Power transmission line risk judgment method and system and storage medium

Abstract

The invention discloses a power transmission line risk judgment method and system and a storage medium. The method comprises the following steps: acquiring a shot image of a target insulator in a power transmission line; inputting a shot image of a target insulator in the power transmission line into the trained multi-scale feature extraction model to obtain a first scale feature, a second scale feature, a third scale feature and a fourth scale feature; performing feature fusion on the first scale feature, the second scale feature and the third scale feature by using a fusion layer to obtain a first fusion feature; inputting the fourth scale feature and the first fusion feature into a trained feature fusion model for processing to obtain a defect fusion feature of the target insulator; searching a preset database according to the defect fusion feature of the target insulator, and determining a line risk; characteristic defects can be effectively identified, and the power transmission line risk judgment efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a power transmission line risk judgment method and system and a storage medium. BACKGROUND

[0002] In the power transmission line, the insulator plays an indispensable role. It not only provides stable support for the conductive part, but also realizes efficient isolation to reliably separate the live wire from the grounded part such as the tower, thereby ensuring the safety and stability of power transmission. It is one of the core components to ensure the reliable operation of the power grid. However, the power transmission line is exposed to complex and changeable natural environment for a long time, and needs to withstand strong wind, heavy rain, snow, lightning and day and night temperature difference and other extreme conditions. The problems caused by flashover damage, spontaneous combustion, aging and brittle fracture occur frequently. These problems often lead to a sudden drop or even loss of insulation performance, thereby causing line tripping, power failure accidents, and even secondary disasters such as forest fires. This makes the accurate detection of insulator defect features a key link to ensure the safe operation of the power transmission line - through real-time monitoring of the appearance defects, dirt accumulation, material degradation and other features of the insulator, potential risks can be warned in advance, and scientific basis can be provided for operation and maintenance decision.

[0003] At present, the feature recognition method for insulators in the power transmission line mainly relies on traditional image processing methods such as Hough transform, Canny edge extraction, ant colony clustering, etc. However, these methods have obvious limitations: their core assumptions are often based on ideal scenarios where the image background is simple and the insulator defect features are prominent, and recognition is achieved through preset rules or local feature matching. However, in actual aerial inspection, images are often disturbed by uneven lighting, cloud and fog blocking, camera shaking and other noise, and complex elements such as trees and buildings in the background are highly similar to the insulator structure, and the defect types (such as cracks, pollution flashover, damage) are significantly different in shape, size and texture, making it difficult for traditional algorithms to extract stable features, resulting in high false detection rate and missed detection rate, and thus low efficiency of line risk judgment, which cannot meet the needs of large-scale power grid intelligent operation and maintenance.

[0004] In addition, the existing feature recognition method for insulators in the power transmission line also uses the YOLOv8 model to recognize the features of the insulators in the power transmission line, which can improve the recognition speed and generalization ability with the help of its efficient target detection framework, but the detection accuracy of the model for small defects (such as early cracks of insulators) is limited, and false detection is also likely to occur under complex background interference, affecting the efficiency of line risk judgment. SUMMARY

[0005] The purpose of the present application is to provide a power transmission line risk judgment method and system and a storage medium, which can accurately recognize feature defects through a target model and improve the efficiency of power transmission line risk judgment.

[0006] To achieve the above object, the present application adopts the following technical solutions: In a first aspect, the present application provides a power transmission line risk judgment method, comprising: Obtaining a photographed image of a target insulator in a power transmission line; Inputting the photographed image of the target insulator in the power transmission line into a trained target model for processing to obtain a defect fusion feature of the target insulator; wherein the target model comprises a multi-scale feature extraction model, a fusion layer and a feature fusion model; the processing comprises: Inputting the photographed image of the target insulator in the power transmission line into the trained multi-scale feature extraction model to obtain a first scale feature, a second scale feature, a third scale feature and a fourth scale feature; Using the fusion layer to perform feature fusion on the first scale feature, the second scale feature and the third scale feature to obtain a first fusion feature; Inputting the fourth scale feature and the first fusion feature into the trained feature fusion model for processing to obtain the defect fusion feature of the target insulator; According to the defect fusion feature of the target insulator, searching a preset database to determine the line risk.

[0007] Optionally, it further comprises: before inputting the photographed image of the target insulator in the power transmission line into the trained target model for processing, performing preprocessing, the preprocessing comprises: Performing denoising processing on the photographed image to obtain a denoised photographed image; Performing geometric correction on the denoised photographed image to obtain a preliminarily processed photographed image; Inputting the preliminarily processed photographed image into a preset ResNet model for image decomposition to obtain a first initial image set; For each initial image in the first initial image set, performing denoising processing on the initial image by a U-Net attention model to obtain a denoised initial image; Fusing all the denoised initial images to obtain a preprocessed photographed image.

[0008] Optionally, the multi-scale feature extraction model is improved according to a backbone network of a YOLOv8 model, the backbone network of the YOLOv8 model comprises a first CBS module, a second CBS module, a third C2f module, a fourth CBS module, a fifth C2f module, a sixth CBS module, a seventh C2f module, an eighth CBS module, a ninth C2f module and a tenth SPPF module connected in sequence; the improvement comprises: Replacing the second CBS module, the fourth CBS module, the sixth CBS module and the eighth CBS module with a second deep convolution module, a fourth deep convolution module, a sixth deep convolution module and an eighth deep convolution module having the same structure, respectively; replacing the tenth SPPF module with a pooling module; adding an attention module after the backbone network.

[0009] Optionally, the input of the photographed image of the target insulator in the power transmission line into the trained multi-scale feature extraction model obtains a first scale feature, a second scale feature, a third scale feature, and a fourth scale feature, including: The photographed image is sequentially processed by a first CBS module, a second deep convolution module, a third C2f module, the fourth deep convolution module, and the fifth C2f module to obtain a first scale feature; The first scale feature is sequentially processed by a sixth deep convolution module and a seventh C2f module to obtain a second scale feature; The second scale feature is sequentially processed by an eighth deep convolution module, a ninth C2f module, and a pooling module to obtain a third scale feature; The third scale feature is processed by the attention module to obtain a fourth scale feature.

[0010] Optionally, the structures of the second deep convolution module, the fourth deep convolution module, the sixth deep convolution module, and the eighth deep convolution module each include a two-dimensional convolution layer and a first fusion layer connected with the two-dimensional convolution layer; each deep convolution module is used for: obtaining an input feature, inputting the input feature into the two-dimensional convolution layer for feature extraction to obtain a first convolution feature; inputting the first convolution feature and the input feature into the first fusion layer for fusion to obtain an output feature.

[0011] Optionally, the pooling module includes a first maximum pooling layer, a second maximum pooling layer, and a third maximum pooling layer connected in sequence, a second fusion layer connected with the first maximum pooling layer, the second maximum pooling layer, and the third maximum pooling layer, and a fourth maximum pooling layer connected with the second fusion layer; the pooling module is used for: substituting the input of the pooling module into the first maximum pooling layer to obtain a first pooling feature; substituting the first pooling feature into the second maximum pooling layer to obtain a second pooling feature; substituting the second pooling feature into the third maximum pooling layer to obtain a third pooling feature; fusing the first pooling feature, the second pooling feature, and the third pooling feature by using the second fusion layer to obtain a fused pooling feature; substituting the fused pooling feature into the fourth maximum pooling layer to obtain the output of the pooling module.

[0012] Optionally, the feature fusion model comprises: a first max-pooling layer, a first average-pooling layer, a second max-pooling layer, a second average-pooling layer, a third fusion layer connected with the first max-pooling layer and the first average-pooling layer, a fourth fusion layer connected with the second max-pooling layer and the second average-pooling layer, and a fifth fusion layer connected with the third fusion layer and the fourth fusion layer; and the trained feature fusion model is used for: inputting the fourth scale feature into the first max-pooling layer and the first average-pooling layer respectively, performing max-pooling and average-pooling along a width dimension to obtain a width dimension max-pooling feature and a width dimension average-pooling feature; inputting the fourth scale feature into the second max-pooling layer and the second average-pooling layer respectively, performing max-pooling and average-pooling along a height dimension to obtain a height dimension max-pooling feature and a height dimension average-pooling feature; inputting the width dimension max-pooling feature and the width dimension average-pooling feature into the third fusion layer to obtain a width dimension fusion feature; inputting the height dimension max-pooling feature and the height dimension average-pooling feature into the fourth fusion layer to obtain a height dimension fusion feature; inputting the width dimension fusion feature, the height dimension fusion feature and the first fusion feature into the fifth fusion layer to obtain a defect fusion feature of the target insulator.

[0013] Optionally, the training process of the target model comprises: obtaining a set of photographed images of insulators in a power transmission line; preprocessing images in the set of photographed images to obtain a set of preprocessed images; annotating defect types of the set of preprocessed images to obtain a training set of the target model inputting the training set into the target model to train the target model, and obtaining optimal model parameters of the target model; determining a trained target model based on the optimal model parameters.

[0014] In a second aspect, the present application provides a power transmission line risk judgment system, comprising: a data acquisition module configured to obtain photographed images of target insulators in a power transmission line; a data processing module configured to input the photographed images of the target insulators in the power transmission line into a trained target model to process the photographed images and obtain defect fusion features of the target insulators; wherein the target model comprises a multi-scale feature extraction model, a fusion layer and a feature fusion model; and the data processing module is further configured to: input the photographed images of the target insulators in the power transmission line into the trained multi-scale feature extraction model to obtain first scale features, second scale features, third scale features and fourth scale features; The first fused feature is obtained by fusing the first-scale feature, the second-scale feature, and the third-scale feature using a fusion layer; The fourth-scale feature and the first fusion feature are input into the trained feature fusion model for processing to obtain the defect fusion feature of the target insulator. The risk assessment module is used to search a preset database based on the defect fusion characteristics of the target insulator to determine the line risk. Thirdly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed, implements the transmission line risk assessment method described in the first aspect.

[0015] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: This invention provides a method, system, and storage medium for assessing the risks of transmission lines. The method inputs captured images of target insulators in the transmission line into a trained target model for processing, obtaining defect fusion features of the target insulators. The target model includes a multi-scale feature extraction model and a feature fusion model connected to the multi-scale feature extraction model. Through the multi-scale feature extraction model, defect features in the captured images are first extracted at multiple scales to obtain first-scale features, second-scale features, third-scale features, and fourth-scale features, which can more comprehensively reflect the local and global features of the captured images. Then, a fusion layer is used to combine the first three scale features. The output features are extracted and fused. Then, the fused features and fourth-scale features are further fused through a feature fusion model to obtain defect fusion features. This further integrates information from different levels and scales to form a more representative and discriminative feature representation, improving the ability to detect defect features in captured images. Finally, the risk of transmission lines is determined by searching a preset database based on the defect fusion features. This invention solves the problem of false detection and false negative rates caused by significant differences in the morphology, scale, and texture of cracks, flashover, and damage to insulators. It can more accurately identify the defect features of target insulators and effectively improve the efficiency of risk assessment for transmission lines.

[0016] This invention provides a method, system, and storage medium for assessing the risks of transmission lines. Before inputting the captured images of target insulators in the transmission line into the trained target model, the method performs preprocessing. By utilizing ResNet and U-Net attention models, it can effectively remove noise, blur, and other interference factors in the images, improving image clarity and contrast. High-quality captured images provide a better data foundation for subsequent feature extraction and analysis, helping to extract effective information from the captured images more accurately, thereby improving the accuracy of insulator defect feature detection and assessment.

[0017] The application provides a power transmission line risk judgment method and system and a storage medium.

[0018] The application provides a power transmission line risk judgment method and system and a storage medium, wherein the feature fusion model uses a maximum pooling layer and an average pooling layer to perform maximum pooling and average pooling in the width dimension and the height dimension respectively, and performs layer-by-layer fusion on the results, and finally fuses the first fused features; all-round information of image multi-scale and multi-direction is gathered to form highly comprehensive and more discriminative feature representation, so that more rich and accurate features are provided for subsequent power transmission line risk judgment, and the accuracy and reliability of the judgment are improved. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 Fig. 1 shows a flowchart of a power transmission line risk judgment method according to an embodiment of the application; Figure 2 Fig. 3 shows a processing process diagram of a trained target model according to an embodiment of the application; Figure 3 Fig. 4 shows a preprocessing flowchart according to an embodiment of the application; Figure 4 Fig. 5 shows a target model structure diagram according to an embodiment of the application. DETAILED DESCRIPTION

[0020] The application will be further described below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the application, and cannot be used to limit the protection scope of the application.

[0021] Embodiment 1

[0022] As shown in Fig. 1, the embodiment of the application introduces a power transmission line risk judgment method, which comprises the following steps: Figure 1 S1: obtaining a photographed image of a target insulator in a power transmission line; S2: inputting the photographed image of the target insulator in the power transmission line into a trained target model for processing to obtain defect fusion features of the target insulator; S3: searching a preset database according to the defect fusion features of the target insulator to determine the line risk.

[0023] ​The target model includes a multi-scale feature extraction model, a fusion layer, and a feature fusion model. For a specific structure, refer to Figure 4 ; Specifically, as shown in Figure 2 , the processing process of the target model includes: S21: input the photographed image of the target insulator in the power transmission line into the trained multi-scale feature extraction model to obtain a first scale feature, a second scale feature, a third scale feature, and a fourth scale feature; S22: obtain a first fusion feature by performing feature fusion on the first scale feature, the second scale feature, and the third scale feature; S23: input the fourth scale feature and the first fusion feature into the trained feature fusion model for processing to obtain a defect fusion feature of the target insulator.

[0024] In this embodiment, based on the power transmission line risk judgment method provided in the embodiment of the application, the multi-scale feature extraction model is used to perform multi-scale feature extraction on the photographed image of the target insulator, which can more comprehensively reflect the local and global features of the image target. Then, the fusion layer is used to integrate the feature information of different scales to obtain a first fusion feature, so that the feature contains richer semantic and detail information. Then, the fourth scale feature and the first fusion feature are substituted into the feature fusion model to obtain a defect fusion feature of the target insulator, which further integrates information of different levels and scales to form a more representative and discriminative feature representation, improves the feature detection capability of the target object, and finally determines the risk of the power transmission line according to the defect fusion feature and the preset database, thereby improving the accuracy of power transmission line fault detection (insulator defect recognition) and improving the risk judgment efficiency.

[0025] In this embodiment, before the photographed image of the target insulator in the power transmission line is input into the trained target model for processing in step S2, preprocessing is also required. As shown in Figure 3 , the preprocessing process includes: S01: perform denoising processing on the photographed image to obtain a denoised photographed image; S02: perform geometric correction on the denoised photographed image to obtain a preliminarily processed photographed image; S03: input the preliminarily processed photographed image into a preset ResNet model for image decomposition to obtain a first initial image set; S04: for each initial image in the first initial image set, perform denoising processing on the initial image by a U-Net attention model to obtain a denoised initial image; S05: fuse all the denoised initial images to obtain a preprocessed photographed image.

[0026] Specifically, the means for denoising the captured image include median filtering, Gaussian filtering and other algorithms, which can remove salt and pepper noise, Gaussian noise and the like in the image; geometric correction is based on control point matching technology to correct image distortion caused by shooting angle and lens distortion, so as to obtain a clear and accurate image.

[0027] Specifically, the preliminary processed captured image is decomposed to obtain a first set of initial images using a preset ResNet model. The ResNet model, as a kind of deep residual network, has strong feature extraction and expression ability, and can capture image features and structural information at different levels, segment the image into multiple sub-images, and each sub-image may contain different local features or structural parts of the image. This decomposition method helps to more carefully process each part of the image; and the deep structure and residual connection design of ResNet can effectively avoid the gradient vanishing problem, so that the network can be trained deeper, thereby extracting richer and higher level image features to provide a good foundation for subsequent processing.

[0028] Specifically, the preset ResNet model used in the embodiment includes a starting layer, a four-layer convolution framework and a post-processing layer, wherein the starting layer includes a 7x7 size convolution layer (channel 64, step 2) and a 3x3 size max pooling layer (step 2); the four-layer convolution framework has the same structure and includes three convolution layers, which are 1x1 convolution layer, 3x3 convolution layer and 1x1 convolution layer; the post-processing layer includes an average pooling layer and a full connection layer; the ResNet model and the U-Net attention model are network models commonly used by those skilled in the art, and their detailed structures are not specifically described here.

[0029] Specifically, step S05 fuses all the denoising initial images to obtain a preprocessed captured image. Through fusion processing, the advantages of each denoising initial image can be integrated, and the clear information of different parts can be integrated to restore a complete and high-quality captured image. This fusion method not only can further remove possible noise, but also can enhance the overall consistency and integrity of the image, so that the captured image contains more rich and accurate information. The high-quality captured image is of great significance to the subsequent feature extraction task, which can improve the accuracy of detection and recognition and reduce the false and missed cases.

[0030] This embodiment combines the processing methods of the ResNet model and the U-Net attention model, enabling it to adapt to images of varying types and complexities. ResNet's image decomposition can handle image diversity, while the U-Net attention model's noise reduction processing can perform targeted operations based on the characteristics of different sub-images. The overall process improves the robustness of the target model to various noises and image variations, enabling effective image processing under different shooting conditions and environmental factors. This ensures the reliability and stability of the processing results, providing a solid image foundation for subsequent transmission line risk assessment.

[0031] In this embodiment, the multi-scale feature extraction model is obtained by improving the backbone network of the YOLOv8 model. The backbone network of the YOLOv8 model includes a first CBS module, a second CBS module, a third C2f module, a fourth CBS module, a fifth C2f module, a sixth CBS module, a seventh C2f module, an eighth CBS module, a ninth C2f module, and a tenth SPPF module connected in sequence. Improvements to the backbone network include: (1) Replace the second CBS module, the fourth CBS module, the sixth CBS module, and the eighth CBS module with the second depthwise convolutional module, the fourth depthwise convolutional module, the sixth depthwise convolutional module, and the eighth depthwise convolutional module with the same structure, respectively; (2) Replace the tenth SPPF module with a pooling module; (3) Add an attention module after the backbone network.

[0032] like Figure 4 As shown, the multi-scale feature extraction model includes a first CBS module, a second deep convolutional module, a third C2f module, a fourth deep convolutional module, a fifth C2f module, a sixth deep convolutional module, a seventh C2f module, an eighth deep convolutional module, a ninth C2f module, a pooling module, and an attention module connected in sequence. The processing of images of target insulators based on a multi-scale feature extraction model includes: The captured image is processed sequentially through the first CBS module, the second depthwise convolution module, the third C2f module, the fourth depthwise convolution module, and the fifth C2f module to obtain the first scale feature; The first-scale features are processed sequentially by the sixth depthwise convolution module and the seventh C2f module to obtain the second-scale features. The second-scale features are processed sequentially by the eighth depthwise convolution module, the ninth C2f module, and the pooling module to obtain the third-scale features. The third-scale features are input into the attention module for processing to obtain the fourth-scale features.

[0033] Specifically, a well-trained multi-scale feature extraction model can extract image features from different scales, namely, first-scale features, second-scale features, third-scale features, and fourth-scale features. Features at different scales contain information about the image at different levels of detail. Small-scale features can capture local details and edge information of the image, while large-scale features can reflect the overall structure and semantic information of the image. The comprehensive use of multi-scale features can more comprehensively describe the image content and enhance the model's adaptability and robustness to various situations.

[0034] Specifically, the first CBS module serves as the starting point, performing preliminary feature extraction on the captured image and quickly capturing its basic feature information. The subsequent second, fourth, sixth, and eighth depth convolutional layers, compared to the previous CBS module, significantly reduce computational load while deeply mining features at different levels of the image. This combination ensures both comprehensive feature extraction and improved computational efficiency, enabling the target model to quickly process image data and meet the speed requirements of feature recognition.

[0035] The second, fourth, sixth, and eighth deep convolutional modules all include a two-dimensional convolutional layer and a first fusion layer connected to the two-dimensional convolutional layer. The connection structure of the two-dimensional convolutional layer is the same as that of the CBS module. The deep fusion of the two-dimensional convolutional layer and the first fusion layer can integrate different characteristics, so that the output features contain richer information and more comprehensively represent the content of the input image, which is beneficial for subsequent tasks such as insulator defect detection to capture details and overall features.

[0036] The working process of the depthwise convolution module includes: The input features are obtained and then fed into a two-dimensional convolutional layer for feature extraction to obtain the first convolutional features. The first convolutional feature and the input feature are fed into the first fusion layer and fused to obtain the output feature.

[0037] In this embodiment, compared to some complex multi-layer convolutional structures, the deep convolutional module, which performs convolution followed by simple copying and fusion, increases feature diversity and improves the cost-effectiveness of feature extraction without significantly increasing computational load. It ensures detection accuracy while maintaining computational efficiency, meeting the dual requirements of speed and performance in practical applications.

[0038] Specifically, layers three, five, seven, and nine (C2f) are modules in the YOLOv8 model that enable cross-layer feature fusion. By fusing features from different stages and scales, the model can comprehensively utilize shallow detail features and deep semantic features, enriching feature representation. For example, in insulator detection, it focuses on both the edges and textures of the insulator and grasps its overall structural information, improving the ability to identify insulator defects.

[0039] Specifically, the pooling module can downsample the feature map, reducing the feature dimensionality and computational cost. Simultaneously, pooling operations make the model more robust to changes in feature position and scale; even if the position and size of the insulator in the image change slightly, it can accurately extract effective features, improving the model's generalization ability.

[0040] The pooling module includes a first max pooling layer, a second max pooling layer, and a third max pooling layer connected in sequence; a second fusion layer connected to the first max pooling layer, the second max pooling layer, and the third max pooling layer; and a fourth max pooling layer connected to the second fusion layer. The working process of the pooling module includes: Substitute the input of the pooling module into the first max pooling layer to obtain the first pooling feature; Substitute the first pooling feature into the second max pooling layer to obtain the second pooling feature; Substitute the second pooling feature into the third max pooling layer to obtain the third pooling feature; The first pooling feature, the second pooling feature, and the third pooling feature are fused using a second fusion layer to obtain a fused pooling feature; The fusion pooling features are substituted into the fourth max pooling layer to obtain the output of the pooling module.

[0041] In this embodiment, the input features are sequentially passed through the first, second, and third max-pooling layers. Each pooling layer filters features at different scales, progressively extracting key features at different levels, from finer granularity to coarser granularity, thus comprehensively covering image feature information and aiding in the identification of insulator defects at various scales. The second fusion layer fuses the pooling features from each level, combining the advantages of features at different scales, making the fused pooling features more representative and robust, capable of handling changes in insulator position and orientation in the image. The features are then further refined by the fourth max-pooling layer, making the final output features more stable and reliable, improving the accuracy and generalization ability of the target model for insulator defect detection.

[0042] Specifically, the attention module employs the SimAM attention mechanism, which enables the target model to automatically focus on key regions in the image relevant to insulator defect detection, suppressing interference from irrelevant information. For example, in complex backgrounds, it highlights the insulator portion, reduces the impact of background noise on detection, improves the targeting and accuracy of feature extraction, and thus enhances the precision of defect detection.

[0043] In this embodiment, as Figure 4 As shown, the first fused feature is obtained by fusing the first-scale feature, the second-scale feature and the third-scale feature through the fusion layer; the feature information of different scales is integrated so that the feature contains richer semantic and detailed information.

[0044] In this embodiment, as Figure 4 As shown, the feature fusion model includes: max pooling layer 1, average pooling layer 1, max pooling layer 2, average pooling layer 2, a third fusion layer connected to max pooling layer 1 and average pooling layer 1, a fourth fusion layer connected to max pooling layer 2 and average pooling layer 2, and a fifth fusion layer connected to the third fusion layer and the fourth fusion layer. Specifically, step S23, which involves inputting the fourth-scale feature and the first fusion feature into the trained feature fusion model for processing to obtain the defect fusion features of the target insulator, includes: The fourth-scale features are input into the first max pooling layer and the first average pooling layer, respectively, and max pooling and average pooling are performed along the width dimension to obtain the width dimension max pooling features and the width dimension average pooling features. The fourth-scale features are input into the second max pooling layer and the second average pooling layer, respectively, and max pooling and average pooling are performed along the height dimension to obtain the height dimension max pooling features and the height dimension average pooling features. The width-dimension max pooling feature and the width-dimension average pooling feature are input into the third fusion layer and fused to obtain the width-dimension fused feature. The high-dimensional max pooling feature and the high-dimensional average pooling feature are input into the fourth fusion layer and fused to obtain the high-dimensional fused feature; The width dimension fusion feature, the height dimension fusion feature, and the first fusion feature are input into the fifth fusion layer for fusion to obtain the defect fusion feature of the target insulator.

[0045] Specifically, in this embodiment, max pooling and average pooling are performed in the width and height dimensions, respectively. Max pooling highlights the maximum value in the feature map, capturing the most significant feature information, such as the area with the most obvious defects in the captured image; average pooling considers the overall average of the feature map, smoothing the features and retaining more global information. Combining the two, features are obtained from different perspectives, complementing each other, allowing the resulting pooled features in the width and height dimensions to more comprehensively reflect the image information. Pooling operations can reduce the feature dimensionality and computational load. Simultaneously, it can mitigate the impact of noise to some extent, making the features more robust. For example, some random noise present in the captured image will be weakened after pooling, which is beneficial for subsequent accurate feature analysis.

[0046] Specifically, in this embodiment, the max pooling feature and average pooling feature in the width dimension are fused to obtain the fused feature in the width dimension, and the same applies to the height dimension. This fusion further integrates the features obtained by different types of pooling operations in the same direction, strengthens the expressive power of features in the width and height dimensions, and enables the model to more accurately grasp the structure and characteristics of the image in the horizontal and vertical directions.

[0047] Specifically, in this embodiment, the width dimension fusion feature, the height dimension fusion feature, and the first fusion feature are fused to obtain the defect fusion feature of the target insulator. This is a further integration of features from different directions with the first fusion feature from the previous multi-scale fusion, comprehensively gathering information from multiple scales and directions of the image to form a highly integrated and more discriminative feature representation. This provides richer and more accurate feature basis for subsequent transmission line risk assessment, improving the accuracy and reliability of the assessment.

[0048] In this embodiment, the fusion process for all fusion layers is to average the input.

[0049] In this embodiment, the combination of multi-scale feature extraction and feature fusion methods enables the target model to better adapt to images under different scenes and conditions, and to handle various complex image situations. This improves the model's generalization ability, allowing it to maintain good detection and judgment performance under different shooting conditions, environmental factors, and changes in target objects, reducing misjudgments or missed judgments caused by specific condition limitations.

[0050] In this embodiment, the target model needs to be trained before processing the captured images of the target insulator. The process includes: Acquire a set of images of insulators in transmission lines; The images in the captured image set are preprocessed to obtain a preprocessed image set; Defect type labeling is performed on the preprocessed image set to obtain the training set for the target model. The training set is input into the target model for training to obtain the optimal model parameters of the target model; The target model is determined based on the optimal model parameters.

[0051] In this embodiment, the preset database stores defect fusion features corresponding to different types of insulator defects, as well as the probability of risk (set based on different severity levels); the defect fusion features in the preset database are also obtained based on the target model.

[0052] In this embodiment, searching a preset database based on the defect fusion characteristics of the target insulator includes: The cosine similarity between the defect fusion features of the target insulator and the defect fusion features in the preset database is calculated. If the calculated cosine similarity value is greater than the preset threshold, the two features are considered similar, and the risk type and probability are obtained. The preset threshold is determined by technical personnel.

[0053] Specifically, the risk of the line is determined by searching a pre-set database based on the defect fusion characteristics of the target insulator. Since the defect fusion characteristics of the target insulator contain multi-scale and multi-level information of the image, they can more accurately describe the state and characteristics of the insulator. By comparing them with the standard or reference data in the pre-set database, it is possible to more accurately determine whether there is a risk to the line and the degree of the risk. This provides a scientific and reliable basis for the maintenance and management of the line, helps to discover potential problems in a timely manner and take corresponding measures to ensure the safe operation of the line.

[0054] Example 2

[0055] Based on the target model in the transmission line risk assessment method introduced in Example 1 and the YOLOv8 model in the existing insulator defect identification method, this example conducts a comparative experiment: Experimental training data: The dataset consists of 5000 images of transmission lines containing insulator defects, with 3500 images in the training set, 500 in the validation set, and 1000 in the test set. The data covers different lighting conditions (strong light, cloudy days, nighttime), complex backgrounds (vegetation cover, tower structures), and multi-scale targets (long-distance insulators, close-range defect details). Experimental parameters: The target model is embedded with ResNet decomposition, U-Net denoising, and SimAM attention mechanism; Training configuration: 100 iterations, learning rate 0.001, batch size 16, optimizer AdamW; Evaluation metrics: mAP50 (mean accuracy @ IoU=0.5), inference speed (FPS), noise robustness; Experimental results: The results comparison table for the target model and the YOLOv8 model is shown below:

[0056] As can be seen from the table, compared with YOLOv8, the improved target model's mAP50 has increased from 68.2% to 75.8%, showing a significant advantage, especially in small target detection and complex background scenes. The mAP50 retention rate in low-quality images has increased by 9.1%, adapting to common problems in captured images such as uneven lighting and image blurring. The combination of multi-scale feature fusion and attention mechanism enables the model to capture local details more accurately (such as insulator cracks), providing more reliable feature basis for transmission line risk assessment. The improved target model enhances the analysis capability of transmission line images.

[0057] Example 3

[0058] This embodiment provides a transmission line risk assessment system, including: The data acquisition module is used to acquire images of target insulators in transmission lines; The data processing module is used to input the captured images of the target insulators in the transmission line into the trained target model for processing, and to obtain the defect fusion features of the target insulators. The target model includes a multi-scale feature extraction model, a fusion layer, and a feature fusion model; the data processing module is also used for: The captured images of the target insulators in the transmission line are input into the trained multi-scale feature extraction model to obtain the first-scale feature, the second-scale feature, the third-scale feature and the fourth-scale feature; The first fused feature is obtained by fusing the first-scale feature, the second-scale feature, and the third-scale feature using a fusion layer; The fourth-scale feature and the first fusion feature are input into the trained feature fusion model for processing to obtain the defect fusion feature of the target insulator. The risk assessment module is used to search a preset database based on the defect fusion characteristics of the target insulator to determine the line risk.

[0059] Example 4

[0060] This embodiment provides a computer-readable storage medium storing a computer program that, when executed, implements the transmission line risk assessment method described in Embodiment 1.

[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for assessing the risk of transmission lines, characterized in that, include: Acquire images of target insulators in power transmission lines; Images of target insulators in transmission lines are input into a trained target model for processing to obtain defect fusion features of the target insulators. The target model includes a multi-scale feature extraction model, a fusion layer, and a feature fusion model. The processing includes: The captured images of the target insulators in the transmission line are input into the trained multi-scale feature extraction model to obtain the first-scale feature, the second-scale feature, the third-scale feature and the fourth-scale feature; The first fused feature is obtained by fusing the first-scale feature, the second-scale feature, and the third-scale feature using a fusion layer; The fourth-scale feature and the first fusion feature are input into the trained feature fusion model for processing to obtain the defect fusion feature of the target insulator. Based on the defect fusion characteristics of the target insulator, a pre-set database is searched to determine the line risk.

2. The method for assessing the risk of transmission lines according to claim 1, characterized in that, Also includes: Before inputting the captured images of target insulators in the transmission line into the trained target model for processing, preprocessing is performed, which includes: The captured image is denoised to obtain a denoised captured image; After denoising, the captured image is geometrically corrected to obtain a pre-processed captured image. The pre-processed captured images are input into a preset ResNet model for image decomposition to obtain the first initial image set; For each initial image in the first initial image set, a denoised initial image is obtained by denoising the initial image using the U-Net attention model. All the initial noise-reduced images are fused together to obtain the preprocessed captured image.

3. The method for assessing the risk of transmission lines according to claim 2, characterized in that, The multi-scale feature extraction model is derived from an improvement on the backbone network of the YOLOv8 model. The YOLOv8 model backbone network comprises, in sequence, a first CBS module, a second CBS module, a third C2f module, a fourth CBS module, a fifth C2f module, a sixth CBS module, a seventh C2f module, an eighth CBS module, a ninth C2f module, and a tenth SPPF module. The improvement includes: Replace the second, fourth, sixth, and eighth CBS modules with the same structure: a second, fourth, sixth, and eighth depthwise convolutional module. Replace the tenth SPPF module with a pooling module; An attention module is added after the backbone network.

4. The method for assessing the risk of transmission lines according to claim 3, characterized in that, The process involves inputting captured images of target insulators in transmission lines into a trained multi-scale feature extraction model to obtain first-scale features, second-scale features, third-scale features, and fourth-scale features, including: The captured image is processed sequentially through the first CBS module, the second depth convolution module, the third C2f module, the fourth depth convolution module, and the fifth C2f module to obtain the first scale feature; The first scale features are processed sequentially through the sixth depthwise convolution module and the seventh C2f module to obtain the second scale features; The second-scale features are processed sequentially through the eighth depthwise convolution module, the ninth C2f module, and the pooling module to obtain the third-scale features; The third-scale feature is processed by the attention module to obtain the fourth-scale feature.

5. The method for assessing the risk of transmission lines according to claim 4, characterized in that, The second, fourth, sixth, and eighth depthwise convolutional modules all include a two-dimensional convolutional layer and a first fusion layer connected to the two-dimensional convolutional layer; each depthwise convolutional module is used for: The input features are obtained and then fed into a two-dimensional convolutional layer for feature extraction to obtain the first convolutional features. The first convolutional feature and the input feature are fed into the first fusion layer and fused to obtain the output feature.

6. The method for assessing the risk of transmission lines according to claim 4, characterized in that, The pooling module includes a first max pooling layer, a second max pooling layer, and a third max pooling layer connected in sequence; a second fusion layer connected to the first max pooling layer, the second max pooling layer, and the third max pooling layer; and a fourth max pooling layer connected to the second fusion layer; the pooling module is used for: Substitute the input of the pooling module into the first max pooling layer to obtain the first pooling feature; Substitute the first pooling feature into the second max pooling layer to obtain the second pooling feature; Substitute the second pooling feature into the third max pooling layer to obtain the third pooling feature; The first pooling feature, the second pooling feature, and the third pooling feature are fused using a second fusion layer to obtain a fused pooling feature; The fusion pooling features are substituted into the fourth max pooling layer to obtain the output of the pooling module.

7. The method for assessing the risk of transmission lines according to claim 1, characterized in that, The feature fusion model includes: a first max pooling layer, a first average pooling layer, a second max pooling layer, a second average pooling layer, a third fusion layer connected to the first max pooling layer and the first average pooling layer, a fourth fusion layer connected to the second max pooling layer and the second average pooling layer, and a fifth fusion layer connected to the third fusion layer and the fourth fusion layer; the trained feature fusion model is used for: The fourth-scale features are input into the first max pooling layer and the first average pooling layer, respectively, and max pooling and average pooling are performed along the width dimension to obtain the width dimension max pooling features and the width dimension average pooling features. The fourth-scale features are input into the second max pooling layer and the second average pooling layer, respectively, and max pooling and average pooling are performed along the height dimension to obtain the height dimension max pooling features and the height dimension average pooling features. The width-dimension max pooling feature and the width-dimension average pooling feature are input into the third fusion layer and fused to obtain the width-dimension fused feature. The high-dimensional max pooling feature and the high-dimensional average pooling feature are input into the fourth fusion layer and fused to obtain the high-dimensional fused feature; The width dimension fusion feature, the height dimension fusion feature, and the first fusion feature are input into the fifth fusion layer for fusion to obtain the defect fusion feature of the target insulator.

8. The method for assessing the risk of transmission lines according to claim 1, characterized in that, The training process of the target model includes: Acquire a set of images of insulators in transmission lines; The images in the captured image set are preprocessed to obtain a preprocessed image set; Defect type labeling is performed on the preprocessed image set to obtain the training set for the target model. The training set is input into the target model for training to obtain the optimal model parameters of the target model; The target model is determined based on the optimal model parameters.

9. A transmission line risk assessment system, characterized in that, include: The data acquisition module is used to acquire images of target insulators in transmission lines; The data processing module is used to input captured images of target insulators in transmission lines into a trained target model for processing, thereby obtaining defect fusion features of the target insulators. The target model includes a multi-scale feature extraction model, a fusion layer, and a feature fusion model. The data processing module is also used for: The captured images of the target insulators in the transmission line are input into the trained multi-scale feature extraction model to obtain the first-scale feature, the second-scale feature, the third-scale feature and the fourth-scale feature; The first fused feature is obtained by fusing the first-scale feature, the second-scale feature, and the third-scale feature using a fusion layer; The fourth-scale feature and the first fusion feature are input into the trained feature fusion model for processing to obtain the defect fusion feature of the target insulator. The risk assessment module is used to search a preset database based on the defect fusion characteristics of the target insulator to determine the line risk.

10. A computer-readable storage medium storing a computer program that, when executed, implements the transmission line risk assessment method according to any one of claims 1-8.

Images