Photovoltaic panel longitude and latitude positioning method and device based on unmanned aerial vehicle photography

Abstract

The invention belongs to the technical field of photovoltaic power station intelligent operation and maintenance, and particularly relates to a photovoltaic panel longitude and latitude positioning method and device based on unmanned aerial vehicle photography, and the method comprises the steps: obtaining a DOM image and a DSM image based on an unmanned aerial vehicle photography image, constructing a photovoltaic string detection model, so as to detect a photovoltaic string in the unmanned aerial vehicle photography image and the DOM image, and carrying out the detection of the photovoltaic string. The position of the photovoltaic string is described, and a photovoltaic string position descriptor is constructed; and based on the photovoltaic string position descriptor, matching the unmanned aerial vehicle photographic image with the photovoltaic string in the DOM image, and converting a central point pixel coordinate of the matched photovoltaic string in the DOM image into a ground latitude and longitude coordinate so as to realize latitude and longitude positioning of the photovoltaic string in the unmanned aerial vehicle image. The method solves the problems that a traditional positioning method depends on manpower, is low in efficiency, is insufficient in precision, remarkably improves the positioning efficiency and accuracy of the photovoltaic string in a complex terrain, and is suitable for the intelligent operation and maintenance management of a large-scale photovoltaic field station.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent operation and maintenance of photovoltaic power stations, and specifically relates to a method and device for locating the longitude and latitude of photovoltaic panels based on drone photography. Background Art

[0002] PV stations typically consist of thousands to tens of thousands of PV strings, distributed over a wide area and in complex terrain. Accurately locating the longitude and latitude coordinates of PV strings is crucial for troubleshooting, performance monitoring, and maintenance navigation. Traditional positioning methods rely primarily on manual measurement or GPS single-point positioning, which presents the following issues: Low positioning efficiency: Traditional methods rely on manual measurement or simple image processing, making it difficult to quickly and accurately obtain the longitude and latitude of PV strings; Insufficient accuracy: The lack of accurate detection of the location of photovoltaic strings makes it difficult for positioning accuracy to meet actual needs; Poor adaptability: Existing methods are generally difficult to adapt to the positioning requirements of complex terrain and large-scale photovoltaic sites.

[0003] Current methods for locating PV strings in photovoltaic (PV) plants primarily rely on drones for data collection, followed by manual identification or a combination of drone data to determine the longitude and latitude of the PV strings. This manual identification approach relies entirely on the experience of on-site maintenance personnel, resulting in low efficiency when it comes to inspecting and locating PV strings in large-scale PV plants, severely impacting maintenance efficiency. Alternatively, drone images can be combined with gimbal information, such as GPS signals and gimbal pitch angles, to automatically locate the longitude and latitude of PV strings in inspection images. However, current PV string positioning methods often rely on accurate drone pose data, which is easily affected by the actual image acquisition environment, resulting in reduced positioning accuracy. Furthermore, in PV plants with uneven terrain and large elevation differences, relying solely on drone images and gimbal data can even be ineffective. Therefore, using drone photography to locate the longitude and latitude of PV strings remains a challenging task.

[0004] Currently, some work has achieved certain results in photovoltaic string positioning tasks at photovoltaic stations. The relevant technologies are as follows: Chinese patent document CN119006552A discloses a method, terminal, and storage medium for zoning and locating photovoltaic panels at a photovoltaic station. The method includes: obtaining orthographic images, collecting orthographic images of the photovoltaic station via drone, and reconstructing and synthesizing the orthographic images into orthographic images with coordinate system information; structuring the photovoltaic panel positioning data in layers, dividing the photovoltaic station plan drawings according to requirements, and determining the specific positions of the photovoltaic panels on the photovoltaic station plan drawings; merging the data, placing the photovoltaic station plan drawings into the coordinate system information of the orthographic images, and matching the orthographic images with the photovoltaic station plan drawings so that the photovoltaic panels in the photovoltaic station plan drawings correspond to the photovoltaic panels in the orthographic images and obtain corresponding coordinate information and image information. This enables convenient retrieval and query of photovoltaic panels through a control platform, clarifying the specific location information and image information of abnormal photovoltaic panels.

[0005] Chinese patent document CN118261067A discloses a method, system, device, and medium for locating defective photovoltaic panels, relating to the field of computer vision technology. The method includes: using a preset defect detection model to perform defect detection on a pre-detection image to determine the coordinates of the defective pixel based on the defect detection results. Feature data of the pre-detection image is extracted; the feature data includes the resolution of the pre-detection image, the longitude and latitude corresponding to the center pixel, and the height of the drone relative to the ground. The unit pixel distance is obtained based on the focal length and imaging pixel spacing of the drone's camera, as well as the height of the drone relative to the ground; the unit pixel distance is the distance between adjacent unit pixels in the pre-detection image. The longitude and latitude of the defective pixel are determined based on the unit pixel distance, the defective pixel coordinates, the resolution of the pre-detection image, and the longitude and latitude corresponding to the center pixel. This solution can quickly and accurately locate the geographic location of defective photovoltaic panels in photovoltaic stations.

[0006] Chinese patent document CN117911497A discloses a method, system, device, and medium for locating photovoltaic panels based on drone imaging, relating to the field of computer vision technology. The method includes: obtaining the spatial rectangular coordinates of a target photovoltaic panel; the spatial rectangular coordinates are determined using data obtained by multi-angle photography and ranging of the target photovoltaic panel using a drone; converting the spatial rectangular coordinates to a geodetic coordinate system to obtain the longitude and latitude of the target photovoltaic panel; photographing the photovoltaic string where the target photovoltaic panel is located based on the longitude and latitude of the target photovoltaic panel to obtain a top-down view of the corresponding photovoltaic string; performing image segmentation processing on the top-down view of the photovoltaic string, and determining the position of the target photovoltaic panel within the photovoltaic string based on the segmentation results; and generating positioning information for the target photovoltaic panel based on the longitude and latitude of the target photovoltaic panel and its position within the photovoltaic string. This solution facilitates the rapid and accurate positioning of specific photovoltaic panels within a photovoltaic station.

[0007] Although the above-mentioned document CN119006552A performs zoning positioning of photovoltaics in photovoltaic stations, its application scenarios are limited. It requires the known image position of the photovoltaic strings to be positioned, and manual participation in the positioning process, which affects the inspection efficiency of photovoltaic stations. The method in document CN118261067A locates defective photovoltaic panels. This method uses the principle of camera imaging and uses the camera and the height of the drone to locate the longitude and latitude of the defective photovoltaic panels. This method relies on accurate internal parameters of the camera for geometric calculations. In addition, this method relies on the height relative to the ground when the drone is photographing. However, this height is difficult to obtain in the actual operation and maintenance environment of photovoltaic stations, especially in mountainous photovoltaic station environments with large altitude differences, which affects the accuracy of the longitude and latitude positioning of photovoltaic panels. The method disclosed in document CN117911497A uses data obtained from multi-angle photography and ranging to locate the longitude and latitude of photovoltaic panels. This method uses three-dimensional spatial information, and the longitude and latitude positioning accuracy of photovoltaic panels is relatively high, but it also brings a relatively high positioning cost. The positioning of a single photovoltaic panel requires the participation of multi-angle photography and ranging, which reduces the positioning efficiency during photovoltaic station inspections and is not conducive to application in actual production environments.

[0008] In summary, how to solve the above key issues and achieve high-precision, low-cost, and automated longitude and latitude positioning of photovoltaic strings based on drone photography is one of the problems that technicians in this field urgently need to solve. Summary of the Invention

[0009] The present invention aims to overcome at least one of the defects of the above-mentioned prior art and provide a photovoltaic panel latitude and longitude positioning method based on drone photography to solve the problems of low efficiency, insufficient accuracy and difficulty in adapting to complex terrain in the current operation and maintenance management of photovoltaic power stations. It provides photovoltaic power station operation and maintenance personnel with a reliable photovoltaic string latitude and longitude positioning method, significantly improving the operation and maintenance management efficiency of photovoltaic power stations.

[0010] The present invention also discloses a device loaded with a photovoltaic panel latitude and longitude positioning method based on drone photography.

[0011] The detailed technical solutions of the present invention are as follows: A photovoltaic panel latitude and longitude positioning method based on drone photography, the method comprising: S1. Obtain digital orthophoto DOM images and digital surface model DSM images based on drone photography images; S2. Constructing a photovoltaic string detection model to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, and describing the positions of the detected photovoltaic strings to construct photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image; S3. Based on the photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image, the photovoltaic strings in the drone photography image and the digital orthophoto DOM image are matched, and the center point pixel coordinates of the photovoltaic strings in the matched digital orthophoto DOM image are converted into ground longitude and latitude coordinates to achieve longitude and latitude positioning of the photovoltaic strings in the drone image.

[0012] Preferably, in S2, a photovoltaic string detection model is constructed to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, specifically comprising: Obtain digital orthophoto DOM images and drone photography images of photovoltaic stations, and use annotation tools to annotate the photovoltaic strings in them to construct a training sample set; Constructing a photovoltaic string detection model based on a deep learning algorithm and training it using the training sample set; Inputting the drone photography image and the digital orthophoto DOM image into the trained photovoltaic string detection model to obtain a detection result including the location and bounding box information of the photovoltaic string; The detection results are post-processed to filter out falsely detected and missed photovoltaic string information, and obtain the final location and bounding box information of the photovoltaic strings in the drone photography image and the digital orthophoto DOM image.

[0013] According to a preferred embodiment of the present invention, in S2, the detected positions of the photovoltaic strings are described to construct a photovoltaic string position descriptor in the drone photographic image, specifically including: Calculate the center coordinates of the drone photography image, that is, let the height of the drone photography image be , the width is w, then the coordinates of its center point for: ; Traverse each photovoltaic string in the drone photography image and calculate the center point coordinates of each photovoltaic string for: (2); In formula (2): They represent the first The minimum and maximum values ​​of the horizontal coordinates of each photovoltaic string; They represent the first The minimum and maximum values ​​of the vertical coordinates of each photovoltaic string; Taking the center point of the drone photography image as the first pole, establish the first polar coordinate system and calculate the relative offset vector of each photovoltaic string for: ; In formula (3): Indicates the first The relative offset of the horizontal and vertical coordinates of each photovoltaic string; Traverse the center point of each photovoltaic string in the drone photography image and calculate the distance and angle from the center point to the first pole: ; (5); In formulas (4) and (5): Indicates the first The distance from the center point coordinates of each photovoltaic string to the first pole; Indicates the first The angle between the line connecting the first photovoltaic string and the first pole and the polar axis of the first polar coordinate system; The calculated angle and distance ( As the position descriptor of each photovoltaic string in the drone photography image ,Right now: (6); In formula (6): is the number of PV strings.

[0014] According to a preferred embodiment of the present invention, in S2, the detected positions of the photovoltaic strings are described to construct photovoltaic string position descriptors in the digital orthophoto DOM image, specifically including: Traverse each photovoltaic string in the digital orthophoto DOM image and calculate the center point coordinates of each photovoltaic string for: (7); In formula (7): Respectively represent the first The minimum and maximum values ​​of the horizontal coordinates of each photovoltaic string; Respectively represent the first The minimum and maximum values ​​of the vertical coordinates of each photovoltaic string; The ground point projected on the ground by the center point of the drone photography image The second pole , establish the second polar coordinate system and calculate the relative offset vector of each photovoltaic string for: ; In formula (8): Indicates the first The relative offset of the horizontal and vertical coordinates of each photovoltaic string; Traverse the center point of each photovoltaic string in the digital orthophoto DOM image and calculate the distance and angle from the center point to the second pole: (9); (10); In formulas (9) and (10): Indicates the first The distance from the center point coordinates of each photovoltaic string to the second pole; Indicates the first The angle between the line connecting the first photovoltaic string and the second pole and the polar axis of the second polar coordinate system; The calculated angle and distance ( As the position descriptor of each photovoltaic string in the digital orthophoto DOM image ,Right now: (11); In formula (11): is the number of PV strings.

[0015] According to the preferred embodiment of the present invention, in S2, the ground point corresponding to the center of the drone photography picture is The calculation is based on the digital surface model (DSM) image of the photovoltaic station and the drone gimbal data, wherein the drone gimbal data includes the latitude and longitude coordinates of the drone body, the flight altitude, and the pitch angle, yaw angle, and roll angle of the camera gimbal; Calculate the ground point where the center point of the drone photography image is projected onto the ground , specifically including: Based on the engineering coordinates in the digital surface model (DSM) image of the photovoltaic station, the longitude and latitude coordinates of the UAV body are converted into DSM pixel coordinates O; Based on the yaw angle of the drone camera gimbal, the spatial coordinate point of the drone body is set as T, and a ray is drawn from the DSM pixel coordinate point O of the drone body. Starting from point O, all pixel points P on the ray are traversed to obtain the ground point corresponding to each pixel point. altitude and longitude and latitude; Calculate the ground point corresponding to each pixel The height difference between the spatial coordinate point T and the drone body , and use the Haversine formula to calculate the ground point corresponding to each pixel The distance from the DSM pixel coordinate point O of the drone body , based on the height difference and distance Calculate the ground point corresponding to each pixel The angle between the line connecting the spatial coordinate point T of the drone body and the plumb line of the drone body ,Right now: (12); Select the angle closest to the pitch angle of the drone camera gimbal Corresponding pixel points , and its corresponding ground point It is the ground point where the center point of the drone photography image is projected onto the ground.

[0016] According to a preferred embodiment of the present invention, in S3, matching the drone photographic image with the photovoltaic strings in the digital orthophoto DOM image specifically includes: The position descriptor of each photovoltaic string in the UAV photography image ( Sort by distance value from small to large; The position descriptor of each photovoltaic string in the digital orthophoto DOM image ( Sort by distance value from small to large; Traverse the position descriptors of each photovoltaic string in the sorted drone photography image ( , and calculate the position descriptor of each photovoltaic string in the sorted digital orthophoto DOM image ( Losses between In a single matching, the photovoltaic string with the least loss in the digital orthophoto DOM image is selected as the matching result, and forms a matching pair with the corresponding photovoltaic string in the drone photography image; Use the bidirectional optimal search method to iterate the single matching results, that is: For the position descriptors of each photovoltaic string in the sorted drone photography image ( and the position descriptor of each photovoltaic string in the digital orthophoto DOM image ( , build a double-ended priority queue; Use a bidirectional greedy matching strategy for matching: From the sorted position description subsequences of the source end of the drone photography image and the target end of the digital orthophoto DOM image, matching is started from the minimum angle or distance value end and the maximum angle or distance value end of the double-ended priority queue at the same time, and the current matching pointers in the two directions are maintained respectively, including the head pointer and the tail pointer; For the head pointer descriptor on the source side, find the most similar descriptor within the current head and tail pointer range on the target side, calculate the angle difference and distance similarity respectively, and select the one with the highest comprehensive score as the candidate match; at the same time, perform the same operation in reverse from the tail pointer on the target side; The matching is terminated when at least one end of the double-ended priority queue is empty or the angle ranges of the remaining unmatched descriptors do not overlap; During the matching process, an error matching mechanism is introduced, and the polar coordinate transformation model of the matching pairs is fitted using the RANSAC geometric verification method, and abnormal matching pairs that deviate from this model are eliminated. Using topological consistency check, a Delaunay triangulation is constructed for the retained matching pairs, and matching pairs that destroy the spatial adjacency relationship are removed to obtain the final matching results.

[0017] According to the preferred embodiment of the present invention, in S3, polar coordinate-scale adaptive loss is adopted. As the position descriptor of each photovoltaic string in the drone photography image ( Compared with the position descriptor of each photovoltaic string in the digital orthophoto DOM image ( Losses between The polar-scale adaptive loss Including orientation consistency loss , scale invariance loss and topology constraint loss ,Right now: (13); In formula (13): 、 、 All are adaptive adjustment coefficients; Among them, the orientation consistency loss Designed to resist rotation interference, specifically: (14); The scale invariance loss Designed to achieve cross-resolution adaptation, specifically: (15); The topology constraint loss Designed to maintain spatial relationships, specifically: (16); In formula (16): is the number of PV strings, Respectively represent the indivual PV strings; The adaptive adjustment coefficient 、 、 The initial value is set to: , which is dynamically optimized through the following formula: .

[0018] In another aspect of the present invention, a device for implementing a method for locating the latitude and longitude of a photovoltaic panel based on drone photography is provided, the device comprising: An acquisition module is used to acquire digital orthophoto DOM images and digital surface model DSM images based on drone photography images; A construction module is used to construct a photovoltaic string detection model to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, and describe the positions of the detected photovoltaic strings to construct photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image; A positioning module is used to match the photovoltaic strings in the drone photography image and the digital orthophoto DOM image based on the photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image, and convert the center point pixel coordinates of the photovoltaic strings in the matched digital orthophoto DOM image into ground longitude and latitude coordinates to achieve longitude and latitude positioning of the photovoltaic strings in the drone image; The deployment process of the photovoltaic string detection model is as follows: export the photovoltaic string detection model as an ONNX model and use the ONNX optimization tool to simplify the exported ONNX model; then use the RKNN Toolkit to load the optimized ONNX model, quantize the ONNX model, and convert the floating-point ONNX model into a fixed-point RKNN model; finally, deploy the generated RKNN model on the device platform.

[0019] In another aspect of the present invention, an electronic device is provided, comprising: at least one processor; and A memory storing instructions, which, when executed by the at least one processor, enables the at least one processor to execute the photovoltaic panel latitude and longitude positioning method based on drone photography as described above.

[0020] In another aspect of the present invention, a machine-readable storage medium is provided, which stores executable instructions. When the instructions are executed, the machine executes the photovoltaic panel latitude and longitude positioning method based on drone photography as described above.

[0021] Compared with the prior art, the present invention has the following beneficial effects: The present invention discloses and implements an efficient, low-cost, stable and accurate, terminal-deployable method for the longitude and latitude positioning of photovoltaic strings by drones. Compared with traditional photovoltaic string longitude and latitude methods, this method does not require human intervention and can automatically complete the longitude and latitude positioning of photovoltaic strings in photovoltaic stations at low cost, greatly reducing the cost of locating faulty photovoltaic strings during the operation and maintenance of photovoltaic stations, and improving the operation and maintenance efficiency of large-scale photovoltaic stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of the photovoltaic panel latitude and longitude positioning method based on drone photography described in the present invention.

[0023] Figure 2 It is a schematic diagram of a digital orthophoto DOM image in Example 1 of the present invention.

[0024] Figure 3 It is a schematic diagram of a digital surface model DSM image in Example 1 of the present invention.

[0025] Figure 4 It is a top view of the center point of the drone photography image projected onto the ground in Example 1 of the present invention.

[0026] Figure 5 This is a side view of the center point of the drone photography image projected onto the ground in Example 1 of the present invention.

[0027] Figure 6 Schematic diagram of the bounding box of the photovoltaic string in the digital orthophoto DOM image in Example 1 of the present invention.

[0028] Figure 7 Schematic diagram of the bounding box of a photovoltaic string in an image taken by a drone in Example 1 of the present invention.

[0029] Figure 8 Schematic diagram of the position descriptors of each photovoltaic string in the drone photography image in Example 1 of the present invention.

[0030] Figure 9 Schematic diagram of the matching results of the UAV photography image and the photovoltaic strings in the digital orthophoto DOM image in Example 1 of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0033] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0034] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0035] To address the current problems of low efficiency, insufficient accuracy, and difficulty in adapting to complex terrain in photovoltaic string positioning in the operation and maintenance of photovoltaic power stations, the present invention provides a photovoltaic panel latitude and longitude positioning method based on drone photography. The method uses a drone to obtain an orthophoto map (DOM) and digital surface model (DSM) of the photovoltaic station, combines pan-tilt attitude data with a deep learning model to accurately detect the position of photovoltaic strings; constructs a photovoltaic string position descriptor to match the photovoltaic strings in the drone image with the ground DOM image; based on the DOM geographic reference information, the pixel coordinates are converted into longitude and latitude coordinates; and finally, the above method is deployed on an edge computing terminal to achieve high-precision automated positioning of photovoltaic strings.

[0036] The present invention significantly improves the positioning efficiency and accuracy of photovoltaic strings in complex terrains and is suitable for intelligent operation and maintenance management of large-scale photovoltaic stations.

[0037] The photovoltaic panel latitude and longitude positioning method and device based on drone photography of the present invention will be further described below in conjunction with specific embodiments.

[0038] Example 1 Ginseng Figure 1 This embodiment provides a method for locating the longitude and latitude of photovoltaic panels based on drone photography, the method comprising: S1. Obtain digital orthophoto DOM images and digital surface model DSM images based on drone photography images; S2. Constructing a photovoltaic string detection model to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, and describing the positions of the detected photovoltaic strings to construct photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image; S3. Based on the photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image, the photovoltaic strings in the drone photography image and the digital orthophoto DOM image are matched, and the center point pixel coordinates of the photovoltaic strings in the matched digital orthophoto DOM image are converted into ground longitude and latitude coordinates to achieve longitude and latitude positioning of the photovoltaic strings in the drone image.

[0039] In this embodiment, a large photovoltaic station is taken as an example to further illustrate the above method.

[0040] In S1, the digital orthophoto DOM image and the digital surface model DSM image are obtained based on the drone photography image, specifically: Use the high-resolution camera on the drone to take aerial photos of the photovoltaic station. The drone flight control system ensures that the camera's posture is stable during the aerial photography process to avoid image distortion.

[0041] The image sequences collected by the drone were then preprocessed, including denoising, image enhancement, and geometric correction, to ensure image quality met the requirements of subsequent analysis. Software was then used to synthesize digital orthophoto (DOM) images and digital surface model (DSM) images of the photovoltaic station.

[0042] Among them, the generated digital orthophoto DOM image parameters Figure 2 As shown; the generated digital surface model DSM image parameters Figure 3 shown.

[0043] In S2, a photovoltaic string detection model is constructed to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, which specifically includes the following steps.

[0044] S2-1. Obtain digital orthophoto DOM images and drone photography images of photovoltaic stations, and use annotation tools to annotate the photovoltaic strings in them to construct a training sample set.

[0045] In this embodiment, a large number of DOM images and drone photography images of photovoltaic stations are collected to construct a training sample set; and labeling tools, such as the LabelImg tool, are used to label the photovoltaic strings in the training sample set to ensure that the model can accurately identify the photovoltaic strings.

[0046] S2-2. Construct a photovoltaic string detection model based on a deep learning algorithm, and train it using the training sample set.

[0047] Preferably, a deep learning algorithm is used to construct a photovoltaic string detection model, the model input is a DOM image or a drone photography image, and the output is the location and bounding box of the photovoltaic string.

[0048] The photovoltaic string detection model is trained through the training sample set to optimize the model parameters and improve the detection accuracy.

[0049] S2-3. Input the drone photography image and the digital orthophoto DOM image into the trained photovoltaic string detection model to obtain a detection result including the location and bounding box information of the photovoltaic string.

[0050] That is, the drone photography images and ground DOM images are input into the trained photovoltaic string detection model, and the model detection outputs the location and bounding box information of the photovoltaic strings.

[0051] S2-4. Post-process the detection results to filter out falsely detected and missed photovoltaic string information, and obtain the final positions and bounding box information of the photovoltaic strings in the UAV photography image and the digital orthophoto DOM image.

[0052] Specifically, detection results can be post-processed using non-maximum suppression (NMS) or soft-NMS algorithms to remove redundant detection frames with high overlap and reduce false detections. Furthermore, a confidence threshold can be set to filter out low-confidence detection results, further reducing the false detection rate and eliminating falsely detected and missed PV strings, ensuring the accuracy of detection results.

[0053] The bounding box information parameters of the photovoltaic strings in the final digital orthophoto DOM image are Figure 6 As shown in the figure, the bounding box information of the photovoltaic string in the drone photography image is Figure 7 shown.

[0054] Furthermore, in S2, the detected positions of the photovoltaic strings are described to construct a photovoltaic string position descriptor in the drone photography image, which specifically includes the following steps.

[0055] S2-5. Calculate the coordinates of the center point of the drone photography image, that is, assume the height of the drone photography image is , the width is w, then the coordinates of its center point for: .

[0056] S2-6, traverse each photovoltaic string in the drone photography image, and calculate the center point coordinates of each photovoltaic string for: (2); In formula (2): They represent the first The minimum and maximum values ​​of the horizontal coordinates of each photovoltaic string; They represent the first The minimum and maximum values ​​of the vertical coordinates of each photovoltaic string.

[0057] S2-7, using the center point of the drone photography image as the first pole, establish a first polar coordinate system and calculate the relative offset vector of each photovoltaic string for: ; In formula (3): Indicates the first The relative offset of the horizontal and vertical coordinates of each photovoltaic string.

[0058] S2-8. Traverse the center point of each photovoltaic string in the drone photography image and calculate the distance and angle from the center point to the first pole: ; (5); In formulas (4) and (5): Indicates the first The distance from the center point coordinates of each photovoltaic string to the first pole; Indicates the first The angle between the line connecting the first photovoltaic string and the first pole and the polar axis of the first polar coordinate system.

[0059] S2-9, calculate the angle and distance ( As the position descriptor of each photovoltaic string in the drone photography image ,Right now: (6); In formula (6): is the number of PV strings.

[0060] Furthermore, in S2, the detected positions of the photovoltaic strings are described to construct photovoltaic string position descriptors in the digital orthophoto DOM image, which specifically includes the following steps.

[0061] S2-10, traverse each photovoltaic string in the digital orthophoto DOM image, and calculate the center point coordinates of each photovoltaic string for: (7); In formula (7): Respectively represent the first The minimum and maximum values ​​of the horizontal coordinates of each photovoltaic string; Respectively represent the first The minimum and maximum values ​​of the vertical coordinates of each photovoltaic string.

[0062] S2-11. Projecting the center point of the drone's photographic image onto the ground The second pole , establish the second polar coordinate system and calculate the relative offset vector of each photovoltaic string for: ; In formula (8): Indicates the first The relative offset of the horizontal and vertical coordinates of each photovoltaic string.

[0063] S2-12, traverse the center point of each photovoltaic string in the digital orthophoto DOM image, and calculate the distance and angle from the center point to the second pole respectively: (9); (10); In formulas (9) and (10): Indicates the first The distance from the center point coordinates of each photovoltaic string to the second pole; Indicates the first The angle between the line connecting the first photovoltaic string and the second pole and the polar axis of the second polar coordinate system.

[0064] S2-13, calculate the angle and distance ( As the position descriptor of each photovoltaic string in the digital orthophoto DOM image ,Right now: (11); In formula (11): is the number of PV strings.

[0065] Ginseng Figure 8 , shows a schematic image of the position descriptor of each photovoltaic string in the drone photography image.

[0066] Furthermore, in S2, the ground point corresponding to the center of the drone photography picture It is calculated based on the digital surface model (DSM) image of the photovoltaic station and the drone gimbal data, which includes the longitude and latitude coordinates of the drone body, the flight altitude, and the pitch, yaw, and roll angles of the camera gimbal.

[0067] In this embodiment, the ground point projected by the center point of the drone photography image on the ground is calculated. , specifically including: 1) Based on the engineering coordinates in the digital surface model (DSM) image of the photovoltaic station, the longitude and latitude coordinates of the UAV body are converted into DSM pixel coordinates O; 2) Based on the yaw angle of the drone camera gimbal, set the spatial coordinate point of the drone body as T, draw a ray starting from the DSM pixel coordinate point O of the drone body, and traverse all the pixel points P on the ray along point O to obtain the ground point corresponding to each pixel point in the digital surface model DSM image altitude and longitude and latitude; 3) Calculate the ground point corresponding to each pixel The height difference between the spatial coordinate point T and the drone body , and use the Haversine formula to calculate the ground point corresponding to each pixel The distance from the DSM pixel coordinate point O of the drone body , then based on the height difference and distance Calculate the ground point corresponding to each pixel The angle between the line connecting the spatial coordinate point T of the drone body and the plumb line of the drone body ,Right now: (12); The above process can be referred to Figure 4 and Figure 5 As shown; 4) Select the angle closest to the pitch angle of the drone camera gimbal Corresponding pixel points , and its corresponding ground point It is the ground point where the center point of the drone photography image is projected onto the ground.

[0068] In the above S3, based on the photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image, the photovoltaic strings in the drone photography image and the digital orthophoto DOM image are matched, which specifically includes the following steps.

[0069] S3-1, the position descriptor of each photovoltaic string in the drone photography image ( Sort by distance value from small to large.

[0070] S3-2, the position descriptor of each photovoltaic string in the digital orthophoto DOM image ( Sort by distance value from small to large.

[0071] S3-3, traverse the position descriptors of each photovoltaic string in the sorted drone photography image ( , and calculate the position descriptor of each photovoltaic string in the sorted digital orthophoto DOM image ( The loss between.

[0072] To this end, the method of this embodiment designs the following "Polar-Scale Adaptive Loss" (PSAL). This loss function combines the set and topological characteristics, and is more suitable for drone oblique photography to match the set graphics between the perspective image and the DOMG orthophoto image. Loss of orientation consistency , scale invariance loss and topology constraint loss Composition, namely: (13); In formula (13): 、 、 Both are adaptive adjustment coefficients.

[0073] Among them, the orientation consistency loss Designed to resist rotation interference, specifically: (14); This part of the loss function uses the annular distance metric to solve the problem of large periodic angle differences across 360°.

[0074] The scale invariance loss Designed to achieve cross-resolution adaptation, specifically: (15); The loss function design of this part is normalized by ground sampling distance (GSD) to eliminate the scale difference between drone and DOM images.

[0075] The topology constraint loss Designed to maintain spatial relationships, specifically: (16); In formula (16): is the number of PV strings, Respectively represent the indivual PV strings.

[0076] The design of this partial loss function can maintain the relative distance ratio between photovoltaic strings and enhance the matching robustness.

[0077] Furthermore, the adaptive adjustment coefficient 、 、 The initial value is set to: , which can be dynamically optimized according to the matching confidence, as follows: .

[0078] S3-4. For a single match, the photovoltaic string with the smallest loss in the digital orthophoto DOM image is selected as the matching result, and a matching pair is formed with the corresponding photovoltaic string in the drone photography image. Figure 9 shown.

[0079] S3-5. Use the bidirectional optimal search method to iterate the single matching results, namely: 1) For the position descriptors of each photovoltaic string in the sorted drone photography image ( and the position descriptor of each photovoltaic string in the digital orthophoto DOM image ( , build a double-ended priority queue; 2) Use a bidirectional greedy matching strategy for matching: From the sorted position description subsequences of the UAV image (source) and the DOM image (destination), matching is initiated simultaneously from the head and tail ends (i.e., the minimum and maximum polar angle or distance values) of the double-ended priority queue, and the current matching pointers (head pointer and tail pointer) in both directions are maintained respectively; For the head pointer descriptor on the source side, the most similar descriptor is found within the range of the current head and tail pointers on the target side (e.g., the head pointer ± k neighborhood), and the polar angle difference and distance similarity (e.g., Euclidean distance) are calculated. The one with the highest comprehensive score is selected as the candidate match. At the same time, the same operation is performed in reverse from the tail pointer on the target side. The matching is terminated when at least one end of the double-ended priority queue is empty or the polar angle ranges of the remaining unmatched descriptors do not overlap.

[0080] 3) During the matching process, a false match proposal mechanism is introduced, and a polar coordinate transformation model of the matching pairs is fitted using RANSAC geometric verification. Abnormal matching pairs that deviate from this model are eliminated. 4) Use topological consistency check to construct Delaunay triangulation for the retained matching pairs, remove matching pairs that destroy the spatial adjacency relationship, and obtain the final matching results.

[0081] Based on the above, after completing the matching between the drone photography image and the ground DOM photovoltaic strings, the pixel coordinates of the center point of the matched ground DOM strings are converted into ground longitude and latitude coordinates according to the engineering coordinate system stored in the DOM image, thereby realizing the longitude and latitude positioning function of the photovoltaic strings in the drone image.

[0082] In summary, the present invention discloses a method for determining the ground projection of the center point of a drone image, which is based on the drone posture information and GPS information, and innovatively combines the longitude and latitude and elevation information in the DSM image of the photovoltaic station. With the help of the prior data of the photovoltaic station, the actual ground longitude and latitude position corresponding to the center point of the drone photography image is first determined, and then the photovoltaic strings on the drone image side are linked to the photovoltaic strings on the ground side, and the photovoltaic string longitude and latitude positioning task is converted into a photovoltaic string matching problem, which greatly reduces the difficulty of photovoltaic string longitude and latitude positioning and improves the efficiency of photovoltaic string longitude and latitude positioning.

[0083] The present invention also discloses a photovoltaic string matching and positioning method, which converts the longitude and latitude positioning task of the UAV photovoltaic strings into a photovoltaic string matching task. By calculating the descriptors of the photovoltaic strings on the UAV screen side and the DOM ground side, the positions of the photovoltaic strings on the UAV screen side and the DOM ground side are quantitatively described, and the relative position description of the photovoltaic strings is converted into the distance loss between the descriptors, thereby realizing simple and efficient matching of the UAV screen and the ground photovoltaic strings, and then completing the longitude and latitude positioning of the photovoltaic strings.

[0084] Example 2 This embodiment provides a device for implementing a method for longitude and latitude positioning of photovoltaic panels based on drone photography, that is, deploying the above method on a terminal hardware device, so that the device can autonomously complete the positioning of photovoltaic modules in a photovoltaic station.

[0085] The device comprises: An acquisition module is used to acquire digital orthophoto DOM images and digital surface model DSM images based on drone photography images; A construction module is used to construct a photovoltaic string detection model to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, and describe the positions of the detected photovoltaic strings to construct photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image; A positioning module is used to match the photovoltaic strings in the drone photography image and the digital orthophoto DOM image based on the photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image, and convert the center point pixel coordinates of the photovoltaic strings in the matched digital orthophoto DOM image into ground longitude and latitude coordinates to achieve longitude and latitude positioning of the photovoltaic strings in the drone image; The deployment process of the photovoltaic string detection model is as follows: First, the photovoltaic string detection model is exported as an ONNX model in the onnx format, and an ONNX optimization tool, such as the onnx-simplifier tool, is used to simplify the exported ONNX model, remove redundant operations, and reduce model complexity; Subsequently, the optimized ONNX model was loaded using the RKNN Toolkit and quantized to convert the floating-point ONNX model into a fixed-point RKNN model, aiming to improve the inference speed of the model on the RV1126 terminal device. Subsequently, the generated RKNN model is deployed to the device platform, and the model or hardware configuration is further optimized based on the actual operating performance (such as inference speed, memory usage, etc.).

[0086] At the same time, the other parts of the method are converted into terminal programs and deployed to the RV1126 platform to enable the terminal platform to autonomously perform photovoltaic string positioning.

[0087] The effectiveness of the method of the present invention is further verified by combining specific application examples as follows: 1. Use drone photography to collect orthophoto image sequences of the photovoltaic station, and use step S1 of this method to obtain DOM and DSM images of the photovoltaic station; 2. Calculate the position of the center point of the photographic image on the ground using step S2 of this method; 3. Constructing a photovoltaic string detection model according to step S2 of the present method to detect photovoltaic strings in the drone photography image and the ground DOM image; 4. Constructing a photovoltaic string position descriptor according to step S2 of this method; 5. According to the method in step S3 of this method, the drone image is matched with the ground photovoltaic strings and the longitude and latitude positioning is achieved; 6. Deploy the above method to the terminal device to realize autonomous operation of the terminal.

[0088] Based on the above, the method and device of the present invention are further described below in combination with an application scenario of a 50MW photovoltaic station located in the desert area of ​​Xinjiang.

[0089] A 50MW photovoltaic power station located in the desert region of Xinjiang, containing 12,000 photovoltaic strings, covers an area of ​​approximately 30 hectares and includes 3,200 photovoltaic strings. It is necessary to locate the longitude and latitude of the photovoltaic strings during drone inspections. The specific implementation process of the method of the present invention is as follows: (1) Obtaining DOM and DSM images of photovoltaic stations: A DJI M300 RTK drone equipped with a Hasselblad H6D camera was used to collect orthophoto sequences of the site at an altitude of 80 meters and an 80% overlap rate. Geometric correction and stitching were performed using DJI Maps software to generate DOM and DSM images with a resolution of 2 cm / pixel, using the WGS84-UTM Zone 48N coordinate system.

[0090] (2) Calculate the position of the center point of the photographic image on the ground: The pixel coordinates of the center point of the drone's photographic image are calculated (3000, 2000). The pitch angle of the DJI M300RTK drone gimbal is read as 62.5°, the yaw angle is 0°, and the drone's GPS coordinates are (81.30xxxxE, 40.71xxxxN). The engineering coordinate system in the DOM and DSM images is read, and the GPS longitude and latitude coordinates of the drone are mapped to DSM pixel coordinates. According to step S2 of this method, the DSM pixel coordinates of the center point of the photographic image on the ground are calculated and converted into longitude and latitude coordinates using the engineering coordinate system.

[0091] (3) Construct a PV string detection model using drones to detect PV strings in drone photography images and ground DOM images: The YOLOv8 model was trained on a dataset consisting of 10,000 labeled drone images and segmented DOM images (covering complex terrain scenes such as deserts and mountains). Training parameters included an initial learning rate of 0.001, a batch size of 16, and 300 epochs. The model achieved a mAP@0.5 score of 98.2% on the test set.

[0092] After the model is deployed on the terminal, it takes 0.5 seconds per detection to detect photovoltaic strings in drone images, outputting the photovoltaic string bounding boxes on each frame of the drone image, with an error detection rate of less than 3%. For ground-based DOM images, the model detects photovoltaic strings in the DOM images in blocks, taking 5 seconds per detection and outputting 3,200 sets of photovoltaic string bounding boxes on the DOM images, with an error detection rate of less than 5%. Manual filling is used for misdetected and missed photovoltaic strings.

[0093] (4) Construct a photovoltaic string descriptor and match the photovoltaic strings in the drone photography image and the DOM image: PV string position descriptors were constructed in the UAV photography images and the ground DOM images according to step S2 of this method, and the UAV images were matched with the ground PV strings according to step S3 of this method. All PV strings in the UAV photography images were matched. For the 500 image sequences captured by UAV photography, 453 PV strings were correctly matched, with a PV string matching success rate of 95%.

[0094] The pixel coordinates of the matched ground photovoltaic strings are mapped into the latitude and longitude coordinates of the photovoltaic strings using the engineering coordinate system saved in the DOM image. The coordinate mapping process is point-by-point, and the mapping accuracy is 100%.

[0095] (5) Terminal deployment and actual measurement.

[0096] Deployed on Rockchip RV1126 platform: Model quantization: FP32 → INT8, compressing the model size from 103MB to 23MB; Memory usage: Peak memory <512MB; End-to-end latency: from DOM and DSM input to positioning result output, the entire process is less than 15 seconds; Accuracy: After the terminal hardware is deployed, the comprehensive positioning accuracy of this method is 95%.

[0097] In summary, the terminal deployment solution for the longitude and latitude positioning of photovoltaic strings based on drone photography disclosed in the present invention deploys the method on the terminal device. During the inspection of the photovoltaic station by the drone, the position of the photovoltaic strings in the picture can be located without cloud data transmission, and the longitude and latitude coordinates of each photovoltaic string can be output, thereby improving the operation and maintenance efficiency of the photovoltaic station and further liberating productivity.

[0098] Furthermore, the present invention was deployed and tested in a 150mW-class photovoltaic station, and only a low-cost edge computing platform was required to achieve autonomous longitude and latitude positioning of photovoltaic strings, providing a technical solution for the subsequent precise positioning of defective photovoltaic strings.

[0099] Example 3 This embodiment further provides an electronic device, including: at least one processor; and A memory storing instructions, which, when executed by the at least one processor, enables the at least one processor to execute the photovoltaic panel latitude and longitude positioning method based on drone photography as described above.

[0100] In this embodiment, electronic devices may include, but are not limited to: personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile computing devices, smart phones, tablet computers, cellular phones, personal digital assistants (PDAs), handheld devices, messaging devices, wearable computing devices, consumer electronic devices, and the like.

[0101] Example 4 This embodiment also provides a machine-readable storage medium storing executable instructions, which, when executed, enable the machine to execute the photovoltaic panel latitude and longitude positioning method based on drone photography as described above.

[0102] Specifically, a system or device equipped with a readable storage medium can be provided, on which software program codes that implement the functions of any of the above-mentioned embodiments are stored, and a computer or processor of the system or device can read and execute instructions stored in the readable storage medium.

[0103] In this case, the program code itself read from the machine-readable medium can implement the functions of any one of the above embodiments, and thus the machine-readable code and the machine-readable storage medium storing the machine-readable code constitute part of this specification.

[0104] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (e.g., CD-ROMs, CD-Rs, CD-RWs, DVD-ROMs, DVD-RAMs, DVD-RWs, DVD-RWs), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code may be downloaded from a server computer or a cloud via a communication network.

[0105] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0106] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0107] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1The function specified in one or more boxes.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0109] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the technical solutions of the present invention, and are not intended to limit the specific implementation methods of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A photovoltaic panel latitude and longitude positioning method based on drone photography, characterized in that: The method comprises: S1. Obtain digital orthophoto DOM images and digital surface model DSM images based on drone photography images; S2. Constructing a photovoltaic string detection model to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, and describing the positions of the detected photovoltaic strings to construct photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image; S3. Based on the photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image, the photovoltaic strings in the drone photography image and the digital orthophoto DOM image are matched, and the center point pixel coordinates of the photovoltaic strings in the matched digital orthophoto DOM image are converted into ground longitude and latitude coordinates to achieve longitude and latitude positioning of the photovoltaic strings in the drone image.

2. The photovoltaic panel latitude and longitude positioning method based on drone photography according to claim 1 is characterized in that: In S2, a photovoltaic string detection model is constructed to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, specifically including: Obtain digital orthophoto DOM images and drone photography images of photovoltaic stations, and use annotation tools to annotate the photovoltaic strings in them to construct a training sample set; Constructing a photovoltaic string detection model based on a deep learning algorithm and training it using the training sample set; Inputting the drone photography image and the digital orthophoto DOM image into the trained photovoltaic string detection model to obtain a detection result including the location and bounding box information of the photovoltaic string; The detection results are post-processed to filter out falsely detected and missed photovoltaic string information, and obtain the final location and bounding box information of the photovoltaic strings in the drone photography image and the digital orthophoto DOM image.

3. The photovoltaic panel latitude and longitude positioning method based on drone photography according to claim 1 is characterized in that: In S2, the detected positions of the photovoltaic strings are described to construct a photovoltaic string position descriptor in the drone photography image, specifically including: Calculate the center coordinates of the drone photography image, and set the height of the drone photography image to be , the width is w, then the coordinates of its center point for: ； Traverse each photovoltaic string in the drone photography image and calculate the center point coordinates of each photovoltaic string for: (2)； In formula (2): They represent the first The minimum and maximum values ​​of the horizontal coordinates of each photovoltaic string; They represent the first The minimum and maximum values ​​of the vertical coordinates of each photovoltaic string; Taking the center point of the drone photography image as the first pole, establish the first polar coordinate system and calculate the relative offset vector of each photovoltaic string for: ； In formula (3): Indicates the first The relative offset of the horizontal and vertical coordinates of each photovoltaic string; Traverse the center point of each photovoltaic string in the drone photography image and calculate the distance and angle from the center point to the first pole: ； (5)； In formulas (4) and (5): Indicates the first The distance from the center point coordinates of each photovoltaic string to the first pole; Indicates the first The angle between the line connecting the first photovoltaic string and the first pole and the polar axis of the first polar coordinate system; The calculated angle and distance ( As the position descriptor of each photovoltaic string in the drone photography image : (6)； In formula (6): is the number of PV strings.

4. The photovoltaic panel latitude and longitude positioning method based on drone photography according to claim 3 is characterized in that: In S2, the detected positions of the photovoltaic strings are described to construct a photovoltaic string position descriptor in the digital orthophoto DOM image, specifically including: Traverse each photovoltaic string in the digital orthophoto DOM image and calculate the center point coordinates of each photovoltaic string for: (7)； In formula (7): Respectively represent the first The minimum and maximum values ​​of the horizontal coordinates of each photovoltaic string; Respectively represent the first The minimum and maximum values ​​of the vertical coordinates of each photovoltaic string; The ground point projected on the ground by the center point of the drone photography image The second pole , establish the second polar coordinate system and calculate the relative offset vector of each photovoltaic string for: ； In formula (8): Indicates the first The relative offset of the horizontal and vertical coordinates of each photovoltaic string; Traverse the center point of each photovoltaic string in the digital orthophoto DOM image and calculate the distance and angle from the center point to the second pole: (9)； (10)； In formulas (9) and (10): Indicates the first The distance from the center point coordinates of each photovoltaic string to the second pole; Indicates the first The angle between the line connecting the first photovoltaic string and the second pole and the polar axis of the second polar coordinate system; The calculated angle and distance ( As the position descriptor of each photovoltaic string in the digital orthophoto DOM image : (11)； In formula (11): is the number of PV strings.

5. The photovoltaic panel latitude and longitude positioning method based on drone photography according to claim 4 is characterized in that: In S2, the ground point corresponding to the center of the drone photography image The calculation is based on the digital surface model (DSM) image of the photovoltaic station and the drone gimbal data, wherein the drone gimbal data includes the latitude and longitude coordinates of the drone body, the flight altitude, and the pitch angle, yaw angle, and roll angle of the camera gimbal; Calculate the ground point where the center point of the drone photography image is projected onto the ground , specifically including: Based on the engineering coordinates in the digital surface model (DSM) image of the photovoltaic station, the longitude and latitude coordinates of the UAV body are converted into DSM pixel coordinates O; Based on the yaw angle of the drone camera gimbal, the spatial coordinate point of the drone body is set as T, and a ray is drawn from the DSM pixel coordinate point O of the drone body. Starting from point O, all pixel points P on the ray are traversed to obtain the ground point corresponding to each pixel point. altitude and longitude and latitude; Calculate the ground point corresponding to each pixel The height difference between the spatial coordinate point T and the drone body , and use the Haversine formula to calculate the ground point corresponding to each pixel The distance from the DSM pixel coordinate point O of the drone body , based on the height difference and distance Calculate the ground point corresponding to each pixel The angle between the line connecting the spatial coordinate point T of the drone body and the plumb line of the drone body : (12)； Select the angle closest to the pitch angle of the drone camera gimbal Corresponding pixel points , and its corresponding ground point It is the ground point where the center point of the drone photography image is projected onto the ground.

6. The photovoltaic panel latitude and longitude positioning method based on drone photography according to claim 4 is characterized in that: In S3, matching the drone image with the photovoltaic strings in the digital orthophoto DOM image specifically includes: The position descriptor of each photovoltaic string in the UAV photography image ( Sort by distance value from small to large; The position descriptor of each photovoltaic string in the digital orthophoto DOM image ( Sort by distance value from small to large; Traverse the position descriptors of each photovoltaic string in the sorted drone photography image ( , and calculate the position descriptor of each photovoltaic string in the sorted digital orthophoto DOM image ( Losses between In a single matching, the photovoltaic string with the least loss in the digital orthophoto DOM image is selected as the matching result, and forms a matching pair with the corresponding photovoltaic string in the drone photography image; Use a bidirectional optimal search method to iterate single matching results, including: For the position descriptors of each photovoltaic string in the sorted drone photography image ( and the position descriptor of each photovoltaic string in the digital orthophoto DOM image ( , build a double-ended priority queue; A bidirectional greedy matching strategy is used for matching. From the sorted position descriptor sequences of the source end of the drone photography image and the target end of the digital orthophoto DOM image, matching is started from the minimum angle or distance value end and the maximum angle or distance value end of the double-ended priority queue at the same time. The current matching pointers in two directions, including the head pointer and the tail pointer, are maintained respectively. For the head pointer descriptor on the source end, the most similar descriptor is found within the current head and tail pointer range of the target end, the angle difference and distance similarity are calculated respectively, and the one with the highest comprehensive score is selected as the candidate match. At the same time, the same operation is performed in reverse from the tail pointer on the target end. When at least one end of the double-ended priority queue is empty or the angle ranges of the remaining unmatched descriptors do not overlap, the matching is terminated. During the matching process, an error matching mechanism is introduced, and the polar coordinate transformation model of the matching pairs is fitted using the RANSAC geometric verification method, and abnormal matching pairs that deviate from this model are eliminated. Using topological consistency check, a Delaunay triangulation is constructed for the retained matching pairs, and matching pairs that destroy the spatial adjacency relationship are removed to obtain the final matching results.

7. The photovoltaic panel latitude and longitude positioning method based on drone photography according to claim 6, characterized in that: In S3, polar coordinate-scale adaptive loss is adopted As the position descriptor of each photovoltaic string in the drone photography image ( Compared with the position descriptor of each photovoltaic string in the digital orthophoto DOM image ( Losses between The polar-scale adaptive loss Including orientation consistency loss , scale invariance loss and topology constraint loss , the formula is: (13)； In formula (13): 、 、 All are adaptive adjustment coefficients; Among them, the orientation consistency loss Designed to resist rotation interference, specifically: (14)； The scale invariance loss Designed to achieve cross-resolution adaptation, specifically: (15)； The topology constraint loss Designed to maintain spatial relationships, specifically: (16)； In formula (16): is the number of PV strings, Respectively represent the indivual PV strings; The adaptive adjustment coefficient 、 、 The initial value is set to: , which is dynamically optimized through the following formula: 。 8. A device for implementing a method for locating the longitude and latitude of photovoltaic panels based on drone photography, characterized in that: The device comprises: An acquisition module is used to acquire digital orthophoto DOM images and digital surface model DSM images based on drone photography images; A construction module is used to construct a photovoltaic string detection model to detect photovoltaic strings in the drone photography image and the digital orthophoto DOM image, and describe the positions of the detected photovoltaic strings to construct photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image; A positioning module is used to match the photovoltaic strings in the drone photography image and the digital orthophoto DOM image based on the photovoltaic string position descriptors in the drone photography image and the digital orthophoto DOM image, and convert the center point pixel coordinates of the photovoltaic strings in the matched digital orthophoto DOM image into ground longitude and latitude coordinates to achieve longitude and latitude positioning of the photovoltaic strings in the drone image; The deployment process of the photovoltaic string detection model is as follows: export the photovoltaic string detection model as an ONNX model and use the ONNX optimization tool to simplify the exported ONNX model; then use the RKNN Toolkit to load the optimized ONNX model, quantize the ONNX model, and convert the floating-point ONNX model into a fixed-point RKNN model; finally, deploy the generated RKNN model on the device platform.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and A memory storing instructions, which, when executed by the at least one processor, causes the at least one processor to execute the photovoltaic panel latitude and longitude positioning method based on drone photography as described in any one of claims 1 to 7.

10. A machine-readable storage medium, characterized in that The machine-readable storage medium stores executable instructions, which, when executed, enable the machine to execute the photovoltaic panel latitude and longitude positioning method based on drone photography as described in any one of claims 1 to 7.

Images