Power transmission line stockbridge damper defect detection method based on multiple tasks

Abstract

The invention provides a multi-task-based power transmission line stockbridge damper defect detection method, which comprises the following steps: acquiring an original power transmission line stockbridge damper image, and generating a data sample close to a real fault sample; a multi-task detection network model is established, a classification loss function and a segmentation loss function are constructed for the multi-task detection network model, and the established multi-task detection network model comprises a shared feature layer, classification branches and segmentation branches; based on a data sample, training the multi-task detection network model by adopting an alternate training mode to obtain a trained multi-task detection network model, and in the training process, adaptively adjusting weight parameters by adopting an adaptive dynamic weighting strategy; and inputting a to-be-detected image of the vibration damper of the power transmission line into the trained multi-task detection network model to obtain segmentation positioning and defect detection results of the vibration damper of the power transmission line. According to the invention, the problem of poor robustness and detection performance in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a multi-task based method for detecting defects of a power transmission line anti-vibration hammer. Background Art

[0002] Intelligent detection of key components of transmission lines is the key to improving the level of intelligent power grids. As a common component of transmission lines, the shock-absorbing hammer plays a role in reducing the vibration of the wires caused by wind. Due to long-term exposure to the outdoor environment, the shock-absorbing hammer is prone to various faults. Once the shock-absorbing hammer fails, the wires will bend periodically during multiple vibrations, which can easily cause fatigue damage, leading to power outages. In severe cases, large-scale power outages will occur. Therefore, in order to ensure the normal operation of the shock-absorbing hammer, it is necessary to perform defect detection on the shock-absorbing hammer.

[0003] The existing technology mainly uses manual operations to detect defects on shock-absorbing hammers, which is difficult to operate and has low detection efficiency. Although there are some defect detection models based on image processing, there are still problems with robustness and poor detection performance. Summary of the invention

[0004] The object of the present invention is to provide a multi-task based method for detecting defects of power transmission line anti-vibration hammers, so as to solve the problems of poor robustness and detection performance of the prior art.

[0005] A multi-task based transmission line anti-vibration hammer defect detection method, comprising: Step S1, obtaining an original transmission line anti-vibration hammer image, performing supervised data enhancement processing on the original transmission line anti-vibration hammer image, and then sending the processed data into a generative adversarial network to generate a data sample close to a real fault sample; Step S2, establishing a multi-task detection network model, and constructing a classification loss function and a segmentation loss function for the multi-task detection network model, the established multi-task detection network model includes a shared feature layer, a classification branch and a segmentation branch, the shared feature layer includes a MobileNetV3 network and an improved void space pyramid pooling module, the classification branch adopts a YOLOv5s classification network, and the segmentation branch adopts a DeepLabV3+ segmentation network, the shared feature layer is used to extract image features, and the extracted image features are respectively sent to the classification branch and the segmentation branch to realize border regression and obtain image classification results and image segmentation results; Step S3, based on the data samples, the multi-task detection network model is trained by an alternating training method to obtain a trained multi-task detection network model, and during the training process, an adaptive dynamic weighting strategy is used to adaptively adjust the weight parameters; Step S4, inputting the transmission line anti-vibration hammer image to be detected into the trained multi-task detection network model to obtain the segmentation positioning and defect detection results of the transmission line anti-vibration hammer.

[0006] According to the transmission line anti-vibration hammer defect detection method provided by the present invention, the multi-task detection network model includes a shared feature layer, a classification branch and a segmentation branch. The shared feature layer in the present invention includes a MobileNetV3 network and an improved atrous space pyramid pooling (ASPP) module, which reduces network parameters, increases the detection receptive field, speeds up detection efficiency, and strengthens feature extraction capabilities. The multi-task detection network model integrates the YOLOv5s classification network and the DeepLabV3+ segmentation network, and designs a classification loss function and a segmentation loss function, which can better capture boundary and shape features and improve detection robustness and detection performance. In addition, in the process of training the multi-task detection network model, the present invention adopts an adaptive dynamic weighting strategy to adaptively adjust weight parameters, which can effectively balance the learning between different tasks, avoid the subjectivity of artificial weighting, improve the training effect, and further improve the detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A schematic flow chart of the method for detecting defects of a shock-absorbing hammer in a power transmission line provided by the present invention. DETAILED DESCRIPTION

[0008] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0009] See also Figure 1 The embodiment of the present invention provides a method for detecting shock-absorbing hammer defects in a power transmission line, comprising steps S1 to S4.

[0010] Step S1, obtaining an original transmission line anti-vibration hammer image, performing supervised data enhancement processing on the original transmission line anti-vibration hammer image, and then sending the processed data into a generative adversarial network to generate a data sample close to a real fault sample.

[0011] Specifically, the original transmission line anti-shock hammer image is obtained, and the original transmission line anti-shock hammer image is extended by brightening, adding noise, and randomly rotating. The processed data is then sent to a generative adversarial network to generate data samples that are close to real fault samples.

[0012] Step S2, establish a multi-task detection network model, and construct a classification loss function and a segmentation loss function for the multi-task detection network model. The established multi-task detection network model includes a shared feature layer, a classification branch and a segmentation branch. The shared feature layer includes a MobileNetV3 network and an improved hole space pyramid pooling module. The classification branch adopts a YOLOv5s classification network, and the segmentation branch adopts a DeepLabV3+ segmentation network. The shared feature layer is used to extract image features, and the extracted image features are respectively sent to the classification branch and the segmentation branch to realize border regression and obtain image classification results and image segmentation results.

[0013] Among them, in the shared feature layer, the data sample is sent to the MobileNetV3 network to extract the feature map of the data sample, and then the feature map of the data sample is sent to the improved void space pyramid pooling module to extract multi-scale features. Among them, the improved void space pyramid pooling module includes 1 1×1 convolution, 4 void convolutions and 1 pooling layer, and the void rates of the 4 void convolutions are set to 7, 10, 13, and 16 respectively, so as to enhance the ability of the model to extract multi-scale features, and perform feature fusion and dimensionality compression on the feature map of the data sample through 1×1 convolution, so as to extract image features.

[0014] Specifically, the extracted image features are sent to the YOLOv5s classification network to fuse features at different levels, and the bounding box regression is realized through the output layer of the YOLOv5s classification network to obtain the image classification result; The extracted image features are sent to the DeepLabV3+ segmentation network to obtain multi-scale fused feature information, and then the image segmentation results are output after channel adjustment and size adjustment.

[0015] Among them, the expression of the classification loss function is: ; ; in, L c represents the classification loss function, A represents the intersection area of ​​the predicted box and the real box of the shock-absorbing hammer, B represents the union area of ​​the predicted box and the true box of the shock-absorbing hammer, is the coverage area correlation coefficient, is the center point distance correlation coefficient, is the aspect ratio correlation coefficient, c Represents the diagonal length distance of the minimum enclosed area that contains both the predicted box and the true box. Represents the Euclidean distance between the center point of the predicted box and the center point of the real box, vis the aspect ratio consistency measurement parameter, Indicates the width of the real box, Indicates the height of the real frame, w Indicates the width of the prediction box, h Indicates the height of the prediction box, IOU It is the intersection ratio of the intersection of the predicted box and the true box of the shock-absorbing hammer and the union of the predicted box and the true box of the shock-absorbing hammer.

[0016] In this embodiment, the segmentation loss function introduces weighted loss on the basis of the multi-category cross entropy loss function, which can solve the problem of poor segmentation effect caused by the scarcity of defect samples in the anti-vibration hammer data set. Specifically, the expression of the segmentation loss function is: ; in, L se represents the segmentation loss function, N Represents the total number of pixels. C Indicates the total number of shock-absorbing hammer categories, Indicates the type of shock-absorbing hammer is j No. i Pixel annotations, Indicates the type of shock-absorbing hammer is j No. i Pixel prediction, w j Indicates the type of shock-absorbing hammer is j The weight coefficient of .

[0017] Step S3, based on the data samples, the multi-task detection network model is trained by an alternating training method to obtain a trained multi-task detection network model, and during the training process, an adaptive dynamic weighting strategy is used to adaptively adjust the weight parameters.

[0018] Among them, the total loss function of the multi-task detection network model is L The expression is: ; in, is the weight parameter of the classification branch, is the weight parameter of the segmentation branch; In the process of adaptively adjusting weight parameters using adaptive dynamic weighting strategy, the following equation is satisfied: ; ; in, For the t The weight parameter of the classification branch at the iteration, For the tThe weight parameter of the split branch at the iteration, For the classification branch t -1 The ratio between the validation set loss and the training set loss at iteration 1, specifically t -1 iteration, the validation set loss is divided by the t -1 training set loss, To separate the branches t -1 The ratio between the validation set loss and the training set loss at iteration 1, specifically t -1 iteration, the validation set loss is divided by t -1 training set loss at iteration 1.

[0019] In this embodiment, the multi-task detection network model is trained by an alternating training method, which specifically includes: First, the training set of the classification branch and the training set of the segmentation branch are constructed. The training set of the classification branch contains the defect target frame and image label of the shockproof hammer image, and the training set of the segmentation branch includes the defect area. Then, the number of training iterations is determined. Each iteration training process is sampled from the training set of the classification branch. m Minimum batch samples L Y Train the classification branch by minimizing the classification loss function L c Update the model parameters of the classification branch, and then sample from the training set of the segmentation branch m Minimum batch samples L D Train the segmentation branch by minimizing the segmentation loss function L se Update the model parameters of the segmentation branch; finally, complete the alternating training through multiple iterations.

[0020] Step S4, inputting the transmission line anti-vibration hammer image to be detected into the trained multi-task detection network model to obtain the segmentation positioning and defect detection results of the transmission line anti-vibration hammer.

[0021] In summary, according to the transmission line anti-vibration hammer defect detection method provided by the present invention, the multi-task detection network model includes a shared feature layer, a classification branch and a segmentation branch. The shared feature layer in the present invention includes a MobileNetV3 network and an improved atrous space pyramid pooling (ASPP) module, which reduces network parameters, increases the receptive field of detection, speeds up detection efficiency, and enhances feature extraction capabilities. The multi-task detection network model integrates the YOLOv5s classification network and the DeepLabV3+ segmentation network, and designs a classification loss function and a segmentation loss function, which can better capture boundary and shape features and improve detection robustness and detection performance. In addition, in the process of training the multi-task detection network model, the present invention adopts an adaptive dynamic weighting strategy to adaptively adjust weight parameters, which can effectively balance the learning between different tasks, avoid the subjectivity of artificial weighting, improve the training effect, and further improve the detection performance.

[0022] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0023] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A multi-task based transmission line anti-vibration hammer defect detection method, characterized in that: include: Step S1, obtaining an original transmission line anti-vibration hammer image, performing supervised data enhancement processing on the original transmission line anti-vibration hammer image, and then sending the processed data into a generative adversarial network to generate a data sample close to a real fault sample; Step S2, establishing a multi-task detection network model, and constructing a classification loss function and a segmentation loss function for the multi-task detection network model, the established multi-task detection network model includes a shared feature layer, a classification branch and a segmentation branch, the shared feature layer includes a MobileNetV3 network and an improved void space pyramid pooling module, the classification branch adopts a YOLOv5s classification network, and the segmentation branch adopts a DeepLabV3+ segmentation network, the shared feature layer is used to extract image features, and the extracted image features are respectively sent to the classification branch and the segmentation branch to realize border regression and obtain image classification results and image segmentation results; Step S3, based on the data samples, the multi-task detection network model is trained by an alternating training method to obtain a trained multi-task detection network model, and during the training process, an adaptive dynamic weighting strategy is used to adaptively adjust the weight parameters; Step S4, inputting the transmission line anti-vibration hammer image to be detected into the trained multi-task detection network model to obtain the segmentation positioning and defect detection results of the transmission line anti-vibration hammer.

2. The multi-task based transmission line anti-vibration hammer defect detection method according to claim 1 is characterized in that: In the shared feature layer, the data samples are sent to the MobileNetV3 network to extract the feature maps of the data samples, and then the feature maps of the data samples are sent to the improved dilated space pyramid pooling module to extract multi-scale features, wherein the improved dilated space pyramid pooling module includes 1 1×1 convolution, 4 dilated convolutions and 1 pooling layer, and the dilated rates of the 4 dilated convolutions are set to 7, 10, 13, and 16, respectively. The feature maps of the data samples are fused and dimensionally compressed through 1×1 convolution, thereby extracting image features.

3. The multi-task based transmission line anti-vibration hammer defect detection method according to claim 2 is characterized in that: The extracted image features are sent to the YOLOv5s classification network to fuse features at different levels, and the bounding box regression is realized through the output layer of the YOLOv5s classification network to obtain the image classification result; The extracted image features are sent to the DeepLabV3+ segmentation network to obtain multi-scale fused feature information, and then the image segmentation results are output after channel adjustment and size adjustment.

4. The multi-task based transmission line anti-vibration hammer defect detection method according to claim 3 is characterized in that: The expression of the classification loss function is: ； ； in, L c represents the classification loss function, A represents the intersection area of ​​the predicted box and the real box of the shock-absorbing hammer, B represents the union area of ​​the predicted box and the true box of the shock-absorbing hammer, is the coverage area correlation coefficient, is the center point distance correlation coefficient, is the aspect ratio correlation coefficient, c Represents the diagonal length distance of the minimum enclosed area that contains both the predicted box and the true box. Represents the Euclidean distance between the center point of the predicted box and the center point of the real box, v is the aspect ratio consistency measurement parameter, Indicates the width of the real box, Indicates the height of the real frame, w Indicates the width of the prediction box, h Indicates the height of the prediction box, IOU It is the intersection ratio of the intersection of the predicted box and the true box of the shock-absorbing hammer and the union of the predicted box and the true box of the shock-absorbing hammer.

5. The multi-task based transmission line anti-vibration hammer defect detection method according to claim 4 is characterized in that: The expression of the segmentation loss function is: ； in, L se represents the segmentation loss function, N Represents the total number of pixels. C Indicates the total number of shock-absorbing hammer categories, Indicates the type of shock-absorbing hammer is j No. i Pixel annotations, Indicates the type of shock-absorbing hammer is j No. i Pixel prediction, w j Indicates the type of shock-absorbing hammer is j The weight coefficient of .

6. The multi-task based transmission line anti-vibration hammer defect detection method according to claim 5, characterized in that: Total loss function of the multi-task detection network model L The expression is: ； in, is the weight parameter of the classification branch, is the weight parameter of the segmentation branch; In the process of adaptively adjusting weight parameters using adaptive dynamic weighting strategy, the following equation is satisfied: ； ； in, For the t The weight parameter of the classification branch at the iteration, For the t The weight parameter of the split branch at the iteration, For the classification branch t -The ratio between the validation set loss and the training set loss at 1 iteration, To separate the branches t -The ratio between the validation set loss and the training set loss at 1 iteration.

7. The multi-task based transmission line anti-vibration hammer defect detection method according to claim 6 is characterized in that: The multi-task detection network model is trained by adopting an alternating training method, specifically including: First, the training set of the classification branch and the training set of the segmentation branch are constructed. The training set of the classification branch contains the defect target frame and image label of the shockproof hammer image, and the training set of the segmentation branch includes the defect area. Then, the number of training iterations is determined. Each iteration training process is sampled from the training set of the classification branch. m Minimum batch samples L Y Train the classification branch by minimizing the classification loss function L c Update the model parameters of the classification branch, and then sample from the training set of the segmentation branch m Minimum batch samples L D Train the segmentation branch by minimizing the segmentation loss function L se Update the model parameters of the segmentation branch; finally, complete the alternating training through multiple iterations.

Images