Photogrammetry-based photovoltaic power station map generation method and system

Abstract

The application discloses a photogrammetry-based photovoltaic power station map generation method and system, and relates to the technical field of map generation. The method comprises the following steps: performing dynamic optimization on a reference flight line based on a photogrammetry attention array to obtain a risk-aware inspection flight line; collecting spatial pose data, an initial visible light image stream and an initial infrared thermal imaging stream; acquiring supplementary pose data, a supplementary visible light image stream and a supplementary infrared thermal imaging stream; obtaining a high-confidence TDOM map through joint three-dimensional reconstruction; and performing multispectral fault feature extraction to differentiate and locate K surface fault photovoltaic panels and Q electrical fault photovoltaic panels. The technical problems of insufficient photovoltaic power station map generation precision, inaccurate fault positioning and difficulty in fully covering complex terrain areas in the prior art are solved, the technical effects of generating a high-confidence map, improving the data coverage integrity of complex terrain areas and realizing accurate differentiation and positioning of surface fault and electrical fault photovoltaic panels are achieved.

Description

Technical Field

[0001] This invention relates to the field of map generation technology, and more specifically to a method and system for generating photovoltaic power plant maps based on photogrammetry. Background Technology

[0002] With the continuous advancement of large-scale and clustered construction of photovoltaic power plants, the operation monitoring and maintenance management of photovoltaic arrays is gradually shifting from manual inspection to unmanned and intelligent methods. Among these methods, power plant map generation and defect detection technology based on UAV photogrammetry has become an important means. Existing technologies typically use UAVs with fixed flight paths to collect visible light or infrared images of photovoltaic power plants and combine them with photogrammetric methods to reconstruct two-dimensional orthophotos or three-dimensional models. However, there are still certain limitations in practical applications: On the one hand, existing flight path planning is mostly based on regular grids or experience settings, lacking the dynamic perception capability for complex terrain, array occlusion, and differences in equipment distribution, which can easily lead to insufficient or redundant data collection in local areas; on the other hand, in complex terrain environments such as mountains and hills, factors such as component occlusion, terrain undulation, and dense array arrangement can cause missing image information or matching errors, affecting the accuracy and completeness of map reconstruction; in addition, existing methods mostly use single-viewpoint or single-spectral data for analysis, making it difficult to effectively distinguish and accurately locate surface defects and electrical faults of photovoltaic panels while achieving high-precision spatial reconstruction, resulting in insufficient reliability of fault identification results. Summary of the Invention

[0003] This application provides a method and system for generating photovoltaic power plant maps based on photogrammetry, which solves the technical problems of insufficient accuracy in generating photovoltaic power plant maps, inaccurate fault location, and difficulty in fully covering complex terrain areas in the prior art.

[0004] The first aspect of this application provides a method for generating photovoltaic power plant maps based on photogrammetry, the method comprising: A risk-aware inspection route is obtained by dynamically optimizing the baseline route based on a photogrammetric attention array. During the automatic flight of the main fixed-wing UAV along the risk-aware inspection route over the photovoltaic power station, the onboard georeferenced multispectral acquisition module collects spatial pose data, initial visible light image stream, and initial infrared thermal image stream. Based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and initial visible light image stream, a subordinate rotary-wing UAV is activated to perform local low-altitude multi-angle supplementary scanning to obtain supplementary pose data, supplementary visible light image stream, and supplementary infrared thermal image stream. The supplementary pose data and supplementary visible light image stream are fused with the spatial pose data and initial visible light image stream, and a high-confidence TDOM map is obtained through joint 3D reconstruction using cross-viewpoint dense matching. Using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal image stream and initial infrared thermal image stream are fused with the high-confidence TDOM map to extract multispectral fault features, thereby differentiating and locating K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels.

[0005] A second aspect of this application provides a photovoltaic power plant map generation system based on photogrammetry, the system comprising: Flight path optimization component: Dynamically optimizes the baseline flight path based on photogrammetric attention array to obtain a risk-aware inspection flight path; Data acquisition component: Controls the main fixed-wing UAV to automatically fly along the risk-aware inspection flight path in the photovoltaic power station, and the onboard georeferenced multispectral acquisition module collects spatial pose data, initial visible light image stream, and initial infrared thermal image stream; Data supplementation component: Based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and initial visible light image stream, initiates a slave rotary-wing UAV to perform local low-altitude multi-angle supplementary scanning to obtain supplementary pose data. The system includes: data, supplementary visible light image stream, and supplementary infrared thermal image stream; a 3D reconstruction component: fusing the supplementary pose data and supplementary visible light image stream to the spatial pose data and initial visible light image stream, and obtaining a high-confidence TDOM map through joint 3D reconstruction with cross-viewpoint dense matching; and a fault location component: using the photogrammetric attention array as a spatial weighting operator, fusing the supplementary infrared thermal image stream and initial infrared thermal image stream to the high-confidence TDOM map, and performing multispectral fault feature extraction to differentiate and locate K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, the baseline flight path is dynamically optimized using a photogrammetric attention array to obtain a risk-aware inspection route. Next, during automatic flight of the main fixed-wing UAV along the risk-aware inspection route over the photovoltaic power station, the onboard georeferenced multispectral acquisition module collects spatial pose data, initial visible light image streams, and initial infrared thermal image streams. Further, based on terrain occlusion analysis of the spatial pose data and initial visible light image streams to locate the terrain-occluded areas, a subordinate rotary-wing UAV is activated to perform local low-altitude multi-angle supplementary scanning, acquiring supplementary pose data, supplementary visible light image streams, and supplementary infrared thermal image streams. Then, the supplementary pose data and supplementary visible light image streams are fused with the spatial pose data and initial visible light image streams. Through joint 3D reconstruction using cross-viewpoint dense matching, a high-confidence TDOM map is obtained. Finally, using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal image streams and initial infrared thermal image streams are fused into the high-confidence TDOM map for multispectral fault feature extraction to differentiate and locate K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels. This invention solves the technical problems of insufficient accuracy in generating photovoltaic power plant maps, inaccurate fault location, and difficulty in fully covering complex terrain areas in existing technologies. It achieves the technical effects of generating high-confidence maps, improving the data coverage integrity of complex terrain areas, and realizing the precise differentiation and location of photovoltaic panels for surface faults and electrical faults. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A schematic diagram of the photovoltaic power plant map generation method based on photogrammetry provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of a photovoltaic power plant map generation system based on photogrammetry, provided in an embodiment of this application.

[0009] Explanation of reference numerals in the attached diagram: Route optimization component 11, data acquisition component 12, data supplementation component 13, 3D reconstruction component 14, fault location component 15. Detailed Implementation

[0010] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0011] Example 1, as Figure 1As shown, this application provides a method for generating photovoltaic power plant maps based on photogrammetry, wherein the method includes: The baseline route is dynamically optimized based on a photogrammetric attention array to obtain a risk-aware inspection route.

[0012] Furthermore, dynamic optimization based on photogrammetric attention arrays is performed on the baseline route to obtain a risk-aware inspection route, including: The real-time operation stream transmitted back from the photovoltaic array is retrieved to quantify the probability of abnormal equipment status. After obtaining the quantified fault risk array, surface integrity fault filtering is performed to construct a photogrammetric attention array. Based on the photogrammetric attention array, the baseline route is dynamically optimized to obtain the risk perception inspection route.

[0013] First, the real-time operation streams transmitted from the photovoltaic array via the string monitoring unit, combiner box monitoring unit, inverter monitoring unit, and board-level acquisition terminal are retrieved. These real-time operation streams are then split at the board level according to the photovoltaic panel number, resulting in a set of operating parameters corresponding to each photovoltaic panel. This set of operating parameters includes at least output current, output power, module temperature, power variation between adjacent time periods, irradiance response deviation, and temperature-power coupling deviation. Subsequently, the operating parameter sets for each photovoltaic panel are denoised, missing data is filled in, and dimensions are standardized. Based on a preset time window, stability features, abrupt change features, and deviation features are extracted to construct an operating feature vector for each photovoltaic panel. This operating feature vector is then input into a preset fault probability model to calculate the anomaly probability value and fault type probability distribution for each photovoltaic panel. Combining the geographical coordinates of each photovoltaic panel, the array layout position, and the region number, each anomaly probability value is mapped to the geographic space of the photovoltaic power station, generating a quantified fault risk array.

[0014] The fault probability model is used to output the probability of occurrence of various fault types based on the panel-level operating parameters of the photovoltaic panel.

[0015] The fault probability model is constructed by: retrieving a historical operating sample set, which includes normal photovoltaic panel samples and photovoltaic panel samples with various fault types, and each sample corresponds to the current dispersion coefficient, power drop frequency, temperature-power deviation, and irradiance response deviation within a preset time window; after imputing missing values, removing outliers, and normalizing the panel-level operating parameters of each sample, a sample feature set is constructed, and each sample is labeled with the corresponding fault type based on manual inspection records, maintenance conclusions, or historical fault files, forming a fault label set that corresponds one-to-one with the sample feature set; the sample feature set and fault label set are divided into a training subset and a validation subset according to a preset ratio, and the probability output classification model is iteratively trained based on the training subset, so that the model learns the correspondence between different combinations of panel-level operating parameters and different fault types; in each round of training, the training subset is input into the current model, and the multi-fault type probability distribution results corresponding to each sample are output, and the probability distribution is... The deviation between the output results and the corresponding fault labels is calculated, and the deviation results are used as the basis for parameter updates to iteratively adjust the internal parameters of the model. After completing a single round of parameter updates, the validation subset is input into the updated model to calculate the validation accuracy, validation recall, and validation loss, which are used to evaluate the current model's ability to distinguish different fault types. When the validation loss is lower than a preset convergence threshold, or the improvement in validation performance after multiple rounds of training is lower than a preset improvement threshold, the model training is terminated, and the fault probability model is obtained. After the model training is completed, the board-level operating parameters extracted from the real-time operation stream are input into the fault probability model to output multiple sets of fault type probabilities for each photovoltaic panel. These multiple sets of fault type probabilities are used for subsequent quantification of fault risk array construction. The multiple sets of fault type probabilities include at least the surface contamination fault probability, surface damage fault probability, electrical anomaly fault probability, and normal state probability, thereby providing a probabilistic basis for surface integrity fault filtering and photogrammetric attention array generation.

[0016] After obtaining the quantified fault risk array, surface integrity fault filtering is performed to identify and remove all surface functional anomalies caused by external contaminant adhesion and physical structural damage from the quantified fault risk array. External contaminant adhesion includes at least bird droppings, dust cover, and snow cover; physical structural damage includes at least component breakage, surface scratches, and hail impact damage. This ensures that subsequent route optimization focuses on electrical anomaly risk areas rather than simply areas with visible surface defects. Then, based on the filtered fault risk distribution, combined with the risk level corresponding to the fault type, the degree of continuous clustering in the area, and the degree of matching with historical alarm records, normalized weights are assigned to each spatial location to construct a photogrammetric attention array. Each array unit corresponds to a spatial weight representing the inspection priority. Finally, the baseline route is projected onto the photogrammetric attention array. Within the geographic weighted grid where the LIMITED array is located, the spatial weight distribution of the areas traversed by each baseline flight segment is read. Based on the weight, the waypoint density, flight segment length, and local flight paths are adaptively adjusted to reduce the waypoint spacing, increase image overlap, and achieve denser flight coverage in high-weight areas, while increasing the waypoint spacing and reducing overlapping coverage in low-weight areas. After reconstructing the waypoints, the adjusted waypoint sequence is smoothed for continuity. Flight accessibility is verified by considering the minimum turning capability, climb capability, flight time limit, and safe takeoff altitude requirements of the fixed-wing UAV. When there are local flight segments that do not meet the flight constraints, the corresponding flight segments are locally replanned until the preset flight constraints are met. Finally, a risk-aware inspection route that balances risk-priority coverage, flight path continuity, and flight safety is output to control the main fixed-wing UAV to perform automatic inspection flights of photovoltaic power stations.

[0017] Furthermore, the real-time operation stream transmitted back by the photovoltaic array is retrieved to quantify the probability of abnormal equipment status, resulting in a quantified fault risk array. Surface integrity fault filtering is then performed to construct a photogrammetric attention array, including: The real-time operation flow is decomposed to obtain multiple panel-level operating parameters for multiple photovoltaic panels in the photovoltaic array. These panel-level operating parameters include current dispersion coefficient, power drop frequency, temperature-power deviation, and irradiance response deviation. Multiple operating feature vectors of the multiple panel-level operating parameters are constructed, and multiple sets of fault type probabilities are matched with a preset fault probability model. Based on the geographical coordinates of the multiple photovoltaic panels in the photovoltaic array, the multiple sets of fault type probabilities are spatially mapped to form the quantified fault risk array. After surface integrity fault filtering of the quantified fault risk array based on preset surface fault types, normalized weighting based on fault type risk level is performed to obtain the photogrammetric attention array.

[0018] Preferably, the real-time operation stream undergoes data decomposition and time alignment processing to extract current, power, temperature, and irradiance data for each photovoltaic panel within a continuous time window. Based on a sliding time window, the current dispersion coefficient is calculated to characterize the degree of current fluctuation; the frequency of power drop events is statistically analyzed to characterize output stability; the deviation between temperature and power is calculated to characterize thermal anomaly characteristics; and the deviation between irradiance changes and power response is calculated to characterize response consistency, thereby forming multi-dimensional panel-level operating parameters for each photovoltaic panel. On this basis, the panel-level operating parameters are normalized and dimensionless to construct an operating feature vector at a unified scale. This vector is then input into a preset fault probability model for multi-class matching calculation. The fault probability model is trained based on historical labeled data and used for... The system outputs the probability of occurrence of each photovoltaic panel under different fault types. Then, based on the geographical coordinate information of each photovoltaic panel, the probabilities of the multiple fault types are mapped to a spatial grid coordinate system to construct a quantitative fault risk array consistent with the power plant layout, so that each spatial unit carries the fault risk information of the corresponding photovoltaic panel. After the risk array is constructed, the quantitative fault risk array is filtered based on a preset set of surface fault types, retaining only the surface fault risk components caused by contamination or structural damage, and eliminating non-surface fault interference. Finally, the retained risk components are weighted and fused according to the risk level of different fault types, and overall normalization is performed to ensure that the weight values ​​of each spatial location are within a uniform scale range, generating a photogrammetric attention array for guiding route optimization and fault focusing.

[0019] Furthermore, the risk perception inspection route is obtained by dynamically optimizing the baseline route based on the photogrammetric attention array, including: The photogrammetric attention array is mapped to a geographic weighted raster map; after projecting the baseline route onto the geographic weighted raster map, the waypoint density is adaptively adjusted based on the raster weight values ​​to obtain a discrete waypoint sequence; the discrete waypoint sequence is smoothed using a B-spline curve to obtain a smoothed route, and then the ground clearance is verified to generate the risk perception inspection route.

[0020] Preferably, the photogrammetric attention array is mapped to a geographic weighted raster map according to the actual geographic coordinate system of the photovoltaic power station, so that each raster cell corresponds to an inspection priority weight value. The baseline flight path is projected segment by segment onto the geographic weighted raster map, and the weight distribution, weight mean, and local high-weight clustering in the area traversed by each segment are statistically analyzed. Based on this, adaptive adjustment of waypoint density based on raster weight values ​​is performed on the baseline flight path. For segments traversing high-weight areas, the spacing between adjacent waypoints is reduced and supplementary waypoints are inserted in areas with continuous high-weight distribution to improve the image overlap rate and ground feature observation accuracy in that area. For segments traversing low-weight areas, the spacing between adjacent waypoints is appropriately increased to reduce redundant acquisition and control the overall flight path. At the same time, for areas with large weight change gradients, transition waypoints are added on both sides of the weight change boundary to avoid flight control discontinuity caused by abrupt changes in waypoint density. After completing the waypoint reconstruction of each segment, the entire... The discrete waypoint sequences are sequentially connected and smoothed using B-spline curves to ensure a continuous transition between adjacent straight segments and end transition arcs, thereby generating a smooth flight path that meets the turning continuity requirements of fixed-wing UAVs. After obtaining the smooth flight path, the terrain elevation data of the photovoltaic power station area, the array support height data, and the UAV's preset safety clearance requirements are combined to perform ground clearance verification point by point on the smooth flight path. The actual flight altitude of each position on the flight path relative to the ground surface and the top of the components is calculated. When insufficient ground clearance, terrain protrusion interference, or insufficient obstacle crossing safety margin are detected in a local segment, the flight altitude of the corresponding waypoint is raised and corrected. At the same time, the adjacent segments before and after are adjusted with a gentle slope transition to ensure that the entire flight path meets the UAV's climb capability, descent capability, and safe flight requirements. Finally, a risk perception inspection flight path is output that simultaneously takes into account priority coverage of high-risk areas, complete inspection of rectangular areas, smooth fixed-wing flight, and safe obstacle crossing in complex terrain.

[0021] During the automatic flight of the main fixed-wing UAV along the risk perception inspection route in the photovoltaic power station, the onboard geographic reference multispectral acquisition module collects spatial pose data, initial visible light image stream, and initial infrared thermal image stream.

[0022] First, the risk perception inspection route is loaded into the UAV flight control system, and the flight speed, cruising altitude, and heading control strategies are set according to the preset inspection task parameters, enabling the UAV to perform autonomous flight according to the waypoint sequence. During flight, the high-precision positioning module in the geographic reference multispectral acquisition module performs real-time positioning and attitude calculation on the UAV, acquiring spatial pose data including latitude and longitude coordinates, elevation information, roll angle, pitch angle, and heading angle, and continuously recording it according to a unified time reference. Simultaneously, the visible light gimbal camera is controlled to continuously shoot according to preset sampling intervals or trigger conditions, acquiring the initial visible light image stream covering the photovoltaic array area, and the camera attitude is kept stable relative to the ground through gimbal stabilization control. To reduce motion blur and viewpoint shift, during synchronization, the infrared pan-tilt camera is controlled to acquire thermal images at time intervals consistent with or at a preset ratio to obtain an initial infrared thermal image stream reflecting the temperature distribution of the photovoltaic panel. Radiometric correction and temperature calibration are then performed on the infrared images to ensure the comparability and consistency of the thermal data. Furthermore, time-stamp binding and data frame association are applied to the spatial pose data, visible light image stream, and infrared thermal image stream, ensuring that each image frame corresponds to unique spatial pose information, thereby constructing a multimodal spatiotemporal dataset with geographic reference attributes. Finally, the spatial pose data, initial visible light image stream, and initial infrared thermal image stream are output, providing fundamental data support for subsequent terrain analysis, 3D reconstruction, and multispectral fusion.

[0023] Furthermore, the geographic reference multispectral acquisition module includes an RTK high-precision positioning module, a visible light gimbal camera, and an infrared gimbal camera. During the flight of the main fixed-wing UAV, centimeter-level pose synchronization is performed through the RTK high-precision positioning module to acquire spatial pose data, and sequential image acquisition is performed through the visible light gimbal camera and the infrared gimbal camera to obtain the initial visible light image stream and the initial infrared thermal image stream.

[0024] Before takeoff, the differential signal between the RTK high-precision positioning module and the ground reference station is established and initialized for calibration. This enables the UAV to continuously acquire high-precision positioning results during flight. Combined with the onboard inertial measurement unit, the UAV's attitude is calculated in real time, generating continuous spatial pose data containing position and attitude information. During flight, the spatial pose data is timestamped according to a unified time reference and synchronously bound to subsequently acquired image data. Simultaneously, the visible light gimbal camera is controlled to perform continuous shooting according to a preset sampling strategy. This sampling strategy includes acquisition methods triggered by flight distance or image overlap, and the camera is kept pointing towards the ground through gimbal stabilization control. To achieve consistency, a high-resolution visible light image sequence covering the photovoltaic array area is acquired. Simultaneously, an infrared pan-tilt camera is controlled to perform thermal imaging acquisition. By performing radiometric calibration and temperature inversion processing on the output signals of the infrared sensors, an infrared image sequence reflecting the temperature distribution on the photovoltaic panel surface is generated. This sequence is synchronized with the visible light image or registered according to a preset ratio. During data acquisition, each frame of visible light image is frame-level correlated with the corresponding infrared image and the spatial pose data at the same time, constructing a uniformly numbered multimodal data recording unit. This forms an initial visible light image stream and an initial infrared thermal image stream with geographic reference attributes, providing highly consistent input data for subsequent 3D reconstruction and multispectral fusion processing.

[0025] Based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and the initial visible light image stream, the slave rotor UAV is activated to perform local low-altitude multi-angle supplementary scanning to acquire supplementary pose data, supplementary visible light image stream and supplementary infrared thermal image stream.

[0026] Furthermore, based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and the initial visible light image stream, a slave rotorcraft UAV is initiated to perform local low-altitude multi-angle supplementary scanning to acquire supplementary pose data, supplementary visible light image stream, and supplementary infrared thermal image stream, including: Based on the spatial pose data and the initial visible light image stream, motion recovery structure reconstruction is performed to generate an initial TDOM map and an initial DSM map; elevation gradient analysis is performed on the initial DSM map to locate terrain occlusion areas; the elevation gradient distribution of the terrain occlusion areas is retrieved, and combined with the fault weights of the photogrammetric attention array, an adaptive supplementary scanning path is constructed; the slave rotor UAV is controlled to perform local low-altitude multi-angle supplementary scanning along the adaptive supplementary scanning path to acquire the supplementary pose data, supplementary visible light image stream, and supplementary infrared thermal image stream.

[0027] Preferably, the initial visible light image stream is subjected to image screening, distortion correction, and temporal alignment. Then, feature extraction, corresponding point matching, and camera pose optimization are performed using the spatial pose data corresponding to each image to generate an initial sparse point cloud. Based on this, dense point cloud expansion, surface fitting, and orthorectification are performed to obtain an initial TDOM map reflecting surface texture distribution and an initial DSM map reflecting surface elevation undulations, respectively. After obtaining the initial DSM map, elevation gradient analysis is performed, specifically calculating the elevation gradient between adjacent raster cells region by region according to a preset spatial window. The elevation change range, slope change trend, and local undulation intensity are analyzed, and combined with the texture continuity and image coverage integrity of the corresponding areas in the initial TDOM map, clustering is performed on locations with abrupt elevation changes, local steep slopes, shadow occlusion, or array occlusion to form terrain occlusion areas. After locating the terrain occlusion areas, the elevation gradient distribution of each terrain occlusion area is retrieved and jointly analyzed with the fault weights of the corresponding spatial locations in the photogrammetric attention array. Areas with larger elevation gradients and higher fault weights are assigned higher rescanning priority, while areas with gentler elevation gradients and lower fault weights are assigned lower rescanning priority. Areas with lower density are assigned a relatively lower priority for rescanning. Based on this, the rescanning height, rescanning angle, rescanning segment density, and rescanning sequence for each terrain-obstructed area are determined, and an adaptive rescanning flight path is constructed. The adaptive rescanning flight path includes a circumferential observation segment deployed along the edge of the obstructed area, a directional traversal segment facing the interior of the obstructed area, and a multi-angle transition segment for acquiring images from different top and side angles, to ensure that the obstructed area can obtain multi-directional image coverage under low-altitude conditions. Subsequently, the slave rotorcraft UAV is controlled to perform local low-altitude multi-angle rescanning along the adaptive rescanning flight path. During flight, the onboard positioning and attitude measurement unit continuously records flight position, altitude, and attitude angle information to generate supplementary pose data. Simultaneously, onboard visible light cameras and infrared thermal imaging cameras acquire data synchronously or quasi-synchronously according to the flight rhythm of the supplementary scan segment, resulting in supplementary visible light image streams and supplementary infrared thermal image streams, respectively. Each supplementary image frame is then time-stamped and spatially correlated with the corresponding supplementary pose data to form a supplementary multimodal dataset for terrain-occluded areas. This dataset is used for subsequent high-confidence TDOM map reconstruction and multispectral fault fusion analysis.

[0028] By fusing the supplementary pose data and supplementary visible light image stream to the spatial pose data and the initial visible light image stream, a high-confidence TDOM map is obtained through joint 3D reconstruction with cross-viewpoint dense matching.

[0029] Furthermore, by fusing the supplementary pose data and supplementary visible light image stream to the spatial pose data and the initial visible light image stream, and through joint 3D reconstruction using cross-viewpoint dense matching, a high-confidence TDOM map is obtained, including: Based on RTK localization constraints, the supplementary pose data and spatial pose data are unified to a geocentric Cartesian coordinate system to obtain a spatiotemporally registered pose dataset. Spatiotemporal alignment is performed on the supplementary visible light image stream and the initial visible light image stream based on UTC timestamps to obtain spatiotemporally aligned image frames. After SIFT feature point extraction on the spatiotemporally aligned image frames, joint bundle adjustment optimization is performed using the spatial constraints associated with the spatiotemporally registered pose dataset to output an optimized camera pose parameter set. A disparity spatial image is constructed based on the optimized camera pose parameter set, and a pixel-level disparity map is calculated using SGM. The elevation gradient distribution of the pixel-level disparity map and the initial DSM map is fused to perform gradient weighted compensation for the terrain occlusion region, resulting in an occlusion-compensated disparity map. After 3D point cloud transformation of the occlusion-compensated disparity map, Delaunay triangulation is performed by fusing the terrain structure features of the initial DSM map to output an optimized DSM map. The optimized DSM map is used to perform differential correction on the spatiotemporally aligned image frames to output the high-confidence TDOM map.

[0030] Based on RTK positioning constraints, the differential positioning results, attitude calculation results, and camera extrinsic parameter calibration results of the main fixed-wing UAV and the subordinate rotary-wing UAV at each acquisition time are read. The RTK positioning constraints refer to the real-time correction of the raw satellite positioning results output by the onboard positioning modules of the two UAVs using differential correction information sent from the ground reference station. This enables the main fixed-wing UAV and the subordinate rotary-wing UAV to obtain centimeter-level position coordinates within a unified geodetic reference frame. These centimeter-level position coordinates are then used as spatial reference constraints in subsequent multi-view image joint reconstruction to limit the spatial offset range of the camera projection center corresponding to different image frames and the relative position error between adjacent viewpoints. Based on this, the supplementary pose data and spatial pose data are uniformly converted to the same geocentric Cartesian coordinate system, and the pose records of different platforms are synchronized and organized according to a unified time reference, thereby forming a spatiotemporally registered pose dataset that can be used for joint reconstruction. Furthermore, based on UTC timestamps, the supplementary visible light image stream and the initial visible light image stream are further filtered for temporal correspondence and spatial coverage relationships, retaining image frames whose time intervals meet synchronization requirements and whose fields of view overlap, thus obtaining spatiotemporally aligned image frames. Then, SIFT feature points are extracted from these spatiotemporally aligned image frames, and cross-viewpoint candidate matching relationships are established based on feature description information, combined with spatiotemporal registration... The spatial constraints provided by the quasi-pose dataset eliminate mismatched point pairs to ensure that the feature connectivity relationships involved in the reconstruction satisfy the true imaging geometry. Subsequently, based on the filtered matching point pairs and unified pose information, joint bundle adjustment optimization is performed to iteratively correct the camera position, pose, and spatial connectivity coordinates corresponding to each image frame. During the iterative correction process, the centimeter-level position coordinates corrected by RTK are used as external position priors to participate in the error constraint, so that the camera pose optimization result satisfies both the consistency of image feature reprojection and the consistency of actual geographical location, until the reprojection error converges to a preset range. The optimized camera pose parameter set is then output, and the optimization is performed according to the... A parallax spatial image with epipolar correspondence is constructed using the camera pose parameter set. Pixel-level parallax search is performed in the parallax spatial image to recover depth difference information between images from different viewpoints, resulting in a pixel-level parallax map. Considering that terrain undulations and local occlusions can cause discontinuities or missing parallax in some areas, the elevation gradient distribution of the corresponding area in the initial DSM map is further retrieved. Gradient-guided correction is performed on terrain-occluded areas, with areas exhibiting drastic elevation changes, significant parallax abrupt changes, and concentrated reconstruction holes as key compensation targets. Gradient-weighted compensation is then applied to the pixel-level parallax map based on the terrain continuity trend of the surrounding stable areas, resulting in an occlusion-compensated parallax map.Subsequently, the occlusion-compensated disparity map is inversely calculated into 3D spatial point coordinates to generate a compensated dense point cloud. The point cloud is then constrained and optimized by incorporating known terrain structure features from the initial DSM map. A continuous surface triangulation model is constructed using Delaunay triangulation, outputting an optimized DSM map with higher geometric continuity. Finally, the optimized DSM map is used to perform differential correction on spatiotemporally aligned image frames. Each image is projected pixel-by-pixel onto a unified geographic plane according to the optimized surface geometry. Sharpness optimization, edge transition blending, and brightness consistency adjustments are performed on overlapping areas to eliminate geometric distortion and texture breaks caused by terrain occlusion, viewpoint differences, and pose changes. The result is a high-confidence TDOM map with accurate spatial positioning, continuous texture, and sufficient compensation for occlusion areas.

[0031] Using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are fused into the high-confidence TDOM map to extract multispectral fault features, thereby differentiating and locating K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels.

[0032] Furthermore, using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are fused into the high-confidence TDOM to extract multispectral fault features, thereby differentiating and locating K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels, including: The geographic coordinates of the photovoltaic panels in the initial TDOM map are mapped to the high-confidence TDOM map to spatially define the photovoltaic panel detection area. The supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are fused to obtain a temporal infrared temperature field. Using the photovoltaic panel detection area as a spatial reference, the temporal infrared temperature field is registered to the high-confidence TDOM through perspective transformation to generate a geographic reference multispectral fusion map. The photogrammetric attention array is superimposed on the geographic reference multispectral fusion map space to perform photovoltaic fault focusing detection, so as to differentiate and locate the K surface fault photovoltaic panels and Q electrical fault photovoltaic panels.

[0033] Preferably, the defined photovoltaic panel boundary contours and their geographic coordinates in the initial TDOM map are mapped to the high-confidence TDOM map. Spatial alignment of the photovoltaic panel area is achieved through coordinate consistency transformation and boundary relocation, thus forming a corresponding photovoltaic panel detection area in the high-confidence TDOM map. Based on this, the supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are sequentially integrated under a unified time reference. Through time sequence rearrangement, temperature consistency correction, and abnormal temperature value removal, a continuous and stable temporal infrared temperature field is constructed. Using the photovoltaic panel detection area as a spatial reference, the projection relationship between the infrared image and the TDOM map is established using spatial pose data. Perspective transformation is used to perform frame-by-frame spatial registration of the temporal infrared temperature field, ensuring that the infrared temperature information and the visible light orthophoto image are accurately aligned within the same geographic coordinate system. Correspondingly, a multispectral fusion map with geographic reference attributes is generated. Subsequently, the photogrammetric attention array is superimposed on the multispectral fusion map and used as a spatial weight to weight and enhance the temperature anomaly features of different regions, so that the anomaly information of high-risk areas is displayed first. On this basis, the temperature distribution features and visible light texture features within the detection area of ​​each photovoltaic panel are jointly analyzed. According to the spatial distribution pattern of temperature anomalies and its correspondence with surface structure changes, areas with obvious surface contamination, damage or shading and accompanied by local temperature anomalies are identified as surface fault photovoltaic panels, and areas with internal hot spots, series mismatch or electrical anomalies and insignificant surface texture changes are identified as electrical fault photovoltaic panels. Finally, classification statistics and spatial positioning are performed on all photovoltaic panels, and K surface fault photovoltaic panels and Q electrical fault photovoltaic panels are output.

[0034] Furthermore, by overlaying the photogrammetric attention array onto the georeferenced multispectral fusion map space, photovoltaic fault focusing detection is performed to differentiate and locate the K surface-faulted photovoltaic panels and the Q electrical-faulted photovoltaic panels, including: The photogrammetric attention array is superimposed onto the georeferenced multispectral fusion map to generate a fault-focused heatmap; the fault-focused heatmap is then used for dual-path fault differentiation and localization based on the visible light channel and the infrared channel to obtain the K surface-faulted photovoltaic panels and the Q electrical-faulted photovoltaic panels.

[0035] Preferably, the photogrammetric attention array is fused with the georeferenced multispectral array. Figure 1The spatial coordinates are aligned, and the weight values ​​of each position in the attention array are mapped to the corresponding photovoltaic panel detection area. Temperature and texture information in the multispectral fusion map are spatially weighted to generate a fault-focused heatmap reflecting the degree of risk focus, where high-weight areas correspond to areas of concentrated potential faults. Based on this, a dual-path fault differentiation and localization process is performed on the fault-focused heatmap. The visible light channel-based path performs texture anomaly analysis on the photovoltaic panel area, extracting edge features, brightness variations, and structural continuity information to identify texture anomaly areas caused by surface dirt occlusion, dust accumulation, and component misalignment or damage. These anomalies are then combined with their weight distribution in the heatmap to filter and determine candidate photovoltaic panels for surface faults. The system performs temperature anomaly analysis on the photovoltaic panel area based on the infrared channel path. By statistically modeling and comparing the temperature distribution, it identifies abnormally hot spots that are significantly higher than the surrounding environment and shutdown areas with abnormally low overall temperatures. Based on preset hot spot judgment rules, it filters abnormal areas to determine candidate photovoltaic panels with electrical faults. Furthermore, it performs spatial consistency verification and conflict resolution on the detection results of the two paths. For areas with both texture and temperature anomalies, it prioritizes classification based on the dominant abnormal feature. For areas with only a single abnormal feature, it directly classifies them accordingly, thereby completing the fault type determination. Finally, it performs numbering and spatial positioning on all photovoltaic panels, outputting K photovoltaic panels with surface faults and Q photovoltaic panels with electrical faults.

[0036] In summary, the embodiments of this application have at least the following technical effects: First, the baseline flight path is dynamically optimized using a photogrammetric attention array to obtain a risk-aware inspection route. Next, during automatic flight of the main fixed-wing UAV along the risk-aware inspection route over the photovoltaic power station, the onboard georeferenced multispectral acquisition module collects spatial pose data, initial visible light image streams, and initial infrared thermal image streams. Further, based on terrain occlusion analysis of the spatial pose data and initial visible light image streams to locate the terrain-occluded areas, a subordinate rotary-wing UAV is activated to perform local low-altitude multi-angle supplementary scanning, acquiring supplementary pose data, supplementary visible light image streams, and supplementary infrared thermal image streams. Then, the supplementary pose data and supplementary visible light image streams are fused with the spatial pose data and initial visible light image streams. Through joint 3D reconstruction using cross-viewpoint dense matching, a high-confidence TDOM map is obtained. Finally, using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal image streams and initial infrared thermal image streams are fused into the high-confidence TDOM map for multispectral fault feature extraction to differentiate and locate K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels. This invention solves the technical problems of insufficient accuracy in generating photovoltaic power plant maps, inaccurate fault location, and difficulty in fully covering complex terrain areas in existing technologies. It achieves the technical effects of generating high-confidence maps, improving the data coverage integrity of complex terrain areas, and realizing the precise differentiation and location of photovoltaic panels for surface faults and electrical faults.

[0037] Example 2 is based on the same inventive concept as the photovoltaic power plant map generation method based on photogrammetry in the previous examples, such as... Figure 2 As shown, this application provides a photovoltaic power plant map generation system based on photogrammetry, wherein the system includes: Flight path optimization component 11: Dynamically optimizes the baseline flight path based on photogrammetric attention array to obtain a risk-aware inspection flight path; Data acquisition component 12: Controls the main fixed-wing UAV to automatically fly along the risk-aware inspection flight path in the photovoltaic power station, and the onboard georeferenced multispectral acquisition module collects spatial pose data, initial visible light image stream, and initial infrared thermal image stream; Data supplementation component 13: Based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and initial visible light image stream, initiates a subordinate rotary-wing UAV to perform local low-altitude multi-angle supplementary scanning to obtain supplementary position data. The system includes: pose data, supplementary visible light image stream, and supplementary infrared thermal image stream; 3D reconstruction component 14: fuses the supplementary pose data and supplementary visible light image stream to the spatial pose data and initial visible light image stream, and obtains a high-confidence TDOM map through joint 3D reconstruction with cross-view dense matching; fault location component 15: uses the photogrammetric attention array as a spatial weighting operator, fuses the supplementary infrared thermal image stream and initial infrared thermal image stream to the high-confidence TDOM map, performs multispectral fault feature extraction, and differentiates and locates K surface fault photovoltaic panels and Q electrical fault photovoltaic panels.

[0038] Furthermore, the route optimization component 11 is used to perform the following methods: The real-time operation stream transmitted back from the photovoltaic array is retrieved to quantify the probability of abnormal equipment status. After obtaining the quantified fault risk array, surface integrity fault filtering is performed to construct a photogrammetric attention array. Based on the photogrammetric attention array, the baseline route is dynamically optimized to obtain the risk perception inspection route.

[0039] Furthermore, the data supplementation component 13 is used to perform the following method: Based on the spatial pose data and the initial visible light image stream, motion recovery structure reconstruction is performed to generate an initial TDOM map and an initial DSM map; elevation gradient analysis is performed on the initial DSM map to locate terrain occlusion areas; the elevation gradient distribution of the terrain occlusion areas is retrieved, and combined with the fault weights of the photogrammetric attention array, an adaptive supplementary scanning path is constructed; the slave rotor UAV is controlled to perform local low-altitude multi-angle supplementary scanning along the adaptive supplementary scanning path to acquire the supplementary pose data, supplementary visible light image stream, and supplementary infrared thermal image stream.

[0040] Furthermore, the fault location component 15 is used to perform the following method: The geographic coordinates of the photovoltaic panels in the initial TDOM map are mapped to the high-confidence TDOM map to spatially define the photovoltaic panel detection area. The supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are fused to obtain a temporal infrared temperature field. Using the photovoltaic panel detection area as a spatial reference, the temporal infrared temperature field is registered to the high-confidence TDOM through perspective transformation to generate a geographic reference multispectral fusion map. The photogrammetric attention array is superimposed on the geographic reference multispectral fusion map space to perform photovoltaic fault focusing detection, so as to differentiate and locate the K surface fault photovoltaic panels and Q electrical fault photovoltaic panels.

[0041] Furthermore, the fault location component 15 is used to perform the following method: The photogrammetric attention array is superimposed onto the georeferenced multispectral fusion map to generate a fault-focused heatmap; the fault-focused heatmap is then used for dual-path fault differentiation and localization based on the visible light channel and the infrared channel to obtain the K surface-faulted photovoltaic panels and the Q electrical-faulted photovoltaic panels.

[0042] Furthermore, the route optimization component 11 is used to perform the following methods: The real-time operation flow is decomposed to obtain multiple panel-level operating parameters for multiple photovoltaic panels in the photovoltaic array. These panel-level operating parameters include current dispersion coefficient, power drop frequency, temperature-power deviation, and irradiance response deviation. Multiple operating feature vectors of the multiple panel-level operating parameters are constructed, and multiple sets of fault type probabilities are matched with a preset fault probability model. Based on the geographical coordinates of the multiple photovoltaic panels in the photovoltaic array, the multiple sets of fault type probabilities are spatially mapped to form the quantified fault risk array. After surface integrity fault filtering of the quantified fault risk array based on preset surface fault types, normalized weighting based on fault type risk level is performed to obtain the photogrammetric attention array.

[0043] Furthermore, the route optimization component 11 is used to perform the following methods: The photogrammetric attention array is mapped to a geographic weighted raster map; after projecting the baseline route onto the geographic weighted raster map, the waypoint density is adaptively adjusted based on the raster weight values ​​to obtain a discrete waypoint sequence; the discrete waypoint sequence is smoothed using a B-spline curve to obtain a smoothed route, and then the ground clearance is verified to generate the risk perception inspection route.

[0044] Furthermore, the three-dimensional reconstruction component 14 is used to perform the following methods: Based on RTK localization constraints, the supplementary pose data and spatial pose data are unified to a geocentric Cartesian coordinate system to obtain a spatiotemporally registered pose dataset. Spatiotemporal alignment is performed on the supplementary visible light image stream and the initial visible light image stream based on UTC timestamps to obtain spatiotemporally aligned image frames. After SIFT feature point extraction on the spatiotemporally aligned image frames, joint bundle adjustment optimization is performed using the spatial constraints associated with the spatiotemporally registered pose dataset to output an optimized camera pose parameter set. A disparity spatial image is constructed based on the optimized camera pose parameter set, and a pixel-level disparity map is calculated using SGM. The elevation gradient distribution of the pixel-level disparity map and the initial DSM map is fused to perform gradient weighted compensation for the terrain occlusion region, resulting in an occlusion-compensated disparity map. After 3D point cloud transformation of the occlusion-compensated disparity map, Delaunay triangulation is performed by fusing the terrain structure features of the initial DSM map to output an optimized DSM map. The optimized DSM map is used to perform differential correction on the spatiotemporally aligned image frames to output the high-confidence TDOM map.

[0045] Furthermore, the data acquisition component 12 is used to perform the following methods: The geographic reference multispectral acquisition module includes an RTK high-precision positioning module, a visible light gimbal camera, and an infrared gimbal camera. During the flight of the main fixed-wing UAV, the RTK high-precision positioning module performs centimeter-level pose synchronization and acquires spatial pose data. The visible light gimbal camera and the infrared gimbal camera acquire sequential images to obtain the initial visible light image stream and the initial infrared thermal image stream.

[0046] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for generating photovoltaic power plant maps based on photogrammetry, characterized in that, The method includes: The baseline route is dynamically optimized based on a photogrammetric attention array to obtain a risk-aware inspection route; During the automatic flight of the main fixed-wing UAV along the risk perception inspection route in the photovoltaic power station, the onboard geographic reference multispectral acquisition module collects spatial pose data, initial visible light image stream and initial infrared thermal image stream. Based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and the initial visible light image stream, the subordinate rotary-wing UAV is activated to perform local low-altitude multi-angle supplementary scanning to obtain supplementary pose data, supplementary visible light image stream and supplementary infrared thermal image stream. By fusing the supplementary pose data and supplementary visible light image stream to the spatial pose data and the initial visible light image stream, a high-confidence TDOM map is obtained through joint 3D reconstruction with cross-view dense matching. Using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are fused into the high-confidence TDOM map to extract multispectral fault features, thereby differentiating and locating K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels. A risk-aware inspection route is obtained by dynamically optimizing the baseline route based on a photogrammetric attention array. The method includes: The real-time operation stream transmitted back by the photovoltaic array is retrieved to quantify the probability of abnormal equipment status. After obtaining the quantified fault risk array, surface integrity fault filtering is performed to construct a photogrammetric attention array. The risk perception inspection route is obtained by dynamically optimizing the baseline route based on the photogrammetric attention array. The method involves retrieving the real-time operation stream transmitted back from the photovoltaic array, quantifying the probability of abnormal equipment status, obtaining a quantified fault risk array, filtering for surface integrity faults, and constructing a photogrammetric attention array. The real-time operation flow is decomposed to obtain multiple panel-level operating parameters of multiple photovoltaic panels in the photovoltaic array. The panel-level operating parameters include current dispersion coefficient, power drop frequency, temperature-power deviation, and irradiance response deviation. Construct multiple operating feature vectors for the multiple board-level operating parameters, and match multiple sets of fault type probabilities in a preset fault probability model; Based on the geographical coordinates of the multiple photovoltaic panels in the photovoltaic array, the multiple sets of fault type probabilities are spatially mapped to form the quantified fault risk array; After filtering the quantized fault risk array for surface integrity faults based on preset surface fault types, normalized weighting based on fault type risk level is performed to obtain the photogrammetric attention array.

2. The method for generating photovoltaic power plant maps based on photogrammetry as described in claim 1, characterized in that, Based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and the initial visible light image stream, a slave rotary-wing UAV is initiated to perform local low-altitude multi-angle supplementary scanning to acquire supplementary pose data, supplementary visible light image stream, and supplementary infrared thermal image stream. The method includes: Motion recovery structure reconstruction is performed based on the spatial pose data and the initial visible light image stream to generate an initial TDOM map and an initial DSM map. Perform elevation gradient analysis on the initial DSM map to locate terrain-obscured areas; The elevation gradient distribution of the terrain-obstructed area is retrieved, and combined with the fault weights of the photogrammetric attention array, an adaptive rescanning path is constructed. The subordinate rotary-wing UAV is controlled to perform local low-altitude multi-angle supplementary scanning along the adaptive supplementary scanning route to acquire the supplementary pose data, supplementary visible light image stream, and supplementary infrared thermal image stream.

3. The photovoltaic power plant map generation method based on photogrammetry as described in claim 2, characterized in that, Using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are fused into the high-confidence TDOM to extract multispectral fault features, thereby differentiating and locating K surface-faulted photovoltaic panels and Q electrical-faulted photovoltaic panels. The method includes: The geographic coordinates of the photovoltaic panels in the initial TDOM map are mapped to the high-confidence TDOM map to spatially define the photovoltaic panel detection area; The time-series infrared temperature field is obtained by fusing the supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream. Using the photovoltaic panel detection area as a spatial reference, the temporal infrared temperature field is registered to the high-confidence TDOM through perspective transformation to generate a geographic reference multispectral fusion map. The photogrammetric attention array is superimposed on the georeferenced multispectral fusion map space to perform photovoltaic fault focusing detection, thereby differentiating and locating the K surface-faulted photovoltaic panels and the Q electrical-faulted photovoltaic panels.

4. The photovoltaic power plant map generation method based on photogrammetry as described in claim 3, characterized in that, The method involves overlaying the photogrammetric attention array onto the georeferenced multispectral fusion map space to perform photovoltaic fault focusing detection, thereby differentiating and locating the K surface-faulted photovoltaic panels and the Q electrical-faulted photovoltaic panels. The photogrammetric attention array is superimposed onto the georeferenced multispectral fusion map to generate a fault focus heatmap. The fault-focused heatmap is used to perform dual-path fault differentiation and localization based on the visible light channel and the infrared channel to obtain the K surface fault photovoltaic panels and Q electrical fault photovoltaic panels.

5. The photovoltaic power plant map generation method based on photogrammetry as described in claim 1, characterized in that, The method involves dynamically optimizing the baseline route based on the photogrammetric attention array to obtain the risk-aware inspection route, comprising: The photogrammetric attention array is mapped to a geographic weighted raster map; After projecting the baseline route onto the geographic weighted raster map, adaptive adjustment of waypoint density based on raster weight values ​​is performed to obtain a discrete waypoint sequence; The discrete waypoint sequence is smoothed using B-spline curves to obtain a smoothed route. After the altitude above ground is verified, the risk perception inspection route is generated.

6. The method for generating photovoltaic power plant maps based on photogrammetry as described in claim 2, characterized in that, By fusing the supplementary pose data and supplementary visible light image stream to the spatial pose data and the initial visible light image stream, and performing joint 3D reconstruction through cross-viewpoint dense matching, a high-confidence TDOM map is obtained. The method includes: Based on RTK localization constraints, the supplementary pose data and spatial pose data are unified to the geocentric Cartesian coordinate system to obtain a spatiotemporal registration pose dataset. Based on the UTC timestamp, spatiotemporal alignment is performed on the supplementary visible light image stream and the initial visible light image stream to obtain a spatiotemporally aligned image frame; After extracting SIFT feature points from the spatiotemporally aligned image frames, joint bundle adjustment optimization is performed based on the spatial constraints of the associated spatiotemporally registered pose dataset, outputting an optimized set of camera pose parameters. After constructing the disparity spatial image based on the optimized camera pose parameter set, pixel-level disparity maps are calculated using SGM. By fusing the pixel-level disparity map and the elevation gradient distribution of the initial DSM map, gradient weighted compensation is performed on the terrain occlusion area to obtain an occlusion-compensated disparity map. After performing 3D point cloud conversion on the occlusion compensation disparity map, the terrain structure features of the initial DSM map are fused together to perform Delaunay triangulation, and an optimized DSM map is output. The optimized DSM map is used to perform differential correction on the spatiotemporally aligned image frames, and the high-confidence TDOM map is output.

7. The method for generating photovoltaic power plant maps based on photogrammetry as described in claim 1, characterized in that, The geographic reference multispectral acquisition module includes an RTK high-precision positioning module, a visible light gimbal camera, and an infrared gimbal camera. During the flight of the main fixed-wing UAV, centimeter-level pose synchronization is achieved through the RTK high-precision positioning module, spatial pose data is collected, and sequential image acquisition is performed through the visible light gimbal camera and the infrared gimbal camera to obtain the initial visible light image stream and the initial infrared thermal image stream.

8. A photovoltaic power plant map generation system based on photogrammetry, characterized in that, The system for implementing the photovoltaic power plant map generation method based on photogrammetry according to any one of claims 1-7, the system comprising: Route optimization component: Dynamically optimizes the baseline route based on photogrammetric attention array to obtain a risk-aware inspection route; Data acquisition component: During the automatic flight of the main fixed-wing UAV along the risk perception inspection route in the photovoltaic power station, the onboard geographic reference multispectral acquisition module collects spatial pose data, initial visible light image stream and initial infrared thermal image stream; Data supplementation component: Based on the terrain occlusion area located by terrain occlusion analysis of the spatial pose data and the initial visible light image stream, the subordinate rotary-wing UAV is activated to perform local low-altitude multi-angle supplementary scanning to acquire supplementary pose data, supplementary visible light image stream and supplementary infrared thermal image stream. 3D Reconstruction Component: The supplementary pose data and supplementary visible light image stream are fused to the spatial pose data and the initial visible light image stream. Through joint 3D reconstruction with cross-view dense matching, a high-confidence TDOM map is obtained. Fault location component: Using the photogrammetric attention array as a spatial weighting operator, the supplementary infrared thermal imaging stream and the initial infrared thermal imaging stream are fused into the high-confidence TDOM map to extract multispectral fault features, so as to differentiate and locate K surface fault photovoltaic panels and Q electrical fault photovoltaic panels; The route optimization component is used to perform the following methods: The real-time operation stream transmitted back by the photovoltaic array is retrieved to quantify the probability of abnormal equipment status. After obtaining the quantified fault risk array, surface integrity fault filtering is performed to construct a photogrammetric attention array. The risk perception inspection route is obtained by dynamically optimizing the baseline route based on the photogrammetric attention array.

Images