A method and system for identifying key regions in an overhead power line mapping image

Abstract

The application provides a kind of overhead power line surveying and mapping image key area identification processing method and system, wherein the method comprises: obtaining the original image data collected by camera and the geographic position data of camera shooting position;Match the geographic position data with the preset geographic position database, obtain the natural material prior probability data set corresponding to the camera shooting position;Extract polarization feature and image content feature from the original image data, use double-branch attention network, establish material optical property table based on polarization feature in the first branch processing path, fuse natural material prior probability data set and image content feature in the second branch processing path;Based on material optical property table and fusion result, identify key area from original image data;Conduct confidence assessment on key area, generate key area confidence map.The application improves the identification accuracy of natural material key area in outdoor complex scene.

Description

A method and system for identifying and processing key areas in overhead power line mapping images. Technical Field

[0001] This application relates to the fields of computer vision and image processing technology, and in particular to a method and system for identifying and processing key areas in overhead power line mapping images. Background Technology

[0002] In outdoor surveying of overhead power lines and the acquisition of images of the surrounding environment, accurate identification of key natural material areas (such as sky, water, and vegetation) in the images is crucial for improving environmental perception and evaluating the identification of areas around overhead power lines. Due to issues such as changes in lighting and material diversity in natural scenes, traditional methods struggle to reliably distinguish areas of similar materials, necessitating a high-precision identification technology that can combine information from multiple sources.

[0003] Currently, a single-branch convolutional neural network scheme based on deep learning is widely used. This scheme directly classifies pixels in an image by training on a large amount of labeled data, and uses convolutional layers to extract local features and combine them with global context information to predict the material category.

[0004] The scheme relies on a large amount of labeled data, resulting in high training costs and a tendency to misidentify similar materials under complex lighting conditions. Geographic information and polarization optical characteristics are not effectively utilized, leading to insufficient robustness in specific scenarios (such as water reflection and vegetation shadows). Summary of the Invention

[0005] This application provides a method and system for identifying key areas in overhead power line mapping images, in order to solve the problem of low accuracy in identifying key areas of natural materials in complex outdoor scenes in the prior art.

[0006] In a first aspect, this application provides a method for identifying and processing key areas in overhead power line mapping images, including:

[0007] Acquire raw image data captured by the camera and geographic location data of the camera's shooting location;

[0008] The geographic location data is matched with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera shooting location;

[0009] Polarization features and image content features are extracted from the original image data. Using a dual-branch attention network, a material optical property table is established based on the polarization features in the first branch processing path. In the second branch processing path, the natural material prior probability dataset is fused with the image content features.

[0010] Based on the material optical property table and the fusion results, key regions are identified from the original image data;

[0011] A confidence assessment is performed on the key regions to generate a confidence map of the key regions.

[0012] Optionally, the step of establishing a material optical property table based on the polarization features in the first branch processing path, and fusing the natural material prior probability dataset with the image content features in the second branch processing path, includes:

[0013] In the first branch processing path, the polarization feature is decomposed into multiple angles to obtain intensity components in different polarization directions.

[0014] Based on the intensity component, calculate the intensity variation gradient of the same material region in the polarization direction;

[0015] Based on the intensity change gradient, a table of material optical properties is established;

[0016] In the second branch processing path, the natural material prior probability dataset is converted into a spatial probability distribution map;

[0017] The spatial probability distribution map is correlated with the image content features at the pixel level to generate visual features;

[0018] The visual features are fused with the spatial probability distribution map using a dynamic weighting mechanism to obtain the fusion result.

[0019] Optionally, the step of fusing the visual features with the spatial probability distribution map through a dynamic weighting mechanism to obtain a fusion result includes:

[0020] Calculate the magnitude of the probability value change at each pixel location in the spatial probability distribution map;

[0021] Analyze the degree of feature value fluctuation of the visual feature at the same pixel position;

[0022] Based on the proportional relationship between the change range of the probability value and the fluctuation degree of the feature value, corresponding weighting factors are dynamically assigned to the pixel probability value of the spatial probability distribution map and the feature value of the visual feature.

[0023] The pixel probability value and the feature value are weighted and combined according to the corresponding weight factor to generate a fused feature map, which is the fusion result.

[0024] Optionally, the step of weighting and combining the pixel probability value and the feature value according to the corresponding weight factor to generate a fused feature map includes:

[0025] A first weighting sub-factor is applied to the pixel probability value to generate a weighted probability value;

[0026] A second weighting sub-factor is applied to the eigenvalues ​​to generate weighted eigenvalues;

[0027] The weighted probability value and the weighted feature value at the same pixel location are added together to obtain the fused feature value;

[0028] Based on the fused feature values ​​of all pixel locations, a two-dimensional matrix structure is constructed, and the two-dimensional matrix structure is used as the fused feature map.

[0029] Optionally, the step of matching the geographic location data with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera shooting location includes:

[0030] Extract GPS coordinates and shooting time from the geographic location data;

[0031] Determine the region code of the geographic grid region where the GPS coordinates are located from a preset set of geographic grid regions;

[0032] Based on the season corresponding to the shooting time, retrieve historical material distribution records corresponding to the season from the geographic location database;

[0033] The region code is compared and matched with the region markers in the historical material distribution record. Based on the comparison and matching results, the frequency data of all natural materials are extracted from the historical material distribution record.

[0034] Based on the frequency of occurrence data, the probability weight values ​​of key natural materials are calculated, including the sky, water, and vegetation.

[0035] Based on the probability weight values, a prior probability dataset of natural materials is generated.

[0036] Optionally, identifying key regions from the original image data based on the material optical property table and the fusion result includes:

[0037] The material optical property table and the fusion result are combined to form joint feature data;

[0038] Multiple detection windows are set on the original image data, and the matching degree between the joint feature data and the preset material prototype in each detection window is calculated;

[0039] When the matching degree exceeds the preset matching degree threshold, the corresponding detection window is marked as a candidate key region;

[0040] When multiple candidate key regions exist, perform continuity analysis and merging operations on spatially adjacent candidate key regions;

[0041] Repeat the continuity analysis and merging operation until all candidate key regions meet the preset spatial separation conditions to obtain the key regions.

[0042] Optionally, the step of performing a confidence assessment on the key region and generating a confidence map of the key region includes:

[0043] Calculate three confidence factors for each pixel within the key region, where the first confidence factor is the deviation value between the polarization feature and the material optical property table, the second confidence factor is the probability value of the pixel in the spatial probability distribution map, and the third confidence factor is the similarity of material features between adjacent pixels.

[0044] According to the preset weighting rules, the three confidence factors are fused into a pixel-level confidence score;

[0045] Based on the pixel-level confidence level, a confidence map of the key region is generated.

[0046] Secondly, this application provides a key area identification and processing system for overhead power line mapping images, including:

[0047] The acquisition module is used to acquire raw image data captured by the camera and geographic location data of the camera's shooting location;

[0048] The matching module is used to match the geographic location data with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera shooting location;

[0049] The extraction module is used to extract polarization features and image content features from the original image data. Using a dual-branch attention network, a material optical property table is established based on the polarization features in the first branch processing path, and the natural material prior probability dataset is fused with the image content features in the second branch processing path.

[0050] The identification module is used to identify key regions from the original image data based on the material optical property table and the fusion result;

[0051] The evaluation module is used to evaluate the confidence level of the key area and generate a confidence map of the key area.

[0052] Thirdly, this application provides a computing device, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform a key area identification processing method in an overhead power line survey image as described in any of the first aspects.

[0053] Fourthly, this application provides a computer storage medium storing computer program instructions thereon, which, when executed by a processor, implement a key area identification and processing method in an overhead power line mapping image as described in any one of the first aspects.

[0054] This application provides a method for identifying key regions in overhead power line mapping images. The method includes: acquiring raw image data captured by a camera and geographic location data of the camera's shooting location; matching the geographic location data with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera's shooting location; extracting polarization features and image content features from the raw image data; using a dual-branch attention network, establishing a material optical property table based on the polarization features in the first branch processing path, and fusing the natural material prior probability dataset with the image content features in the second branch processing path; identifying key regions from the raw image data based on the material optical property table and the fusion result; and performing confidence assessment on the key regions to generate a key region confidence map.

[0055] The technical solution provided in this application has the following beneficial effects:

[0056] This application provides a multi-source data foundation for subsequent processing, ensuring the comprehensiveness and timeliness of information acquisition. It enhances scene understanding capabilities using prior geographical knowledge, providing regional characteristic references for material identification. It fully mines multi-dimensional information from images, providing rich feature representations for subsequent analysis. It achieves collaborative processing of polarization characteristic analysis and geographic information fusion, improving feature representation capabilities. It accurately delineates target material regions in images, completing the localization task. It quantifies the reliability of the identification results, providing a decision-making basis for subsequent applications.

[0057] Furthermore, this application also employs a dual-branch processing mechanism. In the first branch, the polarization features are decomposed from multiple angles and gradients are calculated to construct a data table characterizing the optical properties of the material. In the second branch, the geographical prior data is converted into a spatial probability map and associated with visual features. Finally, feature fusion is achieved through dynamic weighting.

[0058] Furthermore, this method improves the accuracy and robustness of natural material regions in complex scenarios through dual-path collaborative processing of polarization characteristic analysis and geographic information fusion. It demonstrates superior performance, especially in challenging scenarios such as illumination changes and similar material differentiation. At the same time, confidence assessment provides a reliable basis for the application of the results.

[0059] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 is a flowchart of a key area identification and processing method in an overhead power line mapping image provided in an embodiment of this application;

[0062] Figure 2 is a schematic diagram of a key area identification and processing system in an overhead power line mapping image provided in an embodiment of this application;

[0063] Figure 3 is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0064] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0065] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0066] Existing deep learning-based image recognition methods primarily rely on single visual information for material differentiation, which faces significant limitations in complex outdoor scenes. On one hand, similar materials (such as water reflections and road surface reflections) exhibit similar features in conventional images, leading to misclassification. On the other hand, they lack consideration for the environmental characteristics of the shooting location; the same material may exhibit different characteristics in different geographical areas, further affecting recognition accuracy. These problems stem from the fact that existing solutions fail to effectively combine optical characteristics with geographical information, resulting in insufficient stability of recognition results in variable natural scenes.

[0067] To address the aforementioned issues, this application proposes a method for identifying key areas in overhead power line mapping images. This method simultaneously acquires image data and geographic location information via a camera, utilizes a geographic database to obtain regional material distribution patterns, and analyzes the material reflection characteristics under polarized light at different angles. The two types of information are dynamically fused to complete the identification. Specifically, the system first matches typical local material distribution data based on the shooting location as a priori reference; simultaneously, it extracts unique optical fingerprints of materials through polarized light analysis; finally, it combines the two types of data for collaborative judgment to accurately distinguish easily confused materials. This approach overcomes the limitations of traditional single-vision analysis, improving the reliability of material identification in natural scenes through dual verification of geographic and optical characteristics, especially improving the identification effect under complex conditions such as reflection and shadow.

[0068] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0069] Figure 1 is a flowchart of a key area identification and processing method in an overhead power line mapping image provided by an embodiment of this application. As shown in Figure 1, the method includes:

[0070] Step 101: Obtain the raw image data captured by the camera and the geographic location data of the camera's shooting location.

[0071] In step 101, the raw image data refers to the digital image containing scene optical information directly captured by the camera, and in this application, specifically refers to image data containing multi-angle polarization information. The camera shooting location refers to the geographical location of the camera, obtained by associating the GPS coordinate data of that location with a geographic location database to obtain the prior material probability of the area, rather than the location of the photographed target. Geographic location data refers to data recording the coordinate and time information of the camera's location at the time of shooting, used to associate with a geographic database to obtain regional features.

[0072] In this embodiment of the application, the camera simultaneously records the digital image captured at the current moment, as well as the location coordinates and time information of the shooting location. The image data includes conventional color information and brightness information in multiple polarization directions, while the location data includes longitude, latitude, and a timestamp accurate to the second. These two types of data will serve as the input basis for subsequent processing.

[0073] For example, when shooting a beach scene at location A, the camera obtains an image that includes the sky, sea, and sand, while also recording the shooting coordinates as X degrees north latitude and Y degrees east longitude, and the time as H:00 AM on W day of Z month, 2023. The image data includes brightness information in three polarization directions: 0 degrees, 45 degrees, and 90 degrees. The location data and time information will be used to query the distribution of typical local materials.

[0074] Step 102: Match the geographic location data with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera shooting location.

[0075] In step 102, the natural material prior probability dataset refers to the set of probabilities of various natural materials appearing based on historical statistics of the shooting location, including the distribution probability values ​​of materials such as sky, water, and vegetation.

[0076] In this embodiment of the application, the system first extracts coordinate and time information from the geographic location data, maps the coordinates to a pre-divided geographic grid system to determine the area number, and then retrieves the historical material distribution records of the area from the database according to the season corresponding to the shooting time. By matching the area number, the system obtains the frequency data of the occurrence of each natural material at that location, and finally calculates the probability weight values ​​of each type of material to form a prior probability dataset.

[0077] For example, based on the coordinates of X degrees north latitude and Y degrees east longitude, the location is determined to belong to grid number YLW-XY. By querying the historical summer data of Z month in 2023, it is found that the sky appeared 800 times, water appeared 300 times, and sand appeared 100 times in this grid. The probabilities of the sky, water, and sand are calculated to be 0.67, 0.25, and 0.08, forming a priori probability dataset.

[0078] Step 103: Extract polarization features and image content features from the original image data. Using a dual-branch attention network, establish a material optical property table based on the polarization features in the first branch processing path. In the second branch processing path, fuse the natural material prior probability dataset with the image content features.

[0079] In step 103, polarization characteristics refer to the intensity variation characteristics of light in different polarization directions in the image. Image content characteristics refer to the visual features of the image, such as color and texture. The material optical property table is a data table recording the polarization characteristics of various materials.

[0080] In this embodiment of the application, the original image data is decomposed into polarization direction, the light intensity value in each direction is calculated, the intensity variation law of the same material area in different polarization directions is analyzed, and a data table reflecting the polarization characteristics of the material is established; at the same time, the color and texture features of the image are extracted, the geographical prior probability is converted into a spatial distribution map and associated and fused with visual features.

[0081] For example, the intensity of the sky region in the beach image at polarization directions of 0 degrees, 45 degrees, and 90 degrees is 120, 118, and 119 respectively, and the gradient value is calculated to be 3; the intensity of the seawater region is 85, 40, and 82, and the gradient value is 88; a characteristic table containing these gradient values ​​is established.

[0082] Step 104: Based on the material optical property table and the fusion result, identify key regions from the original image data.

[0083] In step 104, the key region refers to a continuous pixel region in the image that is identified as a specific natural material.

[0084] In this embodiment, the material property table and the fused features are spliced ​​together to form joint features. The matching degree between the features within the window and the material prototype is calculated by sliding the detection window on the image. Windows that exceed the threshold are marked as candidate regions. Adjacent candidate regions are merged, and finally, the key regions that meet the spatial continuity requirements are output.

[0085] For example, if the matching degree between the joint features and the sky prototype in a 32×32 pixel window is 2.8, which is lower than the threshold of 3.0, it is determined to be a sky region; merge three adjacent windows with qualified matching degrees to form a key sky region with x=100 to 160 and y=200 to 240.

[0086] Step 105: Perform a confidence assessment on the key region and generate a confidence map of the key region.

[0087] In step 105, the key region confidence map is a probability distribution map that reflects the reliability of the identification results.

[0088] In this embodiment, three confidence factors—polarization feature matching degree, geographic probability value, and spatial consistency—are calculated for each key region pixel. These factors are then fused according to preset rules to generate pixel-level confidence, ultimately forming a confidence map of the entire image.

[0089] For example, if a pixel in a certain sky region has a polarization matching degree of 0.9, a geographic probability of 0.7, and a neighborhood consistency of 0.8, a confidence value of 0.83 is calculated by weighting them with 0.5, 0.3, and 0.2 respectively.

[0090] This method achieves accurate identification of key materials in natural scenes by integrating geographical prior knowledge and polarization optical characteristics. It exhibits superior stability under complex lighting and similar material scenes. The reliability of the identification results is further improved through confidence assessment, providing a high-quality basis for regional division for subsequent applications.

[0091] To address the issue of accurate natural material recognition in outdoor scenes, in some embodiments, step 103 involves: establishing a material optical property table based on the polarization features in the first branch processing path, and fusing the natural material prior probability dataset with the image content features in the second branch processing path, including:

[0092] Step 201: In the first branch processing path, perform multi-angle polarization state decomposition on the polarization feature to obtain intensity components in different polarization directions.

[0093] In step 201, multi-angle polarization state decomposition is the process of separating the original polarization image into multiple single polarization direction component images. The intensity component refers to the light intensity value of each pixel in a specific polarization direction.

[0094] In this embodiment of the application, the original image containing multiple polarization directions captured by the camera is input into the processing system. The system decomposes the image into several images with single polarization directions according to a preset polarization angle. Each decomposed image retains only the light intensity information of the corresponding polarization direction. These single polarization images will be used for subsequent material property analysis.

[0095] Step 202: Calculate the intensity change gradient of the same material region in the polarization direction based on the intensity component.

[0096] In step 202, the region of the same material refers to a continuous pixel region in the original image that has similar polarization characteristics. This region is determined by analyzing the intensity component variation law of each polarization direction after multi-angle polarization state decomposition. Specifically, it exhibits a continuous and consistent intensity variation gradient feature in the polarization direction. The intensity variation gradient is a quantitative index of the light intensity variation of the same material region in different polarization directions, reflecting the material's response characteristics to polarized light.

[0097] In this embodiment of the application, the system selects a region in the image that has been initially marked as the same material, calculates the average intensity value of the region in the image of each polarization direction, analyzes the law of these intensity values ​​changing with the polarization angle, and obtains the polarization gradient characteristic value of the material through a specific calculation method.

[0098] Step 203: Establish a material optical property table based on the intensity change gradient.

[0099] In this embodiment of the application, the system classifies and organizes the calculated polarization gradient features of various materials, and establishes a query table according to the material type, wherein each entry contains the material name and the corresponding polarization gradient feature range.

[0100] Step 204: In the second branch processing path, the natural material prior probability dataset is converted into a spatial probability distribution map.

[0101] In step 204, the spatial probability distribution map is a way of converting prior probability data into an image representation, where the value of each pixel in the image represents the probability that the point belongs to a certain type of material.

[0102] In this embodiment of the application, the system uses prior probability data obtained from geographic location data queries to allocate probability values ​​to corresponding spatial locations according to image size and shooting range, generating a probability distribution image with the same size as the original image.

[0103] Step 205: Associate the spatial probability distribution map with the image content features at the pixel level to generate visual features.

[0104] In step 205, pixel-level association refers to the process of establishing a correspondence between the probability distribution map and visual features at the pixel level. Visual features are image features that contain prior geographic information after association processing.

[0105] In this embodiment of the application, the system combines the value of each pixel in the probability distribution map with the corresponding image visual feature value to generate a new feature representation. These new features contain both the original visual information and geographical prior knowledge.

[0106] Step 206: The visual features are fused with the spatial probability distribution map through a dynamic weighting mechanism to obtain the fusion result.

[0107] In step 206, the dynamic weighting mechanism refers to a method that automatically adjusts the weights of different features based on data characteristics. The fusion result is the final feature representation obtained after weighted calculation.

[0108] In this embodiment of the application, the system analyzes the reliability index of visual features and probability distribution in real time, assigns appropriate weight ratios to the two types of features based on the analysis results, performs weighted calculations, and outputs an optimized fusion feature map.

[0109] Here is a specific example:

[0110] In the specific implementation of the beach scene at location A, the system first processes the acquired polarized image containing the sky, seawater, and sand, decomposing the image into monochrome images with three polarization directions: 0 degrees, 45 degrees, and 90 degrees. The brightness values ​​of the sky region in the three directions are 120, 118, and 119, respectively. The gradient value is obtained by calculating the sum of the absolute values ​​of the brightness differences between adjacent polarization directions: |120-118|+|118-119|+|119-120|=3. The brightness values ​​of the seawater region are 85, 40, and 82, and the gradient value is calculated as |85-40|+|40-82|+|82-85|=88. These gradient values ​​are recorded in the material optical property table for subsequent identification. Simultaneously, the system converts the retrieved sky probability (0.67), water probability (0.25), and sand probability (0.08) into a probability distribution map of the same size as the image. The probability value of each pixel is determined by its location; pixels closer to the top of the image are assigned a higher sky probability value, while pixels closer to the bottom are assigned a higher water or sand probability value. Next, the system multiplies the probability distribution map with the color features of the original image at corresponding positions to generate a new visual feature map. For example, the blue channel value of 210 in the sky region is multiplied by the sky probability of 0.67 to obtain 140.7. Finally, the system dynamically assigns weights based on the reliability of gradient features and probability features. For seawater areas with obvious gradient features, a higher weight of 0.7 is assigned to the gradient feature and a weight of 0.3 to the probability feature, resulting in a fusion feature value of 0.7×88+0.3×0.25=61.6+0.075=61.675. For sandy areas with indistinct gradient features, a higher weight of 0.6 is assigned to the probability feature and a weight of 0.4 to the gradient feature, resulting in a fusion feature value of 0.4×6+0.6×0.08=2.4+0.048=2.448. This process generates a fusion feature map of the entire image, providing a reliable basis for subsequent key area identification.

[0111] In this embodiment, the method effectively improves the recognition accuracy of natural material areas through dual-path collaborative processing of polarization characteristic analysis and geographic information fusion. In particular, the recognition stability under complex lighting conditions is significantly improved. At the same time, dynamic weight adjustment ensures adaptability in different scenarios, providing reliable technical support for outdoor image analysis.

[0112] To further improve the accuracy of feature fusion, in some embodiments, step 206: fusing the visual features with the spatial probability distribution map through a dynamic weighting mechanism to obtain the fusion result includes:

[0113] Step 301: Calculate the probability value change range of each pixel position in the spatial probability distribution map.

[0114] In step 301, the probability value change range refers to the degree of difference between the probability values ​​of a certain pixel in the spatial probability distribution map and the probability values ​​of its surrounding pixels, reflecting the certainty of the material distribution at that location.

[0115] In this embodiment, the system selects a local region centered on the target pixel, calculates the difference between the maximum and minimum probability values ​​within the region, and uses this difference as an index of the probability change of the pixel to evaluate the reliability of the geographic prior information.

[0116] Step 302: Analyze the degree of feature value fluctuation of the visual feature at the same pixel position.

[0117] In step 302, the degree of feature value fluctuation refers to the degree of drastic change of visual features in a local area, reflecting the complexity of the image content.

[0118] In this embodiment of the application, the system analyzes the changes in the values ​​of each channel of the visual feature in the same location area, and quantifies the degree of fluctuation by calculating the average difference of the feature values ​​of adjacent pixels, so as to evaluate the stability of the visual feature.

[0119] Step 303: Based on the proportional relationship between the change range of the probability value and the fluctuation degree of the feature value, dynamically assign corresponding weight factors to the pixel probability value of the spatial probability distribution map and the feature value of the visual feature.

[0120] In step 303, the proportional relationship refers to the ratio of the probability value change of a pixel in the spatial probability distribution map to the fluctuation of the visual feature at the corresponding location. This ratio is obtained by dividing the probability change by the feature fluctuation, reflecting the relative reliability of geographic prior information and visual features at that location. When the probability change is large and the feature fluctuation is small, it indicates that the geographic prior information is more reliable; conversely, the visual feature is more reliable. The system dynamically adjusts the weight allocation of the two types of features based on this ratio. The weighting factor is a weighting coefficient allocated according to the reliability of the two types of features.

[0121] In this embodiment, the system uses the ratio of probability change magnitude to feature fluctuation degree as the allocation basis. When the probability change magnitude is relatively large, the weight of spatial probability is increased, and when the feature fluctuation is small, the weight of visual features is increased, thereby achieving dynamic adjustment.

[0122] Step 304: The pixel probability value and the feature value are weighted and combined according to the corresponding weighting factor to generate a fused feature map, which is the fusion result.

[0123] In step 304, the fused feature map is the final result obtained through weighted calculation, and each pixel value is a new feature representation after the two types of features are combined according to their weights.

[0124] In this embodiment of the application, the system multiplies the spatial probability value and visual feature value of each pixel by the corresponding weighting factor and then adds them together to generate a new feature map. This feature map not only retains the original feature information but also reflects the guiding role of geographical prior.

[0125] Here is a specific example:

[0126] In the dynamic weighted fusion process of the beach scene at location A, the system first selects the center pixel of the seawater area with coordinates x=150, y=250. It calculates the difference of 0.06 between the maximum (0.28) and minimum (0.22) water probability values ​​within the 3×3 neighborhood of this point, taking this difference as the probability value variation range. Simultaneously, it analyzes the fluctuation of the blue channel value of the visual feature in adjacent pixels, calculating an average fluctuation of 5.2. Based on the ratio 0.06 / 5.2≈0.0115, the system assigns a weight of 0.1 to the spatial probability and 0.9 to the visual feature. Multiplying the water probability value of this pixel (0.25) by the weight 0.1 yields 0.025, and multiplying the visual feature value (82) by 0.9 yields 73.8. The sum of these values ​​results in a fused feature value of 73.825. For the sky region pixel x=100, y=200, its probability change amplitude is calculated to be 0.03, feature fluctuation degree is 2.1, and the ratio is 0.03 / 2.1≈0.0143. Weights are assigned to 0.15 and 0.85. The probability of sky at this point (0.67) is weighted with the visual feature value (215) to calculate: 0.15×0.67+0.85×215≈0.1+182.75=182.85. For the sand region pixel x=180, y=300, the probability change amplitude is 0.02, and the ratio of feature fluctuation degree (3.5) is 0.0057. Weights are assigned to 0.05 and 0.95. The result is 0.05×0.08+0.95×145≈0.004+137.75=137.754. In the final fused feature map, the feature value of the seawater region mainly reflects visual information (73.825), while the feature values ​​of the sky region (182.85) and the sandy region (137.754) incorporate geographical priors to varying degrees, achieving feature optimization tailored to local conditions.

[0127] In this embodiment, the method achieves complementary advantages between geographical prior knowledge and visual features by dynamically evaluating the reliability of different features and intelligently allocating weights. This results in more accurate and stable feature representations in complex natural scenes, laying a solid foundation for subsequent key area identification, while avoiding feature bias problems caused by fixed weights.

[0128] To further improve the accuracy and reliability of feature fusion, in some embodiments, step 304: weighting the pixel probability value and the feature value according to the corresponding weight factor to generate a fused feature map, includes:

[0129] Step 401: Apply a first weighting sub-factor to the pixel probability value to generate a weighted probability value.

[0130] In step 401, the first weight sub-factor refers to the weighting coefficient assigned to the pixel probability value, reflecting the contribution of geographic prior information in the fusion. The weighted probability value is the result of adjusting the spatial probability value by weights.

[0131] In this embodiment of the application, the system multiplies the spatial probability value of each pixel with the corresponding first weight sub-factor according to the dynamic weight allocation result to obtain the adjusted probability value. This value retains the guiding role of geographic information but is appropriately scaled according to reliability.

[0132] Step 402: Apply a second weighting sub-factor to the eigenvalues ​​to generate weighted eigenvalues.

[0133] In step 402, the second weighting sub-factor refers to the weighting coefficient assigned to the visual feature values, reflecting the degree of contribution of image features in the fusion. The weighted feature values ​​are the result of the visual features after weight adjustment.

[0134] In this embodiment, the system multiplies the visual feature value of each pixel with the corresponding second weighting factor to obtain an adjusted feature value. This value retains the main features of the original visual information but is optimized according to the characteristics of the scene.

[0135] Step 403: Add the weighted probability value and the weighted feature value at the same pixel location to obtain the fused feature value.

[0136] In step 403, the fused feature value is the result of adding the weighted probability value and the weighted feature value at the same position, representing the final fused feature representation.

[0137] In this embodiment of the application, the system adds the probability value and feature value of each pixel after weight adjustment to generate a new feature value, which contains both geographical prior guidance and visual feature information.

[0138] Step 404: Based on the fused feature values ​​of all pixel locations, construct a two-dimensional matrix structure and use the two-dimensional matrix structure as the fused feature map.

[0139] In step 404, the two-dimensional matrix structure refers to a numerical array arranged in rows and columns of the image, where each element corresponds to the fused feature value of a pixel.

[0140] In this embodiment, the system fills the fused feature values ​​of all pixels into the corresponding positions according to the size and pixel arrangement of the original image, and constructs a complete feature matrix. This matrix can be directly used for subsequent key area recognition processing.

[0141] Here is a specific example:

[0142] In the feature fusion process for the beach scene at location A, the system first processes the pixel x=150, y=250 in the seawater area. This pixel has a water probability value of 0.25 and a visual feature value of 82. Based on the dynamic weight allocation, the first weight sub-factor is 0.1, and the second weight sub-factor is 0.9. The system multiplies the probability value 0.25 by 0.1 to obtain a weighted probability value of 0.025, and simultaneously multiplies the feature value 82 by 0.9 to obtain a weighted feature value of 73.8. The sum of these two values ​​(0.025 + 73.8) equals 73.825, which is used as the fusion feature value for this pixel. For the pixel x=100, y=200 in the sky area, its sky probability value 0.67 is multiplied by the first weight sub-factor 0.15 to obtain 0.1005, and its visual feature value 215 is multiplied by the second weight sub-factor 0.85 to obtain 182.75. The sum of these two values ​​is 182.8505. The probability value of sandy land for pixel x=180, y=300 is 0.08 multiplied by 0.05, resulting in 0.004. The feature value is 145 multiplied by 0.95, resulting in 137.75. Adding these together yields 137.754. The system constructs a two-dimensional matrix with 640 rows and 480 columns according to the image size. The calculated fusion feature values ​​are stored in the corresponding elements of the matrix according to their coordinate positions. Specifically, row 150, column 250 stores 73.825; row 100, column 200 stores 182.8505; and row 180, column 300 stores 137.754. The resulting 640×480 digital matrix is ​​the fusion feature map. In this image, the value 73.825 for the sea area primarily reflects visual features, the value 182.8505 for the sky area moderately incorporates geographical priors, and the value 137.754 for the sandy area is primarily driven by visual features, achieving differentiated feature fusion across different regions.

[0143] In this embodiment of the application, the method achieves a scientific fusion of geographical priors and visual features through step-by-step weighting and matrix organization. The generated fused feature map not only preserves image details but also reflects scene characteristics, providing a high-quality feature representation for subsequent processing. At the same time, the matrix structure ensures processing efficiency and data regularity.

[0144] To further improve the accuracy of geographic information matching, in some embodiments, step 102: matching the geographic location data with a preset geographic location database to obtain a prior probability dataset of natural materials corresponding to the camera's shooting location includes:

[0145] Step 501: Extract GPS coordinates and shooting time from the geographic location data.

[0146] In step 501, GPS coordinates refer to the latitude and longitude location information recorded by the camera when taking the picture. The shooting time refers to the specific time point when the image was acquired.

[0147] In this embodiment, the system parses longitude, latitude, and timestamp accurate to the second from the metadata collected by the camera. This data will serve as key index information for querying the geographic database.

[0148] Step 502: Determine the area code of the geographic grid region where the GPS coordinates are located from the preset geographic grid region set.

[0149] In step 502, a geographic grid region is a system that divides geographic space into regular units. A region code is a unique identifier for each grid.

[0150] In this embodiment of the application, the system matches the obtained latitude and longitude coordinates with a predefined grid division scheme, determines the grid to which it belongs by judging the coordinate range, and extracts the standard code of the grid as a query condition.

[0151] Step 503: Based on the season corresponding to the shooting time, retrieve the historical material distribution records corresponding to the season from the geographic location database.

[0152] In step 503, the historical material distribution record is a statistical data on the occurrence of materials classified by season stored in the database, reflecting the regional characteristics of different seasons.

[0153] In this embodiment of the application, the system determines the season based on the shooting time and retrieves the historical data set of the corresponding season from the database. The data set contains the material distribution statistics of each grid area in the same season.

[0154] Step 504: Compare and match the region code with the region marker in the historical material distribution record, and extract the occurrence frequency data of all natural materials from the historical material distribution record based on the comparison and matching results.

[0155] In step 504, natural materials specifically include common outdoor natural environment material categories such as sky, water, vegetation, sand, rocks, soil, and snow. In this solution, key natural materials specifically refer to the three categories of sky, water, and vegetation. Frequency data refers to the statistical count of the number of times a specific material appears in historical records, reflecting the distribution pattern of the material.

[0156] In this embodiment of the application, the system finds the corresponding historical records by matching grid codes, and extracts the occurrence frequency data of materials such as sky, water, and vegetation as the basis for calculating probabilities.

[0157] Step 505: Based on the occurrence frequency data, calculate the probability weight value of key natural materials, including the sky, water and vegetation.

[0158] In step 505, the probability weight value is the result of converting the occurrence frequency into a probability distribution, which is used to quantify the probability of the occurrence of various materials.

[0159] In this embodiment, the system divides the number of occurrences of each type of material by the total number of observations to obtain a normalized probability value, with a focus on retaining the weights of the three key materials: sky, water, and vegetation.

[0160] Step 506: Generate a prior probability dataset of natural materials based on the probability weight values.

[0161] In this embodiment of the application, the system organizes and stores the calculated probability values ​​of various materials in a standard format to form a data set containing material types and corresponding probability values.

[0162] Here is a specific example:

[0163] In the geographic information matching process for the beach scene at location A, the system first extracts the shooting coordinates (X degrees North latitude, Y degrees East longitude) and the time (Z month, W day, H hour, 2023) from the metadata recorded by the camera, where X, Y, Z, W, and H are specific values. The coordinates are then input into a pre-divided geographic grid system to determine that the location belongs to the area of ​​grid number YLW-XY. Based on the shooting time being July, which is summer, the system retrieves the historical summer observation records for this grid from the database for the past three years. Statistics show that in images taken in summer, the sky was marked 800 times, water 300 times, and sand 100 times. The system calculates the total number of observations as 800 + 300 + 100 = 1200. The probability of the sky appearing (800 divided by 1200) is 0.67, the probability of water (300 divided by 1200) is 0.25, and the probability of sand (100 divided by 1200) is 0.08. The sky probability of 0.67 is used for weight allocation in polarization feature analysis, the water probability of 0.25 is used in the feature fusion calculation of the seawater area, and the sandy area probability of 0.08 serves as a reference for sandy area identification. The final generated natural material prior probability dataset contains the probability values ​​of these three key materials.

[0164] In this embodiment of the application, the method obtains the material distribution pattern reflecting the regional characteristics through precise spatiotemporal matching and historical data analysis, providing reliable geographical prior knowledge for image recognition, effectively improving the accuracy of material recognition in complex scenes, and avoiding the misjudgment problem that is easily caused by single visual analysis.

[0165] To further improve the accuracy of key region identification, in some embodiments, step 104: identifying key regions from the original image data based on the material optical property table and the fusion result, includes:

[0166] Step 601: Combine the material optical property table with the fusion result to form joint feature data.

[0167] In step 601, joint feature data refers to a new feature representation formed by combining material polarization characteristics with fused features, which includes comprehensive features of optical properties and geographic information.

[0168] In this embodiment, the system concatenates the gradient features in the material optical property table with the feature values ​​in the fused feature map according to the pixel positions, generating new feature data that simultaneously reflects the material optical properties and spatial distribution patterns.

[0169] Step 602: Set multiple detection windows on the original image data, and calculate the matching degree between the joint feature data and the preset material prototype in each detection window.

[0170] In step 602, the detection window refers to the analysis area that slides across the image. The preset material prototype refers to a set of reference data representing the standard characteristics of various natural materials, obtained through statistical analysis of typical samples. This includes the standard optical properties and spatial distribution characteristics of materials such as the sky, water, and vegetation. The matching degree is an indicator that measures the similarity between features within the window and the standard material.

[0171] In this embodiment of the application, the system moves a square analysis window on the image at fixed steps, calculates the similarity between the joint feature data in each window and the pre-stored material standard features such as sky and water, and obtains a quantified matching score.

[0172] Step 603: When the matching degree exceeds the preset matching degree threshold, the corresponding detection window is marked as a candidate key region.

[0173] In step 603, the candidate key region is an image region that is initially determined to belong to a specific material.

[0174] In this embodiment of the application, the system compares the matching degree of each window with a preset standard value. When the matching degree meets the requirements, the position and range information of the window are recorded as a candidate region.

[0175] Step 604: When there are multiple candidate key regions, perform continuity analysis and merging operations on spatially adjacent candidate key regions.

[0176] In step 604, continuity analysis is a process for evaluating the spatial relationships of candidate regions, and merging operation is a process for integrating adjacent regions that meet the criteria.

[0177] In this embodiment, the system calculates the distance between the center points of adjacent candidate regions and the feature similarity. When the distance is less than a threshold and the features are similar, these regions are merged into a larger continuous region.

[0178] Step 605: Repeat the continuity analysis and merging operation until all candidate key regions meet the preset spatial separation conditions to obtain the key regions.

[0179] In step 605, the preset spatial separation condition refers to the quantitative standard used to determine whether candidate key regions should remain independent or be merged. Specifically, it includes a dual judgment condition of the distance threshold between the center points of adjacent regions and the matching degree difference threshold.

[0180] In this embodiment of the application, the system repeatedly checks the relationship between all candidate regions until the distance between any two regions is greater than a set value and the feature differences are obvious. The region obtained at this time is the final key region.

[0181] Here is a specific example:

[0182] In the key area identification process of the beach scene at location A, the system first concatenates the polarization gradient value 3 of the sky region in the material optical properties table with the corresponding sky feature value 182.85 in the fused feature map to form joint feature data containing [3, 182.85]. A sliding detection is performed on the original image with a window size of 32×32 pixels and a step size of 16 pixels. When a window with the upper left corner coordinates x=96, y=192 is detected, the matching degree between the joint feature data of all pixels within the window and the preset sky prototype [10, 225] is calculated using the Euclidean distance formula. When the preset matching threshold is 45, the matching degree of this window is 42.8, which is less than the threshold, so it is marked as a candidate sky region. Subsequently, the adjacent windows x=112, y=192 with a matching degree of 43.2 and x=96, y=208 with a matching degree of 44.5 were detected. The distance between the center points of these three windows is 16 pixels, which is less than the merging threshold of 25 pixels, and the difference in matching degree is less than 1.0, which meets the merging condition. The system merges them into a larger key region with boundary coordinates of x=96 to 144 and y=192 to 224. The system continues to detect other regions of the image. The same method is used for the seawater region. The matching degree is calculated using the water body prototype [88, 73.825], and finally the complete key region division result is obtained. Among them, the sky region is x=96 to 144 and y=192 to 224, the water region is x=300 to 348 and y=400 to 448, etc. The distance between these regions is greater than 30 pixels and the feature differences are obvious, which meets the spatial separation condition.

[0183] In this embodiment of the application, the method achieves accurate positioning of key materials in natural scenes through feature fusion and multi-scale region analysis. By combining the joint judgment of optical characteristics and geographic information, it effectively improves the accuracy of region recognition in complex scenes. At the same time, it optimizes the spatial continuity of the recognition results through region merging.

[0184] To further improve the reliability assessment of the identification results, in some embodiments, step 105: assessing the confidence level of the key region and generating a key region confidence map includes:

[0185] Step 701: Calculate three confidence factors for each pixel within the key region, wherein the first confidence factor is the deviation value between the polarization feature and the material optical property table, the second confidence factor is the probability value of the pixel in the spatial probability distribution map, and the third confidence factor is the similarity of material features between adjacent pixels.

[0186] In step 701, the first confidence factor refers to the degree of matching between the current pixel's polarization features and the standard material characteristics, obtained by calculating the difference between the two. The second confidence factor reflects the probability of the pixel's occurrence in the geographical prior. The third confidence factor reflects the degree of feature consistency between the pixel and its surrounding pixels.

[0187] In this embodiment, the system first calculates the absolute difference between the pixel polarization gradient value and the standard value in the material property table as the first confidence factor, then reads the probability value of the pixel in the spatial probability distribution map as the second confidence factor, and finally analyzes the similarity of the feature values ​​of adjacent pixels as the third confidence factor.

[0188] Step 702: According to the preset weighting rules, the three confidence factors are fused into a pixel-level confidence score.

[0189] In step 702, the pixel-level confidence score is a reliability score obtained by weighted fusion of three factors, reflecting the credibility of the recognition result.

[0190] In this embodiment of the application, the system calculates the weighted sum of the three confidence factors according to a preset weight ratio, where the first factor has the highest weight, the second factor has the second highest weight, and the third factor has the lowest weight, and finally obtains the confidence score of each pixel.

[0191] Step 703: Generate a confidence map of the key region based on the pixel-level confidence level.

[0192] In this embodiment of the application, the system arranges the confidence values ​​of all pixels according to the image coordinate position to generate a grayscale image with the same size as the original image. The higher the confidence value, the higher the pixel brightness.

[0193] Here is a specific example:

[0194] In the confidence assessment process for the beach scene at location A, the system processes pixels at x=120 and y=220 within the identified key sky regions x=100 to 160 and y=200 to 240. First, the absolute deviation of the polarization gradient value 3 at this point from the standard sky gradient value 10 in the material property table is calculated as |3-10|=7, serving as the first confidence factor. Then, the sky probability value of 0.67 at this point in the spatial probability distribution map is read as the second confidence factor. Next, the average difference in polarization features between this point and its eight neighboring pixels is calculated as 0.2, serving as the third confidence factor. Following a preset weighting rule, the system sets the weight of the first confidence factor to 0.5, the second confidence factor to 0.3, and the third confidence factor to 0.2, performing a weighted calculation: 0.5×7+0.3×0.67+0.2×0.2=3.5+0.201+0.04=3.741. The value was linearly normalized to the range of 0 to 1, resulting in a final confidence score of 0.83 for the pixel. For the water region pixel x=310, y=430, the deviation of its gradient value 88 from the water standard value 90 (2) was calculated as the first confidence factor, the probability of reading water (0.25) as the second confidence factor, and the neighborhood difference (0.3) as the third confidence factor. The weighted calculation was 0.5×2+0.3×0.25+0.2×0.3=1+0.075+0.06=1.135, resulting in a normalized confidence score of 0.72. In the final confidence map, the sky region pixel values ​​were generally above 0.8, indicating high brightness; the water region pixel values ​​were around 0.7, indicating medium brightness; and the sandy region pixel values ​​were below 0.6, indicating low brightness.

[0195] In this embodiment, the method provides reliable quantitative quality indicators for the identification results through multi-dimensional confidence assessment and visualization, which not only helps in decision-making reference for subsequent processing steps, but also intuitively shows the performance of the algorithm in different regions, effectively improving the practicality and interpretability of the system output.

[0196] Figure 2 is a schematic diagram of a key area identification and processing system in an overhead power line mapping image provided in an embodiment of this application. As shown in Figure 2, the system includes:

[0197] The acquisition module 21 is used to acquire the raw image data captured by the camera and the geographical location data of the camera shooting location.

[0198] The matching module 22 is used to match the geographic location data with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera shooting location.

[0199] Extraction module 23 is used to extract polarization features and image content features from the original image data. Using a dual-branch attention network, a material optical property table is established based on the polarization features in the first branch processing path, and the natural material prior probability dataset is fused with the image content features in the second branch processing path.

[0200] The identification module 24 is used to identify key regions from the original image data based on the material optical property table and the fusion result.

[0201] The evaluation module 25 is used to evaluate the confidence level of the key area and generate a confidence map of the key area.

[0202] The key area identification and processing system in overhead power line surveying images shown in Figure 2 can execute the key area identification and processing method in overhead power line surveying images shown in the embodiment of Figure 1. Its implementation principle and technical effects will not be elaborated further. The specific operation methods of each module and unit in the key area identification and processing system in the overhead power line surveying images described in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated further here.

[0203] In one possible design, the key area identification and processing system in the overhead power line mapping image of the embodiment shown in FIG2 can be implemented as a computing device, as shown in FIG3. The computing device may include a storage component 31 and a processing component 32.

[0204] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0205] The processing component 32 is used to execute a key area identification processing method in an overhead power line mapping image according to the embodiment of FIG1 above.

[0206] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described method.

[0207] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0208] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0209] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0210] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0211] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0212] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can implement a method for identifying and processing key areas in an overhead power line mapping image as shown in Figure 1.

[0213] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0214] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0215] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0216] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for identifying and processing key areas in an image of an overhead power line survey, characterized in that, include: Acquire raw image data captured by the camera and geographic location data of the camera's shooting location; The geographic location data is matched with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera shooting location; polarization features and image content features are extracted from the original image data; using a dual-branch attention network, a material optical property table is established based on the polarization features in the first branch processing path; and the natural material prior probability dataset is fused with the image content features in the second branch processing path. Based on the material optical property table and the fusion result, key regions are identified from the original image data; confidence assessment is performed on the key regions to generate a key region confidence map; wherein, in the first branch processing path, a material optical property table is established based on the polarization features, and in the second branch processing path, the natural material prior probability dataset is fused with the image content features, including: in the first branch processing path, multi-angle polarization state decomposition is performed on the polarization features to obtain intensity components in different polarization directions; the intensity change gradient of the same material region in the polarization direction is calculated according to the intensity components; a material optical property table is established according to the intensity change gradient; in the second branch processing path, the natural material prior probability dataset is converted into a spatial probability distribution map; and the spatial probability distribution map is fused with the image content features. The process involves: first, generating visual features through element-level correlation; then, fusing the visual features with the spatial probability distribution map using a dynamic weighting mechanism to obtain a fusion result; and finally, identifying key regions from the original image data based on the material optical property table and the fusion result, which includes: concatenating the material optical property table with the fusion result to form joint feature data; setting multiple detection windows on the original image data and calculating the matching degree between the joint feature data and a preset material prototype within each detection window; marking the corresponding detection window as a candidate key region when the matching degree exceeds a preset matching degree threshold; performing continuous analysis and merging operations on spatially adjacent candidate key regions when multiple candidate key regions exist; repeating the continuous analysis and merging operations until all candidate key regions meet a preset spatial separation condition; and finally obtaining the key regions.

2. The method according to claim 1, characterized in that, The step of fusing the visual features with the spatial probability distribution map through a dynamic weighting mechanism to obtain a fusion result includes: calculating the probability value change amplitude of each pixel position in the spatial probability distribution map; analyzing the feature value fluctuation degree of the visual features at the same pixel position; dynamically assigning corresponding weight factors to the pixel probability values ​​of the spatial probability distribution map and the feature values ​​of the visual features according to the proportional relationship between the probability value change amplitude and the feature value fluctuation degree; and weighting the pixel probability values ​​and the feature values ​​according to the corresponding weight factors to generate a fused feature map, which is the fusion result.

3. The method according to claim 2, characterized in that, The step of weighting and combining the pixel probability values ​​and the feature values ​​according to the corresponding weighting factors to generate a fused feature map includes: applying a first weighting sub-factor to the pixel probability values ​​to generate weighted probability values; applying a second weighting sub-factor to the feature values ​​to generate weighted feature values; adding the weighted probability values ​​and weighted feature values ​​at the same pixel position to obtain fused feature values; and constructing a two-dimensional matrix structure based on the fused feature values ​​at all pixel positions, using the two-dimensional matrix structure as the fused feature map.

4. The method according to claim 1, characterized in that, The step of matching the geographic location data with a preset geographic location database to obtain a prior probability dataset of natural materials corresponding to the camera's shooting location includes: extracting GPS coordinates and shooting time from the geographic location data; determining the region code of the geographic grid region where the GPS coordinates are located from a preset set of geographic grid regions; retrieving historical material distribution records corresponding to the season corresponding to the shooting time from the geographic location database; comparing and matching the region code with the region markers in the historical material distribution records, and extracting the occurrence frequency data of all natural materials from the historical material distribution records based on the comparison and matching results; calculating the probability weight values ​​of key natural materials based on the occurrence frequency data, wherein the key natural materials include the sky, water, and vegetation; and generating a prior probability dataset of natural materials based on the probability weight values.

5. The method according to claim 1, characterized in that, The step of evaluating the confidence level of the key region and generating a confidence map of the key region includes: calculating three confidence factors for each pixel within the key region, wherein the first confidence factor is the deviation value between the polarization feature and the material optical property table, the second confidence factor is the probability value of the pixel in the spatial probability distribution map, and the third confidence factor is the similarity of material features between adjacent pixels; fusing the three confidence factors into a pixel-level confidence level according to a preset weighting rule; and generating a confidence map of the key region based on the pixel-level confidence level.

6. A key area identification and processing system in an overhead power line mapping image, characterized in that, include: The acquisition module is used to acquire raw image data captured by the camera and geographic location data of the camera's shooting location; A matching module is used to match the geographic location data with a preset geographic location database to obtain a natural material prior probability dataset corresponding to the camera shooting location. An extraction module is used to extract polarization features and image content features from the original image data. Using a dual-branch attention network, in the first branch processing path, a material optical property table is established based on the polarization features. In the second branch processing path, the natural material prior probability dataset is fused with the image content features. Specifically, in the first branch processing path, the polarization features are decomposed into multi-angle polarization states to obtain intensity components in different polarization directions. Based on the intensity components, the intensity change gradient of the same material region in the polarization direction is calculated. Based on the intensity change gradient, a material optical property table is established. In the second branch processing path, the natural material prior probability dataset is converted into a spatial probability distribution map. The spatial probability distribution map is then correlated with the image content features at the pixel level to generate visual features. A dynamic weighting mechanism is used to fuse the visual features with the spatial probability distribution map to obtain a fusion result. The recognition module is used to identify key regions from the original image data based on the material optical property table and the fusion result. Specifically, it is used to perform feature stitching between the material optical property table and the fusion result to form joint feature data. Multiple detection windows are set on the original image data, and the matching degree between the joint feature data in each detection window and a preset material prototype is calculated. When the matching degree exceeds a preset matching degree threshold, the corresponding detection window is marked as a candidate key region. When multiple candidate key regions exist, perform continuity analysis and merging operations on spatially adjacent candidate key regions; Repeat the continuity analysis and merging operation until all candidate key regions meet the preset spatial separation conditions; Obtain the key area; The evaluation module is used to evaluate the confidence level of the key area and generate a confidence map of the key area.

7. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the key area identification and processing method in overhead power line mapping images as described in any one of claims 1-5.

8. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a method for identifying and processing key areas in an overhead power line mapping image as described in any one of claims 1-5.

Images