Post porcelain insulator pollution degree detection method and system

Abstract

The invention relates to a post porcelain insulator pollution degree detection method and system, and the method comprises the following steps: obtaining an original image of an insulator, detecting and segmenting an insulator body region in the original image of the insulator through an improved YOLOv8-seg model, and carrying out the filtering processing, and obtaining a visible light image of the insulator body region; extracting RGB three-dimensional color features of the visible light image of the insulator body area; and taking the humidity feature and the RGB three-dimensional color feature as the input of an MLP neural network, and detecting the pollution degree of the insulator body area through the MLP neural network to obtain an insulator pollution degree detection result. Compared with the prior art, the insulator pollution degree identification accuracy is improved, and manpower and material resources are saved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment condition monitoring and maintenance, and particularly to a method and system for detecting the pollution degree of post porcelain insulators. Background Art

[0002] Post porcelain insulators play a crucial role in the power system. They are mainly used to support and isolate the conductors of high-voltage transmission lines. Due to their excellent mechanical strength and electrical insulation performance, porcelain insulators are widely used in various power facilities. The porcelain material can not only withstand extreme weather conditions and environmental pollution but also effectively prevent electric arcs and flashovers. However, with the increase in the service life, the surface of the insulator is prone to accumulate dirt. In rainy weather, its insulation performance will decline, increasing the risk of the power system.

[0003] Currently, the detection of post porcelain insulator pollution is mainly divided into contact detection methods and non-contact detection methods. The traditional contact manual detection method not only consumes a large amount of manpower, material resources and financial resources, but also has certain blindness and lag in maintenance, and cannot meet the requirements of real-time detection and timely alarm.

[0004] With the development of image processing technology and artificial intelligence technology, many non-contact detection methods have emerged in recent years, but they generally have problems such as poor universality, inaccurate image segmentation, and difficulty in efficiently processing a large amount of data. For example, patent application CN115700375A discloses an insulator pollution detection method, device, equipment and medium. The method includes: obtaining an infrared image of the insulator to be detected and environmental data of the corresponding acquisition environment when collecting the infrared image; respectively extracting the infrared feature data of the infrared image and the environmental feature data of the environmental data; performing feature fusion on the infrared feature data and the environmental feature data to obtain target fusion feature data; and determining the target pollution level of the insulator to be detected according to the target fusion feature data. This method improves the accuracy of insulator pollution detection, but it is necessary to fuse different environmental data to obtain fused environmental data and extract environmental feature data from the fused environmental data, which is difficult to efficiently process a large amount of data.

[0005] Therefore, it is necessary to seek a fast and effective algorithm to process the massive data of insulator inspection in real time, accurately identify the pollution degree of insulators, and ensure the reliable operation of insulators. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide a method and system for detecting the pollution degree of post porcelain insulators, which improves the accuracy of identifying the pollution degree of insulators and saves manpower and material resources.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A method for detecting the contamination degree of post insulator, comprising the following steps:

[0009] Obtain the original image of the insulator, detect and segment the insulator body area in the original image of the insulator through an improved YOLOv8-seg model, and perform filtering processing to obtain the visible light image of the insulator body area;

[0010] Extract the RGB three-dimensional color features of the visible light image of the insulator body area, and the RGB three-dimensional color features are obtained by screening and dimension reduction of the RGB image features of the visible light image of the insulator body area through the mRMR feature selection method;

[0011] Take the humidity feature and the RGB three-dimensional color features as the input of the MLP neural network, and detect the contamination degree of the insulator body area through the MLP neural network to obtain the detection result of the insulator contamination degree.

[0012] Further, the humidity feature is an environmental factor, and the humidity feature is obtained through a hygrometer.

[0013] Further, the improved YOLOv8-seg model is provided with an SPPFCSPC module, a CBAM attention module and a plurality of convolutional modules in the backbone part, and the rest except the first convolutional module are variable kernel convolutional modules.

[0014] Further, the CBAM attention module includes a channel attention unit and a spatial attention unit connected in sequence. The channel attention unit is used to enhance the feature representation of the humidity feature and the screened RGB color features in the channel dimension, and the spatial attention unit is used to enhance the feature representation of the humidity feature and the screened RGB color features in the spatial dimension.

[0015] Further, the RGB image features are used to characterize the color difference, and the RGB image features are six-dimensional feature vectors, including the color average values and standard deviations of the red, green, and blue channels.

[0016] Further, the RGB three-dimensional color features include the color average values of the red, green, and blue channels in the visible light image of the insulator body area.

[0017] Further, the output value of the MLP neural network is the equivalent salt deposit density of the contaminated insulator.

[0018] Further, the MLP neural network adopts a single-layer hidden layer network.

[0019] Further, when training the MLP neural network, the ratio range of the equivalent salt deposit density to the non-soluble deposit density on the insulator in the training images is fixed.

[0020] According to another aspect of the present invention, there is provided a pollution degree detection system for post insulators, including:

[0021] A visible light image acquisition module for the insulator body area, which is used to acquire the original image of the insulator, detect and segment the insulator body area in the original image of the insulator through an improved YOLOv8-seg model, and perform filtering processing to obtain the visible light image of the insulator body area;

[0022] A color feature screening and dimensionality reduction module, which is used to extract the RGB three-dimensional color features of the visible light image of the insulator body area, and the RGB three-dimensional color features are obtained by screening and dimensionality reduction of the RGB image features of the visible light image of the insulator body area through the mRMR feature selection method;

[0023] A pollution degree detection module for the insulator, which is used to use the humidity feature and the RGB color features of the visible light image of the insulator body area after screening as the input of the MLP neural network, and detect the pollution degree of the insulator body area through the MLP neural network to obtain the pollution degree detection result of the insulator.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. The present invention detects and segments the insulator body area in the original image of the insulator through an improved YOLOv8-seg model and performs filtering processing to obtain the visible light image of the insulator body area, extracts the RGB color features of the visible light image of the insulator body area, screens and reduces the dimensionality of the RGB color features of the visible light image of the insulator body area through the mRMR feature selection method. Since the pollution color matches the humidity feature, the humidity feature and the RGB color features of the visible light image of the insulator body area after screening are used as the input of the MLP neural network, and the pollution degree of the insulator body area is detected through the MLP neural network to obtain the pollution degree detection result of the insulator, improving the accuracy of insulator pollution degree identification and saving manpower and material resources.

[0026] 2. The lightweight design of the present invention is achieved by replacing the convolutional modules in the YOLOv8-seg model except the first convolutional module with variable kernel convolutional modules. The CBAM attention module is added to the YOLOv8-seg model to enable the network to focus on the key feature regions in the image, improve the detection accuracy, enhance the feature representation of the enhanced humidity feature and the screened RGB color feature in the channel dimension and the spatial dimension, focus on the image feature regions, and replace the SPPF module with the SPPFCSPC module, thereby improving the detection accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 FIG. is a schematic flowchart of a method for detecting the pollution degree of post insulator;

[0028] Figure 2 FIG. is a schematic diagram of the scores of each color feature quantity during mRMR screening;

[0029] Figure 3 FIG. is a schematic diagram of the prediction errors for different numbers of hidden layers and nodes. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives the detailed implementation manner and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0031] English abbreviations involved:

[0032] Convolutional Block Attention Module: CBAM

[0033] Multilayer Perceptron: MLP

[0034] Equivalent Salt Deposit Density: ESDD

[0035] Non Soluble Deposit Density: NSDD

[0036] Mean Square Error: MSE

[0037] Max-Relevance and Min-Redundancy: mRMR

[0038] Mean Average Precision, mAP

[0039] Giga Floating-point Operations Per Second, GFLOPs

[0040] Spatial Pyramid Pooling Fast, SPPF

[0041] Spatial Pyramid Pooling Faster Cross-Stage Partial Concat, SPPFCSPC

[0042] Example 1

[0043] This example provides a method for detecting the contamination degree of post insulator, as Figure 1 shown, including the following steps:

[0044] S1. Obtain the original insulator image, detect and segment the insulator body area in the original insulator image through the improved YOLOv8-seg model, and perform filtering processing to obtain the visible light image of the insulator body area.

[0045] Detect and segment the insulator body area in the original insulator image through the improved YOLOv8-seg model. The improvement process of the improved YOLOv8-seg model includes the following steps:

[0046] Replace the convolutional (Conv) modules in the YOLOv8-seg model except the first convolutional (Conv) module with the adjustable kernel convolutional (AKConv) module to achieve lightweight design.

[0047] Add the CBAM attention module to the YOLOv8-seg model to make the network focus on the key feature areas in the image and improve the detection accuracy. The CBAM attention module includes a channel attention unit and a spatial attention unit. The channel attention unit is used to enhance the feature representation of the humidity feature and the filtered RGB color feature in the channel dimension, and the spatial attention unit is used to enhance the feature representation of the humidity feature and the filtered RGB color feature in the spatial dimension, focusing on the image feature area, thereby improving the detection accuracy.

[0048] Replace the SPPF module with the SPPFCSPC module to increase the model residuals and assist in the optimization and feature extraction of multi-scale object detection. The SPPFCSPC module combines the advantages of spatial pyramid pooling and cross-stage partial network, and through multi-scale pooling and grouped convolution, realizes efficient feature extraction and multi-scale feature fusion.

[0049] For the above three improvement methods, this embodiment conducts strict ablation experiments on them, and uses the same test set to experiment with various combined models, ensuring the fairness of the experiment. The experimental results are shown in Table 1.

[0050] Table 1 Ablation Experiment Results

[0051] Baseline model AKConv CBAM SPPFCSPC mAP Number of parameters GFLOPs √ 0.95693 11779987 42.4 √ √ 0.96806 10292023 41.8 √ √ 0.975 24971219 106.5 √ √ 0.97217 18205331 47.6 √ √ √ 0.98231 23704793 103.3 √ √ √ 0.97522 16892542 45.8 √ √ √ 0.98954 31618101 111.6 √ √ √ √ 0.99395 29385135 110.2

[0052] The mAP evaluation index is used to measure the performance of the object detection algorithm, the number of parameters is used to measure the complexity and computational amount of the model, and GFLOPs refers to the number of billions of floating-point operations per second, which is an important indicator to measure the model's ability to perform floating-point operations.

[0053] As can be seen from Table 1, the detection accuracy of the improved algorithm is 99.4%, which is 4% higher than the 95.6% of the original YOLOv8-seg model.

[0054] S2. Extract the RGB three-dimensional color features of the visible light image of the insulator body area.

[0055] The RGB three-dimensional color features are obtained by screening and dimensionality reduction of the RGB image features of the visible light image of the insulator body area through the mRMR feature selection method.

[0056] Obtain some data of the visible light image of the insulator after image segmentation and conduct digital analysis. Obtain the RGB color features in the RGB color space. The RGB color features are used to characterize the color difference. The RGB color features are six-dimensional feature vectors, including the color averages and standard deviations of the red, green, and blue channels.

[0057] Use the mRMR feature selection algorithm based on regression value mutual information to screen and reduce the dimension of the initial color features, and select a feature subset with high correlation with the target variable and low redundancy among features. The scores of each color feature during mRMR screening are as Figure 2 shown. Obtain the three-dimensional most relevant RGB color features after screening according to the scores, effectively preventing overfitting and improving the generalization ability of the model. The screened RGB color features are three-dimensional eigenvalue, including the color averages of the red, green, and blue channels in the visible light image of the insulator body area.

[0058] S3. Use the humidity feature and the RGB three-dimensional color feature as the input of the MLP neural network, and detect the contamination degree of the insulator body area through the MLP neural network to obtain the detection result of the insulator contamination degree.

[0059] Build an MLP neural network. Since the contamination color matches the humidity feature, use the humidity feature and the average color values of the red, green, and blue channels in the visible light image of the selected insulator body area as the input of the MLP neural network. The humidity feature is the relative humidity, which belongs to environmental factors and can be obtained through a hygrometer. The relative humidity has a significant impact on the contamination conductivity of the insulator surface. When the relative humidity increases, the contamination on the insulator surface is more likely to absorb moisture and form a conductive layer. The output value of the MLP neural network is the equivalent salt deposit density (ESDD) of the insulator contamination layer. ESDD refers to the amount of NaCl equivalent to the content of conductive substances in the contamination attached to each square centimeter of the insulator surface (mg / cm 2 ). ESDD characterizes the conductive ability after the insulator surface contamination is fully dissolved and is an important parameter for evaluating the contamination degree of the external insulation of electrical equipment. The larger the ESDD value, the lower the flashover voltage of the insulator and the more serious the contamination degree.

[0060] When training the MLP neural network, the ratio of the equivalent salt deposit density (ESDD) to the non-soluble deposit density (NSDD) on the contaminated insulator in the training set is 1:6. Test the influence of different hidden layer numbers and hidden layer node numbers of the MLP neural network on the mean square error (MSE) of the model. The test results are as Figure 3 shown. It can be seen from Figure 3 that when the MLP neural network adopts a single hidden layer network with 48 nodes in the hidden layer, the MSE of the network on the test set reaches 0.00514, which can meet the actual detection requirements.

[0061] Example 2

[0062] This example provides a detection system for the contamination degree of post insulators, including:

[0063] A visible light image acquisition module for the insulator body area, which is used to acquire the original insulator image, detect and segment the insulator body area in the original insulator image through an improved YOLOv8-seg model, and perform filtering processing to obtain the visible light image of the insulator body area;

[0064] A color feature screening and dimensionality reduction module, which is used to extract the RGB color features of the visible light image of the insulator body area and perform screening and dimensionality reduction on the RGB color features of the visible light image of the insulator body area through the mRMR feature selection method;

[0065] The insulator contamination degree detection module is used to take the humidity feature and the RGB color feature of the visible light image of the selected insulator body area as the input of the MLP neural network, and detect the contamination degree of the insulator body area through the MLP neural network to obtain the insulator contamination degree detection result.

[0066] The rest is the same as in Embodiment 1.

[0067] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. A method for detecting the pollution degree of post insulator, characterized in that Including the following steps: Obtain the original insulator image, detect and segment the insulator body area in the original insulator image through an improved YOLOv8-seg model, and perform filtering processing to obtain the visible light image of the insulator body area; Extract the RGB three-dimensional color features of the visible light image of the insulator body area, and the RGB three-dimensional color features are obtained by screening and dimensionality reduction of the RGB image features of the visible light image of the insulator body area through the mRMR feature selection method; Use the humidity feature and the RGB three-dimensional color features as the input of the MLP neural network, and detect the pollution degree of the insulator body area through the MLP neural network to obtain the detection result of the insulator pollution degree.

2. The method for detecting the pollution degree of the post insulator according to claim 1, wherein The humidity feature is an environmental factor, and the humidity feature is obtained through a hygrometer.

3. The method for detecting the contamination degree of the post porcelain insulator according to claim 1, characterized in that, In the backbone part of the improved YOLOv8-seg model, there are SPPFCSPC modules, CBAM attention modules and multiple convolutional modules, and the rest are variable kernel convolutional modules except the first convolutional module.

4. The method for detecting the pollution degree of the post insulator according to claim 3, characterized in that, The CBAM attention module includes a channel attention unit and a spatial attention unit connected in sequence. The channel attention unit is used to enhance the feature representation of the humidity feature and the screened RGB color features in the channel dimension, and the spatial attention unit is used to enhance the feature representation of the humidity feature and the screened RGB color features in the spatial dimension.

5. The method for detecting the pollution degree of a post porcelain insulator according to claim 1, wherein The RGB image features are used to characterize the color difference, and the RGB image features are six-dimensional feature vectors, including the color average values and standard deviations of the red, green, and blue channels.

6. The method for detecting the pollution degree of the post insulator according to claim 1, wherein The RGB three-dimensional color features include the color average values of the red, green, and blue channels in the visible light image of the insulator body area.

7. The method for detecting the pollution degree of the post insulator according to claim 1, characterized in that The output value of the MLP neural network is the equivalent salt deposit density of the polluted insulator.

8. The method for detecting the pollution degree of the post porcelain insulator according to claim 1, characterized in that, The MLP neural network adopts a single-layer hidden layer network.

9. The method for detecting the pollution degree of the post insulator according to claim 1, characterized in that, When training the MLP neural network, the ratio range of the equivalent salt deposit density and the non-soluble deposit density on the insulator in the training image is fixed.

10. A pollution degree detection system for post porcelain insulators, characterized in that, Including: A visible light image acquisition module for the insulator body area, which is used to obtain the original insulator image, detect and segment the insulator body area in the original insulator image through an improved YOLOv8-seg model, and perform filtering processing to obtain the visible light image of the insulator body area; A color feature screening and dimensionality reduction module, which is used to extract the RGB three-dimensional color features of the visible light image of the insulator body area, and the RGB three-dimensional color features are obtained by screening and dimensionality reduction of the RGB image features of the visible light image of the insulator body area through the mRMR feature selection method; An insulator pollution degree detection module, which is used to use the humidity feature and the RGB color features of the visible light image of the screened insulator body area as the input of the MLP neural network, and detect the pollution degree of the insulator body area through the MLP neural network to obtain the detection result of the insulator pollution degree.

Images