CN117456198A - Power transmission line fault detection method based on improved Yolov5 neural network - Google Patents
Power transmission line fault detection method based on improved Yolov5 neural network Download PDFInfo
- Publication number
- CN117456198A CN117456198A CN202311035820.XA CN202311035820A CN117456198A CN 117456198 A CN117456198 A CN 117456198A CN 202311035820 A CN202311035820 A CN 202311035820A CN 117456198 A CN117456198 A CN 117456198A
- Authority
- CN
- China
- Prior art keywords
- transmission line
- loss
- power transmission
- fault detection
- neural network
- 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
Links
- 230000005540 biological transmission Effects 0.000 title claims abstract description 77
- 238000001514 detection method Methods 0.000 title claims abstract description 53
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000007689 inspection Methods 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 6
- 238000000034 method Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 13
- 230000004927 fusion Effects 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000000137 annealing Methods 0.000 claims description 3
- 239000012633 leachable Substances 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 239000000203 mixture Substances 0.000 claims description 2
- 238000012423 maintenance Methods 0.000 abstract description 6
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000010606 normalization Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 235000019504 cigarettes Nutrition 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- 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
- G06T7/0004—Industrial image inspection
-
- 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
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Economics (AREA)
- Biophysics (AREA)
- Water Supply & Treatment (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a power transmission line fault detection method based on an improved Yolov5 neural network, which comprises the following steps: step 1, acquiring thousands of transmission line images by using a transmission line inspection information acquisition device; step 2, processing the acquired images to construct a training sample set and a test sample set; step 3, training and testing the convolutional neural network by using the training set sample and the testing set sample successively to obtain a trained YOLOv5 model for power transmission line fault detection; and 4, detecting the power transmission line image to be detected by using the trained power transmission line fault detection YOLOv5 model, and outputting a detection result. The invention can realize accurate fault detection of the power transmission line and obtain the fault category of the power transmission line; the line inspection and maintenance cost can be reduced, and the line inspection and maintenance efficiency can be improved.
Description
Technical Field
The application belongs to the technical field of power transmission line target detection, and particularly relates to a power transmission line fault detection method based on an improved Yolov5 neural network.
Background
The transmission line works in the field throughout the year and can also cross over urban economic zones, roads, bridges and other complex environments. When the urban construction is performed, large-sized mechanical vehicles often invade a transmission line protection area, and accidents such as line short circuit, tripping and power failure caused by discharge and the like due to mechanical touch or insufficient safety distance between the machinery and the transmission line occur. In addition, foreign matter invasion caused by natural factors such as bird flying, tree dumping, branch sliding and the like can also harm the safe and stable operation of the power transmission line. Therefore, the power transmission line needs to be monitored and evaluated, and the safe and stable operation of the line is ensured.
At present, along with the rapid development of intelligent video technology, there are also technologies for monitoring the intrusion of foreign matters into a power transmission line by adopting a video and image mode, wherein the Chinese patent is a power transmission line fault detection method, a power transmission line fault detection system and electronic equipment (application number: 201811547406.6), and the Chinese patent is a power transmission line fault detection method, a power transmission line fault detection device and a mobile terminal (application number: 201910165020.7), which all adopt an image processing mode to detect the faults of the power transmission line, but both lack distinction of fault types and cannot accurately evaluate the harm of different types of faults to the power transmission line; the influence of ice and snow coating, galloping and sag change of the power transmission line on the power transmission line is not eliminated, and a plurality of faults exist, so that improvement is urgently needed.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide the power transmission line fault detection method based on the improved Yolov5 neural network, which can realize the rapid and accurate identification of the fault type, improve the accuracy of power transmission line detection and is beneficial to the later maintenance and repair.
The above object of the present invention is achieved by the following technical solutions:
a transmission line fault detection method based on an improved Yolov5 neural network comprises the following steps:
step 1, acquiring thousands of transmission line images by using a transmission line inspection information acquisition device;
step 2, processing the acquired images to construct a training sample set and a test sample set;
step 3, training and testing the convolutional neural network by using the training set sample and the testing set sample successively to obtain a trained YOLOv5 model for power transmission line fault detection;
and 4, detecting the power transmission line image to be detected by using a trained power transmission line fault detection YOLOv5 model, and outputting a detection result.
In addition, the transmission line inspection information acquisition device in the step 1 adopts an unmanned aerial vehicle carrying shooting equipment.
Furthermore, step 2 comprises:
2.1, respectively carrying out gray scale mask mode treatment on the background in the acquired image;
2.2, marking the position frame of the power transmission line and the fault type of the power transmission line by using an image tool LabelImg, and generating an xml file with the same name as the image name; extracting the position frame of the power transmission line and the fault type information of the power transmission line marked in the XML file by using Python programming, and storing the extracted information into a txt format file, so that the txt format file can be directly identified and trained by a YOLO model to generate a sample data set; the fault types comprise strand breakage, strand scattering, intersection and foreign matter on the lead;
2.3, dividing the produced sample data set, wherein the sample ratio of the training set is 80%, and the sample ratio of the test set is 20%.
In step 3, the YOLOv5 neural network model is configured as follows:
the device comprises an input end, a main network, a neck network and a detection end; the input end is used for carrying out image preprocessing; the backbone network is used for extracting features; the neck network is used for feature fusion; the detection end is used for predicting the image characteristics, generating a boundary box and predicting the category.
Moreover, the backbone network comprises a CBS module, a Bottleneck module, a C3 module and an SPPF feature fusion structure; the C3 module blends the deformable convolution into YOLOv5 for feature extraction, and adds a new module named C3-DCN, which is obtained by modifying the C3 module, wherein Conv in C3 is changed into DCN.
The calculation method for adding the C3-DCN module comprises the following steps:
a convolution kernel R = { (-1, -1), (-1, -1), …, (0, 1) }, a standard convolution as in equation (1), a deformable convolution as in equation (2),
wherein y is 1 (p 0 ) Representing the output of a conventional convolution profile, p n For each position in R, w (p n ) Representing each p n Weights of (2); y is 2 (p 0 ) Feature map output, Δp, representing deformable convolution n Represents a leachable offset, where Δp n |n＝1,…,N},N＝|R|；
Δp during operation of the deformable convolution n Typically representing the fractional part, the formula needs to be modified by bilinear interpolation to the sampling position; as in equation (3),
x(p)＝∑ q G(q,p)·x(q) (3)
G(q,p)＝g(q x ,p x )·g(q y ,p y ) (4)
g(q,p)＝max(0,1-|q-p|) (5)
wherein p=p 0 +p n +Δp n Representing the position coordinates after the convolution kernel is shifted, q representing the spatial position in the feature map x. g (q) x ,p x )·g(q y ,p y ) Convolution kernel representing bilinear interpolation, x (q) is a feature mapThe value at the upper integer position, x (p), represents the bilinear interpolated value.
In the optimization training of the YOLOv5 model, the total loss function is used as a loss function, and the YOLOv5 model is trained by using an optimization gradient and performing back propagation. The total loss function of the model includes classification loss, confidence loss, and target bounding box loss. I.e. the total loss value of the model is:
Loss t ＝α/(α+β+γ)Loss c +β/(α+β+γ)Loss o +γ/(α+β+γ)Loss b (6)
wherein Loss is c The classification Loss is expressed by using BCE Loss, and only the classification Loss of the positive sample is calculated. Loss (Low Density) o Representing confidence Loss, also using BCE Loss, the Loss for all samples was calculated. Loss (Low Density) b The target boundary box loss is represented, the boundary box loss function adopts ciou_loss, and only the boundary box loss of a positive sample is calculated, and alpha, beta and gamma are weight coefficients.
Moreover, training parameters are set as follows: the initial momentum was 0.8, the initial learning rate was 0.01, the weight decay was 0.0005, the cosine annealing learning rate was 0.01, and 100 epochs were trained for each model.
The invention has the advantages and positive effects that:
1. the invention provides a detection method capable of detecting various types of power transmission line faults by utilizing an improved YOLOv5 model, and the method can realize accurate power transmission line fault detection and obtain power transmission line fault types; the line inspection and maintenance cost can be reduced, the line inspection and maintenance efficiency can be improved, and the line inspection and maintenance task completion process is more intelligent; the output end of the YOLOv5 is provided with three detection heads with different sizes, so that the prediction of three different scales can be realized.
2. The invention completes the data set production by using the image labeling tools LabelImg and python programs, and has simple process and easy operation; the training image of the gray background mask is used for manufacturing a sample data set, the image mask is equivalent to shielding the background area, the extraction of the candidate area of the target is completed, and the possible position (candidate area) of the target is highlighted as far as possible; the detection efficiency of the model can be further improved.
3. The improved loss function of the invention can lead the predicted frame to be more approximate to the real frame.
4. The traditional convolutional neural network uses a convolution kernel with a fixed size to extract the features, however, the constraint can lead to poor adaptability of a model and weak generalization, and the deformable convolution is integrated into YOLOv5 to extract the features. Because the integration of the deformable convolution can increase the calculated amount of the model and influence the detection speed, in order to improve the accuracy without excessively reducing the detection speed, a C3 module in the backbone network is replaced by a C3-DCN module, so that the detection speed is improved while the high accuracy is ensured.
Drawings
Fig. 1 is a flow chart of a transmission line fault detection method of the present invention;
FIG. 2 is a schematic diagram of the CBS module of the present invention;
FIG. 3 is a block diagram of a Bottleneck module of the present invention;
FIG. 4 is a schematic illustration of the addition location of the C3-DCN of the present invention in YOLOv 5;
fig. 5 is a schematic diagram of the SPPF module of the present invention.
Detailed Description
The structure of the present invention will be further described by way of examples with reference to the accompanying drawings. It should be noted that the present embodiments are illustrative and not restrictive.
Referring to fig. 1-5, the method for detecting the power transmission line fault based on the improved Yolov5 neural network comprises the following steps:
step 1, acquiring image information of a power transmission line;
thousands of (say about 3000) transmission line images are acquired by using a transmission line inspection information acquisition device, such as an unmanned aerial vehicle carrying shooting equipment.
Step 2, processing the acquired images to construct a training sample set and a test sample set;
1. respectively carrying out gray mask mode treatment on the background in the acquired image; specifically, all images are firstly converted into 3-channel gray images respectively, then the position information of a target object is read, and gray masks outside a rectangular area where targets exist in one image are processed, so that RGB pixel values of the gray masks are kept unchanged from an original image, and a training image of a gray background mask is obtained.
2. Labeling the position frame of the power transmission line and the fault type of the power transmission line by using an image tool LabelImg to generate an xml file with the same name as the image; extracting the position frame of the power transmission line and the fault type information of the power transmission line marked in the XML file by using Python programming, and storing the extracted information into a txt format file, so that the txt format file can be directly identified and trained by a YOLO model to generate a sample data set; the fault types comprise broken strands, scattered strands, crossed strands and foreign matters on the wires;
in the sample data set, the training set sample ratio is 80%, and the test set sample ratio is 20%.
And step 3, training and testing the convolutional neural network by using the training set sample and the testing set sample successively to obtain a trained YOLOv5 model. The optimizer is random gradient descent, the initial momentum is 0.8, the initial learning rate is 0.01, the weight attenuation is 0.0005, the cosine annealing learning rate is 0.01, and 100 epochs are trained by each model; and obtaining the optimal Yolov5 model until the loss, precision and recall rate of the model tend to be stable.
The YOLOv5 neural network model is composed of:
the method mainly comprises four parts, namely an Input end (Input), a Backbone network (Backbone), a Neck network (heck) and a detection end (Head), wherein the Input is an image preprocessing stage, the Backbone is an extraction feature stage, the heck is a feature fusion stage, and the Head predicts image features to generate a boundary frame and a prediction category.
1. Input terminal
The size of the input transmission line pictures is uniformly adjusted to 640 x 640 pixels. The preprocessing process comprises the steps of scaling an input picture to adapt to the input size of a network, and performing normalization, mosaic data enhancement, self-adaptive anchor frame and self-adaptive picture scaling preprocessing.
3. Backbone network
Carrying out convolution processing on the preprocessed transmission line image by utilizing a convolution module group of the feature extraction network to obtain a first feature extraction graph; performing Bottleneck convolution and splicing processing on the first feature extraction graph by using a CBS module of the feature extraction network to obtain a second feature extraction graph; and carrying out pooling and splicing treatment on the second feature extraction graph by utilizing the SPPF module of the feature extraction network to obtain a remarkable feature graph of the power transmission line. The feature map contains the positions and the categories of the defects of the transmission line in the image.
And extracting the characteristics of the power transmission line fault picture through a convolutional neural network. The backbone network mainly consists of several modules, namely a CBS module, a Bottleneck module, a C3 module and an SPPF feature fusion structure.
CBS module: this part consists of convolution, batch normalization and activation functions, see FIG. 2
The Bottleneck module consists of convolution, batch normalization and activation functions, see FIG. 3
And C3, module: the deformable convolution is integrated into YOLOv5 for feature extraction, and a new module named C3-DCN is added, which is obtained by modifying the C3 module, wherein Conv in C3 is changed to DCN.
A convolution kernel R = { (-1, -1), (-1, -1), …, (0, 1) }, a standard convolution as in equation (1), a deformable convolution as in equation (2),
wherein y is 1 (p 0 ) Representing the output of a conventional convolution profile, p n For each position in R, w (p n ) Representing each p n Weights of (2); y is 2 (p 0 ) Feature map output, Δp, representing deformable convolution n Represents a leachable offset, where Δp n |n＝1,…,N},N＝|R|。
Δp during operation of the deformable convolution n Typically representing the fractional part, the equation requires correction of the sampling position using bilinear interpolation. As in equation (3),
x(p)＝∑ q G(q,p)·x(q) (3)
G(q,p)＝g(q x ,p x )·g(q y ,p y ) (4)
g(q,p)＝max(0,1-|q-p|) (5)
wherein p=p 0 +p n +Δp n Representing the position coordinates after the convolution kernel is shifted, q representing the spatial position in the feature map x. g (q) x ,p x )·g(q y ,p y ) A convolution kernel representing bilinear interpolation, x (q) is a value at an integer position on the feature map, and x (p) represents a bilinear interpolated value.
And replacing one C3 module in the backbone network with a C3-DCN module. An alternative position is shown in fig. 4.
SPPF module: the part connects the input features in series, and then splices the input features through three largest convergence layers with the convolution kernel size of 5 multiplied by 5, carries out multi-scale fusion,
3. neck network
The neck network is the critical part between the backbone network and the output. In order to better support a plurality of detection heads at the output end to complete a multi-scale prediction task, the neck network part performs feature fusion on the feature map corresponding to the detection heads, so that the detection capability of the detection heads is improved. In this process, a structure of fpn+pan (Feature Pyramid Network + PathAggregation Network) is adopted. The FPN structure can enhance the fusion of features of different layers, and transfer deep semantic features to a shallow layer, so that semantic expression capacity on multiple scales is enhanced, and multi-scale prediction is realized. Meanwhile, the PAN also realizes feature fusion from bottom to top on the basis of FPN, the positioning information of the shallow layer is transferred to the deep layer, the positioning capability on multiple scales is enhanced, and the detection capability of the model is further improved.
The Neck end adopts a double-tower tactic fusion of the FPN and the PAN to respectively process the feature graphs, so that the output feature graphs have strong semantic features and strong positioning features.
4. An output terminal
The output end of YOLOv5 has three convolution layers with different sizes, and can realize the prediction of three different scales. After the input image passes through the feature extraction layer, 8 times, 16 times and 32 times of downsampling feature images are obtained. In the cigarette appearance failure detection, the sizes of the output feature layers are 80×80×54, 40×40×54, and 20×20×54, respectively.
Wherein, the feature map with the size of 80×80×54 contains the majority of low-level layer features for small target detection; the low-level and high-level characteristic information of the characteristic diagram with the size of 40 multiplied by 54 are equivalent in proportion, and are used for medium target detection; the feature map of size 20 x 54 contains the majority of advanced layer features for large target detection.
5. Loss function
In the optimization training of the YOLOv5 model, the predicted frame is made closer to the real frame. The loss function of the model is optimized in all aspects, namely classification loss, confidence loss and target bounding box loss. I.e. the total loss value of the model is:
Loss t ＝α/(α+β+γ)Loss c +β/(α+β+γ)Loss o +γ/(α+β+γ)Loss b (6)
wherein Loss is c The classification Loss is expressed by using BCE Loss, and only the classification Loss of the positive sample is calculated. Loss (Low Density) o Representing confidence Loss, also using BCE Loss, the Loss for all samples was calculated. Loss (Low Density) b The target boundary box loss is represented, the boundary box loss function adopts ciou_loss, and only the boundary box loss of a positive sample is calculated, and alpha, beta and gamma are weight coefficients.
And step four, detecting the power transmission line image to be detected by using the trained YOLOv5 model, and outputting a detection result.
Configuring an experimental environment; the operating system selects ubuntu20.04; CPU selects Inter (R)CPU E5-2650v4; GPU selects Tesla P100-PCIE; the memory is 12GB; the deep learning framework is pytorch 1.8; python version 3.8.
Although the embodiments of the present invention and the accompanying drawings have been disclosed for illustrative purposes, those skilled in the art will appreciate that: various substitutions, changes and modifications are possible without departing from the spirit of the invention and the appended claims, and therefore the scope of the invention is not limited to the embodiments and the disclosure of the drawings.
Claims (8)
1. A transmission line fault detection method based on an improved Yolov5 neural network comprises the following steps:
step 1, acquiring thousands of transmission line images by using a transmission line inspection information acquisition device;
step 2, processing the acquired images to construct a training sample set and a test sample set;
step 3, training and testing the convolutional neural network sequentially by using the training set sample and the testing set sample to obtain a trained power transmission line fault detection YOLOv5 model;
and 4, detecting the power transmission line image to be detected by using the trained power transmission line fault detection YOLOv5 model, and outputting a detection result.
2. The improved Yolov5 neural network-based transmission line fault detection method of claim 1, wherein the method comprises the steps of: in the step 1, an unmanned aerial vehicle carrying shooting equipment is adopted as the transmission line inspection information acquisition device.
3. The improved Yolov5 neural network-based transmission line fault detection method of claim 1, wherein the method comprises the steps of: the step 2 comprises the following steps:
2.1, respectively carrying out gray scale mask type processing on the background in the acquired image: firstly, all images are respectively converted into 3-channel gray images, then, the position information of a target power transmission line is read, and for a rectangular area with the target power transmission line in one image, RGB pixel values of the rectangular area are kept unchanged from an original image, so that a training image of a gray background mask is obtained.
2.2, marking the position frame of the power transmission line and the fault type of the power transmission line by using an image tool LabelImg, and generating an xml file with the same name as the image name; extracting the position frame of the power transmission line and the fault type information of the power transmission line marked in the XML file by using Python programming, and storing the extracted information into a txt format file, so that the txt format file can be directly identified and trained by a YOLO model to generate a sample data set; the fault types comprise strand breakage, strand scattering, intersection and foreign matter on the lead;
2.3, dividing the produced sample data set, wherein the sample ratio of the training set is 80%, and the sample ratio of the test set is 20%.
4. The improved Yolov5 neural network-based transmission line fault detection method of claim 1, wherein the method comprises the steps of: in step 3, the YOLOv5 neural network model is configured as follows: the device comprises an input end, a main network, a neck network and a detection end; the input end is used for carrying out image preprocessing; the backbone network is used for extracting features; the neck network is used for feature fusion; the detection end is used for predicting the image characteristics, generating a boundary box and predicting the category.
5. The improved Yolov5 neural network-based transmission line fault detection method of claim 4, wherein the method comprises the steps of: the backbone network comprises a CBS module, a Bottleneck module, a C3 module and an SPPF feature fusion structure; the C3 module blends the deformable convolution into YOLOv5 for feature extraction, and adds a new module named C3-DCN, which is obtained by modifying the C3 module, wherein Conv in C3 is changed into DCN.
6. The improved Yolov5 neural network-based transmission line fault detection method of claim 5, wherein the method comprises the steps of: the calculation method for adding the C3-DCN module comprises the following steps:
a convolution kernel R = { (-1, -1), (-1, -1), …, (0, 1) }, a standard convolution as in equation (1), a deformable convolution as in equation (2),
wherein y is 1 (p 0 ) Representing the output of a conventional convolution profile, p n For each position in R, w (p n ) Representing each p n Weights of (2); y is 2 (p 0 ) Feature map output, Δp, representing deformable convolution n Represents a leachable offset, where Δp n |n＝1，…，N}，N＝|R|；
Δp during operation of the deformable convolution n Typically representing the fractional part, the formula needs to be modified by bilinear interpolation to the sampling position; as in equation (3),
x(p)＝∑ q G(q，p)·x(q) (3)
G(q，p)＝g(q x ，p x )·g(q y ，p y ) (4)
g(q，p)＝max(0，1-|q-p|) (5)
wherein p=p 0 +p n +Δp n Representing position coordinates after the convolution kernel is shifted, and q represents a spatial position in the feature map x; g (q) x ，p x )·g(q y ，p y ) A convolution kernel representing bilinear interpolation, x (q) is a value at an integer position on the feature map, and x (p) represents a bilinear interpolated value.
7. The improved Yolov5 neural network-based transmission line fault detection method of claim 1, wherein the method comprises the steps of: in the optimized training of the YOLOv5 model, the total loss function is adopted as a loss function, and the YOLOv5 model is trained by utilizing an optimized gradient and performing back propagation; the total loss function of the model includes classification loss, confidence loss, and target bounding box loss;
Loss t ＝α/(α+β+γ)Loss c +β/(α+β+γ)Loss o +γ/(α+β+γ)Loss b (6)
wherein Loss is c Representing the classification Loss, wherein BCE Loss is adopted, and only the classification Loss of the positive sample is calculated; loss (Low Density) o Representing confidence Loss, also using BCE Loss, calculated as Loss for all samples; loss (Low Density) b The target boundary box loss is represented, the boundary box loss function adopts ciou_loss, and only the boundary box loss of a positive sample is calculated, and alpha, beta and gamma are weight coefficients.
8. The improved Yolov5 neural network-based transmission line fault detection method of claim 1, wherein the method comprises the steps of: the training parameters are set as follows: the initial momentum was 0.8, the initial learning rate was 0.01, the weight decay was 0.0005, the cosine annealing learning rate was 0.01, and 100 epochs were trained for each model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311035820.XA CN117456198A (en) | 2023-08-17 | 2023-08-17 | Power transmission line fault detection method based on improved Yolov5 neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311035820.XA CN117456198A (en) | 2023-08-17 | 2023-08-17 | Power transmission line fault detection method based on improved Yolov5 neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117456198A true CN117456198A (en) | 2024-01-26 |
Family
ID=89578842
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311035820.XA Pending CN117456198A (en) | 2023-08-17 | 2023-08-17 | Power transmission line fault detection method based on improved Yolov5 neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117456198A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117671602A (en) * | 2024-01-31 | 2024-03-08 | 吉林省中农阳光数据有限公司 | Farmland forest smoke fire prevention detection method and device based on image recognition |
-
2023
- 2023-08-17 CN CN202311035820.XA patent/CN117456198A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117671602A (en) * | 2024-01-31 | 2024-03-08 | 吉林省中农阳光数据有限公司 | Farmland forest smoke fire prevention detection method and device based on image recognition |
| CN117671602B (en) * | 2024-01-31 | 2024-04-05 | 吉林省中农阳光数据有限公司 | Farmland forest smoke fire prevention detection method and device based on image recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113705478B (en) | Mangrove single wood target detection method based on improved YOLOv5 | |
| CN113780296B (en) | Remote sensing image semantic segmentation method and system based on multi-scale information fusion | |
| CN111598098B (en) | Water gauge water line detection and effectiveness identification method based on full convolution neural network | |
| Tian et al. | Multiscale building extraction with refined attention pyramid networks | |
| CN115546630A (en) | Construction site extraction method and system based on remote sensing image characteristic target detection | |
| CN110991430B (en) | Ground feature identification and coverage rate calculation method and system based on remote sensing image | |
| CN117456198A (en) | Power transmission line fault detection method based on improved Yolov5 neural network | |
| CN112507861A (en) | Pedestrian detection method based on multilayer convolution feature fusion | |
| CN115713537A (en) | Optical remote sensing image cloud and fog segmentation method based on spectral guidance and depth attention | |
| CN113780270A (en) | Target detection method and device | |
| CN115984543A (en) | Target detection algorithm based on infrared and visible light images | |
| CN111753666A (en) | Method and system for detecting faults of small targets in power transmission line and storage medium | |
| CN113077438B (en) | Cell nucleus region extraction method and imaging method for multi-cell nucleus color image | |
| CN111931721B (en) | Method and device for detecting color and number of annual inspection label and electronic equipment | |
| CN116994161A (en) | Insulator defect detection method based on improved YOLOv5 | |
| CN116543295A (en) | Lightweight underwater target detection method and system based on degradation image enhancement | |
| CN116433661A (en) | Method, device, equipment and medium for detecting semiconductor wafer by multitasking | |
| CN115330703A (en) | Remote sensing image cloud and cloud shadow detection method based on context information fusion | |
| CN115761223A (en) | Remote sensing image instance segmentation method by using data synthesis | |
| CN115641498A (en) | Medium-term rainfall forecast post-processing correction method based on space multi-scale convolutional neural network | |
| CN115861160A (en) | Method and device for detecting surface defects of power interface of mobile phone and storage medium | |
| CN112861869A (en) | Sorghum lodging image segmentation method based on lightweight convolutional neural network | |
| CN116740569B (en) | Deep learning-based snowfall area cloud detection system | |
| CN117557775B (en) | Substation power equipment detection method and system based on infrared and visible light fusion | |
| CN116030347B (en) | High-resolution remote sensing image building extraction method based on attention network |
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 |