Power transmission channel data processing method based on artificial intelligence

Abstract

The power transmission channel data processing method based on artificial intelligence provided by the application comprises the following steps: acquiring a standard image of a patrol area of a power transmission channel and setting an identification anchor point; receiving original images transmitted by different patrol devices, embedding a detection anchor point, identifying and matching the original images with an image set of the same scene; performing data alignment processing on the original images in the image set according to the position of the identification anchor point in the standard image by using a feature alignment algorithm to obtain aligned images; extracting feature fingerprints of the detection anchor points in the aligned images of the same scene of different patrol devices, comparing the feature fingerprints two by two to obtain comparison results; eliminating the aligned images that do not meet the requirements based on the comparison results to form a cross-device patrol image set of the same patrol area; and performing longitudinal data processing and horizontal data comparison based on the identification anchor point and the detection anchor point to obtain the cross-device patrol image set, which can improve the accuracy of subsequent defect identification and ensure the stable operation of the power system.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data processing method for power transmission channels based on artificial intelligence. Background Technology

[0002] Transmission channels are core infrastructure for ensuring the safe and stable operation of the power grid. They encompass the transmission line itself, including towers, conductors, insulators, and fittings, as well as the surrounding environment within the line corridor. The quality of their operation and maintenance directly affects the reliability of the power supply system. With the development of intelligent inspection technology, various types of equipment, such as drones, ground inspection robots, handheld cameras, and infrared thermal imagers, have been widely used in the inspection of transmission channels, enabling the rapid acquisition of massive amounts of inspection images and providing data support for equipment defect identification and condition assessment.

[0003] However, images collected by multiple inspection devices exhibit significant heterogeneity. Different devices have varying shooting angles and resolutions, and their image quality varies due to factors such as lighting, weather, and device precision. Existing data processing methods can only perform simple deduplication and format conversion, lacking a precise alignment mechanism for cross-device data and effective detection and verification methods. This leads to issues such as feature misalignment and perspective deviation in images of the same area collected by different devices. Furthermore, existing methods cannot effectively determine the scene consistency of cross-device images, easily resulting in the inclusion of non-homogeneous or low-quality images and generating a large amount of false anomaly data. This not only increases the workload of subsequent annotation and model training but also reduces the accuracy of defect identification, severely restricting the efficient utilization of transmission channel inspection data and failing to meet the actual needs of intelligent operation and maintenance of the power grid. Summary of the Invention

[0004] In view of this, the present invention proposes a data processing method for power transmission channels based on artificial intelligence, which can perform correction and alignment processing on images acquired across devices, thereby improving the accuracy of subsequent defect identification.

[0005] The technical solution of this invention is implemented as follows: The data processing method for power transmission channels based on artificial intelligence includes the following steps: Step S1: Obtain a standard image of the inspection area of ​​the power transmission channel and set identification anchor points in the standard image; Step S2: Receive raw images transmitted from different inspection devices, embed detection anchor points in the raw images, identify the raw images and match them with image sets of the same scene; Step S3: Use a feature alignment algorithm to perform data alignment processing on the original images in the image set according to the position of the anchor points identified in the standard image, and obtain the aligned image; Step S4: Extract the feature fingerprint set of the detection anchor points in the aligned images of the same scene from different inspection equipment, and perform pairwise comparisons to obtain the comparison results; Step S5: Based on the comparison results, remove the aligned images that do not meet the requirements to form a cross-device inspection image set for the same inspection area.

[0006] Preferably, step S1 includes the following steps: Step S11: Retrieve standard images of the inspection area of ​​the transmission channel from the power system database, and perform noise reduction and brightness equalization preprocessing on the standard images. Step S12: Query the core electrical equipment in the inspection area and set several identification anchor points in the standard image according to the structural characteristics of the core electrical equipment; Step S13: Record the reference pixel coordinates, physical size mapping relationship and relative position parameters of each recognition anchor point in the standard image, and construct a recognition anchor point reference library.

[0007] Preferably, step S2 includes the following specific steps: Step S21: Receive the original images collected by different inspection devices during inspection in the inspection area, and perform noise reduction and brightness equalization preprocessing on the original images. Step S22: Embed detection anchors in the original image based on the inherent weak features of the power equipment, convert the inherent weak features into attribute labels, and package them with the pixel coordinates of the detection anchors as the metadata of the detection anchors. Step S23: Match the original images with the GPS coordinates when the inspection equipment collects the original images to obtain the original images at the same location; Step S24: Calculate the similarity of the original images of the same location from different inspection equipment, and take the original images with a similarity greater than a preset threshold as the image set of the same scene.

[0008] Preferably, the specific steps of step S24 are as follows: extract the core equipment structural features of the original images of the same location of different inspection equipment through a feature extraction algorithm, calculate the similarity of the core equipment structural features using cosine similarity, and take the original images with similarity greater than a preset threshold as the image set of the same scene, wherein the preset threshold is set based on the equipment perspective difference and environmental interference tolerance range of the power inspection scene.

[0009] Preferably, step S3 includes the following specific steps: Step S31: Query the standard image of the same scene as the image set, and extract the reference pixel coordinates of the recognition anchor point of the standard image; Step S32: Extract the recognition anchor points of the original images in the image set using the target detection algorithm to determine the current pixel coordinates; Step S33: Based on the difference between the current pixel coordinates and the reference pixel coordinates, perform angle correction on the original image and obtain the corrected image; Step S34: Adapt the resolution of the corrected image to the resolution of the standard image to obtain the aligned image.

[0010] Preferably, the specific steps of step S33 are as follows: Match the current pixel coordinates with the reference pixel coordinates one by one, and generate a set of anchor point corresponding point pairs; The set of anchor point corresponding to the point pair is input into the pre-trained deep learning homography matrix solving model, and the deep learning homography matrix solving model outputs the optimal homography matrix. Perspective transformation is performed on the original image based on the homography matrix to achieve viewpoint conversion and angle correction, thereby obtaining the corrected image; If the pixel coordinates of the recognition anchor points in the corrected image deviate significantly from the reference pixel coordinates, the angle correction is performed again after adjusting the weights of the loss function of the deep learning homography matrix solving model.

[0011] Preferably, step S4 includes the following specific steps: Step S41: Extract the pixel coordinates and attribute labels of the detection anchor points in the aligned images of the same scene from different inspection devices; Step S42: Extract the feature vectors of the detection anchor points of the aligned image through the attention model, and use the hash algorithm to convert the feature vectors into feature fingerprints to form a feature fingerprint set; Step S43: Perform pairwise comparisons of the feature fingerprints in the feature fingerprint set, and obtain the matching degree of the two sets of feature fingerprints by calculating the cosine similarity. Step S44: Compare the matching degree with the preset matching threshold, and obtain the matching results of feature consistency, acceptable deviation and feature inconsistency.

[0012] Preferably, the matching threshold is set according to the type difference between the two inspection devices corresponding to the matching degree.

[0013] Preferably, step S5 includes the following specific steps: Step S51: Based on the comparison results, remove the aligned images with inconsistent features; Step S52: Mark the differences in the acquisition characteristics of the aligned image with acceptable deviation; Step S53: Integrate the aligned images with acceptable deviations after marking and the aligned images with consistent features to form a cross-equipment inspection image set for the same inspection area.

[0014] Preferred options also include: Step S6: Issue an alarm signal to the inspection equipment corresponding to the aligned image that does not meet the requirements.

[0015] Compared with the prior art, the beneficial effects of the present invention are: The data processing method for power transmission channels based on artificial intelligence of the present invention first acquires standard images of the inspection area of ​​the power transmission channel and sets identification anchor points in the standard images as correction benchmarks. Then, after receiving the original images transmitted by different inspection devices, the original images are identified and classified to obtain an image set of different inspection devices in the same scene. Each original image in the image set is aligned with the standard image based on the identification anchor points through a feature alignment algorithm, thereby realizing the alignment correction of the original images of a single inspection device. In addition, before alignment correction, detection anchor points are embedded in the original image. After the alignment correction of the original image of a single inspection device is completed, the feature fingerprint set of the detection anchor points is extracted for the aligned images of different inspection devices in the same scene. Then, the feature fingerprints are compared, and the aligned images are filtered based on the comparison results. The finally left aligned images are the clear inspection images of different inspection devices in the same scene, forming a cross-device inspection image set. Based on the cross-device inspection image set, the power equipment of the power transmission channel can be accurately assessed for defects, ensuring the stable operation of the power system. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the power transmission channel data processing method based on artificial intelligence according to the present invention. Detailed Implementation

[0018] To better understand the technical content of this invention, a specific embodiment is provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0019] See Figure 1 The power transmission channel data processing method based on artificial intelligence provided by this invention includes the following steps: Step S1: Obtain a standard image of the inspection area of ​​the power transmission channel and set identification anchor points in the standard image; Step S2: Receive raw images transmitted from different inspection devices, embed detection anchor points in the raw images, identify the raw images and match them with image sets of the same scene; Step S3: Use a feature alignment algorithm to perform data alignment processing on the original images in the image set according to the position of the anchor points identified in the standard image, and obtain the aligned image; Step S4: Extract the feature fingerprint set of the detection anchor points in the aligned images of the same scene from different inspection equipment, and perform pairwise comparisons to obtain the comparison results; Step S5: Based on the comparison results, remove the aligned images that do not meet the requirements to form a cross-equipment inspection image set for the same inspection area; Step S6: Issue an alarm signal to the inspection equipment corresponding to the aligned image that does not meet the requirements.

[0020] The present invention provides an artificial intelligence-based data processing method for power transmission channels, which is used to correct and align original inspection images collected from different inspection areas of power transmission channels in order to generate clear inspection images. Before data processing, the power transmission channel is divided into several inspection areas according to its specific location, and a standard image of each inspection area is obtained, including clear and bright images of different locations within the inspection area. Then, identification anchor points are set according to some key features in the standard images, and the identification anchor points serve as the reference for subsequent alignment and correction.

[0021] When different inspection devices are used to select parts for the same inspection area, the inspection devices will collect original inspection images of different scenes in the inspection area. Due to environmental factors and differences in inspection devices, the collected original inspection images will differ. Therefore, it is necessary to filter and process the original inspection images. The standard image is the benchmark for data processing. When processing data, it is necessary to ensure that the original image matches the standard image. Therefore, it is necessary to identify and match the original images collected by different inspection devices to obtain the original image that matches the scene of the standard image. After integrating the original images of all inspection devices in the same scene, an image set can be obtained. Since there are multiple inspection scenes, multiple image sets can be formed. Each individual image set will undergo subsequent correction, alignment, and feature comparison.

[0022] After determining the image set, the positions of recognition anchor points can be extracted from the original images in the image set. A feature alignment algorithm is then used to align the positions of the recognition anchor points in the original images with those in the standard images, resulting in aligned images. This process is the processing of the original images for a single inspection device, i.e., the vertical data processing process. After alignment, each original image in the image set can yield a corresponding aligned image. At this point, horizontal data comparison is performed on the aligned images of different inspection devices. Before alignment calibration, detection anchor points are embedded in each original image. These detection anchor points are used for comparison of aligned images across different inspection devices. The system extracts feature fingerprint sets from each image and then compares the feature fingerprint sets of aligned images in the same scene to obtain comparison results. Based on the comparison results, the aligned images that do not meet the requirements are removed, and the remaining aligned images form a cross-equipment inspection image set. The cross-equipment inspection image set can be used to make a comprehensive judgment on defects in power equipment, improve the accuracy of defect judgment, and ensure the stable operation of the power system. For aligned images that do not meet the requirements, it may be due to faults in the inspection equipment or other reasons. Based on the inspection equipment corresponding to the aligned images that do not meet the requirements, alarm commands can be generated and sent out to facilitate the maintenance and repair of the inspection equipment by the staff.

[0023] Preferably, step S1 includes the following steps: Step S11: Retrieve standard images of the inspection area of ​​the transmission channel from the power system database, and perform noise reduction and brightness equalization preprocessing on the standard images. Step S12: Query the core electrical equipment in the inspection area and set several identification anchor points in the standard image according to the structural characteristics of the core electrical equipment; Step S13: Record the reference pixel coordinates, physical size mapping relationship and relative position parameters of each recognition anchor point in the standard image, and construct a recognition anchor point reference library.

[0024] In the long-term operation and maintenance of power systems, standard images can be obtained under favorable external conditions and with the help of high-precision equipment. These standard images are stored in the power system database. Standard images can be directly extracted from the power system database. After extraction, the standard images can be preprocessed to facilitate the setting of identification anchor points. The identification anchor points need to be set according to the characteristics of the transmission equipment. Specific selections include tower apex, insulator string connection point, and hardware fixing hole. After setting the identification anchor points, the reference pixel coordinates, physical size mapping relationship, and relative position parameters of each identification anchor point are recorded, and an identification anchor point reference library is constructed to facilitate subsequent alignment and correction.

[0025] Preferably, step S2 includes the following specific steps: Step S21: Receive the original images collected by different inspection devices during inspection in the inspection area, and perform noise reduction and brightness equalization preprocessing on the original images. Step S22: Embed detection anchors in the original image based on the inherent weak features of the power equipment, convert the inherent weak features into attribute labels, and package them with the pixel coordinates of the detection anchors as the metadata of the detection anchors. Step S23: Match the original images with the GPS coordinates when the inspection equipment collects the original images to obtain the original images at the same location; Step S24: Calculate the similarity of the original images of the same location from different inspection equipment, and take the original images with a similarity greater than a preset threshold as the image set of the same scene.

[0026] After the anchor point benchmark library is built, data processing can begin. This includes the generation of numerous raw inspection images by various inspection devices such as drones, robots, infrared meters, and manual cameras during inspections of the inspection area. These raw images are also preprocessed, and then anchor points are embedded based on the inherent weak features of the power equipment in the inspection area to facilitate subsequent horizontal data comparison. These inherent weak features include insulator skirt spacing, tower bolt hole arrangement, and nameplate thermal texture. Embedding anchor points using inherent weak features has the advantage of high stability, can be stably extracted across devices and scenarios, and is compatible with multi-source heterogeneous inspection devices. The inherent weak features are then converted into attribute tags, such as insulator string-skirt spacing, and the attribute tags and pixel coordinates of the anchor points are packaged as the metadata of the anchor points for subsequent retrieval.

[0027] After embedding the detection anchor points, the original images can be identified. First, the original images collected by different inspection devices at the same location are initially screened based on the GPS coordinates when the original images are collected. Then, the similarity of the original images at the same location is calculated. When the similarity is greater than the preset threshold, it can be determined that the original images were collected at the same scene and can be integrated into an image set.

[0028] Preferably, the specific steps of step S24 are as follows: extract the core equipment structural features of the original images of the same location of different inspection equipment through a feature extraction algorithm, calculate the similarity of the core equipment structural features using cosine similarity, and take the original images with similarity greater than a preset threshold as the image set of the same scene, wherein the preset threshold is set based on the equipment perspective difference and environmental interference tolerance range of the power inspection scene.

[0029] The similarity calculation adopts the cosine similarity calculation method. First, the core equipment structural features of the original images at the same location are extracted by the feature extraction algorithm, such as the tower outline and the arrangement of insulator strings. Then, the similarity value can be obtained by calculating the cosine similarity. The similarity value is compared with a preset threshold. If the similarity between the two original images is greater than the preset threshold, the corresponding original images can be sent to the image set of the same scene for storage.

[0030] Preferably, step S3 includes the following specific steps: Step S31: Query the standard image of the same scene as the image set, and extract the reference pixel coordinates of the recognition anchor point of the standard image; Step S32: Extract the recognition anchor points of the original images in the image set using the target detection algorithm to determine the current pixel coordinates; Step S33: Based on the difference between the current pixel coordinates and the reference pixel coordinates, perform angle correction on the original image and obtain the corrected image; Step S34: Adapt the resolution of the corrected image to the resolution of the standard image to obtain the aligned image.

[0031] During alignment correction, it is necessary to first query a standard image that is in the same scene as the image set, and then extract the reference pixel coordinates of the recognition anchor points in the standard image. Each original image in the image set has its recognition anchor points extracted by the target detection algorithm, and the current pixel coordinates of the recognition anchor points are obtained. Finally, using the reference pixel coordinates as a reference, the original images are corrected according to the difference between the current pixel coordinates and the reference pixel coordinates, so as to obtain the corrected image of each original image. In addition, since the images collected by different inspection devices have different resolutions, resolution unification processing is also required. According to the resolution of the standard image, the resolution of the corrected image is adapted by a super-resolution model to obtain the aligned image.

[0032] Preferably, the specific steps of step S33 are as follows: Match the current pixel coordinates with the reference pixel coordinates one by one, and generate a set of anchor point corresponding point pairs; The set of anchor point corresponding to the point pair is input into the pre-trained deep learning homography matrix solving model, and the deep learning homography matrix solving model outputs the optimal homography matrix. Perspective transformation is performed on the original image based on the homography matrix to achieve viewpoint conversion and angle correction, thereby obtaining the corrected image; If the pixel coordinates of the recognition anchor points in the corrected image deviate significantly from the reference pixel coordinates, the angle correction is performed again after adjusting the weights of the loss function of the deep learning homography matrix solving model.

[0033] The correction process of this invention is achieved through a deep learning homography matrix solving model. This model, through adaptive training on power transmission channel equipment scenarios, accurately learns the image angle deviation patterns. It matches the current pixel coordinates of the identification anchor points in each original image with the reference pixel coordinates of the standard image, obtaining a set of anchor point pairs corresponding to each inspection device. This set of anchor point pairs can be input into the trained deep learning homography matrix solving model, which generates the optimal homography matrix. Then, the original image can be corrected based on the deep learning homography matrix solving model. For example, perspective transformation mainly involves viewpoint conversion and angle correction. After perspective transformation, a corrected image can be obtained. Then, the corrected image is further verified. If the pixel coordinates of the recognition anchor points in the corrected image deviate significantly from the reference pixel coordinates, it needs to be corrected again. At this time, the weights of the loss function of the deep learning homography matrix solving model can be adjusted. Then, the set of anchor point corresponding point pairs is re-inputted into the deep learning homography matrix solving model until the deviation between the pixel coordinates of the recognition anchor points in the corrected image and the reference pixel coordinates is within the set value. Finally, the corrected image is output for resolution adaptation.

[0034] Preferably, step S4 includes the following specific steps: Step S41: Extract the pixel coordinates and attribute labels of the detection anchor points in the aligned images of the same scene from different inspection devices; Step S42: Extract the feature vectors of the detection anchor points of the aligned image through the attention model, and use the hash algorithm to convert the feature vectors into feature fingerprints to form a feature fingerprint set; Step S43: Perform pairwise comparisons of the feature fingerprints in the feature fingerprint set, and obtain the matching degree of the two sets of feature fingerprints by calculating the cosine similarity. Step S44: Compare the matching degree with the matching threshold set according to the type difference between the two inspection devices corresponding to the matching degree, and obtain the comparison results of consistent features, acceptable deviation, and inconsistent features.

[0035] After vertical data processing, the original images in the image set are aligned. Then, horizontal data comparison is performed on the aligned images in the same image set. First, the pixel coordinates and attribute labels of the detection anchor points in the aligned images are extracted. Then, the feature vectors of the detection anchor points are extracted through the attention model, and the feature vectors are converted into feature fingerprints through the LSH hash algorithm to form the feature fingerprint set of each aligned image. Finally, according to the pairwise matching of inspection devices, the cosine similarity of the feature fingerprint set is calculated, and the cosine similarity is used as the matching degree. If the matching degree is less than half of the set matching threshold, the comparison result of inconsistent features is obtained. If the matching degree is greater than half of the set matching threshold but less than the matching threshold, the comparison result of acceptable deviation is obtained. If the matching degree is greater than the matching threshold, the comparison result of consistent features is obtained. The matching threshold is set differently according to the different types of the two inspection devices.

[0036] Preferably, step S5 includes the following specific steps: Step S51: Based on the comparison results, remove the aligned images with inconsistent features; Step S52: Mark the differences in the acquisition characteristics of the aligned image with acceptable deviation; Step S53: Integrate the aligned images with acceptable deviations after marking and the aligned images with consistent features to form a cross-equipment inspection image set for the same inspection area.

[0037] For aligned images with inconsistent features, they can be directly rejected, and an alarm command will be issued to the corresponding inspection equipment. For aligned images with acceptable deviations, the differences in features collected by the equipment will be marked on them, and they will be integrated with aligned images with consistent features into a cross-equipment inspection image set to facilitate the subsequent defect identification process.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data processing method for power transmission channels based on artificial intelligence, characterized in that, Includes the following steps: Step S1: Obtain a standard image of the inspection area of ​​the power transmission channel and set identification anchor points in the standard image; Step S2: Receive raw images transmitted from different inspection devices, embed detection anchor points in the raw images, identify the raw images and match them with image sets of the same scene; Step S3: Use a feature alignment algorithm to perform data alignment processing on the original images in the image set according to the position of the anchor points identified in the standard image, and obtain the aligned image; Step S4: Extract the feature fingerprint set of the detection anchor points in the aligned images of the same scene from different inspection equipment, and perform pairwise comparisons to obtain the comparison results; Step S5: Based on the comparison results, remove the aligned images that do not meet the requirements to form a cross-device inspection image set for the same inspection area.

2. The data processing method for power transmission channels based on artificial intelligence according to claim 1, characterized in that, The specific steps of step S1 include: Step S11: Retrieve standard images of the inspection area of ​​the transmission channel from the power system database, and perform noise reduction and brightness equalization preprocessing on the standard images. Step S12: Query the core electrical equipment in the inspection area and set several identification anchor points in the standard image according to the structural characteristics of the core electrical equipment; Step S13: Record the reference pixel coordinates, physical size mapping relationship and relative position parameters of each recognition anchor point in the standard image, and construct the recognition anchor point reference library.

3. The data processing method for power transmission channels based on artificial intelligence according to claim 1, characterized in that, The specific steps of step S2 include: Step S21: Receive the original images collected by different inspection devices during inspection in the inspection area, and perform noise reduction and brightness equalization preprocessing on the original images. Step S22: Embed detection anchors in the original image based on the inherent weak features of the power equipment, convert the inherent weak features into attribute labels, and package them with the pixel coordinates of the detection anchors as the metadata of the detection anchors. Step S23: Match the original images with the GPS coordinates when the inspection equipment collects the original images to obtain the original images at the same location; Step S24: Calculate the similarity of the original images of the same location from different inspection equipment, and take the original images with similarity greater than a preset threshold as the image set of the same scene.

4. The data processing method for power transmission channels based on artificial intelligence according to claim 3, characterized in that, The specific steps of step S24 are as follows: extract the core equipment structure features of the original images of the same location of different inspection equipment through the feature extraction algorithm, calculate the similarity of the core equipment structure features using cosine similarity, and take the original images with similarity greater than a preset threshold as the image set of the same scene. The preset threshold is set based on the equipment perspective difference and environmental interference tolerance range of the power inspection scene.

5. The data processing method for power transmission channels based on artificial intelligence according to claim 1, characterized in that, The specific steps of step S3 include: Step S31: Query the standard image of the same scene as the image set, and extract the reference pixel coordinates of the recognition anchor point of the standard image; Step S32: Extract the recognition anchor points of the original images in the image set using the target detection algorithm to determine the current pixel coordinates; Step S33: Based on the difference between the current pixel coordinates and the reference pixel coordinates, perform angle correction on the original image and obtain the corrected image; Step S34: Adapt the resolution of the corrected image to the resolution of the standard image to obtain the aligned image.

6. The data processing method for power transmission channels based on artificial intelligence according to claim 5, characterized in that, The specific steps of step S33 are as follows: Match the current pixel coordinates with the reference pixel coordinates one by one, and generate a set of anchor point corresponding point pairs; The set of anchor point corresponding to the point pair is input into the pre-trained deep learning homography matrix solving model, and the deep learning homography matrix solving model outputs the optimal homography matrix. Perspective transformation is performed on the original image based on the homography matrix to achieve viewpoint conversion and angle correction, thereby obtaining the corrected image; If the pixel coordinates of the recognition anchor points in the corrected image deviate significantly from the reference pixel coordinates, the angle correction is performed again after adjusting the weights of the loss function of the deep learning homography matrix solving model.

7. The data processing method for power transmission channels based on artificial intelligence according to claim 1, characterized in that, The specific steps of step S4 include: Step S41: Extract the pixel coordinates and attribute labels of the detection anchor points in the aligned images of the same scene from different inspection devices; Step S42: Extract the feature vectors of the detection anchor points of the aligned image through the attention model, and use the hash algorithm to convert the feature vectors into feature fingerprints to form a feature fingerprint set; Step S43: Perform pairwise comparisons of the feature fingerprints in the feature fingerprint set, and obtain the matching degree of the two sets of feature fingerprints by calculating the cosine similarity. Step S44: Compare the matching degree with the preset matching threshold, and obtain the matching results of feature consistency, acceptable deviation and feature inconsistency.

8. The data processing method for power transmission channels based on artificial intelligence according to claim 7, characterized in that, The matching threshold is set based on the type difference between the two inspection devices corresponding to the matching degree.

9. The data processing method for power transmission channels based on artificial intelligence according to claim 7, characterized in that, The specific steps of step S5 include: Step S51: Based on the comparison results, remove the aligned images with inconsistent features; Step S52: Mark the differences in the acquisition characteristics of the aligned image with acceptable deviation; Step S53: Integrate the aligned images with acceptable deviations after marking and the aligned images with consistent features to form a cross-equipment inspection image set for the same inspection area.

10. The data processing method for power transmission channels based on artificial intelligence according to claim 1, characterized in that, Also includes: Step S6: Issue an alarm signal to the inspection equipment corresponding to the aligned image that does not meet the requirements.

Images