Electric power tower signboard damage identification system based on comparative learning

Abstract

The invention provides an electric power tower signboard damage identification system based on comparative learning, and the system comprises the following parts: an image sample library creation module which obtains an electric power tower signboard image sample, and creates an image sample library; the image acquisition module is used for acquiring an electric pole number plate image in real time; the image preprocessing module is used for preprocessing the electric pole number plate image; the image feature extraction module is used for extracting local features, including key points, edges, textures and shapes, of the electric pole number plate image by adopting an image feature extraction algorithm; the image recognition module is used for calculating the feature similarity between the collected power pole number plate image and all image samples in the image sample library by adopting cosine similarity, and marking the damage degree of the power pole number plate image according to the feature similarity; and the visualization module is used for displaying the quantitative proportion of the images of the electric pole number plates with different damage degrees through a pie chart. Through extraction of four types of features, local details of the image are reserved, and the recognition precision is improved.

Description

Technical Field

[0001] The invention relates to a power pole tower signboard damage recognition technology, and in particular to a power pole tower signboard damage recognition system based on contrast learning. Background Art

[0002] Power pole tower signs, such as pole number plates and warning signs, are important components of the power system. Their damage or loss may cause many problems in the management of power facilities, affecting safety inspections, maintenance efficiency, and post-disaster recovery. Therefore, timely and accurate identification of the damage status of pole number plates and warning signs is of great significance to the safe operation and maintenance of power facilities.

[0003] Most existing damage recognition methods rely on traditional image processing or deep learning-based image classification methods. However, traditional methods have problems such as low recognition accuracy and large errors. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a power pole tower signboard damage recognition system based on comparative learning.

[0005] The power pole tower signboard damage recognition system based on contrastive learning provided by the present invention includes the following:

[0006] Image sample library creation module 11: used to obtain normal, slightly damaged, moderately damaged and severely damaged power pole tower sign image samples and create an image sample library;

[0007] Image acquisition module 12: used for real-time acquisition of power pole number plate images;

[0008] Image preprocessing module 13: used for preprocessing the power pole number plate image;

[0009] Image feature extraction module 14: used to extract local features of the power pole number plate image using an image feature extraction algorithm, including key points, edges, textures and shapes;

[0010] Image recognition module 15: used to calculate the feature similarity between the collected power pole number plate image and all image samples in the image sample library by using cosine similarity, and identify the degree of damage of the collected power pole number plate image according to the feature similarity;

[0011] Visualization module 16: used to display the quantity ratios between normal, slightly damaged, moderately damaged and severely damaged power pole number plate images through a pie chart.

[0012] Preferably, the image preprocessing module includes:

[0013] Image denoising unit: uses Gaussian filtering to denoise the collected image data of power tower signboards;

[0014] Image enhancement unit: enhances the contrast and brightness of power pole tower signs through contrast enhancement algorithm;

[0015] Image normalization unit: normalizes the pixel values ​​of the power tower sign to a specific range;

[0016] Image cropping and scaling unit: cropping and scaling the power pole tower sign according to the size of the power pole tower sign.

[0017] Preferably, the image feature extraction module includes:

[0018] Key point detection unit: uses image feature extraction algorithm to detect key points in the power pole number plate image, including corner points and edge intersections;

[0019] Feature description unit: describes the detected key points and generates feature vectors;

[0020] Edge detection unit: uses the Canny edge detection algorithm to extract edge information from the power pole number plate image;

[0021] Texture analysis unit: uses texture analysis method to extract texture features in the power pole number plate image;

[0022] Shape analysis unit: extracts shape features from the power pole number plate image through contour extraction and shape context algorithm;

[0023] Feature fusion unit: fuses the extracted key points, edges, textures and shape features to generate a comprehensive feature vector.

[0024] Preferably, the texture analysis unit comprises:

[0025] Image block subunit: divides the power pole license plate image into several image blocks of equal size. The size and number of blocks can be adjusted according to the image resolution and the distribution information of texture features;

[0026] Gabor filter application subunit: Gabor filters are used to apply Gabor filters of different directions and scales to each image block, the texture responses of the image blocks at different directions and scales are extracted, and the texture feature vector is generated by calculating the statistical characteristics of the filter response.

[0027] Preferably, the shape analysis unit comprises:

[0028] Contour extraction subunit: obtains edge information from the edge detection unit and extracts contours in the image through a contour detection algorithm;

[0029] Contour screening subunit: Screen all extracted contours, remove noise contours and irrelevant contours, and retain contours related to the power pole number plate;

[0030] Contour point sampling subunit: uniformly samples a number of contour points from the contour;

[0031] Polar coordinate system construction subunit: construct a polar coordinate system with each contour point as the center, and calculate the positions of other contour points in the polar coordinate system;

[0032] Shape context description subunit: counts the distribution information of each contour point in the polar coordinate system and generates a shape context histogram;

[0033] Shape context feature vector generation subunit: concatenates the shape context histograms of all contour points to generate a shape context feature vector.

[0034] Preferably, the feature fusion unit comprises:

[0035] Feature vector receiving subunit: used to receive feature vectors output by the key point detection unit, edge detection unit, texture analysis unit and shape analysis unit;

[0036] Feature alignment subunit: aligns different types of feature vectors;

[0037] Feature weighting subunit: used to assign weights to different types of feature vectors;

[0038] Feature splicing subunit: splices the weighted key point, edge, texture and shape feature vectors to generate a comprehensive feature vector;

[0039] Feature normalization subunit: normalizes the concatenated comprehensive feature vector;

[0040] Result output subunit: outputs the normalized comprehensive feature vector for use by the image recognition module.

[0041] Preferably, the image recognition module includes

[0042] Cosine similarity calculation unit: used to calculate the similarity between the feature vector of the collected power pole number plate image and the feature vector of each image sample in the image sample library by using cosine similarity;

[0043] Similarity sorting unit: used to sort the calculated cosine similarities and select several image samples with the highest similarity to the collected power pole number plate images;

[0044] Damage degree recognition unit: used to filter out the damage degree label of the image sample according to the similarity sorting unit, and determine the damage degree of the collected power pole number plate image;

[0045] Damaged area marking unit: marks the damaged area in the power pole number plate image based on the damage degree recognition result.

[0046] Preferably, the visualization module comprises:

[0047] Data statistics unit: used to count the recognition results and calculate the number of normal, slightly damaged, moderately damaged and severely damaged power pole license plate images respectively;

[0048] Pie chart drawing unit: used to draw a pie chart based on the statistical results, and divide the pie chart into four areas corresponding to four damage degrees;

[0049] Color mapping unit: used to assign corresponding colors to each area;

[0050] Label marking unit: used to mark the damage degree category and its corresponding quantity in each area of ​​the pie chart;

[0051] Dynamic update unit: used to support dynamic update of pie charts. When new power pole number plate image recognition results are generated, the quantity ratio and area size in the pie chart are updated;

[0052] Result display unit: used to display the drawn pie chart on a visualization interface, through which users can view the quantity distribution of power pole number plate images with different degrees of damage.

[0053] The power pole tower sign damage recognition system based on contrastive learning provided by the present invention can collect and store various power pole tower sign images with different degrees of damage, establish a standardized data set, and thus provide high-quality data support for subsequent damage recognition. In combination with the image feature extraction algorithm, the key features in the image can be effectively identified. By extracting four types of features, namely key points, edges, textures and shapes, not only the local details of the image, such as corner points, edge intersections, and global structures such as texture and shape information, are retained, but also the image features under different degrees of damage are covered, thereby improving the recognition accuracy. The multi-dimensional features are integrated into a comprehensive feature vector by using a feature fusion unit. Through weighted and normalized processing, the importance of different features is balanced, and the ability to describe image features in complex scenes is enhanced. Through image segmentation and Gabor filter application, texture features are extracted from different directions and scales to capture small changes in the image surface, which is particularly suitable for the detection of slight damage and makes up for the shortcomings of traditional methods in texture analysis. Through contour extraction, screening, sampling and shape context algorithms, the system comprehensively describes the shape information in the image. Shape context description enhances the recognition ability of complex contours, especially for severely damaged signboards, thereby improving the accuracy of damage identification; by real-time acquisition of power pole signboard images, and through the image preprocessing module to clarify and standardize the acquired images, by calculating the similarity between the image and the sample library image, the degree of damage of the signboard can be automatically identified and the results can be marked, thereby realizing the automatic and intelligent detection of power pole signboards. By using cosine similarity to calculate the feature similarity between images, the similarity between images with different degrees of damage can be more accurately judged, thereby improving the accuracy of damage identification, especially for images with slight damage or blurred edges, it can effectively reduce the possibility of misidentification. Through the visualization module, the number of images of different degrees of damage of power pole signboards can be displayed in the form of pie charts, which is convenient for managers to quickly understand the distribution and trend of various damage conditions, take maintenance measures in time, effectively improve the efficiency of damage detection of power pole signboards, reduce the workload of manual inspection, and discover damage in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 The architecture diagram of the power tower signboard damage identification system based on contrastive learning provided by the present invention;

[0056] Figure 2 A unit structure diagram of the image preprocessing module provided by the present invention;

[0057] Figure 3 A unit structure diagram of the image feature extraction module provided by the present invention;

[0058] Figure 4 A subunit structure diagram of a texture analysis unit in an image feature extraction module provided by the present invention;

[0059] Figure 5 A subunit structure diagram of a shape analysis unit in an image feature extraction module provided by the present invention;

[0060] Figure 6 A subunit structure diagram of a feature fusion unit in an image feature extraction module provided by the present invention;

[0061] Figure 7 A unit structure diagram of the image recognition module provided by the present invention;

[0062] Figure 8 A unit structure diagram of the visualization module provided by the present invention; DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. 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.

[0064] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0065] refer to Figure 1 The power pole tower signboard damage recognition system based on contrastive learning provided by the present invention includes the following:

[0066] Image sample library creation module 11: used to obtain normal, slightly damaged, moderately damaged and severely damaged power pole tower sign image samples and create an image sample library;

[0067] Image acquisition module 12: used for real-time acquisition of power pole number plate images; wherein the power pole number plate images include power pole number plates and warning signs.

[0068] Image preprocessing module 13: used for preprocessing the power pole number plate image;

[0069] Image feature extraction module 14: used to extract local features of the power pole number plate image using an image feature extraction algorithm, including key points, edges, textures, shapes and other information.

[0070] Image recognition module 15: used to calculate the feature similarity between the collected power pole number plate image and all image samples in the image sample library by using cosine similarity, and identify the degree of damage of the collected power pole number plate image according to the feature similarity;

[0071] Visualization module 16: used to display the quantity ratios between normal, slightly damaged, moderately damaged and severely damaged power pole number plate images through a pie chart.

[0072] The power pole tower signboard damage recognition system based on contrastive learning provided by the present invention can collect and store images of power pole tower signboards with various degrees of damage, establish a standardized data set, and thus provide high-quality data support for subsequent damage recognition. Combined with the image feature extraction algorithm, it can effectively identify the key features in the image, thereby improving the accuracy of damage recognition. By real-time acquisition of power pole number plate images, and by using an image preprocessing module to clarify and standardize the acquired images, the degree of damage of the signboard can be automatically identified by calculating the similarity between the image and the sample library image, and the result can be marked, thereby realizing the automated and intelligent detection of power pole tower signboards. By using cosine similarity to calculate the feature similarity between images, the similarity between images with different degrees of damage can be judged more accurately, thereby improving the accuracy of damage identification, especially for images with slight damage or blurred edges, which can effectively reduce the possibility of misidentification. The number of images of power pole tower signs with different degrees of damage can be displayed in the form of a pie chart through the visualization module, which is convenient for managers to quickly understand the distribution and trend of various types of damage, take maintenance measures in time, effectively improve the efficiency of damage detection of power pole tower signs, reduce the workload of manual inspection, and detect damage in time.

[0073] refer to Figure 2 In a preferred embodiment, the image preprocessing module 13 includes:

[0074] Image denoising unit 131: uses Gaussian filtering to perform denoising on the collected power pole tower sign image data to remove noise interference in the image and improve image quality;

[0075] Image enhancement unit 132: enhances the contrast and brightness of the power pole tower sign through a contrast enhancement algorithm, making the damaged area more obvious for subsequent identification;

[0076] Image normalization unit 133: normalizes the pixel value of the power tower sign to a specific range, such as between 0 and 1, to eliminate the influence of different lighting conditions on image recognition and ensure the consistency of image data;

[0077] The image cropping and scaling unit 134 crops and scales the power pole tower sign according to the size of the power pole tower sign; ensures that only the target area is included in the image, and unifies the size of the image.

[0078] refer to Figure 3 In a preferred embodiment, the image feature extraction module 14 includes:

[0079] Key point detection unit 141: uses SIFT, SURF or ORB algorithm to detect key points in the power pole number plate image, including corner points and edge intersection points; these key points are usually areas with significant changes in the image, such as corner points, edge intersection points, etc.

[0080] Feature description unit 142: performs feature description on the detected key points and generates feature vectors; if SIFT and SURF algorithms are used, the features are described by constructing a Gaussian difference pyramid and calculating a local gradient direction histogram.

[0081] Edge detection unit 143: uses the Canny edge detection algorithm to extract edge information in the power pole number plate image, which helps to identify the contour and boundary in the image, thereby assisting in determining the damaged area.

[0082] Specifically, the power pole number plate image is converted into a grayscale image to reduce the computational complexity and highlight the edge information. The grayscale method can use the weighted average method or directly take the grayscale value of a certain channel to perform Gaussian filtering on the grayscale image to smooth the image and remove noise. The kernel size and standard deviation of the Gaussian filter can be adjusted according to the image noise level and edge detail requirements. The Sobel operator is used to calculate the gradient amplitude and direction of the image. The gradient amplitude represents the degree of change of the grayscale value in the image, and the gradient direction represents the direction of change of the grayscale value. The gradient amplitude image is non-maximum suppressed, the local maximum value in the gradient direction is retained, non-edge pixels are removed, and the edges are refined. Two thresholds (high threshold and low threshold) are set, and edge detection is performed on the gradient amplitude image after non-maximum suppression. The specific steps are as follows: pixels with gradient amplitude greater than the high threshold are marked as strong edges; pixels with gradient amplitude between the high threshold and the low threshold are marked as weak edges; pixels with gradient amplitude less than the low threshold are marked as non-edges. Edge connection processing is performed on weak edge pixels, and the adjacent strong edge pixels are connected to form a complete edge. The edge connection method can adopt 8-neighborhood connection or 4-neighborhood connection. The detected edge information is output for use by the subsequent shape analysis unit and feature fusion unit.

[0083] Texture analysis unit 144: uses texture analysis method to extract texture features in the power pole number plate image. Texture features can reflect the local structure and pattern of the image and help identify subtle changes in the damaged area.

[0084] Shape analysis unit 145: extracts shape features from the power pole number plate image through contour extraction and shape description algorithms; shape features can describe the geometric properties of objects in the image and help identify shape changes in damaged areas.

[0085] Feature fusion unit 146: fuses the extracted key points, edges, textures and shape features to generate a comprehensive feature vector, which can more comprehensively describe the local features of the power pole number plate image and improve the accuracy of damage recognition.

[0086] refer to Figure 4 In a preferred embodiment, the texture analysis unit 144 comprises:

[0087] Image block subunit 1441: divides the power pole number plate image into a number of image blocks of equal size, so as to extract local texture features for each image block, and the size and number of the extracted blocks can be adjusted according to the image resolution and the distribution information of the texture features;

[0088] Gabor filter application subunit 1442: uses Gabor filter to apply Gabor filters of different directions and scales to each image block, extracts texture responses of the image block at different directions and scales, and generates texture feature vectors by calculating statistical features of the filter response including mean, variance, energy, etc.

[0089] refer to Figure 5 In a preferred embodiment, the shape analysis unit 145 comprises:

[0090] Contour extraction subunit 1451: obtain edge information in the edge detection unit, and extract the contour in the image through the contour detection algorithm including the findContours function in OpenCV. The contour is the boundary line of the object in the power pole number plate image;

[0091] Contour screening subunit 1452: screening all extracted contours, removing noise contours and irrelevant contours, and retaining contours related to the power pole number plate. The screening method can be based on features such as the area, perimeter or shape complexity of the contour;

[0092] Contour point sampling subunit 1453: uniformly samples a number of contour points from the contour as a basis for calculating the shape context;

[0093] Polar coordinate system construction subunit 1454: constructs a polar coordinate system with each contour point as the center, and calculates the positions of other contour points in the polar coordinate system, including distance and angle;

[0094] Shape context description subunit 1455: Counts the distribution information of each contour point in the polar coordinate system and generates a shape context histogram; each bin of the shape context histogram represents the distribution of the contour point within a specific distance and angle range.

[0095] Shape context feature vector generation subunit 1456: concatenates the shape context histograms of all contour points to generate a shape context feature vector.

[0096] refer to Figure 6 In a preferred embodiment, the feature fusion unit 146 includes:

[0097] Feature vector receiving subunit 1461: used to receive feature vectors output by the key point detection unit, the edge detection unit, the texture analysis unit and the shape analysis unit;

[0098] Feature alignment subunit 1462: performs alignment processing on different types of feature vectors to ensure that the dimensions of each feature vector are consistent. The alignment method may include feature dimensionality reduction such as PCA dimensionality reduction, feature filling such as zero filling, or feature interpolation.

[0099] Feature weighting subunit 1463: used to assign weights to different types of feature vectors to reflect their importance in the comprehensive features. The weights can be set according to the contribution of the features or experimental results, such as determined by cross-validation or feature importance analysis.

[0100] Feature splicing subunit 1464: Splicing the weighted key point, edge, texture and shape feature vectors to generate a comprehensive feature vector. The splicing method can be simple vector splicing or more complex feature fusion methods such as feature cascading or feature pooling.

[0101] Feature normalization subunit 1465: normalizes the combined feature vectors to ensure the consistency of the feature vectors in the numerical range. The normalization method may be minimum-maximum normalization, Z-score normalization, or L2 regularization.

[0102] Result output subunit 1466: outputs the normalized comprehensive feature vector for use by the image recognition module.

[0103] refer to Figure 7 In a preferred embodiment, the image recognition module 15 includes

[0104] The cosine similarity calculation unit 151 is used to calculate the similarity between the feature vector of the collected power pole number plate image and the feature vector of each image sample in the image sample library by using cosine similarity.

[0105] The similarity sorting unit 152 is used to sort the calculated cosine similarities and select a number of image samples with the highest similarity to the collected power pole number plate image.

[0106] The damage degree identification unit 153 is used to filter out the damage degree labels of the image samples according to the similarity sorting unit, and determine the damage degree of the collected power pole number plate image.

[0107] Specifically, if the damage degree of the high-similarity image sample is "normal", the collected image is judged to be normal; if it is "slightly damaged", "moderately damaged" or "severely damaged", it is judged to be the corresponding damage degree respectively.

[0108] The damaged area marking unit 154 marks the damaged area in the power pole number plate image according to the damage degree recognition result. The marking method uses a rectangular frame and a polygonal frame, and a text description of the damage degree is added inside or outside the marked frame.

[0109] Result output unit 155: outputs the recognition result and the annotated image for subsequent processing or display. The recognition result may include information such as damage category, similarity score, and annotated area coordinates.

[0110] refer to Figure 8 In a preferred embodiment, the visualization module 16 includes

[0111] The data statistics unit 161 performs statistics on the recognition results, and respectively calculates the number of normal, slightly damaged, moderately damaged, and severely damaged power pole number plate images.

[0112] The pie chart drawing unit 162 draws a pie chart according to the statistical results, and divides the pie chart into four areas corresponding to four damage degrees:

[0113] Green area: represents the number of normal power pole license plate images, which is the ratio of the number of normal images to the total number of images.

[0114] Yellow area: indicates the number of slightly damaged power pole license plate images, which is the ratio of the number of slightly damaged images to the total number of images.

[0115] Purple area: represents the number of images of moderately damaged power pole license plate, which is the ratio of the number of moderately damaged images to the total number of images.

[0116] Red area: indicates the number of images of severely damaged power pole license plates, which is the ratio of the number of severely damaged images to the total number of images.

[0117] Color mapping unit 163: assigns a corresponding color to each region:

[0118] Normal: Green

[0119] Slightly damaged: yellow

[0120] Moderate damage: purple

[0121] Severe damage: red

[0122] The label marking unit 164 is used to mark the damage degree category and its corresponding quantity in each area of ​​the pie chart.

[0123] Dynamic update unit 164: supports dynamic update of the pie chart, and automatically updates the quantity ratio and area size in the pie chart when new power pole number plate image recognition results are generated.

[0124] Result display unit 165: displays the drawn pie chart on a visualization interface, through which the user can view the quantity distribution of power pole number plate images with different degrees of damage. The pie chart can be exported as a picture or PDF file, so that the user can save and share the statistical results.

Claims

1. A power tower signboard damage recognition system based on contrastive learning, characterized in that: These include: Image sample library creation module: used to obtain normal, slightly damaged, moderately damaged and severely damaged power pole tower sign image samples and create an image sample library; Image acquisition module: used to collect power pole number plate images in real time; Image preprocessing module: used for preprocessing the power pole number plate image; Image feature extraction module: used to extract local features of the power pole number plate image using an image feature extraction algorithm, including key points, edges, textures and shapes; Image recognition module: used to calculate the feature similarity between the collected power pole number plate image and all image samples in the image sample library by using cosine similarity, and identify the degree of damage of the collected power pole number plate image according to the feature similarity; Visualization module: used to display the quantity ratio between normal, slightly damaged, moderately damaged and severely damaged power pole number plate images through pie charts.

2. The power pole tower signboard damage recognition system based on contrastive learning according to claim 1 is characterized in that: The image preprocessing module comprises: Image denoising unit: uses Gaussian filtering to denoise the collected image data of power tower signboards; Image enhancement unit: enhances the contrast and brightness of power pole tower signs through contrast enhancement algorithm; Image normalization unit: normalizes the pixel values ​​of the power tower sign to a specific range; Image cropping and scaling unit: cropping and scaling the power pole tower sign according to the size of the power pole tower sign.

3. The power pole tower signboard damage recognition system based on contrastive learning according to claim 1 is characterized in that: The image feature extraction module comprises: Key point detection unit: uses image feature extraction algorithm to detect key points in the power pole number plate image, including corner points and edge intersection points; Feature description unit: describes the detected key points and generates feature vectors; Edge detection unit: uses the Canny edge detection algorithm to extract edge information from the power pole number plate image; Texture analysis unit: uses texture analysis method to extract texture features in the power pole number plate image; Shape analysis unit: extracts shape features from the power pole number plate image through contour extraction and shape context algorithm; Feature fusion unit: fuses the extracted key points, edges, textures and shape features to generate a comprehensive feature vector.

4. The power pole tower signboard damage recognition system based on contrastive learning according to claim 3 is characterized in that: The texture analysis unit comprises: Image block subunit: divides the power pole license plate image into several image blocks of equal size. The size and number of blocks can be adjusted according to the image resolution and the distribution information of texture features; Gabor filter application subunit: Gabor filters are used to apply Gabor filters of different directions and scales to each image block, the texture responses of the image blocks at different directions and scales are extracted, and the texture feature vector is generated by calculating the statistical characteristics of the filter response.

5. The power pole tower signboard damage recognition system based on contrastive learning according to claim 3 is characterized in that: The shape analysis unit comprises: Contour extraction subunit: obtains edge information from the edge detection unit and extracts contours in the image through a contour detection algorithm; Contour screening subunit: Screen all extracted contours, remove noise contours and irrelevant contours, and retain contours related to the power pole number plate; Contour point sampling subunit: uniformly samples a number of contour points from the contour; Polar coordinate system construction subunit: construct a polar coordinate system with each contour point as the center, and calculate the positions of other contour points in the polar coordinate system; Shape context description subunit: counts the distribution information of each contour point in the polar coordinate system and generates a shape context histogram; Shape context feature vector generation subunit: concatenates the shape context histograms of all contour points to generate a shape context feature vector.

6. The power pole tower signboard damage identification system based on contrastive learning according to claim 3 is characterized in that: The feature fusion unit comprises: Feature vector receiving subunit: used to receive feature vectors output by the key point detection unit, edge detection unit, texture analysis unit and shape analysis unit; Feature alignment subunit: aligns different types of feature vectors; Feature weighting subunit: used to assign weights to different types of feature vectors; Feature splicing subunit: splices the weighted key point, edge, texture and shape feature vectors to generate a comprehensive feature vector; Feature normalization subunit: normalizes the concatenated comprehensive feature vector; Result output subunit: outputs the normalized comprehensive feature vector for use by the image recognition module.

7. The power pole tower signboard damage recognition system based on contrastive learning according to claim 1 is characterized in that: The image recognition module includes Cosine similarity calculation unit: used to calculate the similarity between the feature vector of the collected power pole number plate image and the feature vector of each image sample in the image sample library by using cosine similarity; Similarity sorting unit: used to sort the calculated cosine similarities and select several image samples with the highest similarity to the collected power pole number plate images; Damage degree recognition unit: used to filter out the damage degree label of the image sample according to the similarity sorting unit, and determine the damage degree of the collected power pole number plate image; Damaged area marking unit: marks the damaged area in the power pole number plate image based on the damage degree recognition result.

8. The power pole tower signboard damage recognition system based on contrastive learning according to claim 1 is characterized in that: The visualization module includes: Data statistics unit: used to count the recognition results and calculate the number of normal, slightly damaged, moderately damaged and severely damaged power pole license plate images; Pie chart drawing unit: used to draw a pie chart based on the statistical results, and divide the pie chart into four areas corresponding to four damage degrees; Color mapping unit: used to assign corresponding colors to each area; Label marking unit: used to mark the damage degree category and its corresponding quantity in each area of ​​the pie chart; Dynamic update unit: used to support dynamic update of pie charts. When new power pole number plate image recognition results are generated, the quantity ratio and area size in the pie chart are updated; Result display unit: used to display the drawn pie chart on a visualization interface, through which users can view the quantity distribution of power pole number plate images with different degrees of damage.

Images