Intelligent tower detection method and satellite image processing system

Abstract

The invention relates to the technical field of intelligent operation inspection of power transmission lines, in particular to an intelligent tower detection method and a satellite image processing system, and the method comprises the steps: obtaining tower space information through constructing a tower information database, carrying out the complexity judgment of the tower space information, and carrying out the labeling processing of the tower space information according to a judgment result, the method comprises the following steps: acquiring tower marking data, training a tower identification model according to the acquired tower marking data, then performing correction fusion on tower detection information, acquiring correction detection information, identifying the correction detection information according to the tower identification model, and outputting a tower identification result; through the method, the problems of low identification precision, low efficiency and high marking cost of a traditional tower identification method in a large-scale and high-complexity scene are effectively solved, the tower detection precision is improved, the adaptability of a model in a complex environment is enhanced, and the intelligent level and the operation and maintenance efficiency of electric power inspection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of transmission lines, and particularly to an intelligent detection method for poles and towers and a satellite image processing system. Background Art

[0002] With the rapid development of the power system, the scale of transmission lines has been continuously expanding. The traditional manual inspection method has problems such as low efficiency, high cost, and high danger, and it is difficult to meet the efficient and accurate operation and maintenance requirements of modern power grids. With the rapid development of satellite remote sensing technology, the technology of detecting poles and towers based on satellite remote sensing has emerged. Using the data information provided by satellites and artificial intelligence algorithms, it is possible to achieve automatic and intelligent identification and monitoring of transmission line poles and towers.

[0003] However, the existing methods for detecting poles and towers based on satellite remote sensing still face many challenges. First, the shapes and spatial distributions of poles and towers are complex and diverse. Especially in areas with complex terrain or dense poles and towers, the difficulty of identifying poles and towers increases significantly. Second, the data information is affected by factors such as atmospheric conditions, light changes, and geometric distortions, resulting in unstable single-source data information, which in turn affects the accuracy of pole and tower detection.

[0004] The information disclosed in this background art section is only intended to deepen the understanding of the overall background art of the present disclosure, and should not be regarded as an admission or any form of implication that this information constitutes the prior art known to those skilled in the art. Summary of the Invention

[0005] The present invention provides an intelligent detection method for poles and towers and a satellite image processing system, which can effectively solve the problems in the background art.

[0006] In order to achieve the above object, the technical solutions adopted by the present invention are as follows: Construct a pole and tower information database, and obtain the spatial information of poles and towers according to the pole and tower information database; Judge the complexity of the pole and tower spatial information, perform annotation processing on the pole and tower spatial information according to the complexity judgment result, and obtain pole and tower annotation data; Construct a pole and tower recognition model, and train the pole and tower recognition model according to the pole and tower annotation data; Collect pole and tower detection information, perform calibration and fusion on the pole and tower detection information to obtain calibrated detection information, and identify the calibrated detection information according to the trained pole and tower recognition model, and output a pole and tower recognition result.

[0007] Further, identifying the calibrated detection information according to the trained pole and tower recognition model and outputting a pole and tower recognition result includes: Perform instance segmentation on the corrected detection information according to the trained tower identification model to obtain a number of image detection results. The image detection results include mask information, and the image detection results correspond one-to-one with the mask information; Perform class screening on a number of the image detection results, determine whether the image detection results are towers, and extract the tower contours according to the mask information for the tower judgment results to obtain a number of tower image data; Calculate the confidence levels of a number of the tower image data respectively, and filter the tower image data based on the confidence levels to obtain a pixel-level segmentation result; Identify the tower detection information according to the pixel-level segmentation result to obtain a tower identification result.

[0008] Furthermore, calculating the confidence levels of a number of the tower image data respectively includes: Calculate the class probabilities of a number of the tower image data respectively. The class probabilities are used to represent the possibility that the detection results belong to the tower class; Calculate the corresponding target scores according to the tower image data. The target scores indicate whether the tower image data contains real tower objects; Perform weighted combination according to the class probabilities and the target scores to obtain the confidence levels of the tower image data. The confidence levels are used to measure the credibility of the tower image data as the tower.

[0009] Furthermore, perform annotation processing on the tower spatial information according to the complexity determination result to obtain tower annotation data, including: Obtain regular tower information according to the tower spatial information; Determine the boundary features of each tower according to the regular tower information, and generate an initial boundary block diagram corresponding to the tower; Perform adaptive adjustment on the initial rectangular block diagram according to the structure and spatial distribution of the tower, and perform class annotation to obtain an annotated boundary block diagram; Convert the format of the annotated boundary block diagram to obtain tower annotation data.

[0010] Furthermore, perform annotation processing on the tower spatial information according to the complexity determination result to obtain tower annotation data, including: Obtain special-shaped tower information according to the tower spatial information. The special-shaped tower information includes overlapping and irregular tower images; Generate a number of initial masks according to the special-shaped tower information, extract the tower boundaries respectively according to the number of the initial masks, and optimize the mask contours based on the tower boundaries to obtain a number of optimized masks; Perform overlapping detection on several of the optimized masks, calculate the mask overlapping pixels, determine the mask overlapping state based on the mask overlapping pixels, and adjust the mask boundaries based on the mask overlapping state; Convert the format of the adjusted optimized mask to obtain pole and tower annotation data.

[0011] Further, perform calibration and fusion on the pole and tower detection information to obtain calibrated detection information, including: Obtain the sensor calibration coefficient according to the remote sensing satellite, and obtain the radiance according to the sensor calibration coefficient and the pole and tower detection information; Construct an atmospheric correction model, correct atmospheric scattering and atmospheric absorption according to the atmospheric correction model, and correct the solar altitude angle; Calculate the true reflectance of the ground object according to the radiance, perform consistency adjustment on the illumination conditions according to the true reflectance of the ground object, and obtain a radiometrically corrected image; Perform geometric correction on the radiometrically corrected image to obtain several geometrically corrected images, and perform image fusion on the several geometrically corrected images to obtain calibrated detection information.

[0012] Further, perform geometric correction on the radiometrically corrected image to obtain several geometrically corrected images, including: Extract geometric distortion information from the radiometrically corrected image, select ground control points based on the geometric distortion information, and obtain geometric deviations according to the radiometrically corrected image and the ground control points; Construct a geometric transformation matrix according to the geometric deviations, perform image reprojection according to the geometric transformation matrix to obtain a reprojected image, and the image reprojection is used to unify the geographic coordinates; Perform accuracy verification on the reprojected image to obtain geometrically corrected images.

[0013] Further, construct a pole and tower information database, including: Collect historical remote sensing image information, and extract historical pole and tower images and historical pole and tower spatial information based on the historical remote sensing image information; Preprocess the historical pole and tower images, and correspond the preprocessed historical pole and tower images with the historical pole and tower spatial information one by one to construct a pole and tower spatial mapping relationship; Construct an index for the pole and tower spatial mapping relationship to establish a pole and tower information database.

[0014] A satellite image processing system, the system includes: A pole and tower database construction module, which constructs a pole and tower information database and obtains pole and tower spatial information according to the pole and tower information database; The tower marking optimization module determines the complexity of the tower spatial information, performs marking processing on the tower spatial information according to the complexity determination result, and obtains tower marking data; The tower model training module constructs a tower recognition model and trains the tower recognition model according to the tower marking data; The remote sensing image recognition module collects tower detection information, corrects and fuses the tower detection information to obtain corrected detection information, and recognizes the corrected detection information according to the trained tower recognition model, and outputs a tower recognition result.

[0015] Furthermore, the tower marking optimization module includes: The regular tower extraction unit obtains regular tower information according to the tower spatial information; The tower boundary generation unit determines the boundary features of each tower according to the regular tower information and generates an initial boundary block diagram corresponding to the tower; The boundary optimization marking unit adaptively adjusts the initial rectangular block diagram according to the structure and spatial distribution of the tower, and performs category marking to obtain a marked boundary block diagram; The marking data conversion unit converts the format of the marked boundary block diagram to obtain tower marking data.

[0016] Through the technical solution of the present invention, the following technical effects can be achieved: Effectively solves the problems of low recognition accuracy and low efficiency of traditional methods in large-scale and high-complexity scenarios. By constructing a tower information database, combining instance segmentation and confidence calculation of deep learning to optimize the recognition result, the accuracy of tower detection is improved; adopting a complexity determination strategy, separately optimizing the marking for regular towers and special-shaped towers, enhancing the adaptability of the model in complex environments; using radiation correction and geometric correction methods to improve the processing accuracy of remote sensing images, ensuring the accuracy of tower position information. Overall, this method realizes efficient, accurate and automated tower recognition, improving the intelligent level and operation and maintenance efficiency of power inspection.

[0017] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following specifically gives the specific implementation manners of this application. Brief Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a schematic flow diagram of the intelligent detection method for the pole tower; Figure 2 It is a schematic flow diagram for obtaining the pole tower identification result; Figure 3 It is a schematic flow diagram for obtaining the confidence level of the pole tower image; Figure 4 It is a schematic architecture diagram for obtaining the pole tower identification result. Detailed implementation manners

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the description of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0022] Embodiment 1; As Figure 1 shown, the present application provides an intelligent detection method for a pole tower, and the method includes: S100: Construct a pole tower information database, and obtain the pole tower spatial information according to the pole tower information database; S200: Determine the complexity of the pole tower spatial information, perform annotation processing on the pole tower spatial information according to the complexity determination result, and obtain the pole tower annotation data; S300: Construct a pole tower recognition model, and train the pole tower recognition model according to the pole tower annotation data; S400: Collect the pole tower detection information, perform calibration and fusion on the pole tower detection information to obtain the calibrated detection information, and identify the calibrated detection information according to the trained pole tower recognition model, and output the pole tower recognition result.

[0023] Specifically, first, a pole and tower information database is established to store the spatial information of pole and towers. The stored information includes: the geographical location information of pole and towers, the historical detection data of pole and towers, the pole and tower category information (such as straight pole and tower, corner pole and tower, terminal pole and tower, etc.), the surrounding environment information of pole and towers (such as terrain features, vegetation coverage), and other relevant pole and tower characteristic data. Subsequently, the complexity of the pole and tower spatial information is determined to judge the regularity of the data information related to pole and towers. In some embodiments, shape matching technology and scale-invariant feature transform are applied to detect the regularity of the pole and tower shape and the distribution of key points. Finally, spatial relationship analysis is carried out. The distance and arrangement between pole and towers can be analyzed using GIS tools, and clustering algorithms can be used to detect the clustering pattern, so as to effectively evaluate the complexity of the pole and tower data information. Then, annotation is carried out according to the complexity determination result to obtain the pole and tower annotation data. On this basis, a pole and tower recognition model can be constructed based on deep learning technology, such as convolutional neural network, and the annotation data is used for deep learning training. The training process includes steps such as data augmentation, feature extraction, object detection and instance segmentation optimization. In at least one embodiment, transfer learning technology is adopted to enable the model to adapt to remote sensing images with different resolutions, improve the generalization ability of the model, and optimize the object positioning accuracy in combination with the intersection over union loss. Finally, the corrected detection information is detected by the trained pole and tower recognition model, and the pole and tower recognition result is output.

[0024] Through the technical solution of the present invention, the problems of low recognition accuracy and low efficiency of traditional methods in large-scale and high-complexity scenarios are effectively solved. By constructing a pole and tower information database, the accuracy of pole and tower detection is improved; a complexity determination strategy is adopted to optimize the annotation for regular pole and towers and special-shaped pole and towers respectively, enhancing the adaptability of the model in complex environments; radiation correction and geometric correction methods are used to improve the processing accuracy of remote sensing images, ensuring the accuracy of the pole and tower position information. Overall, this method realizes efficient, accurate and automatic pole and tower recognition, improving the intelligent level and operation and maintenance efficiency of power inspection.

[0025] Furthermore, as Figure 2 and Figure 4 shown, the corrected detection information is recognized according to the trained pole and tower recognition model, and the pole and tower recognition result is output, including: S410: Perform instance segmentation on the corrected detection information according to the trained pole and tower recognition model to obtain a number of image detection results. The image detection results contain mask information, and the image detection results correspond to the mask information one by one; S420: Perform category screening on a number of image detection results to judge whether the image detection results are pole and towers, and extract the pole and tower contours according to the mask information for the pole and tower judgment results to obtain a number of pole and tower image data; S430: Calculate the confidence levels of several pole tower image data respectively, filter the pole tower image data based on the confidence levels, and obtain the pixel-level segmentation result; S440: Identify the pole tower detection information according to the pixel-level segmentation result, and obtain the pole tower identification result.

[0026] As an optimization of the above embodiment, first, perform instance segmentation on the calibrated detection information according to the trained pole tower identification model. The instance segmentation can be carried out in the following manner: preprocess the collected satellite remote sensing image, including adjusting the image size and performing color correction to meet the input requirements of the model. Then, input the processed image into the trained pole tower identification model. The model first identifies potential pole tower candidate regions in the image through the region proposal network. For each candidate region, the model further uses a classifier to determine whether it is a pole tower and precisely adjusts its position through a bounding box regressor. At the same time, use the mask branch in the pole tower identification model to generate the corresponding pixel-level mask; the pixel-level mask is a two-dimensional array, where the value of each pixel point is used to distinguish whether the pixel belongs to the pole tower object. For example, the area with a pixel value of 1 (or other specific value) represents the pixel range of the pole tower, while the area with a pixel value of 0 represents the background or non-pole tower area. The pixel-level mask can accurately describe the contour of the pole tower. Finally, optimize and refine the mask by applying non-maximum suppression and morphological operations, and output the final pole tower detection result and segmentation result; then, by analyzing the image detection result, the morphological features of each pole tower instance can be calculated, including shape integrity, size ratio, edge sharpness, etc., and matched with the standard pole tower structure; if the shape of a certain detection result seriously deviates from the normal structure of the pole tower (such as being too slender, having too small or too large an area, etc.), then this result may be a false detection and needs to be further reviewed or directly excluded to obtain several pole tower image data; afterwards, calculate the confidence levels of several pole tower image data respectively, filter the pole tower image data based on the confidence levels, obtain the pixel-level segmentation result, and identify the pole tower detection information according to the pixel-level segmentation result to obtain the pole tower identification result.

[0027] Furthermore, as Figure 3 shown, calculating the confidence levels of several pole tower image data respectively includes: S431: Calculate the class probabilities of several pole tower image data respectively. The class probabilities are used to represent the possibility that the detection result belongs to the pole tower class; S432: Calculate the corresponding target scores according to the pole tower image data. The target scores represent whether the pole tower image data contains real pole tower objects; S433: Perform weighted combination according to the class probabilities and the target scores to obtain the confidence levels of the pole tower image data. The confidence levels are used to measure the credibility of the pole tower image data being a pole tower.

[0028] In this embodiment, first, the class probabilities of several pole and tower image data can be obtained based on a deep learning network. For example, in object detection models such as Mask R-CNN or YOLO, after the network recognizes a pole and tower object, it outputs a class probability value indicating the likelihood that the object belongs to the pole and tower category. The class probability is usually output by the softmax layer of the deep learning model and ranges from 0 to 1. The higher the class probability, the greater the likelihood that the detection result belongs to the pole and tower. Subsequently, the corresponding target score is calculated based on the pole and tower image data. The target score can be calculated based on multiple factors, including: whether the target morphological features (the geometric structure of the pole and tower, such as height, shape, contour, etc.) conform to the pole and tower features, whether the target texture features (such as the edges of the pole and tower, material reflection characteristics, etc.) conform to the pole and tower texture patterns, and whether the confidence of the target area is higher than a set threshold. The calculation of the target score can adopt a target feature extraction method based on a convolutional neural network to extract the morphological and texture features of the pole and tower, and calculate the target score through a classification network (such as an SVM or a fully connected layer). Then, the confidence is calculated by calculating the obtained class probability and the target score. The calculation formula is as follows: , is the confidence of the pole and tower image data, is the class probability, is the target score, , is the weight coefficient, where the weight coefficient is set based on historical pole and tower recognition results; the confidence of the pole and tower image data is obtained according to the calculation formula.

[0029] Furthermore, according to the complexity determination result, the pole and tower spatial information is labeled and processed to obtain pole and tower labeling data, including: Obtain regular pole and tower information according to the pole and tower spatial information; According to the regular pole and tower information, determine the boundary features of each pole and tower, and generate an initial boundary block diagram corresponding to the pole and tower; Make an adaptive adjustment to the initial rectangular block diagram according to the structure and spatial distribution of the pole and tower, and perform class labeling to obtain a labeled boundary block diagram; Convert the format of the labeled boundary block diagram to obtain pole and tower labeling data.

[0030] Specifically, first, according to the pole tower spatial information and complexity judgment rules, regular pole tower images are determined. After obtaining the regular pole tower information, it is necessary to generate an initial bounding box for each pole tower, that is, determine its circumscribed contour in the image. Then, based on the center coordinates of the pole tower, analyze the spatial footprint of the pole tower, and obtain the outer contour of the pole tower through image segmentation (such as DeepLabV3, U-Net) or edge detection (Canny, Sobel), calculate the minimum circumscribed rectangle of the pole tower, and obtain the initial bounding box of the pole tower. For different types of pole towers, different initial bounding boxes need to be set. For example, transmission towers: usually have a quadrilateral structure, and the minimum circumscribed rectangle can be directly used. Cement poles: due to their large aspect ratio, the size of the bounding box needs to be adjusted according to the height of the pole tower. Subsequently, based on the pole tower spatial information, obtain the pole tower spatial distribution and pole tower structure. If the pole towers are densely distributed, it is necessary to avoid overlapping of the bounding boxes, and DBSCAN clustering can be used to optimize the distance between the bounding boxes. If the pole towers exist in isolation, the bounding boxes are appropriately increased to ensure complete coverage of the pole tower contour, thereby obtaining the bounding box diagram. Then, based on the bounding box diagram of the pole tower, category labeling of the pole tower is performed: the following method can be used for category labeling: calculate the aspect ratio of the pole tower, then use an edge detection algorithm to extract the top contour of the pole tower, and finally combine the structure information of the pole tower to perform category labeling of the pole tower to obtain the labeled bounding box diagram of the pole tower. Finally, in order to adapt to the training requirements of different models, it is necessary to convert the annotation format, and the format conversion is usually implemented through Python or annotation tools (such as LabelImg, Roboflow) to obtain the pole tower annotation data.

[0031] Furthermore, according to the complexity judgment result, the pole tower spatial information is annotated and processed to obtain the pole tower annotation data, including: Obtain the information of special-shaped pole towers according to the pole tower spatial information, and the information of special-shaped pole towers includes overlapping and irregular pole tower images; Generate several initial masks according to the information of special-shaped pole towers, extract the boundaries of the pole towers respectively based on the several initial masks, and optimize the mask contours based on the pole tower boundaries to obtain several optimized masks; Perform overlapping detection on the several optimized masks, calculate the overlapping pixels of the masks, determine the mask overlapping state according to the overlapping pixels of the masks, and adjust the mask boundaries based on the mask overlapping state; Convert the format of the adjusted optimized masks to obtain the pole tower annotation data.

[0032] As a preference of the above embodiments, it is first necessary to obtain special-shaped tower data based on the spatial information of the towers. The special-shaped towers include overlapping towers and tower images with irregular shapes. Since the structures of the special-shaped towers are complex and the image representation forms are diverse, it is impossible to directly use the conventional bounding box annotation method for processing. Preferably, an image segmentation method, such as the GrabCut algorithm based on edge detection, is used to obtain the optimized mask of the tower. After obtaining the optimized mask, overlapping detection is further performed to identify whether there is an overlap of multiple tower mask regions. Specifically, the overlapping area between each mask is calculated, and the number of overlapping pixels of the masks is counted. If the number of pixels in the overlapping area accounts for a large proportion, it is determined that there may be overlapping targets in the tower area and segmentation optimization is required. For this purpose, a mask segmentation method, such as an image segmentation method based on the morphological watershed algorithm or K-means clustering, can be used to split the mask of the overlapping area so that the mask area of each tower is more independent. In addition, the geometric features of the tower can be combined, and by analyzing the height of the tower, the image illumination characteristics, and the spatial distribution pattern of the tower, the mask segmentation effect can be further optimized to ensure that the tower targets are not mis-merged or split. Finally, after the mask optimization is completed, it is necessary to perform format conversion to adapt to the requirements of subsequent model training or database storage. Specifically, the optimized tower mask can be converted into a standard annotation format, such as the COCO format (a general annotation format for deep learning object detection and segmentation tasks) or the GeoJSON format (suitable for spatial data storage and processing in GIS systems). This format conversion process includes steps such as normalizing the mask coordinates, converting the mask to the standard image size, and generating polygon contour data to ensure the reusability and compatibility of the annotation data and obtain the tower annotation data.

[0033] Furthermore, the tower detection information is corrected and fused to obtain the corrected detection information, including: Obtain the sensor calibration coefficient according to the remote sensing satellite, and obtain the radiance according to the sensor calibration coefficient and the tower detection information; Construct an atmospheric correction model, correct the atmospheric scattering and atmospheric absorption according to the atmospheric correction model, and correct the solar altitude angle; Calculate the true reflectance of the ground object according to the radiance, and perform consistency adjustment on the illumination conditions according to the true reflectance of the ground object to obtain the radiance-corrected image; Perform geometric correction on the radiance-corrected image to obtain a number of geometrically corrected images, and fuse the number of geometrically corrected images to obtain the corrected detection information.

[0034] In this embodiment, first, according to the sensor characteristics of the remote sensing satellite, the sensor calibration coefficients are obtained, including parameters such as the spectral response characteristics, gain, and offset value of the sensor. These calibration coefficients are used to perform radiometric calibration on the pole and tower detection information: converting the digital values of the image into radiance values. The formula is as follows: , where \(L\) is the radiance, and are the gain and offset coefficients of the sensor respectively, is the pixel value of the original image, and the pixel value of the original image can be obtained from the pole and tower detection information; through this conversion, the influence of the sensor characteristics of the remote sensing image is eliminated, making the radiance values of different images comparable; subsequently, for the atmospheric interference of the remote sensing image, an atmospheric correction model is constructed to eliminate the influence of atmospheric scattering and absorption on the image spectrum. The 6S radiative transfer model can be used as the atmospheric correction model; based on the atmospheric correction model, first calculate the influence of atmospheric scattering on the image, and remove the influence of aerosol scattering through radiometric calibration, and correct the solar altitude angle. Subsequently, according to the obtained radiance, convert the radiance into the top-of-atmosphere reflectance. The top-of-atmosphere reflectance still contains the influence of atmospheric scattering and absorption. Through the atmospheric correction model, combined with the aerosol type, atmospheric water vapor content, observation geometric parameters, etc. collected by the remote sensing satellite, calculate the atmospheric transmittance and atmospheric path radiance, and calculate the true reflectance of the ground object. Since the reflectance of the ground object should be consistent at different times and in different images, it is also necessary to perform illumination consistency adjustment: histogram matching can be used to adjust the histogram of the image to be processed to the histogram distribution of the reference image to make its brightness and contrast consistent, and complete the consistency adjustment to obtain the radiometrically calibrated image. Subsequently, perform geometric calibration on the radiometrically calibrated image to obtain several geometrically calibrated images, and fuse the several geometrically calibrated images to obtain the calibrated detection information.

[0035] Furthermore, performing geometric calibration on the radiometrically calibrated image to obtain several geometrically calibrated images includes: extracting geometric distortion information from the radiometrically calibrated image, selecting ground control points based on the geometric distortion information, and obtaining geometric deviations according to the radiometrically calibrated image and the ground control points; constructing a geometric transformation matrix according to the geometric deviations, performing image reprojection according to the geometric transformation matrix to obtain a reprojected image, and image reprojection is used to unify the geographic coordinates; performing accuracy verification on the reprojected image to obtain geometrically calibrated images.

[0036] Specifically, first, extract geometric distortion information to identify the geometric deformation suffered by the remote sensing image during the imaging process. Geometric distortion usually stems from factors such as sensor distortion, platform movement, and terrain influence. Sensor distortion can be obtained from the sensor geometric parameters provided by the satellite (such as focal length, principal point offset, tilt angle, etc.). Platform movement distortion results from the attitude changes during satellite or drone image capture and can be extracted from the attitude angles (pitch angle, yaw angle, and roll angle) recorded in the satellite ephemeris data or inertial navigation system (INS). Terrain influence distortion requires combining digital elevation model (DEM) data to analyze the impact of terrain undulation on the geometric shape of the image. After extracting the geometric distortion information, select ground control points according to the high-precision map or GIS database. The ground control points are the reference points for geometric correction. Next, calculate the geometric deviation, that is, the deviation between the ground control points in the remote sensing image and their corresponding true geographical coordinates. Preferably, calibrate the pixel coordinates of the ground control points in the image coordinate system and obtain the geographical longitude and latitude of the corresponding ground control points in the geographical coordinate system, and calculate the error between the image coordinates and the geographical coordinates. These errors constitute the geometric deviation data. Subsequently, based on the calculated geometric deviation, construct a geometric transformation matrix for converting the original image coordinates to the standard geographical coordinate system. An affine transformation or a projection transformation can be used to construct the geometric transformation matrix. Then, according to the constructed transformation matrix, perform image reprojection, that is, remap the pixels of the remote sensing image to the standard geographical coordinate system. Image reprojection usually adopts nearest neighbor interpolation, bilinear interpolation, or cubic convolution interpolation to maintain image quality and spatial accuracy. After completing the reprojection, the root mean square error can be used for accuracy verification to obtain the geometrically corrected image.

[0037] Furthermore, construct a pole and tower information database, including: Collect historical remote sensing image information, and extract historical pole and tower images and historical pole and tower spatial information based on the historical remote sensing image information; Preprocess the historical pole and tower images, and correspond the preprocessed historical pole and tower images with the historical pole and tower spatial information one by one to construct a pole and tower spatial mapping relationship; Construct an index for the pole and tower spatial mapping relationship to establish a pole and tower information database.

[0038] As a preference of the above embodiments, first, historical remote sensing images are collected from the historical image library. Multiple periods of high-resolution remote sensing satellite image data covering different seasons and time periods without cloud or snow cover are selected under different solar altitude angles and azimuth angles. After obtaining the historical remote sensing images, deep learning object detection technologies (such as Faster R-CNN, YOLO) are used, combined with GIS data analysis, to preliminarily detect the pole and tower targets in the images and extract the pole and tower image data. At the same time, combined with the GIS spatial data of the pole and tower (such as the longitude, latitude coordinates and elevation information of the pole and tower), the spatial information of the historical pole and tower is obtained. Due to different sources of historical remote sensing images, problems such as inconsistent resolution, different lighting conditions, and geometric distortion of the images may exist. Therefore, data preprocessing is required, specifically including preprocessing such as radiometric correction, geometric correction and image fusion of the historical pole and tower images. After completing the image preprocessing, the nearest neighbor matching algorithm can be used to find the closest pole and tower position information in the GIS database according to the central coordinates of the pole and tower images, and a mapping relationship can be established. A relational database PostgreSQL + PostGIS or a NoSQL database can be used to construct the database.

[0039] Embodiment 2; Based on the same inventive concept as an intelligent detection method for a pole and tower and a satellite image processing system in the foregoing embodiments, the present invention also provides a satellite image processing system, which includes: A pole and tower database construction module that constructs a pole and tower information database and obtains the spatial information of the pole and tower according to the pole and tower information database; A pole and tower annotation optimization module that determines the complexity of the pole and tower spatial information and performs annotation processing on the pole and tower spatial information according to the complexity determination result to obtain pole and tower annotation data; A pole and tower model training module that constructs a pole and tower recognition model and trains the pole and tower recognition model according to the pole and tower annotation data; A remote sensing image recognition module that collects pole and tower detection information, corrects and fuses the pole and tower detection information to obtain corrected detection information, and recognizes the corrected detection information according to the trained pole and tower recognition model to output a pole and tower recognition result.

[0040] The above adjustment system in the present invention can effectively implement an intelligent detection method for a pole and tower and a satellite image processing system, and the technical effects that can be achieved are as described in the above embodiments and will not be elaborated here.

[0041] Furthermore, the pole and tower annotation optimization module includes: A regular pole and tower extraction unit that obtains regular pole and tower information according to the pole and tower spatial information; A pole and tower boundary generation unit that determines the boundary features of each pole and tower according to the regular pole and tower information and generates an initial boundary block diagram corresponding to the pole and tower; Boundary optimization annotation unit, adaptively adjusts the initial rectangular block diagram according to the structure and spatial distribution of the pole tower, and performs category annotation to obtain an annotated boundary block diagram; Annotation data conversion unit, converts the format of the annotated boundary block diagram to obtain pole tower annotation data.

[0042] Similarly, for the above optimization solutions of the system, the corresponding optimization effects of the methods in Embodiment 1 can also be respectively achieved, and will not be elaborated here again.

[0043] Although the present application has been described in connection with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present application. Accordingly, the present specification and the drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. An intelligent detection method for a pole tower, characterized in that, The method includes: Construct a pole and tower information database, and obtain the spatial information of the pole and tower according to the pole and tower information database; Judge the complexity of the pole and tower spatial information, perform annotation processing on the pole and tower spatial information according to the complexity judgment result, and obtain pole and tower annotation data; Construct a pole and tower recognition model, and train the pole and tower recognition model according to the pole and tower annotation data; Collect pole and tower detection information, perform calibration and fusion on the pole and tower detection information to obtain calibrated detection information, and identify the calibrated detection information according to the trained pole and tower recognition model to output a pole and tower recognition result.

2. The intelligent detection method of the pole tower according to claim 1, wherein, Identifying the calibrated detection information according to the trained pole and tower recognition model and outputting a pole and tower recognition result includes: Performing instance segmentation on the calibrated detection information according to the trained pole and tower recognition model to obtain a plurality of image detection results, the image detection results include mask information, and the image detection results correspond to the mask information one by one; Perform category screening on a plurality of the image detection results, judge whether the image detection results are poles and towers, and extract the pole and tower contours according to the mask information for the pole and tower judgment results to obtain a plurality of pole and tower image data; Calculate the confidence levels of the plurality of pole and tower image data respectively, and filter the pole and tower image data based on the confidence levels to obtain a pixel-level segmentation result; Identify the pole and tower detection information according to the pixel-level segmentation result to obtain a pole and tower recognition result.

3. The intelligent detection method for the pole tower according to claim 2, wherein, Calculating the confidence levels of the plurality of pole and tower image data respectively includes: Calculate the class probabilities of the plurality of pole and tower image data respectively, and the class probabilities are used to represent the possibility that the detection results belong to the pole and tower category; Calculate the corresponding target scores according to the pole and tower image data, and the target scores indicate whether the pole and tower image data contains real pole and tower objects; Perform weighted combination according to the class probability and the target score to obtain the confidence level of the pole and tower image data, and the confidence level is used to measure the credibility of the pole and tower image data as the pole and tower.

4. The intelligent detection method of the pole tower according to claim 1, characterized in that, Performing annotation processing on the pole and tower spatial information according to the complexity judgment result to obtain pole and tower annotation data includes: Obtain regular pole and tower information according to the pole and tower spatial information; According to the regular pole and tower information, determine the boundary features of each pole and tower, and generate an initial boundary block diagram corresponding to the pole and tower; Perform adaptive adjustment on the initial rectangular block diagram according to the structure and spatial distribution of the pole and tower, and perform category annotation to obtain an annotated boundary block diagram; Convert the format of the annotated boundary block diagram to obtain pole and tower annotation data.

5. The intelligent detection method for the pole tower according to claim 1, characterized in that, Performing annotation processing on the pole and tower spatial information according to the complexity judgment result to obtain pole and tower annotation data includes: Obtain special-shaped pole and tower information according to the pole and tower spatial information, and the special-shaped pole and tower information includes overlapping and irregular pole and tower images; Generate a plurality of initial masks according to the special-shaped pole and tower information, extract the pole and tower boundaries respectively according to the plurality of initial masks, and optimize the mask contours based on the pole and tower boundaries to obtain a plurality of optimized masks; Perform overlapping detection on a number of the optimized masks, calculate the mask overlapping pixels, determine the mask overlapping status based on the mask overlapping pixels, and adjust the mask boundaries based on the mask overlapping status; Convert the format of the adjusted optimized mask to obtain pole tower annotation data.

6. The intelligent detection method of the pole tower according to claim 1, wherein Perform calibration and fusion on the pole tower detection information to obtain calibrated detection information, including: Obtain the sensor calibration coefficient according to the remote sensing satellite, and obtain the radiance according to the sensor calibration coefficient and the pole tower detection information; Construct an atmospheric correction model, correct atmospheric scattering and atmospheric absorption according to the atmospheric correction model, and correct the solar altitude angle; Calculate the true reflectance of the ground object according to the radiance, perform consistency adjustment on the lighting conditions according to the true reflectance of the ground object, and obtain a radiometrically corrected image; Perform geometric correction on the radiometrically corrected image to obtain a number of geometrically corrected images, and fuse the number of geometrically corrected images to obtain calibrated detection information.

7. The intelligent detection method of the pole tower according to claim 6, wherein Perform geometric correction on the radiometrically corrected image to obtain a number of geometrically corrected images, including: Extract geometric distortion information from the radiometrically corrected image, select ground control points based on the geometric distortion information, and obtain geometric deviations according to the radiometrically corrected image and the ground control points; Construct a geometric transformation matrix according to the geometric deviations, perform image reprojection according to the geometric transformation matrix to obtain a reprojected image, and the image reprojection is used to unify the geographic coordinates; Perform accuracy verification on the reprojected image to obtain geometrically corrected images.

8. The intelligent detection method for the pole tower according to claim 1, characterized in that, Construct a pole tower information database, including: Collect historical remote sensing image information, and extract historical pole tower images and historical pole tower spatial information based on the historical remote sensing image information; Preprocess the historical pole tower images, and correspond the preprocessed historical pole tower images and the historical pole tower spatial information one by one to construct a pole tower spatial mapping relationship; Construct an index for the pole tower spatial mapping relationship to establish a pole tower information database.

9. A satellite image processing system, characterized in that, The system includes: A pole tower database construction module that constructs a pole tower information database and obtains pole tower spatial information according to the pole tower information database; A pole tower annotation optimization module that determines the complexity of the pole tower spatial information, performs annotation processing on the pole tower spatial information according to the complexity determination result, and obtains pole tower annotation data; A pole tower model training module that constructs a pole tower recognition model and trains the pole tower recognition model according to the pole tower annotation data; A remote sensing image recognition module that collects pole tower detection information, performs calibration and fusion on the pole tower detection information to obtain calibrated detection information, and recognizes the calibrated detection information according to the trained pole tower recognition model to output a pole tower recognition result.

10. The satellite image processing system according to claim 9, wherein The pole tower annotation optimization module includes: A regular pole tower extraction unit that obtains regular pole tower information according to the pole tower spatial information; A pole tower boundary generation unit that determines the boundary features of each pole tower according to the regular pole tower information and generates an initial boundary block diagram corresponding to the pole tower; A boundary optimization annotation unit that adaptively adjusts the initial rectangular block diagram according to the structure and spatial distribution of the pole tower and performs category annotation to obtain an annotated boundary block diagram; Annotation data conversion unit, which converts the format of the annotated boundary block diagram to obtain tower pole annotation data.

Images