Power transmission line fault diagnosis method, device and system and medium

Abstract

The invention relates to the technical field of power transmission line diagnosis, and provides a power transmission line fault diagnosis method, device and system and a medium. According to the implementation scheme, first sample images corresponding to all fault types are expanded, and a first sample image set is obtained; performing feature extraction on each second sample image in the first sample image set to obtain each first feature image; constructing a feature database based on each first feature image to obtain a first feature database; performing integrity verification on the first feature database to obtain a target feature database; and matching the image data to be diagnosed with the image data in the target feature database to obtain a fault diagnosis result. According to the embodiment of the invention, the reliability of power transmission line fault diagnosis can be improved.

Description

Technical Field

[0001] This invention relates to the field of power transmission line diagnostic technology, and in particular to a method, device, system and medium for diagnosing power transmission line faults. Background Technology

[0002] As a crucial component of the power system, the operating status of transmission lines directly affects the safety and stability of the power grid. When lines are exposed to the outdoor environment for extended periods, components such as conductors, fittings, and insulators are susceptible to corrosion, aging, electrical discharge erosion, and external impacts, leading to various types of potential faults.

[0003] Existing fault image sample databases often suffer from problems such as unbalanced sample distribution, insufficient feature extraction, and chaotic feature database structure. Therefore, when the training samples are incomplete or features are missing, the model is prone to mismatch when retrieving similar fault features, which affects the reliability of the final diagnosis results.

[0004] Therefore, how to effectively process images of different fault types in the sample database to improve the reliability of transmission line fault diagnosis is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] This invention provides a method, device, system, and medium for diagnosing power transmission line faults, which can solve at least one of the above-mentioned technical problems.

[0006] In a first aspect, embodiments of the present invention provide a method for diagnosing transmission line faults, including: The first sample images corresponding to each fault type are expanded to obtain a set of first sample images; Feature extraction is performed on each second sample image in the first sample image set to obtain each first feature image; A feature database is constructed based on each of the first feature images to obtain the first feature database; The integrity of the first feature database is verified to obtain the target feature database. The image data to be diagnosed is matched with each image data in the target feature database to obtain the fault diagnosis result.

[0007] Secondly, embodiments of the present invention provide a power transmission line fault diagnosis device, comprising: The sample expansion module is used to expand the first sample images corresponding to each fault type to obtain a set of first sample images; The feature extraction module is used to extract features from each of the second sample images in the first sample image set to obtain each first feature image; The feature database construction module is used to construct a feature database based on each of the first feature images to obtain the first feature database; The integrity verification module is used to perform integrity verification on the first feature database to obtain the target feature database. The matching module is used to match the image data to be diagnosed with each image data in the target feature database to obtain the fault diagnosis result.

[0008] Thirdly, embodiments of the present invention also provide a transmission line fault diagnosis system, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described in any one of the embodiments of the present invention.

[0009] Fourthly, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the method described in any one of the embodiments of the present invention.

[0010] The present invention expands the first sample images corresponding to different fault types to obtain a first sample image set. This effectively reduces the classification bias caused by sample scarcity or imbalance in traditional diagnostic methods. Then, features are extracted from each second sample image in the first sample image set to obtain first feature images, thereby improving the discriminative power of feature representation. Based on this, a feature database is constructed using each first feature image to obtain a first feature database. The integrity of the first feature database is then verified to obtain a target feature database. This avoids the false positives and false negatives caused by missing features or disordered indexes in traditional feature databases. Subsequently, the image to be diagnosed is matched with the image data in the target feature database to obtain the fault diagnosis result. Thus, because the established database has high-quality feature representation and a feature structure that has undergone integrity verification, the accuracy and reliability of the diagnostic results can be significantly improved when matching and recognizing the image data to be diagnosed.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided for a better understanding of this solution and do not constitute a limitation of the invention. Wherein: Figure 1 This is a flowchart of a transmission line fault diagnosis method according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a power transmission line fault diagnosis device according to an embodiment of the present invention; Figure 3 This is a schematic block diagram of an electronic device used to implement the methods of embodiments of the present invention. Detailed Implementation

[0013] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0014] This invention provides a method, apparatus, device, and medium for diagnosing transmission line faults. The entity executing this method can be the transmission line fault diagnosis device provided in this application, or a computer device integrating the transmission line fault diagnosis device. The transmission line fault diagnosis device can be implemented in hardware or software, and the computer device can be a terminal or a server.

[0015] Figure 1 This is a flowchart of a power transmission line fault diagnosis method according to an embodiment of the present invention.

[0016] like Figure 1 As shown, the transmission line fault diagnosis method may include: S110, Expand the first sample image corresponding to each fault type to obtain the first sample image set; S120, extract features from each second sample image in the first sample image set to obtain each first feature image; S130, construct a feature database based on each first feature image to obtain the first feature database; S140, Perform integrity verification on the first feature database to obtain the target feature database; S150, the image data to be diagnosed is matched with each image data in the target feature database to obtain the fault diagnosis result.

[0017] For example, original sample images for each fault type are acquired. These original sample images can be obtained from on-site inspection photos of transmission lines, video frames captured from drone inspections, or historical labeled image libraries. Based on manual or automatic labeling algorithms, the original sample images are labeled with fault types to obtain the fault type corresponding to each image. Then, preprocessing operations are performed on the original sample images. Preprocessing operations may specifically include resolution normalization, noise filtering, and illumination compensation to improve the stability of subsequent image augmentation.

[0018] For example, fault types include rust, breakage, cracks, foreign object entanglement, loose components, etc.

[0019] For example, the first sample image refers to the original base sample image that has been correctly labeled with the fault type, and is used for subsequent data augmentation operations. For instance, an inspection image manually labeled as "slight corrosion" can serve as a first sample image.

[0020] For example, the object of the first sample image can be an insulator or a conductor in a high-voltage transmission line. It can also be an image that includes both insulators and conductors.

[0021] For example, the first sample image set refers to an image set containing a large number of diverse fault samples generated based on each first sample image through data augmentation, used to improve the robustness of feature extraction and subsequent diagnosis. For instance, if the original "light corrosion" samples are only 50 images, after augmentation, 300 images are obtained, and these 300 images constitute the first sample image set corresponding to light corrosion.

[0022] In this example, the first set of sample images may contain both dilated and undilated sample images, or it may consist entirely of dilated sample images.

[0023] For example, the second sample image refers to the input image selected from the first sample image set for performing the feature extraction operation. The second sample image may be the same as the first sample image or may be different from the first sample image. Specifically, if the second sample image is an undilated image, it is the same as the first sample image; if the second sample image is an image that has been dilated based on the first sample image, it is different from the first sample image.

[0024] For example, the first feature image refers to the image feature representation extracted based on the second sample image, which may include texture feature maps, edge feature maps, feature vectors or feature maps output by a deep convolutional network, etc. For instance, after inputting a rusted image into a convolutional neural network, the 128×128-dimensional feature map or 2048-dimensional feature vector output by the intermediate layer of the network is the first feature image corresponding to that image.

[0025] For example, the first feature database refers to a preliminary feature storage structure built upon all first feature images for fault image retrieval and matching. Each record in this database contains at least: fault type, fault level label, and corresponding first feature image (feature vector / graph). For example, a record in the database might contain: fault type: corrosion; fault level: moderate; feature vector: a feature array of length 2048.

[0026] For example, the target feature database refers to the feature database that meets quality requirements and is suitable for fault diagnosis, obtained after the first feature database has undergone integrity verification (such as missing data detection, feature distribution balance check, and index structure integrity check). For instance, if there are missing or incorrectly labeled "severely damaged" samples in the first feature database, and these are corrected or supplemented after verification, the resulting new database is the target feature database.

[0027] For example, the image data to be diagnosed refers to the line operation images that have not yet been labeled with fault types, which are collected in real time by inspection equipment (such as drones, tower cameras, or ground inspection devices), and are the input objects of the diagnostic process. For example, if a drone inspection finds an image of a hardware surface that appears to have rust, this image is the image data to be diagnosed.

[0028] For example, the fault diagnosis result refers to the diagnostic output information consisting of fields such as fault type, fault level, and matching confidence, after the image data to be diagnosed is matched with various image features in the target feature database based on similarity. For example, the system outputs: "Fault type: rust; Fault level: moderate; Matching confidence: 0.92".

[0029] According to the above implementation method, the first sample images for each fault type are first expanded to construct a sufficiently large and evenly distributed set of first sample images. Then, feature extraction is performed on each second sample image from this set to obtain first feature images that can characterize color, texture, and shape. A feature database is then constructed based on all the first feature images to form a structured first feature database. Further, the database is subjected to integrity verification, invalid features are filtered out, and missing rules are corrected, thereby generating a reliable target feature database. Finally, the image to be diagnosed is matched with the target feature database to output the corresponding fault diagnosis result. In this way, a synergistic effect of sample expansion, unified feature representation, reliable database structure, and accurate matching judgment can be achieved, making the transmission line fault diagnosis process more stable and reliable in multi-fault types and multi-image scenarios, thereby effectively improving the reliability of the overall diagnosis result.

[0030] In one implementation, the first sample images corresponding to each fault type are augmented to obtain a first sample image set, including: determining a corresponding augmentation strategy based on the fault type of each first sample image to augment each first sample image, thereby obtaining a first sub-sample image set; enhancing each sub-sample image corresponding to a first fault level in the first sub-sample image set using a support vector machine model to update the first sub-sample image set, thereby obtaining a second sub-sample image set; adjusting each second sub-sample image in the second sub-sample image set based on the fault level, thereby obtaining a third sub-sample image set; calculating the gradient magnitude of each rust sample image in the third sub-sample image set using an edge detection algorithm to obtain the edge sharpness of each rust sample image; and when the edge sharpness of each rust sample image is less than a preset sharpness threshold, performing edge enhancement on each rust sample image based on a Laplacian operator as a convolution kernel to update the third sub-sample image set, thereby obtaining the first sample image set.

[0031] For example, an augmentation strategy refers to an image enhancement or transformation scheme selected based on the differences in image features for sample images of different fault types. For instance, if the fault type is "mild corrosion," the augmentation strategy may include rotation, mirroring, brightness perturbation, etc., to enhance robustness under changes in illumination; if the fault type is "damage and cracks," the augmentation strategy may include random cropping, geometric distortion, etc., to enhance the model's sensitivity to changes in crack direction.

[0032] For example, the first subsample image set refers to the expanded image set formed after each first sample image is expanded using an expansion strategy. For instance, if the original input contains 10 "slightly rusted" images, after applying five expansion strategies such as rotation, brightness perturbation, and size scaling, 50 images can be generated, which constitute the first subsample image set.

[0033] In this example, based on the fault type of the first sample image, corresponding expansion strategies are selected to perform operations such as rotation, scaling, translation, brightness perturbation, and blur enhancement on the image. The expanded image is then format-converted and size-normalized to generate the first subsample image set. For example, the slightly corroded image is rotated by 15°, 30°, and 45°, its brightness is adjusted by +20%, and it is randomly cropped, ultimately expanding from 10 original images to 60 expanded images.

[0034] For example, the first fault level refers to the lowest level of the preset fault severity under the fault type, which is used to prioritize enhancement compensation when the samples are imbalanced. For example, the rust category is usually divided into light / moderate / severe, so light rust is the first fault level; "minor cracks" in the damage category can also be regarded as the first fault level.

[0035] For example, the second subsample image set refers to the image set obtained by enhancing the samples corresponding to the first fault level through a support vector machine model based on the first subsample image set.

[0036] In this example, based on the texture, color, and edge features of the samples corresponding to the first fault level in the first sub-sample image set, these features are input into a support vector machine (SVM) classifier. Synthetic samples are obtained through SVM feature generation or interpolation mechanisms, and these synthetic samples are added to the first sub-sample image set to form the second sub-sample image set. For example, SVM interpolation is performed on the texture feature vectors of lightly corroded images, increasing the number of lightly corroded samples from 20 to 50, thus forming an enhanced second sub-sample image set.

[0037] For example, the third subsample image set refers to the image set obtained by adjusting the sample ratio of the second subsample image set based on the fault level. For instance, if the sample size of mild / moderate / severe faults needs to be maintained in a 2:3:3 ratio, then samples of each level are added, deleted, or resampled based on the second subsample image set to finally meet the specified ratio, which is the third subsample image set.

[0038] In this example, based on the preset sample quantity ratio for each fault level, oversampling, undersampling, or random perturbation resampling is performed on the samples of each level in the second sub-sample image set. The updated sample images are then reordered and labeled to obtain the third sub-sample image set. For example, if the ratio of mild / moderate / severe is set to 2:3:3, oversampling is performed when there are insufficient mild samples in the second sub-sample image set, and undersampling is performed when there are too many severe samples, so that the final sample distribution meets the specified ratio.

[0039] In this example, an edge detection operator is used to calculate the gradient of the third sub-sample image, obtaining the gradient magnitude matrix of each pixel. The maximum gradient value is then extracted as an edge sharpness index to determine whether the image edges are clear. Edge detection operators include, for example, the Sobel operator (Sobel Operator) and the Prewitt operator (PrewittOperator).

[0040] For example, if the maximum gradient value obtained after calculating the gradient of a rusted image is 28, and the preset sharpness threshold is 35, then the image edge sharpness is determined to be insufficient.

[0041] In this example, a convolution operation with the Laplacian operator as the convolution kernel is applied to the rust sample image with edge sharpness less than a preset threshold to enhance the image edge contrast and texture details. The enhanced image is then used to replace the corresponding image in the third subsample image set to generate the first sample image set.

[0042] According to the above implementation method, the first sample images are expanded in multiple levels through a fault type-driven expansion strategy. This includes fault level enhancement based on a support vector machine model, adjustment of image characteristics according to different fault levels, and edge sharpness detection and Laplacian enhancement processing based on gradient magnitude for rust-type sample images. This gradually constructs a first sample image set with higher quality and more balanced distribution. In this way, the first sample image set is not only expanded in quantity but also comprehensively improved in terms of clarity, feature saliency, and fault level discriminability. This helps subsequent feature extraction and diagnostic models achieve more stable and accurate recognition results in multiple fault scenarios, thereby significantly improving the overall reliability of fault diagnosis.

[0043] In one implementation, feature extraction is performed on each of the second sample images in the first sample image set to obtain each of the first feature images, including: standardizing each of the second sample images to obtain each of the standardized images; performing edge sharpness detection on each of the standardized images to obtain an edge sharpness matrix corresponding to each standardized image; performing edge enhancement processing on the standardized images corresponding to the edge sharpness matrix of each standardized image if the maximum gradient value in the edge sharpness matrix is ​​less than a preset gradient threshold to obtain an edge-enhanced image; optimizing each of the edge-enhanced images to obtain each of the optimized feature images; and performing texture extraction based on each of the optimized feature images of the first filter to obtain each of the first feature images.

[0044] For example, the standardization process includes size normalization, pixel value normalization, and brightness alignment. Specifically, the input image is scaled to a fixed resolution, pixel values ​​are normalized from 0 to 1, and the overall brightness is consistent using histogram equalization to generate a standardized image for subsequent calculations. For instance, a 640×480 rust sample image is scaled to 256×256, and the pixel values ​​are converted from 0-255 to 0-1. Simultaneously, histogram equalization is performed to stabilize the brightness distribution in the rust area, thus obtaining a standardized image.

[0045] In this example, a gradient operator is applied to the normalized image to calculate the horizontal and vertical gradients of each pixel. The gradient magnitude of each pixel is obtained by taking the square root of the sum of the squares of the horizontal and vertical gradients. A gradient magnitude matrix is ​​constructed using these gradient magnitudes, and this matrix is ​​used as the edge sharpness matrix. For example, after processing a normalized image with the Sobel operator, a 256×256 gradient magnitude matrix is ​​calculated. The gradient values ​​are largest near the rusted edges; therefore, this gradient magnitude matrix is ​​the corresponding edge sharpness matrix.

[0046] For example, the maximum gradient value of the edge sharpness matrix is ​​compared with a set threshold. When the maximum gradient value is insufficient to reflect obvious structural edges, a second-order gradient operator such as the Laplacian operator is used to perform convolution operation on the standardized image to enhance the image edges. At the same time, the convolution result is fused with the original image in a weighted manner to generate an edge-enhanced image.

[0047] For example, if the maximum gradient value of the sharpness matrix of a standardized rust image is only 9, and the preset threshold is 15, the system performs Laplacian enhancement on the image to make the blurred rust edges clearer, and finally obtains the enhanced image.

[0048] For example, noise suppression, contrast enhancement, and texture smoothing are performed on the edge enhancement image. Texture noise is reduced through bilateral filtering or guided filtering, and the texture of the target region is enhanced through adaptive contrast stretching. This process maintains sharp edges while suppressing unwanted noise, thereby generating an optimized feature image. For instance, if an enhanced image has sharp edges but contains a lot of noise, bilateral filtering can eliminate the noise, and local contrast enhancement can be applied to make the rust texture area more uniform and prominent, thus obtaining an optimized feature image.

[0049] For example, the optimized feature image is subjected to multi-scale, multi-directional convolution using a first filter to extract texture responses in different frequency domains, and the resulting texture intensity matrix is ​​used as the first feature image of the image. The first filter is, for example, a Gabor filter bank.

[0050] For example, by applying Gabor filters in 6 directions and 5 scales to an optimized rust image, 30 sets of texture response images are obtained. Each Gabor response image is used as a first feature image for subsequent similarity matching and feature database construction.

[0051] According to the above implementation method, by sequentially performing standardization, edge sharpness detection, gradient threshold-based adaptive edge enhancement, feature optimization, and texture extraction on the second sample image, each sample image possesses uniform scale characteristics, clearer edge structure, and more stable texture expression before entering the feature construction stage. Edge enhancement is triggered only when the image gradient is insufficient, thus ensuring the targetedness and effectiveness of feature enhancement. Optimization and texture filtering further extract key texture information characterizing the differences in fault morphology, resulting in a first feature image with higher discriminative power and representational completeness. This significantly improves the stability and robustness of feature construction, providing high-quality feature input for subsequent fault diagnosis, thereby enhancing the reliability of the diagnostic results.

[0052] In one implementation, a feature database is constructed based on each first feature image to obtain a first feature database, including: extracting feature vectors from each first feature image to obtain features of each dimension of each first feature image; extracting feature vectors from each first sample image to obtain features of each dimension of each first sample image; determining each second feature image based on the similarity values ​​between the features of each first feature image and the features of each first sample image; clustering each second feature image for each fault type based on each fault level to obtain fault level labels for each second feature image under each fault level; using each fault type as a tree node, each fault level as a child node, and the features of each second feature image and each fault level label as leaf nodes to obtain a B+ tree index structure; and storing the B+ tree index structure in a database file to obtain the first feature database.

[0053] For example, the first feature image is input into the feature vector extraction module, the color histogram feature is obtained through color distribution statistics, the texture features such as texture orientation, contrast and energy are calculated through the gray-level co-occurrence matrix, the shape contour feature is generated through the edge detection result, and then the color feature, texture feature and shape feature are concatenated into feature vectors of each dimension.

[0054] For example, for an image of an insulator with obvious rust, the system obtains color features by statistically analyzing the color ratios of its RGB (Red, Green, Blue) channels, calculates texture roughness as texture features from the gray-level co-occurrence matrix (GLCM), and extracts the shape complexity of the rusted blocks from the edge contours. Finally, the three types of features are combined into a first feature image feature vector with a length of about 120 dimensions.

[0055] It should be noted that the feature vector extraction module includes a color feature extraction submodule, a texture feature extraction submodule, and a shape feature extraction submodule. Specifically, the texture feature extraction submodule calculates the co-occurrence statistics of pixel pairs with different gray values ​​in the input image at preset directions and distances based on the gray-level co-occurrence matrix, and calculates texture features such as orientation, contrast, and energy based on these statistics; the color feature extraction submodule calculates color distribution features based on the color histogram; and the shape feature extraction submodule generates shape contour features based on edge detection results.

[0056] In this example, the first sample image is input into the same feature extraction process as the first feature image, and processed according to the same color space, texture operator and shape descriptor rules to generate the color feature vector, texture feature vector and shape feature vector of the first sample image, ensuring that the feature dimensions are consistent between different samples.

[0057] For example, for a wire clamp sample image in the library that is labeled as "slightly corroded", the system calculates a 32-dimensional color histogram, a 64-dimensional texture feature and a 24-dimensional shape descriptor from the image, and finally constructs a 120-dimensional standardized sample feature vector.

[0058] In this example, for each first feature image, similarity is measured with the color feature vector, texture feature vector, and shape feature vector of all first sample images. The color cosine similarity is calculated as the first similarity value, the normalized Euclidean distance of the texture feature is used as the second similarity value, and the normalized geodesic distance of the shape feature is used as the third similarity value. The three similarities are then weighted and summed to obtain a comprehensive similarity value. The feature images corresponding to the samples with a comprehensive similarity greater than a preset similarity threshold are determined as the second feature images.

[0059] For example, a newly extracted image of a bright erosion spot has high similarity to multiple "severely eroded" images in the sample library in terms of color, texture, and shape. Its overall similarity is 0.87, which is higher than the threshold of 0.80. Therefore, this image is identified as the second feature image.

[0060] In this example, all second feature images under the same fault type are input into a clustering model (e.g., K-means algorithm) according to their feature vectors. K-means clustering or hierarchical clustering is performed based on the sample centers of each fault level. The distance from each second feature image to the cluster center of each level is calculated, and the level corresponding to the cluster center with the smallest distance is taken as the fault level label.

[0061] For example, for the second feature image under "rust failure type", its feature vector is clustered into three categories: "slight rust (L1)", "moderate rust (L2)" and "severe rust (L3)". Among them, a certain second feature image is closest to the L2 cluster center and is therefore assigned the label "moderate rust".

[0062] In this example, "fault type" is used as the first-level index node, and each fault level under each fault type is established as a second-level child node. The feature vector of each second feature image is used as the key, and its corresponding fault level label is used as the associated data. The feature vector is inserted into the leaf nodes of B+ in ascending order, while ensuring that the leaf nodes are linked in the order of the linked list to form an ordered index structure.

[0063] For example, the root node of the tree can be a "rust" or "damage" class, and each class is further divided into "mild", "moderate" and "severe". The feature vector (length 120) of a moderate rust feature image is stored in the leaf node corresponding to the "rust-moderate" path and retains its label information.

[0064] In this example, the complete B+ tree index structure is serialized according to the database file format, the node keys, child node pointers and leaf node contents are encoded, and the serialized data is written to the database file to achieve structured storage and fast loading.

[0065] For example, the system encodes the constructed B+ tree into JavaScript Object Notation (JSON) or a binary structured file and writes it to a database file such as "feature_db.bin". Subsequent diagnostic tasks can directly load this file for efficient index retrieval.

[0066] According to the above implementation method, by extracting multi-dimensional feature vectors from the first feature image and the first sample image respectively, and generating a second feature image based on feature similarity, the association between "original features" and "sample features" is constructed. Then, clustering is performed according to fault type and fault level, mapping each feature image to a clear level label, thus semantically organizing and hierarchizing the features. Finally, a B+ tree index is constructed with fault type as nodes, fault level as child nodes, and feature vectors and level labels as leaf nodes, and written into the database file, giving the database an efficient retrieval structure and a clear hierarchical feature organization method. This significantly improves the structure and retrieval efficiency of the feature library, thereby enhancing the overall reliability and real-time performance of fault diagnosis.

[0067] In one implementation, determining each second feature image based on the similarity values ​​of each dimension feature of each first feature image and each dimension feature of each first sample image includes: acquiring each dimension feature of each first feature image, wherein each dimension feature of the first feature image includes a first color feature, a first texture feature, and a first shape feature; and acquiring each dimension feature of each first sample image, wherein each dimension feature of the first sample image includes a second color feature, a second texture feature, and a second shape feature; performing the following similarity calculations on each first feature image and each second feature image: determining a first similarity value based on the similarity value between the first color feature and the second color feature; determining a second similarity value based on the similarity value between the first texture feature and the second texture feature; and determining a third similarity value based on the similarity value between the first shape feature and the second shape feature; performing a weighted summation of the first similarity value, the second similarity value, and the third similarity value to obtain a comprehensive similarity value between the first feature image and the second feature image; and determining the feature images corresponding to the comprehensive similarity values ​​that are greater than a preset similarity threshold as the second feature images.

[0068] For example, the first feature image is input into a preset feature extraction model, converted into a standard color space, and a statistical color histogram is used as the first color feature. Texture parameters such as contrast, energy, and correlation are calculated through the gray-level co-occurrence matrix as the first texture feature. Then, the multi-scale shape descriptor of the object's shape is obtained using edge detection and contour extraction algorithms as the first shape feature. Finally, the three types of features are combined to form the feature vectors of each dimension of the first feature image.

[0069] For example, for an image of an insulator surface showing a clear rust area, the proportion of its red component is statistically analyzed to obtain the first color feature, the rough texture parameters are obtained through GLCM as the first texture feature, and the rust outline is extracted using the Canny Edge Detection Algorithm (Canny) to generate the first shape feature.

[0070] In this example, the first sample image is input into the same feature extraction process as the first feature image. The second color feature is obtained through the same color space conversion, the second texture feature is obtained through the same texture operator, and the second shape feature is generated through the same contour extraction method to ensure dimensional consistency and comparability between different images.

[0071] For example, for an insulator sample image labeled "slightly damaged" in the library, its color histogram is calculated as the second color feature, GLCM is used to obtain the fine line parameters as the second texture feature, and the shape vector of the damaged area is generated using the edge contour as the second shape feature.

[0072] In this example, for each first feature image, the three types of features are compared sequentially with each first sample image. The first similarity value is obtained by calculating the cosine similarity of the color feature vector, the second similarity value is obtained by normalizing the texture feature vector by Euclidean distance, and the third similarity value is obtained by normalizing the shape feature vector by geodesic distance, so that the three similarity values ​​are all at a comparable and uniform scale.

[0073] For example, the color distribution of a first feature image is very close to that of a "moderately rusted" sample, with a color cosine similarity of 0.92, thus obtaining the first similarity value; the second and third similarity values ​​corresponding to texture and shape are 0.85 and 0.78, respectively.

[0074] In this example, the first similarity value, the second similarity value, and the third similarity value are obtained. They are then linearly weighted according to preset color weight, texture weight, and shape weight. The three weighted similarity values ​​are added together to generate a comprehensive similarity value between the first feature image and the first sample image.

[0075] For example, with similarity weights set to color 0.4, texture 0.4, and shape 0.2, the weighted similarity values ​​of 0.92, 0.85, and 0.78 yield a combined similarity value of 0.856.

[0076] In this example, all comprehensive similarity values ​​are iterated and compared with a preset similarity threshold. If the comprehensive similarity value is greater than the threshold, the corresponding feature image is marked as a second feature image and added to the second feature image set for subsequent clustering and index construction. For example, if the similarity threshold is set to 0.80, and the comprehensive similarity value of 0.856 is greater than the threshold, then the feature image is considered to have a high degree of matching with existing samples and is included in the second feature image set.

[0077] According to the above implementation method, multi-dimensional features such as color, texture, and shape are extracted from the first feature image and the first sample image, respectively. These features are then weighted and fused based on their similarity across dimensions to obtain a comprehensive similarity value that reflects both the appearance and structural differences of the images. By setting a similarity threshold, only feature images with high comprehensive similarity are retained as the second feature images, thus forming a feature set that is more consistent with the distribution of real samples. This effectively improves the accuracy of feature selection, making the generated second feature images more representative and reliable, laying the foundation for building a high-quality feature database.

[0078] In one implementation, the integrity verification of the first feature database to obtain the target feature database includes: performing hierarchical sampling verification on the first feature database to obtain a first verification result; performing segmentation verification on the first feature database based on a first segmentation model to obtain a second verification result; performing coverage verification on the first feature database to obtain a third verification result; and confirming that the first feature database has passed the verification if the first verification result, the second verification result, and the third verification result are all verified to obtain the target feature database.

[0079] For example, the first feature database is layered according to fault type and fault level, and a certain proportion of feature vector samples are randomly selected from each layer. The feature integrity and consistency of the selected samples are calculated, specifically including the validity check of color, texture and shape features, the statistical analysis of the proportion of abnormal or missing features, and the comparison of the verification results with a preset threshold to generate a first verification result.

[0080] For example, for three types of rust faults—light, moderate, and severe—10% of the samples were drawn from each type. It was found that all samples had complete and normal color, texture, and shape characteristics, and the first verification result showed "passed".

[0081] For example, each image feature in the first feature database is input into a pre-trained segmentation model to perform region segmentation on the image and extract key region features. The key region features are compared with the region features stored in the database, and a segmentation consistency index (e.g., intersection-over-union ratio) is calculated. If the consistency index is greater than a preset threshold, the feature is deemed to have passed the verification, and the overall results are summarized to form a second verification result. The pre-trained segmentation model is, for example, a U-shaped convolutional neural network (U-Net).

[0082] For example, if a moderately corroded image segments out rust spots and its intersection-over-union ratio (IoU) is 0.87 compared to the rust spot labeling value in the database, exceeding the threshold of 0.85, then the image passes the verification. The second verification result for all images is "passed".

[0083] For example, the feature vector coverage of various fault types and levels in the first feature database is statistically analyzed, including color distribution, texture pattern, and shape distribution, to check for any uncovered feature space regions (i.e., missing regions). Missing regions are recorded and labeled, coverage metrics are calculated and compared with a preset coverage threshold, and a third verification result is generated.

[0084] For example, the texture feature space coverage of the light corrosion database is 95%, which is higher than the set threshold of 90%. The coverage of the medium and heavy corrosion databases also meets the requirements. Therefore, the third verification result shows "pass".

[0085] For example, a logical AND operation is performed on the first, second, and third verification results. If all three are "passed," the entire first feature database is marked as having passed verification. The database is then formally stored and named the target feature database, serving as a valid basis for subsequent image matching and fault diagnosis. Specifically, if the hierarchical sampling verification, segmentation verification, and coverage verification all pass, the system confirms the database is complete and reliable. The first feature database is then formally output as the target feature database for use in matching images to be diagnosed.

[0086] According to the above implementation method, by sequentially performing hierarchical sampling verification, model segmentation verification, and coverage verification on the first feature database, the balance of sample distribution, the accuracy of feature annotation, and the completeness of fault type and level coverage in the database are verified from different dimensions, forming a multi-angle, complementary verification result. When all three types of verification pass, the database quality can be confirmed to meet the diagnostic requirements, and the target feature database can be generated. In this way, problems such as biased samples, mislabeling, or missing types in the feature database can be effectively avoided, providing more stable and reliable feature support for subsequent image matching and fault diagnosis.

[0087] In one implementation, the fault diagnosis result is obtained by matching the image data to be diagnosed with each image data in the target feature database. This includes: performing multi-dimensional feature extraction on the image data to be diagnosed to obtain the features of each dimension of the image to be diagnosed; searching for samples in the target feature database using a fast nearest neighbor search algorithm based on the features of each dimension of the image to be diagnosed and the features of each image in the target feature database to obtain each target sample; performing weighted voting based on the label information of each target sample to obtain the fault type and fault level of the image data to be diagnosed; generating the location information of the image data to be diagnosed using preset first algorithms; wherein the first algorithms include image segmentation and contour detection algorithms; and determining the fault diagnosis result based on the fault type, fault level, and location information of the image data to be diagnosed.

[0088] For example, the image to be diagnosed is preprocessed, including grayscale conversion, size normalization, noise reduction, and contrast enhancement; color features (such as hue-saturation-brightness histogram), texture features, and shape features (such as edge curvature and contour length) are extracted from the preprocessed image; the extracted features are combined into a multi-dimensional feature vector for subsequent matching.

[0089] For example, an image of an insulator with slight rust, after preprocessing, has color features of [120, 90, 80], texture features of [0.25, 0.30, 0.28], and shape features of 45 pixels. These three features are then combined into a feature vector of length 256.

[0090] For example, the feature vector of the image to be diagnosed is input into a fast nearest neighbor search; several image samples most similar to the feature vector are retrieved from the target feature database, and the similarity values ​​are recorded; the retrieved images and their label information are output as target samples. For example, the feature vector is searched in the database to find the 5 images with the highest similarity, corresponding to light corrosion, light corrosion, heavy corrosion, moderate damage, and light damage, respectively.

[0091] Among them, fast nearest neighbor search, such as K-Dimensional Tree (KD-Tree) or Facebook AI Similarity Search (FAISS).

[0092] For example, the distribution of fault types and levels of the target samples is statistically analyzed, and different samples are assigned weights based on their similarity values ​​(the weights can be set according to actual needs) to perform weighted voting on fault types and levels. The type and level with the highest number of votes or weighted values ​​are selected as the judgment result for the image to be diagnosed.

[0093] For example, in 5 target samples, mild corrosion appeared 3 times (similarity weighted value 2.7), moderate damage appeared once (similarity weighted value 0.5), and severe corrosion appeared once (similarity weighted value 0.3). Finally, the image to be diagnosed was determined to be mild corrosion.

[0094] For example, image segmentation (such as U-Net or thresholding) is performed on the image to be diagnosed to obtain a mask of the defect region; contour detection (such as OpenCV's findContours function) is performed on the mask to extract the boundary coordinates and center position of the defect region to form localization information.

[0095] For example, a lightly corroded area is segmented using U-Net, and contour detection yields the boundary coordinates [(120, 45), (125, 50), ...], with the center point (123, 48). The boundary coordinates and center point are recorded as location information.

[0096] For example, the fault type, fault level, and location information obtained from weighted voting are integrated to generate a complete diagnostic result, including the defect category, severity, specific location, and potentially affected area. For instance, the diagnostic result is: fault type "slight corrosion", fault level "slight", defect area center (123, 48), covering an area of ​​approximately 50 square pixels.

[0097] According to the above implementation method, after multi-dimensional feature extraction of the image data to be diagnosed, a fast nearest neighbor search algorithm is used to efficiently match similar samples in the target feature database. Weighted voting is then performed based on the labels of the matching results to accurately determine the fault type and fault level of the image to be diagnosed. Subsequently, image segmentation and contour detection algorithms are combined to generate precise location information, ultimately forming a complete fault diagnosis result. This not only significantly improves the accuracy and robustness of the diagnostic results but also outputs visualized location information that can be used for operational and maintenance decisions.

[0098] Figure 2 This is a structural block diagram of a power transmission line fault diagnosis device according to an embodiment of the present invention.

[0099] like Figure 2 As shown, the transmission line fault diagnosis device may include: The sample expansion module 510 is used to expand the first sample images corresponding to each fault type to obtain a set of first sample images; Feature extraction module 520 is used to extract features from each second sample image in the first sample image set to obtain each first feature image; The feature database construction module 530 is used to construct a feature database based on each of the first feature images to obtain a first feature database. The integrity verification module 540 is used to perform integrity verification on the first feature database to obtain the target feature database; The matching module 550 is used to match the image data to be diagnosed with each image data in the target feature database to obtain the fault diagnosis result.

[0100] In one embodiment, the sample expansion module includes: An expansion unit is used to determine the corresponding expansion strategy based on the fault type of each first sample image and expand each first sample image to obtain a first sub-sample image set. The subsample image enhancement unit is used to enhance each subsample image corresponding to the first fault level in the first subsample image set through a support vector machine model, so as to update the first subsample image set and obtain the second subsample image set. An adjustment unit is used to adjust each of the second sub-sample images in the second sub-sample image set based on the fault level to obtain a third sub-sample image set; The gradient magnitude calculation unit is used to calculate the gradient magnitude of each rust sample image in the third sub-sample image set through an edge detection algorithm, so as to obtain the edge sharpness of each rust sample image. The first edge enhancement unit is used to perform edge enhancement on each of the rust sample images based on the Laplacian operator as the convolution kernel when the edge sharpness of each of the rust sample images is less than a preset sharpness threshold, so as to update the third sub-sample image set and obtain the first sample image set.

[0101] In one embodiment, the feature extraction module includes: A standardization processing unit is used to perform standardization processing on each of the second sample images to obtain each standardized image; An edge sharpness detection unit is used to perform edge sharpness detection on each of the standardized images to obtain an edge sharpness matrix corresponding to each of the standardized images. The second edge enhancement processing unit is used to perform edge enhancement processing on the standardized image corresponding to the edge sharpness matrix of each standardized image, when the maximum gradient value in the edge sharpness matrix is ​​less than a preset gradient threshold, to obtain an edge-enhanced image. An optimization unit is used to optimize each of the edge enhancement images to obtain each optimized feature image; The texture extraction unit is used to extract textures based on the optimized feature images of the first filter to obtain the first feature images.

[0102] In one implementation, the feature database construction module includes: The first feature vector extraction unit is used to extract feature vectors from each of the first feature images to obtain the features of each of the first feature images in various dimensions. The second feature vector extraction unit is used to extract feature vectors from each of the first sample images to obtain the features of each of the first sample images in various dimensions. The similarity calculation unit is used to determine each second feature image based on the similarity values ​​between each first feature image and each first sample image in each dimension of the features. A clustering unit is used to cluster each of the second feature images based on each fault level for each fault type, so as to obtain the fault level label of each of the second feature images under the fault level. B+ tree index structure unit is used to take each of the fault types as tree nodes, each of the fault levels as child nodes, and each of the dimensional features of the second feature image and each of the fault level labels as leaf nodes to obtain a B+ tree index structure. A storage unit is used to store the B+ tree index structure into a database file to obtain the first feature database.

[0103] In one implementation, the similarity calculation unit is specifically used for: The first feature image is obtained in terms of each dimension, wherein each dimension of the first feature image includes a first color feature, a first texture feature, and a first shape feature; and the first sample image is obtained in terms of each dimension, wherein each dimension of the first sample image includes a second color feature, a second texture feature, and a second shape feature. For each of the first feature images and each of the second feature images, the following similarity calculation is performed: A first similarity value is determined based on the similarity value between the first color feature and the second color feature; a second similarity value is determined based on the similarity value between the first texture feature and the second texture feature; and a third similarity value is determined based on the similarity value between the first shape feature and the second shape feature. The first similarity value, the second similarity value, and the third similarity value are weighted and summed to obtain the comprehensive similarity value between the first feature image and the second feature image; Among the various comprehensive similarity values, the feature images corresponding to the comprehensive similarity values ​​that are greater than the preset similarity threshold are determined as the respective second feature images.

[0104] In one embodiment, the integrity verification module includes: A hierarchical sampling verification unit is used to perform hierarchical sampling verification on the first feature database to obtain a first verification result; The segmentation verification unit is used to perform segmentation verification on the first feature database based on the first segmentation model to obtain a second verification result. The coverage verification unit is used to perform coverage verification on the first feature database to obtain a third verification result; The verification and confirmation unit is used to confirm that the first feature database has passed the verification when the first verification result, the second verification result, and the third verification result are all verified as passed, so as to obtain the target feature database.

[0105] In one implementation, the matching module includes: A multidimensional feature extraction unit is used to extract multidimensional features from the image data to be diagnosed, thereby obtaining the features of each dimension of the image to be diagnosed. The sample search unit is used to search for samples in the target feature database based on the features of each dimension of the image to be diagnosed and the features of each dimension of each image in the target feature database, and obtain each target sample by using a fast nearest search algorithm. The weighted voting unit is used to perform weighted voting based on the label information of each target sample to obtain the fault type and fault level of the image data to be diagnosed. The generation unit is used to generate the positioning information of the image data to be diagnosed by using various preset first algorithms; wherein the first algorithms include image segmentation and contour detection algorithms. The fault diagnosis result determination unit is used to determine the fault diagnosis result based on the fault type, fault level and location information of the image data to be diagnosed.

[0106] The specific functions and examples of each module and submodule of the system in this embodiment of the invention can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0107] The acquisition, storage, and application of user personal information involved in the technical solution of this invention all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0108] This invention also provides a transmission line fault diagnosis system, comprising: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method described in any one of the embodiments of the present invention.

[0109] The beneficial effects of the transmission line fault diagnosis system of this invention are equivalent to the beneficial effects of the above-described transmission line fault diagnosis method, and will not be repeated here.

[0110] This invention also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the method described in any one of the embodiments of this invention.

[0111] The beneficial effects of the storage medium of the present invention are equivalent to the beneficial effects of the above-described transmission line fault diagnosis method, and will not be repeated here.

[0112] Figure 3 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present invention is shown. Electronic device 800 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 800 may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0113] like Figure 3 As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. The RAM 803 may also store various programs and data required for the operation of the electronic device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0114] Multiple components in electronic device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows electronic device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0115] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as a transmission line fault diagnosis method. For example, in some embodiments, the transmission line fault diagnosis method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the transmission line fault diagnosis method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the transmission line fault diagnosis method by any other suitable means (e.g., by means of firmware).

[0116] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0117] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0118] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0119] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0120] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0121] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0122] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0123] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for diagnosing faults in transmission lines, characterized in that, include: The first sample images corresponding to each fault type are expanded to obtain a set of first sample images; Feature extraction is performed on each second sample image in the first sample image set to obtain each first feature image; A feature database is constructed based on each of the first feature images to obtain the first feature database; The integrity of the first feature database is verified to obtain the target feature database. The image data to be diagnosed is matched with each image data in the target feature database to obtain the fault diagnosis result.

2. The method according to claim 1, characterized in that, The first sample image set is obtained by expanding the first sample image corresponding to each fault type, including: Based on the fault type of each first sample image, a corresponding expansion strategy is determined to expand each first sample image to obtain a set of first subsample images. By using a support vector machine model, each sub-sample image corresponding to the first fault level in the first sub-sample image set is enhanced to update the first sub-sample image set and obtain the second sub-sample image set. Based on the fault level, each second sub-sample image in the second sub-sample image set is adjusted to obtain the third sub-sample image set; The gradient magnitude of each rust sample image in the third sub-sample image set is calculated using an edge detection algorithm to obtain the edge sharpness of each rust sample image. If the edge sharpness of each of the corrosion sample images is less than a preset sharpness threshold, edge enhancement is performed on each of the corrosion sample images based on the Laplacian operator as the convolution kernel, so as to update the third sub-sample image set and obtain the first sample image set.

3. The method according to claim 1, characterized in that, The step of extracting features from each of the second sample images in the first sample image set to obtain each of the first feature images includes: Each of the second sample images is standardized to obtain a standardized image. Edge sharpness detection is performed on each of the standardized images to obtain the edge sharpness matrix corresponding to each of the standardized images; For each of the standardized images, if the maximum gradient value in the edge sharpness matrix is ​​less than a preset gradient threshold, edge enhancement processing is performed on the standardized image corresponding to the edge sharpness matrix to obtain an edge-enhanced image. Each of the edge enhancement images is optimized to obtain an optimized feature image; Texture extraction is performed on each of the optimized feature images based on the first filter to obtain each of the first feature images.

4. The method according to claim 1, characterized in that, The construction of the feature database based on each of the first feature images to obtain the first feature database includes: Feature vectors are extracted from each of the first feature images to obtain the features of each of the first feature images in each dimension. Feature vectors are extracted from each of the first sample images to obtain the features of each of the first sample images in various dimensions. Based on the similarity values ​​of each dimension of the first feature image and each dimension of the first sample image, each second feature image is determined. For each of the aforementioned fault types, based on each fault level, each of the second feature images is clustered to obtain a fault level label for each of the second feature images under the fault level. By taking each of the fault types as tree nodes, each of the fault levels as child nodes, and each of the dimensional features of the second feature image and each of the fault level labels as leaf nodes, a B+ tree index structure is obtained. The B+ tree index structure is stored in a database file to obtain the first feature database.

5. The method according to claim 4, characterized in that, Determining each second feature image based on the similarity values ​​between the features of each first feature image and the features of each first sample image in each dimension includes: The first feature image is obtained in terms of each dimension, wherein each dimension of the first feature image includes a first color feature, a first texture feature, and a first shape feature; and the first sample image is obtained in terms of each dimension, wherein each dimension of the first sample image includes a second color feature, a second texture feature, and a second shape feature. For each of the first feature images and each of the second feature images, the following similarity calculation is performed: A first similarity value is determined based on the similarity value between the first color feature and the second color feature; a second similarity value is determined based on the similarity value between the first texture feature and the second texture feature; and a third similarity value is determined based on the similarity value between the first shape feature and the second shape feature. The first similarity value, the second similarity value, and the third similarity value are weighted and summed to obtain the comprehensive similarity value between the first feature image and the second feature image; Among the various comprehensive similarity values, the feature images corresponding to the comprehensive similarity values ​​that are greater than the preset similarity threshold are determined as the respective second feature images.

6. The method according to claim 1, characterized in that, The step of performing integrity verification on the first feature database to obtain the target feature database includes: The first feature database is subjected to hierarchical sampling verification to obtain a first verification result; The first feature database is segmented and verified based on the first segmentation model to obtain the second verification result; A coverage check is performed on the first feature database to obtain a third check result; If the first verification result, the second verification result, and the third verification result are all verified as passed, the first feature database is confirmed to have passed the verification, so as to obtain the target feature database.

7. The method according to claim 1, characterized in that, The step of matching the image data to be diagnosed with each image data in the target feature database to obtain the fault diagnosis result includes: Multidimensional feature extraction is performed on the image data to be diagnosed to obtain the features of the image in each dimension. Based on the features of each dimension of the image to be diagnosed and the features of each dimension of each image in the target feature database, samples are searched in the target feature database using a fast nearest search algorithm to obtain each target sample; The fault type and fault level of the image data to be diagnosed are obtained by weighted voting based on the label information of each target sample. The positioning information of the image data to be diagnosed is generated by various preset first algorithms; wherein, the first algorithm includes image segmentation and contour detection algorithms. Based on the fault type, fault level, and location information of the image data to be diagnosed, the fault diagnosis result is determined.

8. A fault diagnosis device for power transmission lines, characterized in that, include: The sample expansion module is used to expand the first sample images corresponding to each fault type to obtain a set of first sample images; The feature extraction module is used to extract features from each of the second sample images in the first sample image set to obtain each first feature image; The feature database construction module is used to construct a feature database based on each of the first feature images to obtain the first feature database; The integrity verification module is used to perform integrity verification on the first feature database to obtain the target feature database. The matching module is used to match the image data to be diagnosed with each image data in the target feature database to obtain the fault diagnosis result.

9. A transmission line fault diagnosis system, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images