Power grid image intelligent interception method and system based on power grid tower position coordinates

Abstract

The invention relates to an intelligent power grid image interception method and system based on position coordinates of a power grid tower. The method comprises the following steps: acquiring original image data of the power grid tower through image acquisition equipment carried on an inspection aircraft, and preprocessing the original image data to generate a standardized image data set; carrying out feature extraction on tower images in the standardized image data set, identifying tower position coordinates and generating a tower position distribution map; inputting the tower position distribution diagram into an intelligent area determination model, and generating a target interception area range; performing optimization processing on the intercepted boundary to generate a high-precision intercepted boundary; a final interception instruction is generated according to the high-precision interception boundary, and intelligent interception of the tower pole area in the power grid inspection image is achieved; according to the scheme, the power grid inspection image interception precision and adaptability can be improved.

Description

Technical Field

[0001] This application relates to the fields of power grid and information technology, and in particular to a method and system for intelligent cropping of power grid images based on the location coordinates of power grid towers. Background Technology

[0002] In the field of power grid operation and maintenance management, the inspection of power grid towers is crucial. Traditionally, power grid inspections mainly rely on manual on-site checks. This method is not only inefficient, but also extremely difficult and risky in areas with complex geographical environments, such as mountains and forests. With the development of technology, using inspection drones equipped with image acquisition equipment for power grid inspections has gradually become the mainstream approach. Inspection drones can quickly and comprehensively acquire image data of power grid towers, providing strong support for the safe operation of the power grid.

[0003] However, there are currently some technical challenges in processing power grid images acquired by inspection drones. To accurately analyze the condition of power grid towers, it is typically necessary to extract the effective regions containing the towers from a large number of power grid images. Most existing image extraction methods are based on fixed rules or simple threshold judgments, which often cannot adapt to the complexity and diversity of power grid tower distribution. For example, when towers are densely or sparsely distributed, fixed-rule extraction methods may result in inaccurate extraction areas, including too much useless background information or omitting important tower sections.

[0004] To address the aforementioned issues, some studies have proposed image cropping methods based on machine learning. These methods train models to identify the locations of towers and determine the cropping area based on the tower distribution. While these methods improve cropping accuracy to some extent, they still have some limitations. For example, the models may misjudge complex tower distributions and background interference, resulting in low precision in the cropped area. Summary of the Invention

[0005] The main objective of this application is to provide a method and system for intelligent cropping of power grid images based on the location coordinates of power grid poles, which can accurately and efficiently crop the effective area containing poles from power grid inspection images, thereby improving the accuracy and adaptability of the cropped area.

[0006] To achieve the above objectives, embodiments of the present invention provide a method for intelligent cropping of power grid images based on the location coordinates of power grid towers, the method comprising the following steps: The raw image data of the power grid towers is acquired by an image acquisition device mounted on an inspection aircraft, and preprocessing is performed on the raw image data to generate a standardized image dataset; the raw image data includes visible light images of the towers and corresponding spatial location information. Feature extraction is performed on the tower images in the standardized image dataset to identify the precise location coordinates of the towers in the images and generate a tower location distribution map; the tower location distribution map includes the center point coordinates of each tower and its relative distribution information in the image. The distribution map of the pole locations is input into the intelligent region determination model. The model analyzes the distribution map to generate the target interception area. The target interception area is a rectangular or irregular polygonal region calculated based on the pole distribution density and image segmentation logic. Based on the target cropping area, the cropping boundary is optimized to generate a high-precision cropping boundary; the high-precision cropping boundary includes the precise coordinates of the boundary pixels and their transition smoothness information with adjacent areas; A final interception command is generated based on the high-precision interception boundary, and the interception command is sent to the image processing system to realize intelligent interception of the tower area in the power grid inspection image; the final interception command includes the starting coordinates, ending coordinates and corresponding image data index of the interception area.

[0007] Accordingly, embodiments of this application also provide an intelligent power grid image capture system based on the location coordinates of power grid towers, the system comprising: The acquisition module is used to acquire raw image data of power grid towers through image acquisition equipment mounted on the inspection aircraft, and to perform preprocessing operations on the raw image data to generate a standardized image dataset; the raw image data includes visible light images of the towers and corresponding spatial location information; The feature processing module is used to extract features from the tower images in the standardized image dataset, identify the precise location coordinates of the towers in the images, and generate a tower location distribution map; the tower location distribution map includes the center point coordinates of each tower and its relative distribution information in the image. The model analysis module is used to input the tower location distribution map into the intelligent region determination model, and analyze the distribution map through the model to generate the target interception area range; the target interception area range is a rectangular or irregular polygonal region calculated based on the tower distribution density and image segmentation logic. The boundary optimization module is used to optimize the truncated boundary according to the target truncated area range to generate a high-precision truncated boundary; the high-precision truncated boundary includes the precise coordinates of the boundary pixels and their transition smoothness information with adjacent areas; The instruction generation module is used to generate a final interception instruction based on the high-precision interception boundary, and send the interception instruction to the image processing system to realize intelligent interception of the tower area in the power grid inspection image; the final interception instruction includes the starting coordinates, ending coordinates and corresponding image data index of the interception area.

[0008] In summary, the technical solution of this application, by preprocessing the original image data to generate a standardized image dataset, can eliminate image noise, unify resolution, and improve image quality, providing reliable data for subsequent feature extraction. Feature extraction of tower images from the standardized image dataset can accurately identify tower location coordinates, generate a tower location distribution map, and clearly present the tower distribution. The tower location distribution map is input into an intelligent region determination model, which combines tower distribution density with image segmentation logic to generate the target cropping region range, making the cropping region more reasonable. Optimization of the cropping boundary generates a high-precision cropping boundary, enhancing the accuracy and smoothness of the cropping region. Finally, based on the high-precision cropping boundary, a final cropping command is generated to achieve intelligent cropping. This solution can effectively reduce useless background information, improve the accuracy and adaptability of the cropping region, thereby improving the efficiency and accuracy of power grid inspection image analysis. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of a scenario for the intelligent image cropping method for power grids based on the location coordinates of power grid towers in this application embodiment; Figure 2 A flowchart is provided for an embodiment of this application to illustrate a method for intelligent cropping of power grid images based on the location coordinates of power grid towers; Figure 3 A schematic diagram illustrating the process of generating the tower location distribution map provided in the embodiments of this application; Figure 4 A schematic diagram illustrating the process of identifying potential pole areas provided in this application embodiment; Figure 5 A schematic diagram illustrating the process of generating the target capture region range provided in this application embodiment; Figure 6 A flowchart illustrating the optimization of the intercepted region boundary provided in an embodiment of this application; Figure 7 A schematic diagram of the process for integrating the interception boundary provided in an embodiment of this application; Figure 8 This is a schematic diagram of the curve transition smoothing process provided in the embodiments of this application; Figure 9 This is a schematic diagram of the intelligent image capture system for power grids based on the location coordinates of power grid towers provided in this application embodiment; Figure 10 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0010] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0011] This application provides a method and system for intelligent cropping of power grid images based on the location coordinates of power grid towers, which will be described in detail below.

[0012] In this embodiment, the intelligent power grid image cropping method based on the location coordinates of power grid towers is a processing method for power grid inspection images. This method uses the location coordinate information of power grid towers to intelligently analyze and process power grid images collected by inspection drones to accurately crop the effective area containing the towers. Specifically, it first acquires and preprocesses the original image data to generate a standardized image dataset; then, it extracts features from the tower images to determine the tower location coordinates and generate a distribution map; next, it uses an intelligent region determination model to generate the target cropping area range; then, it optimizes the cropping boundary to generate a high-precision cropping boundary; finally, it generates a final cropping command based on the high-precision cropping boundary to achieve intelligent cropping, providing accurate image data for power grid operation and maintenance.

[0013] As shown in Figure 1, a scenario for intelligent image capture of power grid based on the location coordinates of power grid towers is provided. The scenario mainly includes an inspection drone, an image acquisition device, and an image processing system, which are connected to each other via a wireless network.

[0014] Taking a mountain power grid inspection scenario as an example, mountainous terrain is complex, and power grid towers are scattered and irregularly distributed. The inspection drone, equipped with high-resolution image acquisition equipment, flies over the mountainous area according to a pre-set inspection route. During flight, the image acquisition equipment acquires visible light images of the power grid towers at regular time intervals and records the timestamp and geographic coordinates of each image. For example, when the drone flies over a tower located in a valley, the image acquisition equipment immediately acquires an image of that tower and records the acquisition time as 11:15 AM on December 1, 2024, with geographic coordinates of longitude 113.2° and latitude 31.5°.

[0015] The inspection aircraft can transmit the raw image data it acquires to a ground-based image processing system. Due to the complex mountainous environment, the images may contain a large amount of background information such as vegetation and rocks, and some images may be blurry or have abnormal exposure due to factors such as lighting and aircraft vibration. The image processing system can first perform an integrity check on the raw image data, checking for completeness and any data loss or corruption. Image frames with blurriness or abnormal exposure are discarded, generating a pre-selected set of valid images.

[0016] Next, the resolution of the image data in the valid image set is adjusted to unify the image resolution for subsequent processing and analysis. Simultaneously, noise filtering is performed on the images to remove random noise and interference signals, improving image clarity. These preprocessing operations generate a standardized image dataset.

[0017] In standardized image datasets, image processing systems can also extract features from pole images. Specifically, edge detection algorithms and texture analysis methods can be used to extract edge and texture features from pole images, filtering out potential pole regions. Further geometric analysis is performed on these potential pole regions to calculate their center point coordinates and contour information within the image. A unique identifier is assigned to each pole, and the center point coordinates and contour information of each pole are integrated into a pole location distribution map.

[0018] The distribution map of power pole locations is input into the intelligent region determination model. This model analyzes the tower distribution density and image segmentation logic to generate the target cropping area. Since the distribution of power poles in mountainous areas is irregular, the target cropping area may be an irregular polygon. The cropping boundary is optimized to make it smoother and reduce interference from background information. Finally, based on the high-precision cropping boundary, a final cropping instruction is generated to crop the effective area containing the power poles from the standardized image dataset. This provides accurate power pole image data for power grid maintenance personnel, enabling them to promptly identify potential problems with the power poles and ensure the safe operation of the power grid.

[0019] refer to Figure 2 , Figure 2 This is a flowchart illustrating a method for intelligent cropping of power grid images based on the location coordinates of power grid poles, provided in an embodiment of this application. The execution subject of this method can be a computer device (which can be used to implement an image processing system, etc.). The intelligent cropping method for power grid images based on the location coordinates of power grid poles provided in this embodiment specifically includes:

[0020] S10: Obtain raw image data of power grid towers through image acquisition equipment mounted on the inspection aircraft, and perform preprocessing operations on the raw image data to generate a standardized image dataset; the raw image data includes visible light images of the towers and corresponding spatial location information.

[0021] In this embodiment, the raw image data consists of images and related information about power grid towers directly acquired by the image acquisition equipment mounted on the inspection aircraft. Visible light images, captured by the optical sensors of the image acquisition equipment, reflect the appearance characteristics of the towers and can intuitively display their structure, surface condition, etc. Spatial location information is data used to determine the specific location of the towers in real geographic space, and may include latitude and longitude coordinates and altitude. For example, in a large power grid inspection project, the spatial location information of a certain tower might be longitude 110.35°, latitude 25.68°, and altitude 300 meters.

[0022] In one embodiment, when acquiring raw image data, the inspection aircraft flies along a preset inspection route, and the image acquisition device acquires images at regular time intervals. For example, one image is acquired every 3 seconds to ensure comprehensive coverage of the power grid towers. The acquired raw image data may have some problems, such as some images being blurry due to aircraft shaking or abnormal exposure due to poor lighting conditions. Therefore, preprocessing operations are required for the raw image data. Preprocessing operations include integrity verification, i.e., checking whether the image data is complete and whether there is any data loss or corruption; resolution adjustment, unifying the resolution of all images to a preset standard value to facilitate subsequent processing and analysis; and noise filtering, removing random noise and interference signals from the images to improve image clarity. Through these preprocessing operations, a standardized image dataset is generated, which can improve the accuracy and efficiency of subsequent processing and reduce misjudgments caused by data quality issues.

[0023] In one embodiment, step S10 can be implemented as follows: P1: Acquire visible light images of power grid towers at different time periods using a high-resolution camera, and record the timestamp and geographic coordinates of each image.

[0024] In this embodiment, the high-resolution camera has high pixel count and good image quality, enabling it to clearly capture visible light images of power grid towers. A timestamp is used to mark the image acquisition time, and geographic coordinates are used to determine the tower's location. Acquiring images at different time periods allows for the acquisition of information about the tower under different lighting conditions and environments, providing a more comprehensive reflection of the tower's condition.

[0025] For example, during a power grid inspection mission, an inspection drone equipped with a high-resolution camera inspected a power grid in a mountainous area. At 9:00 AM, the camera captured a visible light image of a power pole located at 110.5° longitude and 25.8° latitude, with the timestamp recorded as 09:00:00 on December 20, 2024. At 3:00 PM, when flying over the same pole again, another image was captured, with the timestamp recorded as 15:00:00 on December 20, 2024. These two images from different time periods show differences in the tower's shadows and details due to the different angles of illumination, providing richer information for subsequent analysis.

[0026] In one embodiment, the inspection aircraft can fly along a preset route, with a high-resolution camera capturing images at fixed time intervals. For example, an image is captured every 2 seconds, and a high-precision clock is used to record the timestamp, while geographic coordinate information is obtained through the Global Positioning System (GPS). This ensures that the captured images have accurate time and location information.

[0027] P2: Perform integrity verification on the acquired raw image data, remove image frames that are blurry or have abnormal exposure, and generate a set of effective images after preliminary screening.

[0028] In this embodiment, integrity verification checks whether the image data is complete and whether there is any data loss or corruption. Blurry images may be caused by aircraft shaking, inaccurate camera focusing, etc., while abnormally exposed images may be caused by excessively strong or weak light.

[0029] For example, during the aforementioned power grid inspection in the mountainous area, one of the images collected might be blurry due to airflow affecting the aircraft. Another might be taken at sunrise, where the light is too strong, causing overexposure. By verifying integrity and removing blurry or abnormally exposed image frames, the quality of the image data can be improved.

[0030] In one embodiment, image blurriness or exposure abnormalities can be determined by calculating image sharpness and brightness indices. For example, edge sharpness and contrast are calculated; if the sharpness is below a certain threshold, the image is considered blurry; if the brightness exceeds the normal range, the image is considered abnormally exposed. Unsatisfactory image frames are discarded, generating a pre-screened set of valid images. Assuming 100 original images are acquired, after verification and screening, 10 blurry or abnormally exposed images are removed, resulting in 90 valid images.

[0031] P3: Adjust the resolution of the image data in the effective image set to ensure that the resolution of all images is uniformly set to a preset standard value.

[0032] In this embodiment, images acquired at different time periods may have different resolutions. To facilitate subsequent processing and analysis, the resolution of all images needs to be standardized to a preset standard value. Resolution adjustment can ensure consistency of image data and improve the accuracy of feature extraction and analysis.

[0033] For example, in previous inspections, some images may have been captured in high-resolution mode (4000×3000 pixels), while others may have been captured in low-resolution mode (2000×1500 pixels). The default standard resolution is 1920×1080 pixels, and the resolution of all images needs to be adjusted to this standard value.

[0034] In one embodiment, image scaling algorithms, such as bilinear interpolation or bicubic interpolation, can be used to adjust the image resolution. These algorithms can achieve image scaling while preserving image details. The adjusted image has a uniform resolution, providing a good data foundation for subsequent feature extraction and analysis. Taking a 4000×3000 pixel image as an example, using bilinear interpolation to adjust it to 1920×1080 pixels, although the image size becomes smaller, the algorithm retains as much of the image's key information as possible.

[0035] P4: Perform noise filtering on the image data after resolution adjustment to remove random noise and interference signals from the image, and obtain the image data after noise removal.

[0036] In this embodiment, random noise and interference signals may be caused by factors such as noise from the camera sensor and environmental interference. These noises and interferences can affect image clarity and the accuracy of feature extraction, thus requiring noise filtering.

[0037] For example, in mountainous environments, electromagnetic interference may exist, causing random white or black noise spots to appear in the acquired images. Alternatively, the camera sensor itself may have some noise, which can also appear as small particles in the image. These noise spots can interfere with the identification of tower features.

[0038] In one embodiment, median filtering, Gaussian filtering, and other methods can be used for noise filtering. Median filtering is a non-linear filtering method that can effectively remove salt-and-pepper noise from an image. Gaussian filtering is a linear filtering method that can smooth the image and reduce the impact of noise. For example, for an image containing salt-and-pepper noise, median filtering replaces the gray value of each pixel with the median of the gray values ​​of its neighboring pixels, thereby removing noise. After noise filtering, the resulting image data is improved, enhancing the image quality.

[0039] P5: Integrate the removed image data into a standardized image dataset.

[0040] For example, 90 images, after resolution adjustment and noise filtering, can be organized and stored according to certain rules. Metadata such as the filename, timestamp, and geographic coordinates of each image can be stored together with the image data in a database, forming a standardized dataset.

[0041] S20: Extract features from the tower images in the standardized image dataset, identify the precise location coordinates of the towers in the images, and generate a tower location distribution map; the tower location distribution map includes the center point coordinates of each tower and its relative distribution information in the image.

[0042] In this embodiment, feature extraction is the process of finding information that represents the features of a pole from a standardized image dataset. Pole features include edge features and texture features. Edge features reflect the outline shape of the pole, while texture features can represent the textural details of the pole's surface. For example, the metal structure of a pole may have unique texture features; extracting these features allows for more accurate pole identification.

[0043] In one embodiment, during feature extraction, an edge detection algorithm, such as the Canny edge detection algorithm, can first be used to detect the edge information of the towers in the image, generating an image containing the tower edges. Then, texture analysis methods, such as the gray-level co-occurrence matrix method, are used to extract texture features from the image. Based on these features, potential tower regions are selected. Further geometric analysis is performed on these potential tower regions to calculate their center point coordinates and contour information in the image. The center point coordinates are the coordinates of the geometric center of the tower in the image, accurately representing the tower's position. The contour information describes the tower's external boundary.

[0044] To distinguish different power poles, a unique identifier can be assigned to each pole. The center point coordinates and outline information of each pole are integrated into a pole location distribution map. This map clearly shows the specific location of each pole and its relative distribution relationships, such as the distance and angle between poles. The pole location distribution map is formatted to meet the input requirements of the intelligent region determination model. By generating the pole location distribution map, the distribution of power grid poles can be intuitively understood, providing an important basis for subsequent determination of the intercepting area.

[0045] S30: Input the tower location distribution map into the intelligent region determination model, analyze the distribution map through the model, and generate the target interception area range; the target interception area range is a rectangular or irregular polygonal region calculated based on the tower distribution density and image segmentation logic.

[0046] In this embodiment, the intelligent region determination model is a machine learning or deep learning-based model that, after being trained on a large amount of pole and tower image data, can accurately analyze the distribution map of pole and tower locations. Pole and tower distribution density refers to the ratio of the number of poles and towers within a certain area to the area of ​​that area, reflecting the density of the poles and towers. The image segmentation logic consists of rules and methods for dividing the image into different regions based on the distribution of poles and towers and the characteristics of the image.

[0047] After inputting the tower location distribution map into the intelligent region determination model, the model first extracts the tower distribution density and relative distance information. For example, it calculates the tower distribution density and the relative distance between towers within a certain local area. Through the model's analysis mechanism, this information is comprehensively evaluated to generate an initial intercepted area. The initial intercepted area may be a preliminary division of regions and requires further optimization.

[0048] Based on the initial intercepted area, the minimum distance between each tower and the intercepted boundary is calculated and mapped to a preset distance threshold range. Based on this distance information and the tower's location information, the final target intercepted area is determined. The target intercepted area may be rectangular, suitable for situations where the tower distribution is relatively regular; or it may be an irregular polygon, suitable for situations where the tower distribution is irregular. Generating the target intercepted area using an intelligent region determination model can more accurately determine the area containing towers and reduce interference from useless background information.

[0049] In one embodiment, in step S30, when the intelligent region determines the target interception area range of the model generation, it can also receive externally input task mode instructions; the task mode instructions include at least one of pole and tower fine inspection mode, channel environment inspection mode, and emergency fault inspection mode. The model dynamically adjusts the parameters used to calculate the target interception area range according to different task mode instructions, including: When in the refined inspection mode of the tower, the model prioritizes increasing the proportion of the tower body in the interception area, reduces the preset distance threshold between the interception boundary and the tower outline, and generates an interception area that closely surrounds one or more towers. When in the channel environment inspection mode, the model prioritizes covering the towers and their surrounding channel environment, expands the preset distance threshold, and generates an interception area that can include the corridor connecting the towers and the surrounding environment. When in the emergency fault inspection mode, the model prioritizes the rapid location and interception of potential fault points. It uses pre-stored prior knowledge of fault characteristics to filter the tower distribution map and prioritizes the generation of interception areas containing towers with fault characteristics.

[0050] Based on the target cropping area, the cropping boundary is optimized to generate a high-precision cropping boundary; the high-precision cropping boundary includes the precise coordinates of the boundary pixels and their transition smoothness information with adjacent areas.

[0051] S40: Based on the target cropping area range, optimize the cropping boundary to generate a high-precision cropping boundary; the high-precision cropping boundary includes the precise coordinates of the boundary pixels and their transition smoothness information with adjacent areas.

[0052] In this embodiment, the target cropping area is only a preliminary region division, and its boundaries may be inaccurate or uneven. Therefore, the cropping boundaries need to be optimized. The precise coordinates of the boundary pixels refer to the specific position coordinates of each pixel on the cropping boundary, and these coordinates determine the precise range of the cropping area. Transition smoothness information describes whether the transition between the cropping boundary and adjacent regions is natural and smooth.

[0053] During optimization, the distance information between each tower and the interception boundary is first extracted from the target interception area. The boundary is then initially adjusted based on this information. For example, if a tower is close to the interception boundary, the interception area can be appropriately expanded to ensure the tower is completely contained within it. Smoothness analysis is then performed on the initially adjusted boundary to identify abrupt or discontinuous boundary segments.

[0054] For abrupt or discontinuous boundary segments, boundary optimization processing is performed. Interpolation algorithms, such as cubic spline interpolation, can be used to generate smooth transition boundary curves. Simultaneously, pixel distribution information between the boundary and adjacent regions is considered to ensure a natural transition between the boundary and the background. The final determined boundary curves are then integrated into a high-precision truncation boundary. By optimizing the truncation boundary to generate a high-precision truncation boundary, the accuracy and quality of the truncated area can be improved, making the truncated image more consistent with actual needs.

[0055] S50: Generate a final interception command based on the high-precision interception boundary, and send the interception command to the image processing system to realize intelligent interception of the tower area in the power grid inspection image; the final interception command includes the starting coordinates, ending coordinates and corresponding image data index of the interception area.

[0056] In this embodiment, the final cropping instruction is a command used to instruct the image processing system to crop the image. The start and end coordinates of the cropping region represent the coordinates of the upper left and lower right corners of the cropping region in the image, respectively. These two coordinates determine the specific location and size of the cropping region. The image data index refers to the index of the image data corresponding to the cropping region within the entire image dataset, used for quickly locating and extracting the image data of the cropping region.

[0057] Based on the high-precision cropping boundary, the start and end coordinates of the cropping area are determined, and corresponding image data indexes are generated. The start and end coordinates, along with the image data indexes, are integrated into a final cropping command. This command is sent to the image processing system, which extracts the image data containing the tower from a standardized image dataset according to the command. For example, the image processing system determines the range of the cropping area based on the start and end coordinates, then finds the corresponding image data from the image dataset using the image data index and performs the cropping.

[0058] Through the above methods, this application embodiment achieves intelligent cropping of tower areas in power grid inspection images. Intelligent cropping can accurately extract the effective area containing towers from a large number of power grid inspection images, reducing interference from useless background information, improving the efficiency and accuracy of subsequent data analysis, and providing strong support for power grid operation and maintenance management.

[0059] In one embodiment, reference Figure 3 Step S20 may include steps S21-S25, which will be described in detail below: Step S21: Extract edge features and texture features of the pole images from the standardized image dataset, and filter out potential pole regions based on the edge features and texture features.

[0060] In this embodiment, edge features reflect the outline shape of the tower, while texture features reflect the texture details of the tower surface. By extracting these features and filtering out potential tower areas, the scope of subsequent analysis can be narrowed, improving the accuracy of identification.

[0061] For example, consider an image in a standardized image dataset containing multiple power grid poles and surrounding trees. The metal structures of the poles have distinct edges, which can be extracted using edge detection algorithms. Simultaneously, the metal surfaces of the poles have unique textures, different from those of the trees, which can be distinguished using texture analysis methods. Assuming there are five potential regions in the image, after filtering by edge and texture features, three of these regions are identified as potentially being the power grid pole areas.

[0062] In one embodiment, edge detection algorithms such as the Sobel operator can be used to extract edge features from the image. The Sobel operator detects edges by calculating the gradients of the image in the horizontal and vertical directions. For texture features, the gray-level co-occurrence matrix method can be used, which describes texture by statistically analyzing the spatial distribution of gray levels in the image. Based on the extracted edge and texture features, a certain threshold is set to filter out regions that meet the criteria as potential pole areas.

[0063] In one embodiment, reference Figure 4Step S21 can be implemented in the following way, including steps S211-215, which will be described in detail below: Step S211: Perform an edge detection algorithm on each image in the standardized image dataset to generate an initial edge map containing all edge information in the image.

[0064] In this embodiment, the edge detection algorithm is used to detect the edges of objects in an image. The initial edge map contains edge information of all objects in the image, and this map can be used to initially identify possible pole edges.

[0065] For example, consider an image in a standardized image dataset containing power grid poles and surrounding houses and trees. The Canny edge detection algorithm is used to process this image, detecting the edges of all objects in the image. The metal structure edges of the poles, the outlines of the houses, the branches of the trees, etc., are all detected, forming an initial edge map. Suppose that in the image, the edge of a crossbar of the pole is represented as a continuous white line in the initial edge map, with coordinates from (300, 200) to (400, 200).

[0066] In one embodiment, the Canny edge detection algorithm is used to process each image in the standardized image dataset. The Canny algorithm, through multi-stage processing including Gaussian smoothing, gradient calculation, non-maximum suppression, and double thresholding, accurately detects edges in the image. The resulting initial edge map clearly displays the edge information of all objects in the image.

[0067] Step S212: Extract periodic texture patterns from the image using texture analysis methods to generate texture feature maps related to the tower structure.

[0068] In this embodiment, the texture analysis method is used to identify periodic texture patterns in an image. The structure of a pole or tower possesses certain texture features, such as the texture of a metal surface. Generating a texture feature map associated with the pole or tower structure can help filter out potential pole or tower regions.

[0069] For example, the gray-level co-occurrence matrix (GLCM) method was used for texture analysis of the image above. The metal crossbars of the tower have regular textures, and their gray values ​​exhibit periodic changes within a certain range. The GLCM method was used to calculate the co-occurrence probability between different gray levels in the image, extracting texture features related to the tower structure. In the generated texture feature map, the texture feature value of the area where the tower crossbars are located is significantly different from the surrounding background; for example, the texture feature value of this area is 0.8, while the texture feature value of the surrounding background is between 0.2 and 0.3.

[0070] In one embodiment, texture analysis is performed using the gray-level co-occurrence matrix method. This method calculates the co-occurrence probability between different gray levels in an image to obtain the statistical features of the texture. Based on these features, a texture feature map related to the tower structure is generated. In this map, regions with tower texture features are clearly marked.

[0071] Step S213: Overlay the initial edge map with the texture feature map to identify candidate regions with typical geometric features of towers.

[0072] In this embodiment, the overlay process fuses information from the initial edge map and texture feature map. A candidate region with typical tower geometry refers to a region that simultaneously satisfies both edge and texture features in the overlay map.

[0073] For example, the initial edge map and texture feature map generated above are overlaid. In the overlaid map, the crossbar portion of the tower exhibits both distinct edge lines (from the initial edge map) and specific texture features (from the texture feature map). By setting a certain threshold, regions that simultaneously meet the edge and texture feature conditions are selected as candidate regions. For instance, in the overlaid map, the region with coordinates (300, 200) - (400, 200) simultaneously meets both edge and texture feature requirements and is identified as a candidate region.

[0074] In one embodiment, the initial edge map and texture feature map are added pixel by pixel to obtain an overlay map. Then, based on a preset threshold, regions with higher pixel values ​​in the overlay map are selected as candidate regions with typical geometric features of the tower.

[0075] Step S214: Perform morphological operations on the candidate regions to remove non-target regions that do not meet the tower size and shape constraints.

[0076] In this embodiment, morphological operations involve processing the image such as dilation, erosion, opening, and closing to change the shape and size of objects. Non-target regions that do not conform to the pole's size and shape constraints may be misidentified regions, such as small noise points or objects with significantly different shapes from the pole.

[0077] For example, among the candidate regions identified above, there may be some small noise points or areas that significantly differ from the shape of the tower. An opening operation is performed on the candidate regions; first, an erosion operation is performed to remove small noise points, and then a dilation operation is performed to restore the object's size. Suppose there is a small isolated point in the candidate region with coordinates (310, 210), which is removed using the erosion operation. After morphological operations, non-target regions that do not conform to the tower's size and shape constraints are removed, and the candidate regions better match the characteristics of the tower.

[0078] In one embodiment, the candidate region is first eroded to remove small noise points. Then, an opening operation is performed to remove non-target regions that significantly deviate from the tower's shape. These morphological operations make the candidate region more consistent with the tower's size and shape constraints.

[0079] Step S215: Retain candidate regions that meet the constraints as potential tower regions.

[0080] In this embodiment, candidate regions that meet the constraints are those whose size and shape conform to the characteristics of a tower after morphological operations. Retaining these regions as potential tower areas provides more accurate targets for subsequent analysis.

[0081] For example, after the morphological operations described above, only the region with coordinates in the range (300, 200) - (400, 200) from the original candidate regions meets the size and shape constraints of the tower. This region is retained as a potential tower region. Further analysis can then be performed on this region, such as calculating its center point coordinates and contour information.

[0082] In one embodiment, candidate regions after morphological operations are filtered based on preset tower size and shape constraints. Regions that meet the criteria are marked as potential tower regions for further analysis.

[0083] Step S22: Perform further geometric analysis on the potential tower area and calculate its center point coordinates and contour information in the image.

[0084] In this embodiment, geometric analysis involves analyzing the geometric attributes of the potential tower area, such as its shape and size. The center point coordinates are the coordinates of the geometric center of the tower in the image, accurately representing the tower's location; the contour information describes the tower's external boundary.

[0085] For example, among the three potential tower areas selected above, a geometric analysis is performed on one of them. The center point coordinates are obtained by calculating the sum of the coordinates of all pixels within the area and dividing by the number of pixels. Using a boundary tracking algorithm, such as Freeman chaincode, the positions of the boundary points are recorded sequentially starting from the edge of the area to obtain the contour information. Assuming the center point coordinates of this area are (500, 300), boundary tracking reveals that its contour consists of a series of coordinate points, such as (480, 280), (490, 290), etc.

[0086] In one embodiment, for the selected potential pole tower areas, they are first converted into binary images. Then, the sum of the coordinates of all pixels in the binary image is calculated and divided by the number of pixels to obtain the coordinates of the center point. A boundary tracking algorithm is used to record the positions of boundary points sequentially, starting from the edge of the potential pole tower area, thereby obtaining contour information.

[0087] Step S23: Based on the center point coordinates and contour information, assign a unique identifier to each tower. The identifier is used to distinguish the distribution locations of different towers.

[0088] In this embodiment, the unique identifier is used to accurately distinguish different towers in subsequent processing and analysis. The center point coordinates and outline information of each tower are unique, and the identifier assigned based on this information ensures that each tower has a unique identifier.

[0089] For example, for the three potential tower areas mentioned above, based on their center point coordinates and outline information, the first area is assigned the identifier "Tower_01", the second area is assigned the identifier "Tower_02", and the third area is assigned the identifier "Tower_03". This allows for quick location and processing of specific towers in subsequent analysis.

[0090] In one embodiment, a hash value is generated based on the center point coordinates and outline information of the tower, and this hash value is used as the unique identifier of the tower. The hash value is unique and deterministic, ensuring that the identifier of each tower is different. For example, hashing the center point coordinates (500, 300) and the outline information yields a unique hash value as the identifier of the tower.

[0091] Step S24: Integrate the center point coordinates and outline information of the towers into a tower location distribution map, which includes the specific location of each tower and its relative distribution relationship.

[0092] In this embodiment, the tower location distribution map integrates the center point coordinates and outline information of each tower to visually display the tower distribution. The relative distribution relationships can describe information such as distances and angles between towers.

[0093] For example, the center point coordinates and outline information of the three towers mentioned above can be integrated into a single image. The center point coordinates of "Tower_01" are (500, 300), the center point coordinates of "Tower_02" are (600, 400), and the center point coordinates of "Tower_03" are (700, 350). The distance between "Tower_01" and "Tower_02" can be calculated as √[(600 - 500)² + (400 - 300)²] = 141.4 (pixels), and the angle between them can be calculated using trigonometric functions. This clearly shows the relative distribution relationship between the towers.

[0094] In one embodiment, a graphical tool is used to plot the center point coordinates and outline information of the towers in a coordinate system, generating a tower location distribution map. Different colors or symbols can be used to represent different towers in the map to more clearly show the distribution.

[0095] Step S25: Format the tower location distribution map to adapt it to the input requirements of the intelligent region determination model.

[0096] In this embodiment of the application, the intelligent region determination model has specific input format requirements. Formatting the tower location distribution map can ensure that the model can correctly receive and process the data.

[0097] For example, the intelligent region determination model requires input data in matrix form, where each row represents a tower and each column represents an attribute, such as the x-coordinate and y-coordinate of the center point, or the perimeter of the outline. Transforming the information in the tower location distribution map above, the row data for "Tower_01" becomes [500, 300, 100] (assuming the outline perimeter is 100), the row data for "Tower_02" becomes [600, 400, 120], and the row data for "Tower_03" becomes [700, 350, 110]. This formatting process allows the tower location distribution map to adapt to the model's input requirements.

[0098] In one embodiment, if the intelligent region determination model requires matrix-formatted input data, the center point coordinates and contour information of the towers are converted into a matrix. Each row of the matrix represents a tower, and each column represents an attribute, such as the x-coordinate and y-coordinate of the center point, and the perimeter of the contour. This formatting process allows the tower location distribution map to adapt to the model's input requirements.

[0099] In one embodiment, reference Figure 5 Step S30 may include steps S31-S34, which will be described in detail below: Step S31: Input the tower location distribution map into the feature extraction module of the intelligent region determination model to extract the tower distribution density and relative distance information.

[0100] In this embodiment, the feature extraction module of the intelligent region determination model is used to extract key features from the tower location distribution map. The tower distribution density reflects the density of towers within a certain area, and the relative distance information describes the spatial distance relationship between towers.

[0101] For example, in a map showing the distribution of power pole locations in a city's power grid, there is an area with a relatively dense distribution of poles. After inputting this distribution map into the feature extraction module of the intelligent region determination model, the number of poles per square kilometer in this area is calculated to obtain the pole distribution density. Simultaneously, the Euclidean distance between the center points of every two poles is calculated to obtain relative distance information. Assuming there are 10 poles in an area of ​​1 square kilometer, the pole distribution density is 10 poles / square kilometer. The center point coordinates of poles "Tower_01" and "Tower_02" are (500, 300) and (600, 400) respectively, and the relative distance between them is √[(600 - 500)² + (400 - 300)²] = 141.4 (pixels).

[0102] In one embodiment, the tower location distribution map is divided into several small regions, the number of towers in each small region is counted, and the tower distribution density is calculated. For relative distance information, the Euclidean distance between the center points of every two towers is calculated. This information is extracted from the tower location distribution map using a feature extraction module to provide data support for subsequent analysis.

[0103] Step S32: Through the analysis mechanism of the model, the distribution density and relative distance information of the towers are comprehensively evaluated to generate the initial interception area range.

[0104] In this embodiment, the model's analysis mechanism is a method for processing and analyzing the extracted tower distribution density and relative distance information. A comprehensive evaluation considers the tower distribution and spatial relationships to determine the initial interception area containing the towers.

[0105] For example, in the aforementioned urban power grid area, based on the tower distribution density and relative distance information, the model identified several areas where towers were relatively concentrated. For one of these concentrated areas, where tower density was high and relative distances were close, the model defined this area as an initial cutoff region. Assuming there were 5 towers in this area, with relative distances all within 200 pixels, the model determined a rectangular initial cutoff region based on this information, with its top-left corner coordinates (450, 250) and bottom-right corner coordinates (750, 450).

[0106] In one embodiment, the model can employ cluster analysis to divide the towers into different clusters based on their distribution density and relative distance. Each cluster corresponds to an initial cutoff region, and by adjusting the clustering parameters, a suitable initial cutoff region range can be obtained.

[0107] In one embodiment, reference Figure 6 Step S32 may include steps S321-S325, which will be described in detail below: Step S321: Calculate the variation trend of the tower distribution density in the local area and generate a spatial distribution heat map reflecting the tower density.

[0108] In this embodiment, the variation trend of tower distribution density within a local area can reflect the distribution pattern of towers. The spatial distribution heat map uses the depth of color to represent the density of towers, with darker colors indicating denser towers.

[0109] For example, in a large power grid area, it is divided into several local areas. The tower distribution density is calculated for each local area. For instance, in area A, there are 5 towers per square kilometer; in area B, 10 towers per square kilometer; and in area C, 3 towers per square kilometer. Based on these density values, a spatial distribution heatmap is generated. In the heatmap, area B is the darkest, followed by area A, and area C is the lightest, visually reflecting the density distribution of the towers.

[0110] In one embodiment, the tower location distribution map can be divided into several small local regions, and the tower distribution density within each local region can be calculated. Based on the density value, each local region is represented by a different color, generating a spatial distribution heatmap. A color mapping table can be used to map the density values ​​to different colors, such as from blue (low density) to red (high density).

[0111] Step S322: Based on the spatial distribution heat map, determine the center point of each local area and its corresponding coverage area.

[0112] In this embodiment, the center point of a local area is its geometric center, and the coverage area represents the range of towers included in that area. By determining the center point and the coverage area, the distribution of towers can be described more accurately.

[0113] For example, in the spatial distribution heatmap above, region B is a densely populated area of ​​power poles and towers. Calculating the average coordinates of the center points of all power poles and towers within region B yields the coordinates of the center point of region B as (400, 500). Based on the distribution range of power poles and towers within region B, its coverage area is determined. Assuming a circular area with a radius of 100 pixels centered at the center point, this circular area encompasses all power poles and towers within region B.

[0114] In one embodiment, for each local area, the average coordinates of the center points of all towers are calculated and used as the center point of that local area. Based on the distribution range of the towers, a suitable radius is determined as the coverage area. For example, the maximum distance from the towers to the center point within that area can be calculated and used as the radius of the coverage area.

[0115] Step S323: Combine the relative distance information between the towers to adjust the boundary of the local area to ensure coverage of all key tower locations.

[0116] In this embodiment, the relative distance information between towers can help determine whether the boundary of a local area is reasonable. Adjusting the boundary of the local area can make the coverage area more accurately include the locations of all key towers.

[0117] For example, there is a single tower "Tower_04" near region B. It is close to the boundary of region B and relatively close to other towers within region B. Combining the relative distance information between the towers, the boundary of region B is appropriately expanded to include "Tower_04" within its coverage area. Assuming the original boundary of region B is a circle with a radius of 100 pixels, the radius is expanded to 120 pixels, ensuring that "Tower_04" is included within the new coverage area.

[0118] In one embodiment, for each local area, the poles near its boundary are examined. If a pole is close to the center point of the area but outside the coverage area, and far from the center point of an adjacent area, the boundary of the area is adjusted to include the pole within the coverage area.

[0119] Step S324: Perform a merging operation on the adjusted local area to generate an initial interception area range containing multiple towers.

[0120] In this embodiment, the merging operation combines adjacent and overlapping local areas into a larger area. Generating an initial cutoff area that includes multiple towers can reduce unnecessary segmentation and improve cutting efficiency.

[0121] For example, in a power grid area, there are two adjacent local regions C and D. Region C contains 3 towers, and region D contains 4 towers, with some overlap between the two regions. Merging regions C and D generates an initial intercepted region containing 7 towers. Assuming the boundary coordinates of region C are (100, 100) - (300, 300), and the boundary coordinates of region D are (200, 200) - (400, 400), the boundary coordinates of the merged initial intercepted region are (100, 100) - (400, 400).

[0122] In one embodiment, all adjusted local regions are traversed, and any overlaps are checked. If overlaps are found, these regions are merged into a new region. This process is repeated until all adjacent and overlapping regions are merged, generating an initial cutoff area containing multiple towers.

[0123] Step S325: Perform boundary correction on the initial intercepted area to meet the preset rectangular or irregular polygon format requirements.

[0124] In this embodiment, the preset rectangular or irregular polygon format requirement is to facilitate subsequent image processing and analysis. Boundary correction can make the shape of the initial cropped area conform to a specific format, improving the standardization and accuracy of the cropped area.

[0125] For example, the initial cutoff region obtained after the merging operation may have an irregular shape, while the default requirement is a rectangle. The bounding rectangle of this irregular region is calculated, and this bounding rectangle is used as the corrected cutoff region. Assuming the vertex coordinates of the irregular region are (150, 150), (250, 200), (300, 250), and (200, 300), the top-left corner of its bounding rectangle is (150, 150), and the bottom-right corner is (300, 300). This bounding rectangle is used as the corrected cutoff region.

[0126] In one embodiment, if the required cut-off region is a rectangle, for an irregular initial cut-off region, its circumscribed rectangle is calculated, and the circumscribed rectangle is used as the corrected cut-off region. If an irregular polygon is required, the boundary points of the initial cut-off region are fitted and adjusted using a certain algorithm to make it meet the format requirements of an irregular polygon, such as ensuring that the edges of the polygon are continuous and the interior angles conform to a certain range.

[0127] Step S33: Based on the initial interception area range, calculate the minimum distance between each tower and the interception boundary, and map it to a preset distance threshold range.

[0128] In this embodiment, calculating the minimum distance between each tower and the intercept boundary can assess the proximity of the tower to the intercept area. A preset distance threshold range is used to quantify and classify the distances for subsequent processing.

[0129] For example, given the initial cutoff area, the center point coordinates of tower "Tower_01" are (500, 300). The distances from this center point to the four boundaries of the cutoff area are calculated: 50 pixels to the left boundary, 50 pixels to the top boundary, 250 pixels to the right boundary, and 150 pixels to the bottom boundary. The minimum value of 50 pixels is taken as the minimum distance between the tower and the cutoff boundary. Assuming the preset distance threshold range is [0, 100] pixels, the minimum distance of 50 pixels is mapped to this range.

[0130] In one embodiment, for each tower, the distance from its center point to all points on the intercept boundary is calculated, and the minimum value is taken as the minimum distance between the tower and the intercept boundary. Then, according to a preset distance threshold interval, the minimum distance is linearly mapped so that it falls within the interval.

[0131] Step S34: Associate the minimum distance with the location information of the corresponding tower to generate the target interception area range, which includes the distribution density and boundary distance information of each tower.

[0132] In this embodiment, associating the minimum distance with the location information of the corresponding tower allows for a more comprehensive description of the characteristics of the target interception area. Distribution density and boundary distance information reflect the distribution of towers within the target interception area and their relationship to the boundary.

[0133] For example, in the above example, the minimum distance of 50 pixels for the tower "Tower_01" is associated with its location information (center point coordinates (500, 300)). Simultaneously, combined with the distribution density information of the area where the tower is located, the target interception area is generated. Assuming the distribution density of this area is 8 towers per square kilometer, integrating this information yields a more accurate target interception area, with its upper left corner coordinates adjusted to (430, 230) and lower right corner coordinates adjusted to (770, 470), incorporating both the tower distribution density and boundary distance information.

[0134] In one embodiment, the minimum distance and location information of each tower can be stored in a data structure, such as a dictionary, where the keys are the tower identifiers and the values ​​are the minimum distance and location information. Based on this association information, the initial interception area is adjusted to generate a target interception area that includes the distribution density and boundary distance information of each tower.

[0135] In this embodiment, the intelligent region determination model is a deep learning-based convolutional neural network model, mainly composed of a feature extraction module, an analysis module, and an output module. The following is a detailed description of each module: The feature extraction module extracts tower distribution density and relative distance information from the input tower location distribution map. It consists of multiple convolutional layers, pooling layers, and activation function layers.

[0136] Convolutional layers: These layers use convolutional kernels of varying sizes and numbers to perform sliding convolution operations on the input tower location distribution map to extract local features from the image. For example, using 3x3 or 5x5 kernels, each kernel can learn different feature patterns, such as tower edges and their density. The number of convolutional layers can be adjusted according to the specific situation, typically set to 3-5 layers.

[0137] Pooling layer: Following the convolutional layer, the pooling method is max pooling. The purpose of the pooling layer is to downsample the feature map output by the convolutional layer, reducing the size of the feature map, reducing computational cost, and enhancing the robustness of the features. For example, using a 2x2 max pooling window can reduce the size of the feature map by half.

[0138] Activation function layer: An activation function, such as the ReLU (Rectified Linear Unit) function, is applied after the convolutional layer to introduce non-linearity and increase the model's expressive power. The ReLU function is expressed as follows, and it can effectively alleviate the vanishing gradient problem.

[0139] The analysis module receives feature information output from the feature extraction module, comprehensively evaluates the tower distribution density and relative distance information, and generates an initial interception area. It consists of a fully connected layer and a decision layer.

[0140] Fully connected layers: These layers flatten the feature maps output by the feature extraction module and then further fuse and transform the features through multiple fully connected layers. Each neuron in a fully connected layer is connected to all neurons in the previous layer, enabling it to learn the complex relationships between features. The number of fully connected layers is typically set to 2-3, and the number of neurons in each layer can be adjusted according to the specific requirements.

[0141] Decision layer: The output of the fully connected layer is processed using the Softmax activation function to convert the output into a probability distribution, thereby determining the initial cut-off region.

[0142] The output module, based on the output of the analysis module and the tower location information, generates the final target interception area. It primarily integrates and processes the data, converting the probability distribution output by the analysis module into specific interception area coordinates.

[0143] The model training process is as follows: In training the intelligent region determination model, the first step is to collect a large number of power grid tower location distribution maps. These maps need to cover various tower distribution patterns, such as densely distributed areas and sparsely distributed areas, to ensure that the model can learn diverse features. After collection, each distribution map is meticulously labeled, clearly indicating the specific location of each tower and the corresponding target intercept area. The target intercept area can be a rectangle or an irregular polygon, accurately represented using the vertex coordinates of the polygon. Finally, the dataset is divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set is used for actual model training, allowing the model to learn the features and patterns in the data; the validation set is used to adjust the model's hyperparameters, such as the learning rate and batch size, during training to optimize model performance; and the test set is used to finally evaluate the model's generalization ability on unknown data.

[0144] Next, model initialization is performed, with a series of hyperparameters set appropriately. The learning rate controls the step size of model parameter updates, ranging from 0.001 to 0.01. The number of samples input to the model during each training iteration is set to 16-64, and the number of training epochs is set to 50-100. Simultaneously, the model's weight parameters need to be initialized. For this model using the ReLU activation function, the He initialization method can be used. During training, samples from the training set are input into the model, and the prediction results are obtained through calculations by the feature extraction module, analysis module, and output module. The mean squared error loss function is used to calculate the difference between the predicted result and the labeled target cutoff region. Then, the gradient of the model parameters is calculated based on the loss using the backpropagation algorithm. Finally, the Adam optimization algorithm is used to update the model parameters based on the gradient. The Adam algorithm combines the advantages of AdaGrad and RMSProp, enabling adaptive learning rate and accelerating model convergence.

[0145] After each training round, the model is evaluated using a validation set, calculating metrics such as loss and accuracy on the validation set. Based on the evaluation results, the model's hyperparameters are flexibly adjusted. For example, if the model's loss on the validation set decreases slowly, the learning rate can be adjusted appropriately; if the accuracy does not meet expectations, the batch size can be changed. After the model completes all training rounds, a final evaluation is performed using a test set. Multiple metrics, including loss, accuracy, recall, and F1 score, are calculated on the test set to comprehensively evaluate the model's generalization ability on unknown data. If the evaluation results are unsatisfactory, it may be necessary to re-examine the data quality, adjust the model structure, or further optimize the hyperparameters. Through continuous iterative optimization, the model can achieve optimal performance for better application in real-world scenarios of intelligent power grid image capture.

[0146] In one embodiment, reference Figure 7 Step S30 may include steps S41-S45, which will be described in detail below: Step S41: Extract the distance information between each tower and the interception boundary from the target interception area, and make preliminary adjustments to the boundary based on the distance information.

[0147] In this embodiment, the distance information between each tower and the cutoff boundary reflects the relative positional relationship between the tower and the boundary. Preliminary adjustments to the boundary based on this distance information can make the cutoff area more accurately encompass the towers.

[0148] For example, within the target capture area, there are three towers: "Tower_01" has a minimum distance of 20 pixels from the capture boundary, "Tower_02" has a minimum distance of 50 pixels from the capture boundary, and "Tower_03" has a minimum distance of 10 pixels from the capture boundary. According to preset rules, if the minimum distance is less than 30 pixels, the capture area needs to be expanded. For "Tower_01" and "Tower_03", since they are close to the boundary, the capture area is appropriately expanded in their direction. Assuming the original capture area's top-left corner coordinates are (100, 100) and bottom-right corner coordinates are (500, 500), for "Tower_03" which is closer to the left boundary, the left boundary is expanded 20 pixels to the left, and the new top-left corner coordinates become (80, 100); for "Tower_01" which is closer to the top boundary, the top boundary is expanded 15 pixels upwards, and the new top-left corner coordinates become (80, 85).

[0149] In one embodiment, the distance information between each tower and the interception boundary is extracted from the data structure of the target interception area. Based on preset rules, such as expanding the interception area when the distance is less than a certain threshold and shrinking the interception area when the distance is greater than a certain threshold, the interception boundary is initially adjusted.

[0150] Step S42: Perform smoothness analysis on the initially adjusted boundary to identify boundary segments with abrupt changes or discontinuities.

[0151] In this embodiment, smoothness analysis examines the continuity and smoothness of the boundaries. Abrupt or discontinuous boundary segments may affect the quality of the extracted region and subsequent analysis results.

[0152] For example, there might be abrupt transitions on the boundary of the initially adjusted intercepted area. Suppose there's a line segment on the boundary that moves from coordinates (200, 300) to (220, 350) and then back to (250, 300). At the point (220, 350), there's a significant change in slope, which constitutes an abrupt boundary segment. By calculating the slope change between adjacent boundary points, if the slope change exceeds a certain threshold, the boundary segment is considered to have an abrupt transition.

[0153] In one embodiment, the slope change between adjacent points on the boundary is calculated. If the slope change exceeds a certain threshold, the boundary segment is considered to have an abrupt change. By traversing all points on the boundary, boundary segments with abrupt changes or discontinuities are identified.

[0154] Step S43: For the abrupt or discontinuous boundary segments, perform boundary optimization processing to generate a smooth transition boundary curve.

[0155] In this embodiment, boundary optimization processing corrects abrupt or discontinuous boundary segments to make them smoother. A smooth transition boundary curve can improve the quality and aesthetics of the intercepted region.

[0156] For example, for the boundary segments with abrupt changes mentioned above, a cubic spline interpolation algorithm is used. Assuming the coordinates of the three points in the abrupt boundary segment are (200, 300), (220, 350), and (250, 300), the cubic spline interpolation algorithm inserts new points between these three points, such as at (210, 325) and (230, 330), so that the boundary forms a smooth curve transition between these points.

[0157] In one embodiment, a cubic spline interpolation algorithm is used to process abrupt or discontinuous boundary segments. This algorithm makes the boundaries continuous and smooth by inserting smooth curves between boundary points.

[0158] In one embodiment, reference Figure 8 Step S43 may include steps S431-S435, which will be described in detail below: Step S431: Identify the set of pixel coordinates of the abrupt or discontinuous boundary segment and record its grayscale difference value with the adjacent area.

[0159] In this embodiment, abrupt or discontinuous boundary segments refer to parts on the boundary where there is a sudden change or discontinuity. A set of pixel coordinates is used to determine the specific location of these boundary segments, and the grayscale difference value reflects the difference in grayscale between the boundary segment and adjacent areas, which can be used for subsequent boundary adjustments.

[0160] For example, on the boundary of the initially adjusted cropped area, there is a boundary segment that changes from coordinates (200, 300) to (220, 350) and then back to (250, 300), with a sudden change at (220, 350). The set of pixel coordinates identifying this abrupt boundary segment is [(200, 300), (220, 350), (250, 300)]. Simultaneously, the grayscale difference between this boundary segment and adjacent areas is calculated, assuming the average grayscale value of the boundary segment is 100, the average grayscale value of adjacent areas is 150, and the grayscale difference is 50.

[0161] In one embodiment, the initially adjusted boundary is checked pixel by pixel. When a positional change or grayscale change between adjacent pixels is found to exceed a certain threshold, the pixel and its adjacent pixels are treated as part of abrupt or discontinuous boundary segments, and their coordinates are recorded. Simultaneously, the difference between the average grayscale value of the boundary segment pixel and the average grayscale value of the adjacent region pixels is calculated as the grayscale difference value.

[0162] Step S432: Perform an interpolation algorithm on the set of pixel coordinates to generate a preliminary smooth boundary transition curve.

[0163] In this embodiment, the interpolation algorithm is used to insert new pixels between abrupt or discontinuous boundary segments, making the boundary smoother. The initially smoothed boundary transition curve is a transition curve obtained through the interpolation algorithm, which can improve the continuity of the boundary.

[0164] For example, for the pixel coordinate set [(200, 300), (220, 350), (250, 300)] of the aforementioned abrupt boundary segment, a cubic spline interpolation algorithm is used. First, the coefficients of the interpolation polynomial are calculated based on these points, and then new points are inserted between these points. Assuming points (210, 320) and (230, 330) are inserted, a preliminary smooth boundary transition curve is formed, with coordinates (200, 300), (210, 320), (220, 350), (230, 330), and (250, 300) respectively.

[0165] In one embodiment, a cubic spline interpolation algorithm is used to process the pixel coordinate set. This algorithm constructs a cubic polynomial between adjacent pixels, ensuring the curve is smooth in each interval. Based on the points in the pixel coordinate set, the coefficients of the interpolation polynomial are calculated, thereby generating a preliminary smooth boundary transition curve. Specifically, this can be achieved as follows:

[0166] F1: Obtain all pixel coordinates of the abrupt or discontinuous boundary segment and arrange them in spatial order to generate an ordered coordinate sequence.

[0167] In this embodiment, the pixel coordinates of abrupt or discontinuous boundary segments are key information for determining the boundary location. Arranging these coordinates in spatial order to generate an ordered sequence provides the correct data order for subsequent interpolation algorithms, ensuring the accuracy of the interpolation.

[0168] For example, there is a sudden change region on the boundary, and its pixel coordinates are randomly recorded as [(220, 350), (200, 300), (250, 300)]. Arranged in spatial order from left to right and from top to bottom, the horizontal coordinates are compared first. If the horizontal coordinates are the same, the vertical coordinates are compared to obtain an ordered coordinate sequence [(200, 300), (220, 350), (250, 300)].

[0169] In one embodiment, the pixel coordinates of abrupt or discontinuous boundary segments are sorted according to their position in the image, from left to right and from top to bottom. The horizontal coordinates of the pixels can be compared first; if the horizontal coordinates are the same, then the vertical coordinates are compared, thereby generating an ordered coordinate sequence.

[0170] F2: Perform cubic spline interpolation on the ordered coordinate sequence to generate a preliminary smooth boundary transition curve.

[0171] In this embodiment, the cubic spline interpolation algorithm is an interpolation method that constructs a cubic polynomial between adjacent coordinate points, ensuring the curve is smooth in each interval. Generating a preliminary smooth boundary transition curve can improve the continuity and smoothness of the boundary.

[0172] For example, for the ordered coordinate sequence [(200, 300), (220, 350), (250, 300)] above, a cubic spline interpolation algorithm is used. First, the first or second derivative of the boundary points is determined according to the boundary conditions. It is assumed that the first derivative of the boundary points (200, 300) and (250, 300) is 0. Then, by solving the system of equations, the polynomial coefficients of each interval are obtained. New points are inserted between (200, 300) and (220, 350), and between (220, 350) and (250, 300), such as (210, 320) and (230, 330), to generate preliminary smooth boundary transition curves with coordinates of (200, 300), (210, 320), (220, 350), (230, 330), and (250, 300).

[0173] In one embodiment, the coefficients of a cubic spline interpolation polynomial are calculated based on points in an ordered coordinate sequence. First, the first or second derivatives of the boundary points are determined according to the boundary conditions. Then, the polynomial coefficients for each interval are obtained by solving a system of equations. Finally, a preliminary smooth boundary transition curve is generated based on these coefficients.

[0174] F3: Calculate the rate of change of curvature of the preliminary smooth curve to identify local areas where excessive curvature may exist.

[0175] In this embodiment, the rate of change of curvature reflects the change in the degree of curvature of the curve. Identifying local areas that may be excessively curved can prevent unreasonable curvature of the boundary curve and ensure the rationality of the boundary.

[0176] For example, for the aforementioned preliminary smoothed curve [(200, 300), (210, 320), (220, 350), (230, 330), (250, 300)], the rate of curvature change is obtained by calculating the change in the tangent slope between adjacent points. Near (220, 350), the tangent slope changes significantly, and the rate of curvature change exceeds a preset threshold, identifying this region as a potentially excessively curved local area.

[0177] In one embodiment, the rate of curvature change is calculated point-by-point on the initially smoothed curve. The rate of curvature change can be obtained by calculating the change in the slope of the tangent line between adjacent points. When the rate of curvature change exceeds a certain threshold, the local region is considered to be excessively curved.

[0178] F4: Perform curvature smoothing on the excessively curved local area to reduce its impact on the overall boundary curve.

[0179] In this embodiment, curvature smoothing involves adjusting excessively curved local areas to make their curvature changes more gradual. Reducing its impact on the overall boundary curve ensures the overall quality of the boundary curve.

[0180] For example, for the excessively curved local region near (220, 350) identified above, a local interpolation adjustment method is used. Interpolation points are added within this region, such as at (215, 335) and (225, 340), and the coefficients of the interpolation polynomial are recalculated to reduce the rate of curvature change to a reasonable range. The adjusted boundary curve exhibits less curvature in this region, thus reducing its impact on the overall boundary curve.

[0181] In one embodiment, for identified excessively curved local regions, a local interpolation adjustment method is used to smooth the curvature. Interpolation points can be added or removed within this region, or the coefficients of the interpolation polynomial can be adjusted to reduce the rate of curvature change to a reasonable range.

[0182] F5: Use the processed boundary curve as the initial smoothed boundary transition curve.

[0183] In this embodiment, the boundary curve after curvature smoothing is more reasonable and smoother. Using it as a preliminary smoothed boundary transition curve can provide a better foundation for subsequent boundary adjustments and optimizations.

[0184] For example, after the curvature smoothing process described above, the boundary curve becomes [(200, 300), (210, 320), (215, 335), (220, 350), (225, 340), (230, 330), (250, 300)]. This processed boundary curve is used as the new, initially smoothed boundary transition curve. Further multi-scale analysis and adjustments can be performed on this curve to further improve the quality of the boundary.

[0185] In one embodiment, the pixel coordinate information of the processed boundary curve can be updated as a new, initially smoothed boundary transition curve. Subsequent multi-scale analysis and adjustments can be performed on this curve to further improve the quality of the boundary.

[0186] Step S433: Adjust the curvature of the boundary curve based on the grayscale difference value to make it closer to the actual boundary line between the tower and the background.

[0187] In this embodiment, the grayscale difference value reflects the grayscale changes between the boundary segment and adjacent areas. Adjusting the curvature of the boundary curve can better adapt the boundary to the actual dividing line between the tower and the background, improving the accuracy of the cropping.

[0188] For example, the grayscale difference value of 50 in the aforementioned abrupt boundary segment indicates a significant change in grayscale on both sides of the boundary, which may be the dividing line between the tower and the background. Based on this grayscale difference value, the curvature of the boundary curve in this region can be appropriately increased. Assuming the original boundary curve is relatively flat near (220, 350), the adjustment increases the curvature near that point, making it closer to the actual dividing line between the tower and the background.

[0189] In one embodiment, the curvature of the boundary curve is adjusted based on the magnitude of the grayscale difference. If the grayscale difference exceeds a certain threshold, the curvature of the curve is increased; if the grayscale difference is small, the curvature of the curve is decreased. In this way, the boundary curve more closely approximates the actual boundary between the tower and the background.

[0190] Step S434: Perform multi-scale analysis on the adjusted boundary curve to verify its smoothness and continuity at different resolutions.

[0191] In this embodiment, multi-scale analysis involves observing and analyzing the boundary curves at different resolutions. Verifying the smoothness and continuity of the boundary curves at different resolutions ensures that the boundary maintains good quality under various conditions.

[0192] For example, the adjusted boundary curve is resampled at both high resolution (e.g., 300 dpi) and low resolution (e.g., 72 dpi). At high resolution, the boundary curve is checked to see if it remains smooth without noticeable jagged edges or breaks; at low resolution, the boundary curve is observed to maintain its basic shape and continuity. Suppose the boundary curve is smooth at high resolution, but small jagged edges are found near (220, 350) at low resolution; further adjustments are made to this area to ensure good smoothness and continuity even at low resolution.

[0193] In one embodiment, the adjusted boundary curve is resampled at different resolutions, and its smoothness and continuity at each resolution are then examined. Smoothness and continuity can be evaluated by calculating metrics such as the rate of change of the curve's curvature and the distance between adjacent points. If problems are found, the boundary curve is further adjusted.

[0194] Step S435: Integrate the finalized boundary curve into the high-precision truncation boundary to ensure a natural and accurate transition with adjacent areas.

[0195] In this embodiment, the final determined boundary curve is a high-quality curve obtained after multi-scale analysis and adjustment. Integrating it into the high-precision truncation boundary can improve the quality of the truncation boundary and make the transition between the boundary and adjacent areas more natural and accurate.

[0196] For example, after multi-scale analysis and adjustments, the final boundary curve coordinates are determined to be (200, 300), (210, 320), (220, 350), (230, 330), and (250, 300). These coordinates are integrated into a high-precision boundary-truncation data structure. Simultaneously, based on the pixel characteristics of adjacent regions, the color and transparency of the boundary curves are fine-tuned. Assuming the adjacent region is green vegetation, the color of the boundary curves is slightly adjusted towards green, and the transparency is also adjusted according to the distance from the vegetation, ensuring a natural and precise transition between the boundary and the adjacent region.

[0197] In one embodiment, the pixel coordinate information of the finally determined boundary curve is merged with the existing information of the high-precision truncated boundary. Simultaneously, based on the pixel characteristics of adjacent regions, the color and transparency of the boundary curve are fine-tuned to ensure a natural and precise transition with adjacent regions.

[0198] Step S44: Match the smooth transition boundary curve with the pixel distribution information of the adjacent area to ensure a natural transition between the boundary and the background area.

[0199] In this embodiment, the pixel distribution information of adjacent regions reflects the characteristics of the background. Matching the smoothly transitioning boundary curve with the pixel distribution information can make the transition between the boundary and the background region more natural and reduce obvious segmentation marks.

[0200] For example, the background area near the boundary might be green vegetation, with pixel colors primarily concentrated in the green hue range. For a smoothly transitioning boundary curve, the color and texture features of nearby pixels are analyzed, and the color and transparency of the boundary curve are adjusted to gradually transition towards green. The transparency is also appropriately adjusted based on the distance from the background, resulting in a more natural transition between the boundary and the background area.

[0201] In one embodiment, pixel distribution characteristics, such as color mean and variance, are calculated around each point on the boundary curve. Based on these characteristics, the color and transparency of the boundary curve are adjusted to make its transition with the background area natural.

[0202] Step S45: Integrate the final determined boundary curves into a high-precision truncation boundary.

[0203] In this embodiment, the high-precision cropping boundary is a boundary determined after optimization, possessing high accuracy and quality. Integrating the finally determined boundary curves into the high-precision cropping boundary can provide an accurate basis for subsequent image cropping.

[0204] For example, after the above series of processes, a smooth boundary curve with a natural transition to the background is obtained, with boundary point coordinates of (80, 85), (210, 325), (230, 330), and (500, 500), respectively. These boundary point coordinates are integrated into a data structure, such as a list, to form a high-precision cropping boundary. Subsequent image processing systems can then accurately crop the image based on this high-precision boundary.

[0205] In one embodiment, the coordinate information of the finally determined boundary curve can be converted into a format suitable for image processing systems, such as a list of vertex coordinates of a polygon. This coordinate information is then integrated to form a high-precision truncated boundary.

[0206] Accordingly, to better implement the above methods, this application also provides an intelligent power grid image capture system based on the location coordinates of power grid towers. For example... Figure 9 As shown, the intelligent power grid image capture system 80 based on the location coordinates of power grid towers includes:

[0207] The acquisition module 801 is used to acquire raw image data of power grid towers through image acquisition equipment mounted on the inspection aircraft, and to perform preprocessing operations on the raw image data to generate a standardized image dataset; the raw image data includes visible light images of the towers and corresponding spatial location information; Feature processing module 802 is used to extract features from the tower images in the standardized image dataset, identify the precise position coordinates of the towers in the images, and generate a tower location distribution map; the tower location distribution map includes the center point coordinates of each tower and its relative distribution information in the image; The model analysis module 803 is used to input the tower location distribution map into the intelligent region determination model, analyze the distribution map through the model, and generate the target interception area range; the target interception area range is a rectangular or irregular polygonal region calculated based on the tower distribution density and image segmentation logic. The boundary optimization module 804 is used to optimize the truncated boundary according to the target truncated area range to generate a high-precision truncated boundary; the high-precision truncated boundary includes the precise coordinates of the boundary pixels and their transition smoothness information with adjacent areas; The instruction generation module 805 is used to generate a final interception instruction based on the high-precision interception boundary and send the interception instruction to the image processing system to realize intelligent interception of the tower area in the power grid inspection image; the final interception instruction includes the starting coordinates, ending coordinates and corresponding image data index of the interception area.

[0208] The implementation details of each module are available in the preceding method embodiments and will not be repeated here. The technical effects achieved by each module and device are described in the foregoing method embodiments.

[0209] like Figure 10 As shown, this application embodiment also provides a computer device 90, which includes a processor 901 and a memory 902, wherein the memory 902 stores a computer program, and when the computer program is executed by the processor 901, the processor 901 performs the steps of any of the methods described above.

[0210] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application are still within the scope of this application.

Claims

1. A method for intelligent cropping of power grid images based on the location coordinates of power grid towers, characterized in that, Includes the following steps: The raw image data of the power grid towers is acquired by an image acquisition device mounted on an inspection aircraft, and preprocessing is performed on the raw image data to generate a standardized image dataset; the raw image data includes visible light images of the towers and corresponding spatial location information. Feature extraction is performed on the tower images in the standardized image dataset to identify the precise location coordinates of the towers in the images and generate a tower location distribution map; the tower location distribution map includes the center point coordinates of each tower and its relative distribution information in the image. The distribution map of the pole locations is input into the intelligent region determination model. The model analyzes the distribution map to generate the target interception area. The target interception area is a rectangular or irregular polygonal region calculated based on the pole distribution density and image segmentation logic. Based on the target cropping area, the cropping boundary is optimized to generate a high-precision cropping boundary; the high-precision cropping boundary includes the precise coordinates of the boundary pixels and their transition smoothness information with adjacent areas; A final interception command is generated based on the high-precision interception boundary, and the interception command is sent to the image processing system to realize intelligent interception of the tower area in the power grid inspection image; the final interception command includes the starting coordinates, ending coordinates and corresponding image data index of the interception area.

2. The method according to claim 1, characterized in that, Raw image data of power grid towers is acquired using image acquisition equipment mounted on an inspection aircraft, and preprocessing is performed on the raw image data to generate a standardized image dataset, including: Visible light images of power grid towers were captured by high-resolution cameras at different times, and the timestamp and geographic coordinates of each image were recorded. The integrity of the acquired raw image data is verified, and image frames with blurriness or abnormal exposure are removed to generate a set of effective images after preliminary screening. The resolution of the image data in the effective image set is adjusted to ensure that the resolution of all images is uniformly set to a preset standard value; The image data after resolution adjustment is subjected to noise filtering to remove random noise and interference signals from the image, resulting in the image data after noise removal. The removed image data is integrated into a standardized image dataset.

3. The method according to claim 1, characterized in that, The step of extracting features from the tower images in the standardized image dataset, identifying the precise location coordinates of the towers in the images, and generating a tower location distribution map includes: Edge and texture features of pole images are extracted from the standardized image dataset, and potential pole regions are selected based on the edge and texture features. Further geometric analysis is performed on the potential tower area to calculate its center point coordinates and contour information in the image; Based on the center point coordinates and contour information, a unique identifier is assigned to each tower, and the identifier is used to distinguish the distribution location of different towers; The center point coordinates and outline information of the towers are integrated into a tower location distribution map, which includes the specific location of each tower and its relative distribution relationship. The tower location distribution map is formatted to adapt it to the input requirements of the intelligent region determination model.

4. The method according to claim 1, characterized in that, The process involves inputting the tower location distribution map into the intelligent region determination model, analyzing the distribution map through the model, and generating the target interception area range, including: The tower location distribution map is input into the feature extraction module of the intelligent region determination model to extract the tower distribution density and relative distance information. The analysis mechanism of the model is used to comprehensively evaluate the tower distribution density and relative distance information to generate an initial interception area. Based on the initial interception area, calculate the minimum distance between each tower and the interception boundary, and map it to a preset distance threshold range; The minimum distance is associated with the location information of the corresponding tower to generate a target interception area, which includes the distribution density and boundary distance information of each tower.

5. The method according to claim 1, characterized in that, The step of optimizing the interception boundary based on the target interception area range to generate a high-precision interception boundary includes: Extract the distance information between each tower and the interception boundary from the target interception area, and make preliminary adjustments to the boundary based on the distance information; Smoothness analysis is performed on the initially adjusted boundaries to identify boundary segments with abrupt changes or discontinuities; For the abrupt or discontinuous boundary segments, perform boundary optimization processing to generate boundary curves with smooth transitions; The smooth transition boundary curve is matched with the pixel distribution information of the adjacent area to ensure a natural transition between the boundary and the background area; The finalized boundary curves are integrated into a high-precision truncation boundary.

6. The method according to claim 3, characterized in that, The step of extracting edge and texture features from the standardized image dataset and filtering potential tower regions based on these edge and texture features includes: An edge detection algorithm is performed on each image in the standardized image dataset to generate an initial edge map containing all edge information in the image; Periodic texture patterns in images are extracted using texture analysis methods to generate texture feature maps related to tower structures; The initial edge map and the texture feature map are overlaid to identify candidate regions with typical geometric features of towers; Perform morphological operations on the candidate regions to remove non-target regions that do not meet the tower size and shape constraints; Candidate regions that meet the constraints are retained as potential pole / tower regions.

7. The method according to claim 4, characterized in that, The analysis mechanism of the model comprehensively evaluates the tower distribution density and relative distance information to generate an initial interception area, including: Calculate the variation trend of the tower distribution density in a local area to generate a spatial distribution heat map reflecting the tower density. Based on the spatial distribution heat map, the center point of each local area and its corresponding coverage area are determined. By combining the relative distance information between the towers, the boundaries of the local area are adjusted to ensure coverage of all key tower locations; The adjusted local area is merged to generate an initial intercepted area range that includes multiple towers; The initial captured area is subjected to boundary correction to meet the preset rectangular or irregular polygon format requirements.

8. The method according to claim 5, characterized in that, The step of performing boundary optimization processing on the abrupt or discontinuous boundary segments to generate smooth transition boundary curves includes: Identify the set of pixel coordinates of the abrupt or discontinuous boundary segment and record its grayscale difference value with the adjacent region; An interpolation algorithm is performed on the set of pixel coordinates to generate a preliminary smooth boundary transition curve; The curvature of the boundary curve is adjusted based on the grayscale difference value to make it closer to the actual boundary between the tower and the background. Multi-scale analysis was performed on the adjusted boundary curve to verify its smoothness and continuity at different resolutions; The finalized boundary curves are integrated into the high-precision truncation boundary to ensure a natural and accurate transition with adjacent areas.

9. The method according to claim 8, characterized in that, The step of performing an interpolation algorithm on the set of pixel coordinates to generate a preliminary smooth boundary transition curve includes: Obtain all pixel coordinates of the abrupt or discontinuous boundary segment, and arrange them in spatial order to generate an ordered coordinate sequence; A cubic spline interpolation algorithm is performed on the ordered coordinate sequence to generate a preliminary smooth boundary transition curve; Calculate the rate of curvature change of the boundary transition curve to identify local regions where excessive curvature may exist; Curvature smoothing is performed on the excessively curved local areas to reduce their impact on the overall boundary curve; The processed boundary curve is used as the initial smoothed boundary transition curve.

10. A smart image capture system for power grids based on the coordinates of power grid tower locations, characterized in that, The system includes: The acquisition module is used to acquire raw image data of power grid towers through image acquisition equipment mounted on the inspection aircraft, and to perform preprocessing operations on the raw image data to generate a standardized image dataset; the raw image data includes visible light images of the towers and corresponding spatial location information; The feature processing module is used to extract features from the tower images in the standardized image dataset, identify the precise location coordinates of the towers in the images, and generate a tower location distribution map; the tower location distribution map includes the center point coordinates of each tower and its relative distribution information in the image. The model analysis module is used to input the tower location distribution map into the intelligent region determination model, and analyze the distribution map through the model to generate the target interception area range; the target interception area range is a rectangular or irregular polygonal region calculated based on the tower distribution density and image segmentation logic. The boundary optimization module is used to optimize the truncated boundary according to the target truncated area range to generate a high-precision truncated boundary; the high-precision truncated boundary includes the precise coordinates of the boundary pixels and their transition smoothness information with adjacent areas; The instruction generation module is used to generate a final interception instruction based on the high-precision interception boundary, and send the interception instruction to the image processing system to realize intelligent interception of the tower area in the power grid inspection image; the final interception instruction includes the starting coordinates, ending coordinates and corresponding image data index of the interception area.

Images