Transmission line inspection data processing system based on satellite images

Abstract

The invention relates to the technical field of power transmission line inspection, in particular to a power transmission line inspection data processing system based on satellite images, and aims to solve the limitation of existing power transmission line inspection on large-scale power transmission line corridor wide-area continuous environment robustness macroscopic monitoring. The system comprises a satellite remote sensing data preprocessing module, a power transmission line channel risk point automatic identification and labeling module, a power transmission line pixel principal axis reduction module based on direction field aggregation stability domain judgment, and a power transmission line space-time database and risk assessment module. By adopting the technical scheme, the inspection efficiency, the coverage range and the risk early warning capability of the power transmission line can be improved, wide-area, continuous and robust situation awareness and dynamic change monitoring are realized, and the operation and maintenance intelligence level and the risk prevention and control capability are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power transmission line inspection technology, and in particular to a power transmission line inspection data processing system based on satellite imagery. Background Technology

[0002] With the continuous expansion of the power grid, traditional manual inspection methods can no longer meet the demand for efficient and accurate inspection of large-scale transmission lines. Transmission line inspection data processing systems based on satellite imagery have gradually become an important inspection method due to their wide coverage and lack of terrain limitations. However, existing transmission line inspection technologies still have shortcomings in areas such as large-scale monitoring, adaptability to complex environments, and data processing capabilities, affecting inspection efficiency and accuracy. A search revealed a UAV-based intelligent transmission line inspection system and method (publication number CN103824340B), which uses UAVs equipped with various sensors to inspect transmission lines and utilizes a scheduling system and monitoring platform to optimize the allocation of inspection resources. However, this technical solution mainly relies on UAV equipment, and its inspection range is limited by the UAV's endurance and flight altitude, making it difficult to achieve comprehensive coverage of large-area transmission line corridors. Furthermore, in complex geographical environments (such as mountainous areas and forests), UAVs may be affected by terrain obstruction or weather conditions, resulting in incomplete or insufficiently clear inspection data. Meanwhile, the plan does not address how to combine satellite imagery for large-scale monitoring and data processing, making it difficult to meet the dynamic monitoring needs of changes in the environment surrounding power transmission lines.

[0003] On the other hand, a transmission line inspection system and data transmission method were disclosed in publication number CN111028377B. This system uses multiple sensors mounted on a drone to transmit inspection data to a ground control module at different data transmission frequencies, thereby reducing electromagnetic interference and improving data transmission efficiency. However, this solution also relies on a drone as the core inspection tool. While it solves the interference problem of the data transmission link, its inspection range and accuracy are still limited by the drone's flight altitude and endurance. Furthermore, this solution does not consider how to utilize satellite imagery for large-scale monitoring of transmission line corridors, nor does it provide image processing and data analysis methods for complex environments, making it difficult to meet the needs of full lifecycle management of transmission lines.

[0004] The aforementioned problems indicate that existing transmission line inspection technologies still have shortcomings in terms of wide-area coverage, adaptability to complex environments, and data processing efficiency. Therefore, this invention provides a satellite imagery-based transmission line inspection data processing system. It aims to achieve comprehensive monitoring and dynamic analysis of transmission lines and their surrounding environment by combining high-resolution satellite imagery with intelligent data processing algorithms, thereby improving inspection efficiency and accuracy and meeting the demands of modern power grids for efficient and intelligent inspection systems. Summary of the Invention

[0005] To achieve the aforementioned objectives, this invention provides a satellite imagery-based data processing system for power transmission line inspection, aiming to overcome the limitations of existing power transmission line inspection technologies in achieving wide-area, continuous, and environmentally robust macroscopic situational awareness and dynamic change monitoring of large-scale power transmission line corridors. This invention integrates high-resolution satellite remote sensing data acquisition, multi-level intelligent information extraction, precise power transmission line positioning, and macroscopic risk dynamic assessment technologies to construct an automated and intelligent data processing framework, thereby improving the efficiency, coverage, and risk early warning capabilities of power transmission line inspection.

[0006] This invention provides a data processing system for power transmission line inspection based on satellite imagery, comprising: a satellite remote sensing data preprocessing module, an automatic identification and labeling module for risk points along the power transmission line corridor, a module for principal axis reconstruction of power transmission line pixels based on direction field aggregation stability domain judgment, and a space-time database and risk assessment module for power transmission lines. These modules work collaboratively to achieve comprehensive monitoring and risk analysis of the power transmission line and its surrounding environment.

[0007] The satellite remote sensing data preprocessing module is designed to process the received raw satellite data into standardized remote sensing image products. The preprocessing process specifically includes format parsing, decompression, data inspection, cataloging, radiometric correction, geometric correction, sensor calibration product production, image thematic information enhancement, image fusion, image mosaicking, image cropping, and data projection and format conversion.

[0008] Specifically, the radiation correction module is responsible for eliminating the influence of sensor errors, atmospheric scattering and absorption effects, and solar elevation angle on image radiation values, thereby obtaining the true spectral reflectance information of ground objects and ensuring the spectral consistency and comparability of image data.

[0009] The geometric correction module specifically establishes a geometric correction model by selecting ground control points (GCPs) evenly distributed throughout the image, and then resampling the image to correct geometric distortions, thereby ensuring a unified spatial reference with historical data. The relative mean square error of the planar position is precisely controlled within 2 pixels, and in complex mountainous terrain conditions, the mean square error of the planar position is controlled within 4 pixels to meet the requirements of accurate georegistration.

[0010] The orthorectification module specifically employs a combination of the RPC (Rational Function Model) file inherent in the high-resolution imagery and a Digital Elevation Model (DEM) to perform high-precision positioning correction on the image data. The RPC file essentially fuses the sensor's orbital parameters and various other physical parameters, and calculates a transformation matrix using ground control point elements to describe the mapping relationship between pixel coordinates and geographic coordinates. This correction method effectively eliminates geometric distortions caused by terrain undulations, generating orthorectified images with accurate geographic locations.

[0011] The image thematic information enhancement module specifically employs a spectral band fusion enhancement method, such as incorporating near-infrared spectral information to improve the correlation between the red, green, and blue bands of the true-color remote sensing image. This enhancement method effectively increases the amount of ground feature information in each band of the true-color image, reduces information redundancy, and optimizes the data structure. This significantly enhances the characteristic differences of various target ground features such as vegetation, buildings, water bodies, and exposed surfaces in the true-color remote sensing image, greatly facilitating subsequent automatic remote sensing identification and human-computer interactive analysis.

[0012] The image fusion module specifically employs a multi-scale decomposition tool to decompose the input images to be fused at multiple scales, thereby obtaining their respective multi-scale transform coefficients. Subsequently, these multi-scale transform coefficients are recombinated according to preset fusion rules to generate new multi-scale transform coefficients. Finally, a multi-scale inverse transform is performed to reconstruct the final fused image. This process can more effectively extract spatial texture and spectral information from the image, achieving a complementary advantage between spectral information and spatial details.

[0013] The image mosaicking and cropping module, specifically during image mosaicking, determines a reference image as the basis for the output mosaicked image. This reference image determines parameters such as contrast matching, pixel size, and data type of the final mosaicked image. To ensure a good mosaicking effect, the two or more images to be mosaicked are required to have the same or similar imaging time to maintain a high degree of color consistency and avoid obvious stitching artifacts. Subsequently, the mosaicked image is precisely cropped according to preset region boundaries.

[0014] The data projection and format conversion module, based on the specific needs of remote sensing data applications, performs customized projection settings on the image data, such as converting the image coordinate system to the CGCS2000 coordinate system. The image data itself is specified as .tif format, and by using format conversion tools in professional geographic information systems such as PCI or ArcGIS, it is converted to the required data format, such as Img or EcW format, to meet the compatibility requirements of different application platforms.

[0015] The satellite remote sensing data preprocessing module sets quality requirements for the processed image data. Specifically, these include: the spatial resolution of the satellite panchromatic image is less than 1 meter; the spatial resolution of the multispectral image (including red, green, blue, and near-infrared bands) is less than 4 meters; the cloud cover of the original image is less than 20%, and the cloud layer must not cover important features in the area; the side tilt angle is less than ±15° to ensure that the spatial resolution of the acquired original image meets the usage requirements; the data acquisition range completely covers the required area, and the distance is calculated according to the boundary extension; the image should be registered and corrected, and meet the requirements of relevant national technical standards; the coordinate system of the image output data adopts CGCS2000, and provides data processing functions to process and adjust the coordinate system, data format, and map segmentation standards of the output data to meet user requirements; the output data must not contain any errors, and must provide data that conforms to... The image data must meet the requirements of the library management application system. The background color value of the output data is set to (0, 0, 0), and no other values ​​are allowed within the valid range of the output data. If a pixel with a value of (0, 0, 0) exists within the valid range, it must be uniformly replaced with (1, 1, 1). The orthophoto data output uses TIFF, IMG, and ECW formats with geolocation information. The image colors are fused to produce true-color images with clear edges and no ghosting or blurring. The image has rich layers, clear texture details, and no obvious noise, spots, bad lines, or seams. The overall brightness of the image is moderate, the color contrast is appropriate, and the brightness value distribution range is as suitable as possible for the computer screen's resolution range. The image needs to undergo color homogenization processing to ensure a consistent overall tone, no obvious color difference at the seams, and a natural transition.

[0016] The automatic identification and labeling module for risk points in transmission line corridors is designed to establish an expanded automatic identification target model library from the perspective of risk factor monitoring, and to construct a sample library and model library of potential major safety hazards in transmission line corridors. This allows for the use of corresponding refined intelligent extraction models and automatic identification algorithms to identify, extract, and label different target groups.

[0017] Specifically, the automated risk point extraction employs deep learning-based information extraction technology. This technology fully integrates spatial texture information with spectral information and has been widely used in high-resolution image information extraction due to its advantages such as high detection and segmentation accuracy and strong robustness. The technology mainly includes two categories: target detection and semantic segmentation. The module adopts corresponding extraction methods for different target features. Specifically, the target detection algorithm is suitable for extracting ground features with regular boundaries, such as greenhouses, trees, and corrugated steel sheets; the semantic segmentation algorithm can detect and segment fine texture structures and is suitable for extracting targets with large areas and irregular boundaries, such as water bodies, bare soil, and dust control nets.

[0018] The target detection algorithm, for example, employs the YOLOv8 network model, whose network structure has undergone multiple improvements and optimizations, enhancing target detection accuracy while maintaining high-speed performance. The YOLOv8 network model specifically comprises the following key components: the backbone network adopts the CSPDarknet53 architecture, which extracts more feature information by cross-connecting information from the forward and backward channels, thereby enhancing the network's understanding of image content. The YOLOv8 network model also integrates a Feature Pyramid Network (FPN) and a Path Aggregation Network (PANet). The FPN facilitates target detection on feature maps of different resolutions, enabling the detection of targets of varying sizes, and fuses feature maps of different resolutions through upsampling and downsampling operations. The PANet neck network effectively fuses feature maps at different levels, allowing the network to better understand the contextual information of the target, thus improving detection accuracy. Furthermore, the YOLOv8 network model utilizes a multi-scale detection head, operating in parallel on feature maps of different scales to simultaneously detect targets of varying sizes, thereby improving the model's robustness to changes in target size. The YOLOv8 network model also employs an ensemble learning approach, combining multiple different detection models to improve detection stability and accuracy. Specifically, it enhances overall performance by combining the prediction results of multiple models.

[0019] The semantic segmentation algorithm, for example, employs the DeepLabV3+ network model. This algorithm, while preserving dilated convolution and spatial pyramid pooling layers, fuses multi-scale information through an encoder-decoder structure. In the encoder stage, the DeepLabV3+ network model uses an Xception network to extract features from the input image, and then fuses these features using a dilated spatial pyramid pooling (ASPP) module to avoid information loss. The ASPP module is a multi-scale pyramid feature extraction module containing multiple dilated convolution dilation rates. In the decoder stage, the DeepLabV3+ network model fuses the low-level and high-level features output from the encoder and uses bilinear interpolation upsampling to obtain a high-precision segmentation result, thereby improving the network's segmentation accuracy. To further enhance the algorithm's accuracy, the DeepLabV3+ network model introduces a Squeeze-and-Excitation (SE) attention mechanism module in the feature extraction part. The specific structure of the SE attention mechanism module is as follows: The input is a C×H×W feature map, where C is the number of channels, and H and W are the dimensions of the feature map. First, a global pooling operation is performed on the feature map to obtain a C×1×1 feature map with a global receptive field. Then, the feature map passes through two fully connected layers (FC). The first FC layer is followed by a ReLU activation function, and the second FC layer is followed by a Sigmoid activation function, resulting in another C×1×1 feature map representing the importance weight of each feature channel. Finally, the original C×H×W feature map is multiplied element-wise with the C×1×1 channel weight feature map (scale operation) to obtain a C×H×W feature map with different channel importance, i.e., a feature map with the attention mechanism. By using the SE attention mechanism to weightedly fuse feature maps of different scales, the segmentation accuracy of the algorithm is further improved.

[0020] The transmission line pixel principal axis reconstruction module based on orientation field aggregation stability domain judgment is innovative in that it constructs a full-image orientation field map in satellite imagery and identifies stable orientation concentration domains based on the degree of orientation aggregation, thereby accurately reconstructing the principal axis direction of the transmission line and achieving precise positioning and linear reconstruction of the transmission line. Unlike traditional methods that rely solely on edge features or local texture alignment, this module emphasizes the extraction of global directional trends and stability domain analysis.

[0021] The transmission line pixel principal axis restoration module based on orientation field aggregation stability domain judgment specifically includes: orientation field map construction unit, stability domain identification unit, and principal axis extraction unit.

[0022] The orientation field map construction unit specifically calculates a principal orientation angle for each pixel in the preprocessed satellite image based on its local gradient information or texture orientation features, and stores the principal orientation angle as the orientation vector of the corresponding pixel, thereby generating a pixel orientation field vector map covering the entire image. Each orientation vector in the map represents the dominant spatial orientation at that pixel.

[0023] The stable region identification unit specifically divides the pixel orientation field vector map into multiple local window regions of a preset size, and within each local window region, calculates the distribution density and standard deviation of orientation differences for all pixel orientation vectors. The identification unit marks regions with high orientation vector distribution density and low standard deviation of orientation differences as stable aggregation regions, and determines these stable aggregation regions to be potential transmission line routing areas. This process effectively filters out random noise and irrelevant texture interference in the image, highlighting the inherent orientation consistency characteristics of the transmission line.

[0024] The principal axis extraction unit specifically performs principal axis extraction on the direction vectors within the stable aggregation region identified by the stable region identification unit. The principal axis extraction operation determines one or more principal direction line segments that best represent these direction vectors by fitting the direction vectors within the stable aggregation region, and uses these principal direction line segments as the baseline for the precise location path of the transmission line. This method can accurately recover the principal axis structure of the transmission line even in the presence of a large amount of interfering texture, ensuring the continuity and directional stability of the extracted line direction, thereby significantly improving the resolution accuracy of transmission lines in high-resolution imagery.

[0025] The transmission line spatial-temporal database and risk assessment module is designed to integrate, store, manage, analyze, and assess the risks of all processed data. Specifically, the module includes a spatial-temporal database management unit, a geographic information system (GIS) fusion and visualization unit, and a multi-dimensional risk assessment and early warning unit.

[0026] The spatial-temporal database management unit is responsible for uniformly storing the standardized remote sensing image products output by the satellite remote sensing data preprocessing module, the identification results of various risk elements (including transmission line pixels, towers, vegetation intrusion areas, construction activity areas, water bodies, bare soil, floating objects, etc.) output by the automatic risk point identification and annotation module for transmission line corridors, the transmission line main axis path information output by the transmission line pixel main axis reconstruction module based on the direction field aggregation stability domain judgment, and the extracted transmission line parameter information. The database adopts a distributed storage architecture to support rapid access and efficient management of large-scale spatial-temporal data. The database also includes historical image data, risk element distribution data at different time points, and detailed attribute data of transmission line infrastructure.

[0027] The Geographic Information System (GIS) fusion and visualization unit efficiently integrates all spatial data and geospatial data (such as topography, administrative divisions, land use types, and hydrological distribution) from the spatial-temporal database management unit. The GIS unit can overlay and visualize pre-processed satellite imagery, identified risk points, reconstructed transmission line main axis paths, and other relevant geographic information within a unified geographic coordinate system, forming an intuitive and comprehensive transmission line inspection status map. The GIS unit provides rich spatial analysis functions, such as buffer analysis, overlay analysis, and path analysis, to support deeper geospatial data mining.

[0028] The core function of the multi-dimensional risk assessment and early warning unit is to comprehensively and dynamically assess potential risks within the transmission line corridor based on the data stored in the space-time database, and generate corresponding early warning information. Specifically, the unit implements risk assessment in the following ways: Based on the vegetation areas (including tree species, height, and growth trend) identified by the automatic risk point identification and labeling module of the transmission line corridor, and combined with the voltage level of the transmission line, safety distance standards, and historical vegetation intrusion event data, a vegetation intrusion risk assessment model is established. The model dynamically calculates the minimum distance between vegetation and the transmission line, the degree of erosion of the safety distance by the vegetation growth rate, and generates risk warnings of high, medium, and low levels based on the assessment results.

[0029] Based on the identified construction activity areas (including large machinery, temporary buildings, earthwork accumulation, etc.), and considering the type, scale, duration, and relative position of the construction activities to the power transmission lines, the model assesses the potential risks of collisions, excavation, and ground settlement that the construction activities may cause to the power transmission lines. The model can provide high-level risk warnings for construction activities that are too close to the power transmission lines or pose a potential threat to line safety.

[0030] By combining surface deformation information (such as cracks and landslide signs) interpreted from satellite imagery, topographic slope, aspect, hydrological conditions, and historical geological hazard data analyzed from DEM data, the geological stability along the transmission line is assessed. The model generates geological hazard risk warnings for transmission line sections located in potential landslide, debris flow, and ground subsidence areas.

[0031] Based on the identified floating objects (such as corrugated steel sheets, plastic greenhouse debris, etc.) and foreign objects, combined with their size, type, location, and relative distance to power transmission lines, the model assesses the potential risks to power transmission lines, such as snagging and short circuits. The model immediately generates early warning information for floating objects or foreign objects with high risks.

[0032] Based on the transmission line pixel principal axis reconstruction method, the main axis information of the line is extracted and combined with its spatial position at different time points. By comparing it with historical data, potential problems such as minute displacement, sag changes, and abnormal vibrations of the line can be monitored. The module can also integrate detailed data of the line body obtained by other platforms (such as drones and ground sensors) to further refine the assessment of conductor wear, insulator damage, tower tilt, etc.

[0033] The multi-dimensional risk assessment and early warning unit comprehensively considers the results of the above-mentioned risk assessments and uses algorithms such as multi-factor weighted summation, fuzzy comprehensive evaluation, or machine learning to form a comprehensive risk level judgment for the entire or local area of ​​the transmission line. When the assessment result exceeds a preset threshold, the unit automatically triggers a multi-level early warning mechanism, including but not limited to generating a visual risk map, sending SMS / email notifications to maintenance personnel, and generating a detailed risk report and handling suggestions. The risk assessment model can be dynamically updated and corrected based on new satellite imagery data, realizing continuous monitoring and real-time early warning of changes in the transmission line corridor environment.

[0034] The satellite remote sensing data preprocessing module is used to process the received raw satellite data into standardized remote sensing image products, including: Format parsing and decompression are used to ensure compatibility and handle standard or proprietary data formats from different satellite data providers; Data inspection and cataloging are used to automatically verify the integrity, validity, and accuracy of data and incorporate them into the catalog management system. Radiometric correction is used to eliminate the influence of sensor characteristics, atmospheric effects, and solar elevation angle on image pixel radiance values, thereby obtaining true spectral reflectance information of ground objects and ensuring the spectral consistency and comparability of image data. Geometric correction processing is used to establish a geometric correction model by selecting ground control points that are evenly distributed throughout the entire image, and then resampling the image to complete the geometric deformation correction of the image, thereby ensuring a unified spatial reference with historical data. Image thematic information enhancement is used to improve the amount of ground feature information in each band of true-color remote sensing images by employing spectral band fusion enhancement methods, and to enhance the feature differences of target ground features such as vegetation, buildings, and water bodies. Image fusion uses a multi-scale decomposition tool to decompose the input images to be fused into multiple scales, and then recombines them according to preset fusion rules to achieve complementary advantages of spectral information and spatial details. Image mosaicking and cropping are used to seamlessly stitch multiple images by determining a reference image as the basis for the output mosaic image, and to precisely crop the mosaicked image according to the preset region boundaries. Data projection and format conversion are used to customize the projection settings of image data and convert the data format according to the application requirements of remote sensing data, so as to meet the compatibility requirements of different application platforms.

[0035] The processing method of the satellite imagery-based power transmission line inspection data processing system includes: Standardize the received raw satellite data; Automatically identify and label various risk factors within the transmission line corridor from the standardized remote sensing image products, including: fallen trees, line icing, and abnormal plant growth; Restore the main axis direction of the transmission line; The standardized remote sensing image products, the risk factor identification results, and the transmission line main axis direction information are integrated, stored, managed, analyzed, and risk-assessed.

[0036] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to describe the method.

[0037] The satellite imagery-based power transmission line inspection data processing system provided by this invention achieves the following beneficial effects through the aforementioned systematic technical solution: 1. This invention utilizes high-resolution satellite imagery to achieve comprehensive, continuous, and dynamic monitoring of cross-regional, long-distance power transmission line channels. The coverage area of ​​a single image far exceeds that of drones, significantly improving the coverage and efficiency of inspections and effectively solving the limitations of traditional and drone inspections in dealing with large-scale power grids.

[0038] 2. This invention ensures the quality and consistency of satellite imagery data through a systematic preprocessing process; and intelligently identifies and finely segments vegetation, construction, floating objects and other ground features through deep learning (YOLOv8, DeepLabV3+ with SE). At the same time, it uses directional field aggregation technology to accurately restore the power transmission line itself, enabling macroscopic early warning capabilities for microscopic changes in the channel environment and the line itself, overcoming the problem of fragmented UAV data that is difficult to integrate.

[0039] 3. Satellite remote sensing technology is far less affected by geographical environment and adverse weather conditions than drones, ensuring the continuity and timeliness of inspections. Furthermore, compared to the high equipment, manpower, and operating costs required for large-scale deployment of drones for corridor inspections, this invention has a significant advantage in terms of economy.

[0040] 4. This invention establishes a multi-dimensional risk assessment model, which combines the identified risk factors with the condition of the transmission line itself to quantitatively assess and predict the trends of potential hazards in the transmission line. It can also generate tiered early warning information in a timely manner, providing strong data support for transmission line management departments to discover potential risks early, optimize resource allocation, and formulate preventive maintenance strategies.

[0041] 5. Through geometric correction, orthorectification and GIS integration, this invention ensures the spatial accuracy and consistency of all data and realizes two-dimensional and three-dimensional visualization on an interactive GIS platform, which greatly enhances the management personnel's intuitive perception of the status of power transmission line channels and improves the scientific nature and efficiency of decision-making.

[0042] 6. This invention adopts a distributed data storage and management architecture, which ensures the stability and scalability of massive remote sensing data processing and storage; at the same time, the modular system design facilitates the introduction of more advanced algorithm models or the integration of new data sources in the future. Attached Figure Description

[0043] Figure 1 System block diagram of the power transmission line inspection data processing system based on satellite imagery of this invention; Figure 2 : A schematic diagram of the internal structure of the satellite remote sensing data preprocessing module described in this invention; Figure 3 : A schematic diagram of the internal structure of the automatic identification and marking module for risk points in transmission line channels described in this invention; Figure 4 : A schematic diagram of the internal structure of the transmission line pixel spindle restoration module based on the direction field aggregation stability domain judgment described in this invention; Figure 5 : A schematic diagram of the internal structure of the transmission line space-time database and risk assessment module described in this invention; Figure 6 This invention presents a schematic diagram of the workflow of a power transmission line inspection data system based on satellite imagery. Detailed Implementation

[0044] This invention provides a satellite imagery-based data processing system for power transmission line inspection, aiming to address the challenges of traditional inspection methods when dealing with large-scale power transmission lines, including low efficiency, limited coverage, and difficulty in continuous, environmentally robust macroscopic situational awareness and dynamic change monitoring. This system integrates advanced technologies such as high-resolution satellite remote sensing data acquisition, multi-level intelligent information extraction, precise power transmission line positioning, and dynamic macroscopic risk assessment to construct an automated and intelligent data processing framework, thereby significantly improving the intelligence level, efficiency, coverage, and risk early warning capabilities of power transmission line inspection.

[0045] The system consists of core components such as a satellite remote sensing data preprocessing module, an automatic identification and labeling module for risk points along transmission line corridors, a module for principal axis reconstruction of transmission line pixels based on directional field aggregation stability domain judgment, and a transmission line spatial-temporal database and risk assessment module. These modules work collaboratively to form a tightly coupled intelligent processing chain, jointly completing comprehensive and detailed monitoring and risk analysis of transmission lines and their surrounding environment.

[0046] In one specific embodiment, the function of the satellite remote sensing data preprocessing module is to transform raw remote sensing data acquired from commercial or scientific satellite platforms (e.g., Gaofen series satellites, Pleiades, WorldView, etc.) into standardized remote sensing image products that meet the requirements of subsequent system processing and analysis. The comprehensiveness and refinement of this preprocessing process directly determine the accuracy and reliability of subsequent information extraction and risk assessment.

[0047] Specifically, the preprocessing process first involves format parsing and decompression to ensure the system is compatible with and can handle various satellite data providers' standard or proprietary data formats (e.g., GeoTIFF, HDF5, DIMAP, etc.). This is followed by data inspection and cataloging, automatically verifying the data's integrity, validity, and metadata accuracy, and incorporating it into the system's unified catalog management system for easy retrieval and traceability.

[0048] Furthermore, the module performs radiometric correction, the core objective of which is to eliminate the influence of sensor characteristics, atmospheric effects (including atmospheric scattering and absorption), and factors such as solar altitude angle and terrain undulation on the image pixel radiance values. This process converts the raw digital quantization (DN) values ​​of the image into the true spectral reflectance or radiance values ​​of ground features at the top of the atmosphere or the surface by applying physical models (e.g., MODTRAN, 6S models) or empirical models. For example, sensor correction performs a linear transformation based on the sensor's own gain and bias parameters; atmospheric correction utilizes the known spectral characteristics of specific ground features such as water bodies and dark vegetation in the image, or combines external atmospheric parameter measurement data, to remove the atmospheric distortion effect on the spectrum through inversion algorithms. Through radiometric correction, it is possible to ensure the consistency and comparability of image data acquired at different times and by different sensors in the spectral dimension, providing a solid foundation for subsequent identification of risk factors such as vegetation and water bodies based on spectral features. In a typical scenario, through rigorous radiometric correction, the relative radiometric accuracy of the image can reach within 5%, and the absolute radiometric accuracy can reach within 10%.

[0049] Next, geometric correction is performed to correct geometric distortions caused by factors such as platform attitude changes, Earth curvature, and nonlinear scanning of sensors during image acquisition. Specifically, the geometric correction module establishes a geometric correction model by selecting ground control points (GCPs) uniformly distributed throughout the entire image. These GCPs are typically obtained from high-precision topographic maps, GPS measured points, or even higher-precision reference images, and their positional accuracy in the reference coordinate system is crucial. After the model is established, the image is resampled. Common resampling methods include nearest neighbor interpolation, bilinear interpolation, and cubic convolution interpolation. Cubic convolution interpolation effectively reduces the blurring effect caused by resampling while maintaining the geometric accuracy of the image. This process ensures that the corrected image and historical data have a unified spatial reference, enabling accurate overlay analysis of data from different periods. In flat areas, the relative error of the planar position is precisely controlled within 2 pixels; while in complex terrain conditions such as mountainous areas, by increasing the number of GCPs and optimizing their distribution, the relative error of the planar position can be controlled within 4 pixels to meet the requirements of accurate georegistration. For example, for satellite imagery with a resolution of 0.5 meters, the correction accuracy can be within 1 meter in flat areas and within 2 meters in mountainous areas.

[0050] As a preferred embodiment of the present invention, the module further includes an orthorectification module. This module, based on geometric correction, performs more refined high-precision positioning correction on the image data by combining the RPC (Rational Polynomial Coefficients) file inherent in the high-resolution image with a high-precision digital elevation model (DEM). The RPC file essentially fuses the sensor's orbital parameters, attitude parameters, and various other physical parameters, and calculates a transformation matrix using a small number of ground control point elements to describe the complex mapping relationship between pixel coordinates and geographic coordinates. This correction method effectively eliminates image geometric distortion (i.e., perspective shrinkage and stretching) caused by terrain undulations, generating orthorectified images with true geographic locations. The orthorectified image can be directly used for precise measurement and map production, with a planar positioning accuracy typically reaching one pixel or even sub-pixel level. For example, for an image with a 0.5-meter resolution, its absolute planar positioning accuracy can be controlled within 0.5 meters.

[0051] After spatial and spectral characteristics are corrected, the module performs image thematic information enhancement. This process aims to improve the discernibility of specific land features to facilitate subsequent automated identification. Specifically, spectral band fusion enhancement methods are employed, such as incorporating near-infrared spectral information to improve the correlation between the red, green, and blue bands of the true-color remote sensing image, thereby enhancing vegetation information. Commonly used methods include IHS (Intensity-Hue-Saturation) transformation, Principal Component Analysis (PCA), and weighted sum or Gram-Schmidt spectral sharpening. Through such enhancements, the method can effectively increase the amount of land feature information in each band of the true-color image, reduce information redundancy, optimize data structure, and thus significantly enhance the characteristic differences of various target land features such as vegetation, buildings, water bodies, and bare surfaces in the true-color remote sensing image, greatly facilitating subsequent automated remote sensing identification and human-computer interactive analysis. For example, by fusing near-infrared bands, the Normalized Difference Vegetation Index (NDVI) of vegetation will be highlighted, increasing its distinguishability from non-vegetated areas by at least 20%.

[0052] To achieve a complementary advantage between high spatial resolution and high spectral resolution, the module also includes an image fusion module. This module specifically employs a multi-scale decomposition tool to decompose the input images to be fused into multi-scale values, thereby obtaining their respective multi-scale transform coefficients. For example, wavelet transform, Laplacian pyramid, or Contourlet transform can be used. Subsequently, the obtained multi-scale transform coefficients are recombinated according to preset fusion rules (e.g., based on local variance, mutual information, energy ratio, etc.) to generate new multi-scale transform coefficients. Finally, an inverse multi-scale transform is performed to reconstruct the final fused image. This process effectively integrates the spatial detail information of high-resolution panchromatic images with the spectral information of low-resolution multispectral images, achieving a complementary advantage between spectral information and spatial detail, producing image products with both high spatial and spectral resolution. For example, fusing a 1-meter resolution panchromatic image with a 4-meter resolution multispectral image can yield a color image with a 1-meter spatial resolution, and the spectral distortion (e.g., SAM index) is controlled within 0.05.

[0053] After processing multiple images, the module performs image mosaicking and cropping to generate a seamless image covering the entire target area. Specifically, during image mosaicking, a reference image is chosen as the basis for the output mosaic image. The reference image determines parameters such as contrast matching, pixel size, and data type of the final mosaic image. To ensure mosaicking quality, the two or more images to be mosaicked are selected with similar or identical imaging times (e.g., the same quarter or adjacent months) to maintain high tonal consistency and avoid obvious stitching artifacts caused by differences in lighting or season. Mosaic algorithms typically employ automatic overlapping area finding, feathering, or color histogram matching to smooth seams and eliminate brightness or tonal differences. Subsequently, the mosaicked image is precisely cropped according to the preset transmission line corridor area boundary (usually defined by GIS vector data), retaining only image data within the target area to reduce data redundancy.

[0054] Finally, the module performs data projection and format conversion to meet the compatibility requirements of different application platforms and the needs of geospatial analysis. Specifically, based on the specific needs of the remote sensing data application, custom projection settings are applied to the image data, such as converting the image coordinate system to CGCS2000 (China Geodetic Coordinate System 2000) or UTM projection. This process involves complex coordinate system conversion algorithms to ensure accurate geographic location correspondence. The image data format itself is specified as a highly complete .tif format (GeoTIFF, containing georeferenced information), and format conversion tools in professional geographic information systems such as PCI Geomatica, ArcGIS, or GDAL are used to convert it to the required data format according to application requirements, such as Img or EcW format (for fast browsing of large amounts of data), to meet the compatibility requirements of different application platforms, such as WebGIS, desktop GIS, or mobile applications.

[0055] To ensure data quality, the satellite remote sensing data preprocessing module sets strict quality requirements for the processed image data. Specifically, these include: the spatial resolution of satellite panchromatic imagery must be less than 1 meter (e.g., 0.5 meters or better), and the spatial resolution of multispectral imagery (including four or more spectral bands such as red, green, blue, and near-infrared) must be less than 4 meters (e.g., 2 meters). The cloud cover in the raw image must be less than 20%, and clouds must not cover important features such as power transmission line corridors or towers; the side tilt angle must be less than ±15° to ensure the spatial resolution of the acquired raw image meets usage requirements and to reduce geometric distortion caused by terrain; the data acquisition range must completely cover the required area, and the data must be extended beyond the boundary by a distance (e.g., 500 meters) to provide sufficient surrounding information; the imagery should undergo precise registration and correction and comply with relevant national technical standards (e.g., DL / T 597-2016). (Technical Specifications for Remote Sensing Monitoring of Transmission Lines); The image data coordinate system adopts CGCS2000 and provides data processing functions to process and adjust the coordinate system, data format, and grading standards of the data to meet user requirements. No data gaps or blank areas are allowed in the data, and tile data (tile maps) that meet the requirements of database management application systems can be provided. The background color value of the data is strictly set to (0, 0, 0) to indicate transparent or invalid areas. No other background values ​​are allowed within the valid range of the data. If a pixel with (0, 0, 0) exists within the valid range (e.g., due to missing data), it must be uniformly replaced with (1, 0, 0). Non-zero values ​​(e.g., 1, 1) are used to ensure data consistency and integrity; orthophoto data is presented in TIFF, IMG, and ECW formats with geolocation information for easy reading on different platforms; the images are true-color after color fusion, with clear edges and no ghosting or blurring; the images have rich tonal range, clear texture details, and no obvious noise, spots, bad lines, or seams; the overall brightness is moderate, the color contrast is appropriate, and the brightness distribution range is as suitable as possible for the computer screen's resolution range; the images undergo color homogenization to ensure consistent overall tone, no obvious color difference at seams, and natural transitions, guaranteeing visual quality and analytical accuracy.

[0056] In one specific embodiment, the automatic risk point identification and labeling module for power transmission line corridors has the core function of establishing an expanded automatic identification target model library from the perspective of risk factor monitoring, and constructing a sample library and model library of potential major safety hazards in power transmission line corridors. This allows for the use of corresponding refined intelligent extraction models and automatic identification algorithms for identification, extraction, and labeling of different target groups. This module can efficiently and accurately identify various ground features and activities that pose a potential threat to the operation of power transmission lines from massive amounts of remote sensing image data.

[0057] Specifically, the automated risk point extraction employs deep learning-based information extraction technology. This technology fully utilizes image spectral information and deeply integrates spatial texture information. Due to its advantages such as high detection and segmentation accuracy and strong robustness in complex scenes, it has been widely used in high-resolution remote sensing image information extraction. The technology mainly includes two categories: target detection and semantic segmentation. For different target features, the module adopts corresponding extraction methods. The target detection algorithm is suitable for extracting ground objects with clear boundaries, regular shapes, and relatively discrete numbers, such as greenhouses, isolated trees, corrugated steel roofs, construction machinery, and large vehicles. The semantic segmentation algorithm can detect and segment subtle texture structures and is suitable for extracting large targets with irregular boundaries, such as large bodies of water, bare soil areas, areas covered by dust nets, and large vegetation communities.

[0058] In one specific embodiment, the object detection algorithm employs, for example, the YOLOv8 network model. YOLOv8 is one of the latest versions of the YouOnly Look Once detector series, and its network structure has undergone multiple improvements and optimizations, significantly enhancing object detection accuracy while maintaining high-speed inference performance. The YOLOv8 network model consists of the following key components: the backbone network adopts the CSPDarknet53 architecture. This architecture effectively reduces computational and memory consumption by cross-connecting information from the forward and backward channels, i.e., introducing cross-stage partial connections, while extracting richer feature information to enhance the network's understanding of image content and feature representation capabilities. The YOLOv8 network model also integrates a Feature Pyramid Network (FPN) and a Path Aggregation Network (PANet), which together constitute the network's neck. FPN, through a top-down path, transmits high-level semantic information to low-level feature maps, facilitating target detection on feature maps of different resolutions and enabling the detection of targets of varying sizes. PANet, on the other hand, uses a bottom-up path to transmit low-level localization information to high-level feature maps and fuses feature maps of different resolutions through upsampling and downsampling operations. This allows the network to better understand the contextual information and spatial details of the target, significantly improving detection accuracy, especially when handling targets with large size differences. The YOLOv8 network model further utilizes a multi-scale detection head, working in parallel on feature maps of different scales to simultaneously detect targets of varying sizes, thereby improving the model's robustness to changes in target size. For example, a high-resolution feature map detection head is used for small targets (such as distant construction vehicles), while a low-resolution feature map detection head is used for large targets (such as large factories). Furthermore, the YOLOv8 network model employs ensemble learning methods, such as Weighted Box Fusion (WBF) or model averaging, to combine multiple detection models with different training epochs, hyperparameters, or initial weights. This improves the stability and accuracy of detection. Specifically, it enhances overall performance by combining the predictions (e.g., bounding boxes, confidence scores) of multiple models, effectively reducing the randomness of individual model predictions. In practical deployments, the YOLOv8 model is typically trained on custom datasets containing hundreds of thousands of labeled images. These datasets cover common target categories within power transmission line corridors, including vegetation (such as trees and shrubs), buildings (such as greenhouses and houses), construction machinery (such as excavators and tower cranes), and corrugated steel sheets.The training process typically employs the AdamW optimizer and a learning rate scheduler (such as Cosine Annealing), and undergoes 200-500 epochs of iteration. The batch size is set to 64 or 128, aiming to achieve a mean average precision (mAP) exceeding 0.75 on the COCO dataset standard and exceeding 0.85 on the transmission line-specific dataset.

[0059] In a specific embodiment, the semantic segmentation algorithm employs, for example, the DeepLabV3+ network model. DeepLabV3+ is a leading model in the field of semantic segmentation. This algorithm, while retaining the advantages of atrous convolution and atrous spatial pyramid pooling (ASPP) layers in multi-scale feature capture, effectively improves segmentation accuracy by fusing multi-scale information through an encoder-decoder structure. In the encoder stage, the DeepLabV3+ network model uses the Xception network as its backbone to extract features from the input image. The Xception network employs depthwise separable convolutions, which significantly reduces the number of model parameters and computational cost compared to traditional convolution operations, while maintaining powerful feature extraction capabilities. Subsequently, the image features are fused using the atrous spatial pyramid pooling (ASPP) module to avoid information loss. The ASPP module is a multi-scale pyramid feature extraction module containing multiple dilated convolution rates (e.g., {1, 6, 12, 18}). It effectively handles the problem of target scale variations in images by expanding the receptive field and capturing contextual information at different scales without increasing parameters or computational cost. In the decoder stage, the DeepLabV3+ network model fuses low-level features (from the Xception backbone network, rich in spatial detail) with high-level features (from the ASPP module, rich in semantic information) from the encoder output. Through bilinear interpolation upsampling, it progressively restores the spatial resolution of the image, ultimately obtaining high-precision segmentation results, thereby improving the network's segmentation accuracy and target boundary recognition ability. To further enhance algorithm accuracy, the DeepLabV3+ network model introduces a Squeeze-and-Excitation (SE) attention mechanism module in the feature extraction part.The specific structure of the SE attention mechanism module is as follows: The input is a C×H×W feature map, where C is the number of channels, and H and W are the dimensions of the feature map. First, a global average pooling operation is performed on the feature map to compress the spatial information of each channel into a global channel descriptor, resulting in a C×1×1 feature map with a global receptive field. Then, the feature map passes through two fully connected layers (FC). The first FC layer is followed by a ReLU activation function for non-linear transformation, and the second FC layer is followed by a Sigmoid activation function, resulting in a C×1×1 feature map that represents the importance weight (between 0 and 1) of each feature channel. Finally, the original C×H×W feature map is multiplied element-wise with the C×1×1 channel weight feature map (scale operation) to obtain a C×H×W feature map with different channel importance, i.e., a feature map with an attention mechanism. By employing a SE attention mechanism to weightedly fuse feature maps at different scales, the algorithm's segmentation accuracy is further improved. This is particularly effective when identifying large, irregular terrain features such as water bodies, bare soil, and dust control nets, as it can more accurately capture their subtle spectral and textural features. The DeepLabV3+ model typically performs well on public datasets such as PASCAL VOC and Cityscapes. On a dedicated power transmission line dataset, trained with pixel-level annotations, its average intersection-over-union (mIoU) can reach over 0.80, and its pixel accuracy can exceed 0.95.

[0060] In one specific embodiment, the transmission line pixel principal axis reconstruction module based on orientation field aggregation stability domain judgment is innovative in that it constructs a pixel orientation field map covering the entire image in the satellite image, and identifies stable orientation concentration domains based on the degree of orientation aggregation, thereby accurately reconstructing the principal axis direction of the transmission line and achieving precise positioning and linear reconstruction of the transmission line. Unlike traditional methods that rely solely on edge features or local texture alignment, this module emphasizes the extraction of global directional trends and stability domain analysis, effectively overcoming the interference of complex background textures and noise in high-resolution images on transmission line identification.

[0061] The transmission line pixel principal axis restoration module based on orientation field aggregation stability domain judgment specifically includes an orientation field map construction unit, a stability domain identification unit, and a principal axis extraction unit.

[0062] The orientation field map construction unit specifically calculates a principal orientation angle for each pixel in the preprocessed satellite image based on its local gradient information or texture orientation features, and stores the principal orientation angle as the orientation vector of the corresponding pixel, thereby generating a pixel orientation field vector map covering the entire image. Specific methods may include: for each pixel, first calculating the grayscale gradient (e.g., using Sobel or Prewitt operators) within its local neighborhood (e.g., a 3x3 or 5x5 window) to obtain the gradient magnitude and gradient direction. The gradient magnitude characterizes the intensity of the image intensity change, while the gradient direction indicates the direction of the change. For slender structures such as power transmission lines, their edges typically exhibit significant gradient directionality. Alternatively, methods based on Gabor filter banks or histogram of oriented gradients (HOG) can be used to extract more robust texture orientation features. The principal orientation angle of each pixel is calculated using the arctangent function to determine the gradient direction and converted into an undirected angle within the range of 0 to 180 degrees to eliminate ambiguity in direction (i.e., 0 degrees and 180 degrees represent the same direction). Each direction vector in the atlas represents the dominant spatial direction and the tendency of its local texture direction at that pixel.

[0063] The stable region identification unit specifically divides the pixel orientation field vector map into multiple local window regions of a preset size (e.g., a 16x16 or 32x32 pixel grid). Within each local window region, the distribution density and standard deviation of orientation vectors for all pixels are statistically analyzed. Specifically, for each window, an orientation histogram can be constructed, dividing the 180-degree range into multiple sectors (e.g., 18 sectors, one every 10 degrees). The orientation distribution density can be measured by calculating the proportion of pixels in the peak sector of the histogram to the total number of pixels in the window. The standard deviation of orientation difference can be calculated using cyclic statistical methods (such as based on Watson or Fisher distribution) to accurately reflect the dispersion of the orientation data (rather than nonlinear data). The identification unit marks regions with high orientation vector distribution density (e.g., more than 70% of the pixel orientations are concentrated within ±15 degrees of a certain main direction) and low standard deviation of orientation difference (e.g., less than 20 degrees) as stable aggregation regions and identifies these stable aggregation regions as potential transmission line routing areas. This process effectively filters out random noise and irrelevant texture interference (such as farmland textures, urban building edges, etc.) in the image because the directional vector distribution of these interference areas is usually more random or irregular, while the inherent slender and continuous characteristics of power transmission lines make them exhibit a high degree of directional consistency in local areas. By setting an appropriate threshold, regions with obvious linear structural features can be accurately selected.

[0064] The principal axis extraction unit specifically performs principal axis extraction on the direction vectors within the stable aggregation regions identified by the stable region identification unit. Within each stable aggregation region, due to the high consistency of pixel orientations, the dominant orientation of the region can be determined using statistical methods (such as calculating the mode or mean of the orientation histogram). Subsequently, robust line fitting algorithms based on Hough Transform, Random Sample Consensus (RANSAC), or least squares are used to fit the pixel points within the stable aggregation region that conform to the dominant orientation into one or more principal orientation line segments that best represent these orientation vectors. Hough Transform can effectively detect straight lines in the image, even if the line segments are incomplete or occluded. The RANSAC algorithm iteratively samples data points and evaluates the fitted model to find the optimal model in the presence of a large number of outliers. To ensure the continuity of the transmission line shape restoration, this unit also implements line segment connection and gap filling strategies, such as connecting adjacent short line segments into longer continuous paths based on the endpoint distance, orientation similarity, and image features of the connecting regions. When necessary, active contour models such as the Snake model or Geodesic Active Contours can be used to further refine the initially extracted line segments, making them better fit the edges of the transmission lines. Finally, the main direction line segment is used as the baseline for the precise positioning path of the transmission lines. It has sub-pixel level positioning accuracy (for example, for 0.5-meter resolution images, the positioning accuracy of the main axis segment can reach 0.25 meters), effectively ensuring the continuity and directional stability of the line direction extraction, thereby significantly improving the resolution accuracy of transmission lines in high-resolution images, especially in the presence of a large amount of interfering textures (such as vegetation and building edges), it can still accurately recover the main axis structure of the transmission lines.

[0065] In one specific embodiment, the transmission line space-time database and risk assessment module is responsible for integrating, storing, managing, analyzing and assessing the risks of all processed data, and is responsible for data aggregation, knowledge refinement and intelligent decision output.

[0066] The module specifically includes a spatial-temporal database management unit, a geographic information system (GIS) fusion and visualization unit, and a multi-dimensional risk assessment and early warning unit.

[0067] The space-time database management unit is responsible for uniformly storing standardized remote sensing image products output by the satellite remote sensing data preprocessing module (e.g., high-resolution images after orthorectification and fusion, stored in tile format), various risk element identification results output by the automatic identification and labeling module for transmission line corridor risk points (including transmission line pixels, towers, vegetation intrusion areas, construction activity areas, water bodies, bare soil, floating objects, etc., usually stored in vector polygon, point or line element form, with accompanying confidence level, category and other attribute information), transmission line main axis path information output by the transmission line pixel main axis restoration module based on direction field aggregation stability domain judgment (stored in vector line elements, including line direction and sag information), and transmission line parameter information extracted from other data sources (such as the State Grid Asset Management System) (such as voltage level, conductor type, tower type, installation year, etc.). The database employs a distributed storage architecture, such as storing large-scale remote sensing imagery based on distributed file systems like Hadoop HDFS or Amazon S3, and combining it with spatial databases like PostgreSQL / PostGIS, MongoDB, or Elasticsearch to manage vector features and metadata, supporting rapid access and efficient management of large-scale spatial-temporal data. The database also includes historical imagery data (for time-series analysis), risk factor distribution data at different time points (for change detection), and detailed attribute data for transmission line infrastructure, ensuring data integrity, consistency, and traceability.

[0068] The Geographic Information System (GIS) Fusion and Visualization Unit efficiently integrates all spatial data from the spatial-temporal database management unit with geospatial background data (such as topographic DEMs, administrative divisions, land use types, hydrological distribution, geological structure maps, etc.). The GIS unit can use pre-processed satellite imagery as a base map to overlay and visualize identified risk points, reconstructed transmission line main axis paths, and other relevant geographic information within a unified geographic coordinate system (e.g., CGCS2000), creating an intuitive and comprehensive transmission line inspection status map. This unit provides rich interactive map operation functions, such as zooming, panning, layer control, and attribute querying. Furthermore, the GIS unit provides powerful spatial analysis functions, such as: buffer zone analysis, used to generate safety warning zones of different widths (e.g., 20 meters on each side of a 220kV line, and 30 meters on each side of a 500kV line) on both sides of the transmission line based on the voltage level and safety standards; overlay analysis, used to overlay identified risk elements (such as vegetated areas and construction activity areas) with safety warning zones to accurately determine which risk elements are located within the safety warning zones and calculate their area or quantity; and path analysis, used to optimize the patrol routes of on-site inspection personnel or analyze the accessibility of risk areas. These spatial analysis functions greatly support deeper geospatial data mining and risk assessment.

[0069] The multi-dimensional risk assessment and early warning unit's core function is to comprehensively and dynamically assess potential risks within transmission line corridors based on various multi-source heterogeneous data stored in a space-time database, and generate corresponding early warning information. The unit specifically achieves risk assessment through the following methods: Based on the vegetation areas (including tree species, height, and growth trends) identified by the automatic risk point identification and labeling module for power transmission line corridors, and combined with the voltage level of the transmission lines, safety distance standards (e.g., minimum safety distance requirements between 220kV lines and the ground and obstacles), and historical vegetation intrusion event data, a vegetation intrusion risk assessment model is established. The model first calculates the minimum horizontal and vertical distances between vegetation and the power transmission lines. Then, it uses a vegetation growth model (e.g., empirical growth curves based on tree species, climate conditions, and seasonal data) to predict future growth trends and dynamically calculate the degree of vegetation erosion of the safety distance. Finally, based on the minimum distance, growth rate, voltage level, and historical risk probability, the risk is classified into high, medium, and low levels (e.g., distance less than 50% of the safety distance is high risk, 50%-80% is medium risk, and 80%-100% is low risk) using a weighted summation or decision tree algorithm, and a risk warning is generated based on the assessment results. For example, for fast-growing trees within 5 meters of the power transmission lines whose height has reached or is close to the safety threshold, the system will immediately issue a high-level warning.

[0070] Based on the identified construction activity areas (including large machinery, temporary buildings, and earthwork accumulation), and considering the type, scale, duration (determined through time-series imagery), as well as their relative location to power transmission lines, the model assesses the potential risks to power transmission lines, such as collisions, excavation, foundation settlement, and high-voltage arc flashovers. The model categorizes risks based on distance thresholds and activity type. For example, areas within 20 meters of the power transmission line centerline containing large equipment such as tower cranes and excavators are assessed as high-risk; areas with large-scale earthwork operations within 20-50 meters are assessed as medium-risk. The model can provide high-level risk warnings for construction activities that are too close to power transmission lines or pose a potential threat to line safety, and includes geographical location and time information.

[0071] By combining surface deformation information interpreted from satellite imagery (such as cracks and landslide signs, obtained through before-and-after image comparison or InSAR technology), topographic slope, aspect, curvature, catchment area, hydrological conditions (river and lake distribution) analyzed from DEM data, and historical geological hazard data, the geological stability along the transmission line is assessed. The model uses multi-factor comprehensive evaluation methods (such as analytic hierarchy process, expert scoring, or logistic regression models) to calculate the probability and potential impact of landslides, debris flows, ground subsidence, and collapses in the areas traversed by the transmission line, classifying the risk into four levels: extremely high, high, medium, and low. The model generates geological hazard risk warnings for transmission line sections located in high-risk areas such as potential landslides, debris flows, and ground subsidence, and indicates the specific impact range. For example, when the slope below the transmission line is greater than 25 degrees and there are obvious surface cracks, a high-level landslide risk warning will be triggered.

[0072] Based on the identified floating objects (such as corrugated steel sheets, remnants of plastic greenhouses, kites, etc.) and foreign objects, the system assesses the potential risks to power lines, including snagging, short circuits, and abrasion, considering their size, type (e.g., conductivity), location, and relative distance. The model performs risk assessment by setting critical distance thresholds and object attributes. For example, for corrugated metal sheets larger than 0.5 square meters within 10 meters of a power line, the system immediately issues the highest-level warning, recommending emergency handling; for foreign objects slightly further away but potentially affected by strong winds, a medium-level warning is issued.

[0073] Based on the transmission line pixel principal axis reconstruction method, the module extracts the line principal axis information and combines it with its spatial location at different time points. By comparing it with historical data, it monitors potential problems such as minute displacements, sag changes (e.g., deviations from the theoretical sag curve), and abnormal vibrations (through indirect features such as tower tilt and support displacement). For example, by comparing the vertical height difference between the current line principal axis and the previous principal axis, if the sag in a certain section increases by more than 0.5 meters, it may indicate abnormal line tension. The module can integrate detailed line body data (such as conductor wear images, insulator infrared thermography data, and tower tilt monitoring data) obtained from other platforms (such as drones and ground sensors) to further refine the assessment of conductor wear, insulator damage, and tower tilt, generating a more comprehensive line body health status report.

[0074] The multi-dimensional risk assessment and early warning unit comprehensively considers the results of the above-mentioned risk assessments and uses algorithms such as multi-factor weighted summation, fuzzy comprehensive evaluation, or machine learning (e.g., classifiers based on random forests, support vector machines, or neural networks) to form a comprehensive risk level judgment for the entire or local areas of the transmission line. This comprehensive assessment model assigns an overall risk score and corresponding risk level to each assessment area. When the assessment results exceed a preset threshold, the unit automatically triggers a multi-level early warning mechanism, including but not limited to generating a visual risk map (highlighting risk areas in the GIS interface and using different colors to represent risk levels), sending SMS / email / App push notifications to maintenance personnel (including risk type, location, level, and preliminary suggestions), and generating a detailed risk report and handling suggestions (e.g., suggesting on-site investigation, vegetation pruning, construction halt, etc.). The risk assessment model can be dynamically updated and corrected based on new satellite imagery data, on-site feedback data, and historical event data, achieving continuous monitoring and real-time early warning of changes in the transmission line corridor environment, thus forming an adaptive and continuously improving risk management closed loop.

[0075] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A data processing system for power transmission line inspection based on satellite imagery, characterized in that, The system includes: The satellite remote sensing data preprocessing module is used to standardize the received raw satellite data; The automatic identification and labeling module for risk points in transmission line corridors is used to automatically identify and label various risk factors in the transmission line corridors from the standardized remote sensing image products, including: fallen trees, ice accumulation on the lines, abnormal plant intrusion, floating objects and foreign objects intrusion. The transmission line pixel main axis restoration module based on the direction field aggregation stability domain is used to restore the main axis direction of the transmission line. The transmission line spatial-temporal database and risk assessment module is used to integrate, store, manage, analyze, and assess the risks of the standardized remote sensing image products, the risk factor identification results, and the transmission line main axis direction information.

2. The data processing system for power transmission line inspection based on satellite imagery according to claim 1, characterized in that, The satellite remote sensing data preprocessing module is used to process the received raw satellite data into standardized remote sensing image products, including: The format parsing and decompression subsystem is used to be compatible with and process standard or proprietary data formats from different satellite data providers; The data inspection and cataloging processing subsystem is used to automatically verify the integrity, validity, and metadata accuracy of data and incorporate them into the catalog management system; The radiometric correction processing subsystem is used to eliminate the influence of sensor characteristics, atmospheric effects, and solar elevation angle on the radiometric values ​​of image pixels, thereby obtaining the true spectral reflectance information of ground objects and ensuring the spectral consistency and comparability of image data. The geometric correction processing subsystem is used to establish a geometric correction model by selecting ground control points that are evenly distributed throughout the entire image, and to perform resampling operations on the image to complete the geometric deformation correction of the image, thereby ensuring that it has a unified spatial reference with historical data. The image thematic information enhancement subsystem is used to enhance the amount of ground feature information in each band of true-color remote sensing images by employing spectral band fusion enhancement methods, and to enhance the feature differences of target ground features such as vegetation, buildings, and water bodies. The image fusion subsystem uses a multi-scale decomposition tool to decompose the input images to be fused into multiple scales, and then recombines them according to preset fusion rules to achieve complementary advantages of spectral information and spatial details. The image mosaicking and cropping subsystem is used to seamlessly stitch multiple images by determining a reference image as the basis for the output mosaic image, and to precisely crop the mosaicked image according to the preset region boundaries. The data projection and format conversion subsystem is used to customize the projection settings of image data and convert the data format according to the application requirements of remote sensing data, so as to meet the compatibility requirements of different application platforms.

3. The data processing system for power transmission line inspection based on satellite imagery according to claim 2, characterized in that, The geometric correction processing subsystem further includes an orthorectification module, which performs high-precision positioning correction on the image data by using the RPC file included in the high-resolution image, which contains a rational function model and combines it with a digital elevation model. The RPC file is used to describe the mapping relationship between pixel coordinates and geographic coordinates, and the orthorectification module is used to eliminate image geometric distortion caused by terrain undulations.

4. The data processing system for power transmission line inspection based on satellite imagery according to claim 2, characterized in that, The satellite remote sensing data preprocessing module sets the following quality requirements for the processed image data, including: The spatial resolution of the satellite panchromatic image is less than 1 meter, and the spatial resolution of the multispectral image including red, green, blue, and near-infrared bands is less than 4 meters. The original image has a cloud cover of less than 20%, and the cloud layer must not cover ground features in important areas; the side tilt angle is less than ±15°; The data acquisition range completely covers the required area and is based on the distance of the outer boundary portion; The image result data coordinate system adopts CGCS2000 and provides data processing functions, which can provide slice data that meets the requirements of database management application systems. The background color value of the result data is set to (0, 0, 0). If there are any pixels with a color of (0, 0, 0) within the effective range of the result data, they are uniformly replaced with (1, 1, 1). The orthophoto data results are in TIFF, IMG and ECW formats with geolocation information; The image colors are then blended into a true-color image.

5. The data processing system for power transmission line inspection based on satellite imagery according to claim 1, characterized in that, The transmission line pixel spindle reconstruction module based on the direction field aggregation stability domain judgment includes: The orientation field map construction unit is used to calculate a principal orientation angle for each pixel in the preprocessed satellite image based on its local gradient information or texture orientation features and store it as the orientation vector of the corresponding pixel, thereby generating a pixel orientation field vector map of the entire map range. A stability region identification unit is used to divide the pixel orientation field vector map into multiple local window regions of a preset size, and within each local window region, to statistically analyze the distribution density and standard deviation of orientation vectors of all pixels, marking regions with high orientation vector distribution density and low standard deviation of orientation difference as stable aggregation regions, and determining that the stable aggregation regions are potential transmission line routing regions; and The main axis extraction unit is used to perform main axis extraction operation on the direction vectors in the stable aggregation region identified by the stable region identification unit. By fitting the direction vectors in the stable aggregation region, one or more main direction line segments that best represent these direction vectors are determined, and the main direction line segments are used as the reference lines for the precise positioning path of the transmission line.

6. The data processing system for power transmission line inspection based on satellite imagery according to claim 1, characterized in that: The transmission line spatial-temporal database and risk assessment module includes a spatial-temporal database management unit and a geographic information system fusion and visualization unit; The space-time database management unit is used to uniformly store the standardized remote sensing image products, the identification results of various risk factors output by the automatic identification and annotation module of risk points in the transmission line channel, the transmission line main axis path information output by the transmission line pixel main axis restoration module based on the direction field aggregation stability domain judgment, and the extracted transmission line parameter information, and adopts a distributed storage architecture. The geographic information system fusion and visualization unit is used to efficiently integrate all spatial data and geospatial background data in the space-time database management unit to form an intuitive and comprehensive transmission line inspection status map, and provides spatial analysis functions such as buffer analysis, overlay analysis, and path analysis to support deeper geospatial data mining.

7. The data processing system for power transmission line inspection based on satellite imagery according to claim 6, characterized in that, The transmission line space-time database and risk assessment module includes a multi-dimensional risk assessment and early warning unit, which performs risk assessment in the following ways: Based on the vegetation areas, voltage levels of transmission lines, safety distance standards, and historical vegetation intrusion event data identified by the automatic identification and labeling module for risk points in the transmission line corridor, a model is established to dynamically calculate the minimum distance between vegetation and transmission lines and the degree of erosion of the safety distance by the vegetation growth rate. Based on the identified construction activity area, type, scale, duration, and relative position to the transmission line, assess the risks of collision, excavation, and foundation settlement that the construction activities may cause to the transmission line. By combining surface deformation information interpreted from satellite imagery, topographic slope, aspect, hydrological conditions, and historical geological disaster data analyzed from DEM data, the geological stability along the transmission line is assessed. Based on the identified floating objects and foreign objects, and considering their size, type, location, and relative distance from the power transmission line, assess the potential risks of snagging or short circuits to the power transmission line. Based on the transmission line pixel principal axis reconstruction method, the line principal axis information is extracted and combined with its spatial position at different time points. By comparing it with historical data, potential problems such as small displacement, sag change and abnormal vibration of the line can be monitored. The multi-dimensional risk assessment and early warning unit comprehensively considers the above-mentioned risk assessment results and uses multi-factor weighted summation, fuzzy comprehensive evaluation or machine learning algorithms to form a comprehensive risk level judgment for the whole or local area of ​​the transmission line. When the assessment result exceeds the preset threshold, the unit automatically triggers a multi-level early warning mechanism. The risk assessment model can be dynamically updated and corrected according to new satellite image data to achieve continuous monitoring and real-time early warning of changes in the transmission line corridor environment.

8. The processing method of the data processing system for transmission line inspection based on satellite imagery according to any one of claims 1-7, characterized in that, include: Standardize the received raw satellite data; Automatically identify and label various risk factors within the transmission line corridor from the standardized remote sensing image products, including: fallen trees, line icing, abnormal plant intrusion, floating objects, and foreign object intrusion. Restore the main axis direction of the transmission line; The standardized remote sensing image products, the risk factor identification results, and the transmission line main axis direction information are integrated, stored, managed, analyzed, and risk-assessed.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of claim 8.

Images