;
step 1.2, cleaning the collected damper image, and removing a blurred and low-quality image;
step 1.3, classifying the removed damper images according to categories; the categories include normal vibration damper, falling of vibration damper, collision of vibration damper, displacement of vibration damper and rust of vibration damper; taking a damper image with normal damper classification as a positive sample;
step 1.4, marking the anti-vibration hammer images of each category by using a target frame, and amplifying the marked anti-vibration hammer images to establish a sample data set; the amplification method comprises small-angle rotation, brightness change, overturning, cutting, sliding window slicing, copy-paste, oversampling and the like.
Wherein, when the sliding window section is adopted for amplification:
counting the number histogram and the width-height distribution of defect targets in the damper image, and determining the size of a sliding window for slicing;
sliding the damper image according to a sliding window with specified width and height and step length from left to right and from top to bottom in sequence; the overlapping rate of the sliding slices is controlled between 0.1 and 0.5;
if the target frame is cut by the sliding window in the slicing process, calculating the intersection ratio of the target frame and the sliding window, reserving the target frame with the intersection ratio larger than 0.6, and limiting the size of the target frame in the range of the sliding window.
Amplification using copy-paste:
randomly selecting a source image and a target image for pasting, wherein the source image can be used as the target image;
randomly selecting and pasting a specific defect target in the source image;
the specific position to paste to the target image is randomly selected.
Oversampling amplification refers to: for a smaller number of defect categories, the corresponding images are repeatedly added to the training multiple times.
Step 2, carrying out cluster analysis on the sample data set by using a K-means clustering algorithm, and predicting the size of an anchor frame;
specifically, the embodiment predicts the anchor frame size through a K-means clustering algorithm, including:
step 2.1, setting the number n of clusters according to the number of anchor frames set by a detector in a defect detection network, and randomly initializing n cluster center points A= { a 1 ,a 2 ,…,a n -a }; in this embodiment, the number of anchor frames of the detector is 9, so the number of clusters is set to n=9;
step 2.2 for each sample data x in the sample data set i Calculating the distance between each clustering center point and sample data by adopting a Euclidean distance formula, and dividing the sample data with a relatively close distance into the class where the clustering center point is located;
step 2.3, judging whether the iteration termination condition is met according to the category of each cluster center, outputting the cluster category if the iteration termination condition is met, and updating the cluster center point of the category if the iteration termination condition is not met, wherein an updating formula is as follows:
wherein a is k Representing the updated cluster center point of the kth cluster category, A k Clustering clusters after clustering sample data for the kth clustering center point, |A k I represents cluster A k The number of sample data;
and 2.4, repeating the steps 2.2 and 2.3 until the cluster center point is not changed or reaches the set iteration times, and obtaining the predicted anchor frame size.
Step 3, constructing a defect detection network as shown in fig. 2, which comprises four parts including: inputting a preprocessing network, a feature extraction network, a feature fusion network and a detector; the specific construction steps comprise:
step 3.1, constructing an input preprocessing network: the response to the defect target is improved, the influence of a complex background is restrained, and a processed image is obtained;
step 3.2, designing a feature extraction network: extracting features of the processed images to obtain feature images with different sizes;
step 3.3, constructing a feature fusion network: fusing the obtained feature graphs with different sizes, and outputting a fused feature graph;
step 3.4, building a detector: performing defect detection according to the fusion feature map, and marking in the damper image by using a prediction frame;
step 3.5, defining an optimized regression loss function: calculating loss when the defect detection network is trained, and iteratively updating parameters of the defect detection network through a back propagation algorithm.
Further, step 3.1 specifically includes:
the input preprocessing network consists of a Conv block, a Gushuffle module, a Concat operation and an up-sampling operation; the configuration of the input pre-processing network is shown in table 1 below:
table 1 input pretreatment network configuration table
Wherein, conv block is composed of convolution layer, batch normalization layer and SiLU activation function;
the GShuffle module is shown in fig. 3, the input channels are equally divided into two groups, one group carries out identity mapping, the other group carries out operation through the Ghost module, the convolution kernel size is 3 multiplied by 3, the step length is 1, and the number of the output channels is reduced to 1/2 of the number of the input channels; then splicing the channels generated by the two groups of operations, and rearranging the channels; and finally outputting through Conv blocks.
The input preprocessing network constructed in the embodiment can effectively improve the response of the model to the defect target and reduce the influence of background information and jitter phenomenon on detection.
Further, step 3.2 specifically includes:
the feature extraction network consists of Conv blocks and OSA_n modules;
as shown in fig. 4, the osa_n module divides each Conv block down into two paths, one path is connected to the Conv block of the next layer, the other path performs identity mapping, and n represents performing n times of such operations; finally, merging and outputting all channels generated after n times of operation;
in order to give consideration to the parameter size and the feature extraction effect, n in OSA_n is taken as 4, the number of input channels in the middle of the OSA_n is constant, the calculation efficiency of the GPU can be effectively improved, and the access cost of the memory is reduced.
The feature extraction network constructed in the embodiment can effectively extract features with better comprehensiveness and semantics by changing the feature map and the channel number to extract the image features.
Further, step 3.3 specifically includes:
the feature fusion network is shown in fig. 5, and a bidirectional fusion module of a target detection algorithm PAFPN from bottom to top and from top to bottom is adopted; the PAFPN designs a fusion module, as shown in fig. 6, which consists of Conv blocks, a max pooling operation and a Concat operation;
the fusion module is used for referencing the ideas of SPPF and CSP structures, can enrich the expression capability of the feature map, is very friendly to small-scale defect targets, effectively enhances the expression capability of the features, and deeply fuses the features to improve the performance of the model.
Further, step 3.4 specifically includes:
the detector considers the model parameter and the target size to set three scales, and each scale sets three anchor frames to carry out regression analysis, namely nine anchor frames are arranged in total, so that the detector is favorable for coping with the situation that the target size difference in the image to be detected is large, and can well meet the detection requirement of a multi-scale target.
Taking the anchor frame size correspondence table of table 2 as an example, when the input image size is set to 640×640, the obtained scales are 20×20, 40×40 and 80×80 respectively, wherein the feature map with the scale of 20×20 has a large receptive field, and can be used for identifying defect targets with large sizes; a feature map with a scale of 40×40 has a receptive field of moderate size, which can be used to identify defect targets of moderate size; the feature map with the scale of 80 multiplied by 80 has a small receptive field, can be used for identifying defect targets with small sizes, and in practical application, the scale should be correspondingly adjusted to meet practical requirements.
Table 2 table of anchor frame size correspondence
Further, step 3.5 specifically includes:
the loss function of the defect detection network adopts an optimized regression loss function, comprising a BIOU positioning loss function, a confidence loss function and a classification loss function, which are expressed as follows:
wherein, K, S 2 B is a detection end, all traversed grid cells and all traversed prediction frames respectively; lambda (lambda) * Weights for the corresponding items;indicating whether the kth detection end, the ith grid unit and the jth prediction frame are positive samples, if so, 1, and if not, 0; alpha k The weight of the detection end under each scale is balanced; l (L) local Positioning the loss for the BIOU; l (L) conf Is a confidence loss; l (L) cls Is a classification loss.
Wherein BIOU positioning loss L local The calculation is carried out through the cross ratio, and the formula is as follows:
wherein B is a prediction frame, B gt For the target frame, IOU is the intersection ratio of the predicted frame and the target frame, b is the center point of the predicted frame, b gt As a center point of the target frame, ρ () represents the euclidean distance between the prediction frame and the center point of the target frame, and c is the diagonal length of the minimum frame covering the prediction frame and the target frame; α is; v. is; w (w) gt ,h gt The width and the height of the target frame are respectively; w and h are the width and height of the prediction frame respectively; in this way, BIOU localization lossThe following effects are lost:
first, regardingAnd->The following is shown:
in the formula (II) above, the formula (III), and->In proportion, therefore, in any case, the two variables w and h can be scaled simultaneously, conforming to the actual regression process;
second, v constructed reflects not only the difference in aspect ratio, but also w gt And w or h gt And h, solving when the width and height of the prediction frame satisfy { (w=kw) gt ,h＝kh gt )|k∈R + Problem of v failure at } R + Is a positive real number;
finally, adoptIn variable form such that most of its values fall within the interval (0, 1]In this case, this is very friendly to the artan form, and the magnitude of the change in artan values is large in this interval, making the loss term more sensitive to the length-width difference.
Confidence constraint L conf Obtained by calculating binary cross entropy BCE, expressed as:
wherein P is p Confidence scores for the prediction frames; p (P) gt IOU values for the corresponding target frames;is a binary cross entropy function; omega conf A positive sample weight for a preset confidence level;
classification loss L cis Obtained by calculating binary cross entropy BCE, expressed as:
wherein C is p A category score for the prediction box; c (C) gt A one-hot representation for the target box class;is a binary cross entropy function; omega cls Is the positive sample weight of the classification.
Step 4, training a defect detection network by utilizing the sample data set and the predicted anchor frame size to obtain a defect detection model;
specifically, the step of training the defect detection network in this embodiment includes:
step 4.1, dividing the sample data set into a training set, a testing set and a verification set according to the proportion; in the embodiment, a sample data set is divided into a training set, a testing set and a verification set according to the proportion of 8:1:1;
step 4.2, setting super parameters of the defect detection network, including initial learning rate, final learning rate, training batch, input image size and the like, wherein the super parameters can be adjusted according to actual training conditions; in this embodiment, the initial learning rate is set to 0.01, so that the network can quickly converge; the final learning rate is set to be 0.001, so that the network can be ensured to converge to an optimal solution; the training batch is set to 300 batches; the input image size is set to 640×640;
step 4.3, training the defect detection network by using a training set, verifying by using a verification set every time one batch of training is performed, performing loss calculation by using an optimized regression loss function, performing iterative updating on parameters of the defect detection network by using a back propagation algorithm, and stopping training when the set iterative parameters are reached;
and 4.4, screening the defect detection network with the best training effect by using the test set to serve as a defect detection model.
Step 5: and carrying out defect detection on the transmission line damper by adopting a defect detection model to obtain a detection result.
In this embodiment, defect detection is performed on the damper image collected in the actual power inspection scene by using the defect detection model, and the defect detection result of the power transmission line damper is shown in fig. 7 (a), which is an original collected damper image, and fig. 7 (b) is a defect detection result, including a defect name and a defect position. The detection result shows that the accuracy of the method for detecting the defects of the damper reaches 98.25 percent, and the method can achieve the effect of real-time detection when deployed on front-end equipment.
It should be noted that, for the sake of simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but it should be understood by those skilled in the art that the present application is not limited by the order of actions described, as some steps may be performed in other order or simultaneously in accordance with the present application.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples are preferred embodiments of the present application, but the embodiments of the present application are not limited to the above examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principle of the present application should be made in the equivalent manner, and the embodiments are included in the protection scope of the present application.

Claims (8)

1. The method for detecting the shock-proof hammer defect of the power transmission line is characterized by comprising the following steps of:
step 1, acquiring a transmission line damper image through unmanned aerial vehicle equipment, preprocessing the transmission line damper image, and establishing a sample data set;
step 2, carrying out cluster analysis on the sample data set by using a K-means clustering algorithm, and predicting the size of an anchor frame;
step 3, constructing a defect detection network; the defect detection network comprises an input preprocessing network, a feature extraction network, a feature fusion network and a detector;
step 4, training a defect detection network by utilizing the sample data set and the predicted anchor frame size to obtain a defect detection model;
and 5, performing defect detection on the transmission line damper by adopting a defect detection model to obtain a detection result.
2. The method for detecting vibration damper defects of transmission lines according to claim 1, wherein the step S1 comprises the steps of:
step 1.1, shooting and collecting vibration-proof hammers of a power transmission line under various complex backgrounds through unmanned aerial vehicle equipment; the background includes sky, forest, field, sand, lake, mine and building; the shooting modes comprise close-range shooting, long-range shooting and multi-angle shooting;
step 1.2, cleaning the collected damper image, and removing a blurred and low-quality image;
step 1.3, classifying the removed damper images according to categories; the categories include normal vibration damper, falling of vibration damper, collision of vibration damper, displacement of vibration damper and rust of vibration damper; setting a damper image of a normal class of damper as a comparison sample;
step 1.4, marking the anti-vibration hammer images of each category by using a target frame, and amplifying the marked anti-vibration hammer images to establish a sample data set; the amplification includes small angle rotation, brightness change, flipping, cropping, sliding window slicing, copy-paste, and oversampling.
3. The method for detecting the vibration damper defect of the power transmission line according to claim 1, wherein the step 3 of constructing the defect detection network comprises:
step 3.1, constructing an input preprocessing network: the response to the defect target is improved, the influence of a complex background is restrained, and a processed image is obtained;
step 3.2, designing a feature extraction network: extracting features of the processed images to obtain feature images with different sizes;
step 3.3, constructing a feature fusion network: fusing the obtained feature graphs with different sizes, and outputting a fused feature graph;
step 3.4, building a detector: performing defect detection according to the fusion feature map, and marking in the damper image by using a prediction frame;
step 3.5, defining an optimized regression loss function: calculating loss when the defect detection network is trained, and iteratively updating parameters of the defect detection network through a back propagation algorithm.
4. The method for detecting the vibration damper defect of the power transmission line according to claim 3, wherein the step 3.1 specifically comprises:
the input preprocessing network consists of a Conv block, a Gusholle module, concat operation and up-sampling operation;
the Conv block is composed of a convolution layer, a batch normalization layer and a SiLU activation function;
the GShuffle module equally divides the input channels into two groups: a group of identity mapping is carried out; the other group is operated by a Ghost module; then splicing the two groups of output channels, and rearranging the channels; and finally outputting through Conv blocks.
5. The method for detecting the vibration damper defect of the power transmission line according to claim 3, wherein the step 3.2 specifically comprises:
the feature extraction network consists of a Conv block and an OSA_n module;
each Conv block in the OSA_n module is divided into two paths downwards, one path is connected to the Conv block of the next layer, the other path carries out identity mapping, and finally output channels of the two paths are combined and output;
where n in the OSA_n module represents that n OSA operations are performed.
6. The method for detecting the vibration damper defect of the power transmission line according to claim 3, wherein the step 3.3 specifically includes:
the feature fusion module adopts a bidirectional fusion module of a target detection algorithm PAFPN from bottom to top and from top to bottom;
the fusion module is composed of Conv blocks, a maximum pooling operation and a Concat operation.
7. The method for detecting the vibration damper defect of the power transmission line according to claim 3, wherein the step 3.4 specifically comprises:
the detector sets three scales according to the model parameter and the target size, three anchor frames are set for each scale to carry out regression analysis, and nine anchor frames are set in total.
8. The method for detecting the vibration damper defect of the power transmission line according to claim 3, wherein the step 3.5 specifically comprises:
the optimized regression loss function includes a BIOU location loss function, a confidence loss function, and a classification loss function, expressed as:
wherein, K, S 2 The method comprises the steps of respectively detecting a terminal, traversing all grid cells and traversing all prediction frames; lambda (lambda) * Weights for the corresponding items;indicating whether the kth detection end, the ith grid unit and the jth prediction frame are positive samples, if so, 1, and if not, 0; alpha k The weight of the detection end under each scale is balanced; l (L) local Positioning the loss for the BIOU; l (L) conf Is a confidence loss; l (L) cls Is a classification loss;
the BIOU positioning loss L local The calculation is carried out through the cross ratio, and the formula is as follows:
wherein B is a prediction frame, B gt For the target frame, IOU is the intersection ratio of the predicted frame and the target frame, b is the center point of the predicted frame, b gt As a center point of the target frame, ρ () represents the euclidean distance between the prediction frame and the center point of the target frame, and c is the diagonal length of the minimum frame covering the prediction frame and the target frame; w (w) gt ,h gt The width and the height of the target frame are respectively; w and h are the width and height of the prediction frame respectively;
the confidence constraint L conf Obtained by calculating the binary cross entropy, expressed as:
wherein P is p Confidence scores for the prediction frames; p (P) gt IOU values for the corresponding target frames;is a binary cross entropy function; omega conf A positive sample weight for a preset confidence level;
the classification loss L cls Obtained by calculating the binary cross entropy, expressed as:
wherein C is p A category score for the prediction box; c (C) gt A one-hot representation for the target box class;is a binary cross entropy function; omega cls Is the positive sample weight of the classification.
CN202310454766.6A 2023-04-24 2023-04-24 Transmission line damper defect detection method Pending CN116805398A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310454766.6A CN116805398A (en) 2023-04-24 2023-04-24 Transmission line damper defect detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310454766.6A CN116805398A (en) 2023-04-24 2023-04-24 Transmission line damper defect detection method

Publications (1)

Publication Number Publication Date
CN116805398A true CN116805398A (en) 2023-09-26

Family

ID=88079067

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310454766.6A Pending CN116805398A (en) 2023-04-24 2023-04-24 Transmission line damper defect detection method

Country Status (1)

Country Link
CN (1) CN116805398A (en)

Similar Documents

Publication Publication Date Title
CN110533084B (en) Multi-scale target detection method based on self-attention mechanism
KR102263397B1 (en) Method for acquiring sample images for inspecting label among auto-labeled images to be used for learning of neural network and sample image acquiring device using the same
KR102280414B1 (en) Method for optimizing hyperparameters of auto-labeling device which auto-labels training images for use in deep learning network to analyze images with high precision, and optimizing device using the same
KR102328731B1 (en) Learning method and learning device for object detector based on cnn, adaptable to customers' requirements such as key performance index, using target object merging network and target region estimating network, and testing method and testing device using the same to be used for multi-camera or surround view monitoring
KR102328734B1 (en) Method for automatically evaluating labeling reliability of training images for use in deep learning network to analyze images, and reliability-evaluating device using the same
CN110889324A (en) Thermal infrared image target identification method based on YOLO V3 terminal-oriented guidance
Mahmoud et al. Object detection using adaptive mask RCNN in optical remote sensing images
CN107103332A (en) A kind of Method Using Relevance Vector Machine sorting technique towards large-scale dataset
CN113111727B (en) Feature alignment-based method for detecting rotating target in remote sensing scene
CN111462193A (en) Object detection method and device for monitoring multiple cameras or surrounding views
CN111126278A (en) Target detection model optimization and acceleration method for few-category scene
CN117746077B (en) Chip defect detection method, device, equipment and storage medium
CN116843691B (en) Photovoltaic panel hot spot detection method, storage medium and electronic equipment
CN114972759A (en) Remote sensing image semantic segmentation method based on hierarchical contour cost function
Wang et al. Deep learning model for target detection in remote sensing images fusing multilevel features
CN114078209A (en) Lightweight target detection method for improving small target detection precision
CN115984543A (en) Target detection algorithm based on infrared and visible light images
CN115375672A (en) Coal gangue detection method based on improved YOLOv4 algorithm
CN114219998A (en) Sonar image real-time detection method based on target detection neural network
Pillai et al. Fine-Tuned EfficientNetB4 Transfer Learning Model for Weather Classification
CN117495891A (en) Point cloud edge detection method and device and electronic equipment
CN117612044A (en) Method for inspecting transmission line insulator by unmanned aerial vehicle in complex scene
CN112597875A (en) Multi-branch network anti-missing detection aerial photography target detection method
CN111368625B (en) Pedestrian target detection method based on cascade optimization
CN117274740A (en) Infrared target detection method and device

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