Unmanned aerial vehicle photovoltaic array inclination intelligent monitoring method
The use of drone image recognition technology for photovoltaic panel tilt monitoring solves the problems of sensor equipment complexity and environmental factors, and achieves high-precision, low-cost photovoltaic panel tilt detection.
Patent Information
- Application Number
- CN202511228330.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-29
- Publication Date
- 2025-10-31
AI Technical Summary
Existing photovoltaic panel tilt sensors and automated equipment require large initial investments and regular calibration and maintenance. Environmental factors may cause equipment failure, increasing system complexity.
By using drones to capture images of photovoltaic arrays and employing image recognition and classification models, the tilt angle of the photovoltaic panels can be monitored in real time, eliminating the need to install sensors on the photovoltaic panels.
It achieves high-precision photovoltaic panel tilt angle monitoring, reduces reliance on sensors, lowers system complexity and maintenance costs, and improves environmental adaptability of monitoring.
Smart Images
Figure CN120872003A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of solar photovoltaic panel technology, and in particular to a method for intelligent monitoring of the tilt of a drone-based photovoltaic array. Background Technology
[0002] Solar photovoltaic (PV) panels convert solar energy into electricity. Their core function is to convert solar radiation into direct current (DC) through the photoelectric effect, providing clean energy for homes, industries, and public facilities. The tilt angle of the PV panels must match the angle of sunlight incidence to reduce light reflection and shading, and increase light absorption per unit area. A proper tilt angle can reduce the risk of snow sliding off and prevent a decrease in power generation efficiency due to water or dirt accumulation.
[0003] In existing technologies, tilt sensors, levels, or automated equipment (such as single-axis / dual-axis trackers) are typically used to measure the angle between the photovoltaic panel and the horizontal plane to ensure compliance with design requirements. However, tilt sensors and automated equipment require significant initial investment and regular calibration and maintenance, increasing system complexity. Strong winds, hail, or snow pressure can cause deformation of the photovoltaic panel supports, potentially malfunctioning monitoring equipment and necessitating additional protective measures.
[0004] In view of this, the present invention is hereby proposed. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides an intelligent monitoring method for the tilt of a photovoltaic array from a drone. By processing images captured by the drone, the tilt angle of the photovoltaic array can be obtained, eliminating the need to install tilt sensors on the photovoltaic array.
[0006] In a first aspect, embodiments of the present invention provide a method for intelligent monitoring of the tilt of a photovoltaic array from a drone, comprising:
[0007] The drone is controlled to fly above the photovoltaic array, and images of the photovoltaic array are captured in real time during the flight. The drone dynamically adjusts its pitch angle and altitude during the flight, so that the adjustment range of the pitch angle covers the maximum tilt angle of the photovoltaic array.
[0008] The drone acquires multiple images captured in real time, along with the drone's pitch angle and spatial position when each image is captured.
[0009] A target photovoltaic panel is selected in the photovoltaic array, and the target photovoltaic panel is identified and segmented from multiple images captured in real time to obtain multiple image blocks of the target photovoltaic panel;
[0010] Multiple target image blocks are selected from multiple image blocks of the target photovoltaic panel according to the classification model; wherein, the multiple target image blocks are captured when the pitch angle of the UAV is the same as the current tilt angle of the target photovoltaic panel and the vertical distance between the UAV and the target photovoltaic panel is the same;
[0011] When capturing multiple target image blocks, the drone acquires multiple spatial positions.
[0012] A linear fit is performed on the multiple spatial locations, and the tilt angle of the fitted line is taken as the current tilt angle of the target photovoltaic panel.
[0013] Optionally, before selecting multiple target image patches from multiple image patches of the target photovoltaic panel based on the classification model, the method further includes:
[0014] Training samples are collected. The positive samples in the training samples include: images of photovoltaic panel samples taken by a UAV at the tilt angle of the photovoltaic panel sample and at a fixed vertical distance from the photovoltaic panel sample, given that the tilt angle of the photovoltaic panel sample is known; the positive samples are grouped according to different vertical distances to obtain multiple positive sample groups; the negative sample groups in the training samples include: images of photovoltaic panel samples taken by a UAV at a different tilt angle of the photovoltaic panel sample or at a different vertical distance from the photovoltaic panel sample, given that the tilt angle of the photovoltaic panel sample is known.
[0015] The classification model is trained based on the training samples.
[0016] Optionally, the classification model includes an input layer, a feature extraction layer, an edge extraction layer, a fusion layer, and a classification layer.
[0017] Optionally, training the classification model based on the training samples includes:
[0018] Construct the loss function l:
[0019]
[0020] Where i is the sample ID, n is the number of samples, m is the number of classification layers, and j is the classification layer ID; y ij It is the true value of the j-th classification layer for the i-th sample, which indicates whether it is a negative or positive sample classification. x is the predicted value of the j-th classification layer for the i-th sample, S is the positive sample group, and x is the predicted value of the j-th classification layer for the i-th sample. i Let x be the i-th positive sample, S′ be the negative sample group, and x be the i-th positive sample. i ′ is the i-th negative sample, and θ is the tilt angle of the photovoltaic panel sample. It is the pitch angle of the drone when the i-th positive sample is captured. It is the pitch angle of the drone when the i-th negative sample is captured;
[0021] The loss function is minimized to iterate over the input layer, feature extraction layer, edge extraction layer, fusion layer, and classification layer corresponding to the positive and negative sample groups in the classification model.
[0022] Optionally, multiple target image patches are selected from multiple image patches of the target photovoltaic panel based on a classification model, including:
[0023] Multiple sets of target image blocks are selected from multiple image blocks of the target photovoltaic panel based on the classification model, and each set of target image blocks includes multiple target image blocks;
[0024] Each group of target image blocks includes multiple target image blocks that were captured when the drone's pitch angle was the same as the target photovoltaic panel's current tilt angle, and the vertical distance between the drone and the target photovoltaic panel was the same; and the vertical distances of the target image blocks in different groups were different when they were captured.
[0025] Optionally, linear fitting is performed on the plurality of spatial locations, and the tilt angle of the fitted line is used as the current tilt angle of the target photovoltaic panel, including:
[0026] Linear fitting is performed on multiple spatial positions when capturing the same set of target image blocks to obtain the inclination angle of the fitted line;
[0027] The current tilt angle of the target photovoltaic panel is obtained by averaging the tilt angles of multiple straight lines.
[0028] Optionally, the input layer is used to convert the input image patch into an embedding vector;
[0029] The feature extraction layer is used to perform multi-level feature extraction on the embedding vector to obtain high-dimensional features;
[0030] The edge extraction layer is used to extract low-dimensional features from the embedding vector to obtain low-dimensional features;
[0031] The fusion layer is used to concatenate the high-dimensional features and low-dimensional features, and then perform a convolution operation, a nonlinear transformation, a bilinear interpolation, and a second convolution operation on the concatenated features in sequence to obtain the fused features.
[0032] The system has multiple classification layers, which are used to classify images taken by a UAV at the tilt angle of the target photovoltaic panel according to the vertical distance based on the fusion features; and to classify images taken by a UAV at a different tilt angle than the target photovoltaic panel sample, or at a different vertical distance from the target photovoltaic panel.
[0033] The embodiments of the present invention have the following technical effects:
[0034] In this embodiment, the tilt angle of the photovoltaic panel is detected by a drone, which does not rely on sensors on the photovoltaic panel and is less affected by the environment. This embodiment uses image recognition and classification to convert the tilt angle of the photovoltaic panel into a fitted linear tilt angle based on the drone's spatial position, thus cleverly obtaining the tilt angle of the photovoltaic panel. The drone's high spatial positioning accuracy, combined with a high-precision classification model and multi-point linear fitting technology (effectively filtering out abnormal spatial positions), ensures high accuracy in monitoring the tilt angle of the photovoltaic panel. Attached Figure Description
[0035] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0036] Figure 1 This is a flowchart of an intelligent monitoring method for the tilt of a UAV photovoltaic array provided in an embodiment of the present invention;
[0037] Figure 2 This is a top view of the UAV provided in an embodiment of the present invention;
[0038] Figure 3 This is a schematic diagram of a drone flying over any photovoltaic panel, provided in an embodiment of the present invention.
[0039] Figure 4 This is a schematic diagram of the classification model provided in an embodiment of the present invention. Detailed Implementation
[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0041] The method provided by this invention is applicable to situations where the tilt angle of a solar photovoltaic panel needs to be detected. Solar photovoltaic panels can be installed in various locations such as rooftops and slopes.
[0042] Figure 1 This is a flowchart of an intelligent monitoring method for the tilt of a UAV photovoltaic array provided by an embodiment of the present invention, including the following operations:
[0043] S110: Control the drone to fly above the photovoltaic array and take real-time images of the photovoltaic array during the flight.
[0044] The photovoltaic array consists of multiple solar photovoltaic panels arranged according to certain rules. The drone is equipped with cameras, positioning devices, gyroscopes, and inertial measurement units, which can detect the drone's spatial position and flight angles (including pitch, yaw, and roll angles) with high precision. Figure 3 This is a schematic diagram of a drone flying over any photovoltaic panel, provided in an embodiment of the present invention.
[0045] The solar panel is fixed to the ground with a bracket, and the angle between the panel and the ground is the tilt angle, which is the angle that this invention needs to monitor. The drone flies from the left side of the solar panel along a wavy line to the right side of the drone. The drone needs to adjust its pitch angle (yaw and roll angles are both 0) and altitude to fly along the wavy line. The camera on the drone captures images of the solar panel in real time. Figure 2 This is a top view of the drone provided in this embodiment of the invention. The gray circle in the middle represents the camera, whose optical axis is parallel to the vertical direction of the drone (i.e., perpendicular to the direction of this document). When the drone's pitch angle is the same as the photovoltaic panel's tilt angle (i.e., the mirror surface of the camera on the drone is parallel to the photovoltaic panel), the drone is in a state of... Figure 3 At positions A and C, the drone was taking pictures directly above the photovoltaic panel, and the panel appears rectangular in the images. At position B, the drone was taking pictures from a slightly above and to the side of the panel; due to the tilted shooting angle, the panel does not appear rectangular in the image. (See [link to image]). Figure 4 Besides the shape, the patterns and markings on the photovoltaic panels also undergo deformation. As the drone flies along a wavy line, its altitude varies, and the vertical distance from the camera's optical center to the solar panel also varies; the greater the vertical distance, the larger the proportion of the photovoltaic panel in the image. When the drone is at... Figure 3 At position A in the diagram, the drone is 1 unit vertically from the photovoltaic panel; when the drone is in... Figure 3 At position B in the diagram, the drone is 2 units vertically from the photovoltaic panel; when the drone is in... Figure 3 At position C, the drone is 3 units vertically from the photovoltaic panel. Vertical distance 1 equals vertical distance 3. The drone has the same pitch angle and vertical distance at positions A and C. As the drone flies along a wavy line, it will have multiple positions with the same pitch angle and vertical distance. The line connecting positions A and C is a straight line, and the tilt angle of this line (i.e., the angle between the line and the ground) is the tilt angle of the photovoltaic panel.
[0046] Based on the above analysis, among the multiple images taken by the drone, it is necessary to utilize the characteristics of the photovoltaic panels in the images to identify which images were taken by the drone at the same pitch angle and vertical distance; then, based on the positioning of the image taken by the drone, multiple spatial positions are obtained, thereby obtaining the tilt angle of the straight line connecting the multiple spatial positions.
[0047] It should be noted that the drone dynamically adjusts its pitch angle during flight, ensuring that the adjustment range covers the maximum tilt angle of the photovoltaic array. Here, "maximum tilt angle" refers to the maximum tilt angle that the photovoltaic panel can be adjusted to, for example, 30 degrees. The drone's pitch angle should be adjusted within the range of 0 to 30 degrees to find images captured when the drone flies at a pitch angle that matches the various tilt angles of the photovoltaic panel.
[0048] S120: Acquire multiple images captured in real time from the drone, along with the drone's pitch angle and spatial position when each image is captured.
[0049] While acquiring images from the camera, the drone's processor also obtains the drone's current pitch angle and (3D) spatial position from the inertial measurement unit, positioning device, and gyroscope. Currently, a joint positioning strategy using the inertial measurement unit, positioning device, and gyroscope can achieve centimeter-level spatial positioning, meeting the measurement accuracy requirements for photovoltaic panel tilt angles.
[0050] S130. Select a target photovoltaic panel in the photovoltaic array, and identify and segment the target photovoltaic panel in multiple images captured in real time to obtain multiple image blocks of the target photovoltaic panel.
[0051] A photovoltaic array consists of multiple photovoltaic panels, each of which may have a different tilt angle. In this embodiment, the tilt angle is monitored for one photovoltaic panel (referred to as the target photovoltaic panel). The same method is used for the other photovoltaic panels, which will not be described in detail here.
[0052] Optionally, the photovoltaic panel number located in the upper right corner or side of the photovoltaic panel is a unique identifier for the panel. For example, if the target photovoltaic panel is selected, number ID05, then the photovoltaic panel corresponding to ID05 will be identified from multiple images. Specifically, multiple images are input into the target recognition model to obtain the rectangular frame of the photovoltaic panel corresponding to ID05. This rectangular frame is then cropped to obtain multiple image blocks of the target photovoltaic panel.
[0053] Assuming a total of 1000 images were captured, after removing images with poor clarity, 900 rectangular frames corresponding to the photovoltaic panel with ID05 were identified, resulting in 900 image blocks.
[0054] The purpose of image block recognition and segmentation is to reduce the impact of background images and other photovoltaic panel images on subsequent processing.
[0055] S140. Select multiple target image blocks from multiple image blocks of the target photovoltaic panel according to the classification model.
[0056] Multiple target image patches were captured when the drone's pitch angle was the same as the target photovoltaic panel's current tilt angle, and the vertical distance between the drone and the target photovoltaic panel was the same. In other words, the image patches segmented from the images captured by the drone at positions A and C constitute the target image patches.
[0057] Based on the above analysis, the photovoltaic panels in the target image blocks share the same morphological characteristics and are distinct from non-target image blocks. This embodiment uses a classification model to identify these target image blocks. The classification model is characterized by high accuracy and speed, eliminating the need for manual classification. The structure and training process of the classification model will be described in detail in the following embodiments.
[0058] S150. When capturing multiple target image blocks, the multiple spatial positions of the UAV are obtained.
[0059] For example, obtaining the positions A and C of the drone within the target image patch. The spatial position (i.e., the drone's center of mass position) can be represented using a geodetic coordinate system or a world coordinate system.
[0060] S160. Perform linear fitting on multiple spatial locations, and use the tilt angle of the fitted line as the current tilt angle of the target photovoltaic panel.
[0061] Fitting multiple spatial locations to a straight line, i.e. Figure 3 The green line represents the current tilt angle of the target photovoltaic panel. The angle between this line and the ground (i.e., the tilt angle) is the current tilt angle of the target photovoltaic panel.
[0062] In this embodiment, the tilt angle of the photovoltaic panel is detected by a drone, which does not rely on sensors on the photovoltaic panel and is less affected by the environment. This embodiment uses image recognition and classification to convert the tilt angle of the photovoltaic panel into a fitted linear tilt angle based on the drone's spatial position, thus cleverly obtaining the tilt angle of the photovoltaic panel. The drone's high spatial positioning accuracy, combined with a high-precision classification model and multi-point linear fitting technology (effectively filtering out abnormal spatial positions), ensures high accuracy in monitoring the tilt angle of the photovoltaic panel.
[0063] The structure and training process of the classification model are described in detail below.
[0064] See Figure 4 The classification model consists of an input layer, a feature extraction layer, an edge extraction layer, a fusion layer, and a classification layer. The input layer is used to convert multiple input image patches into embedding vectors.
[0065] The feature extraction layer performs multi-level feature extraction on the embedding vector to obtain high-dimensional features. For example, the feature extraction layer includes multiple convolutional layers and non-linear activation layers. The edge extraction layer performs low-dimensional feature extraction on the embedding vector to obtain low-dimensional features. For example, the edge extraction layer includes one convolutional layer. The fusion layer concatenates the high-dimensional and low-dimensional features, and then performs a convolution operation, a non-linear transformation, bilinear interpolation, and a second convolution operation on the concatenated features to obtain fused features.
[0066] Multiple classification layers are used to classify images taken by a drone flying at the target photovoltaic panel's tilt angle according to vertical distance, based on fused features; and to classify images taken by a drone flying at a different tilt angle than the target photovoltaic panel, or at a different vertical distance than the target photovoltaic panel. Assume there are three classification layers (each integrating a softmax function): Classification Layer 1, Classification Layer 2, and Classification Layer 3. Classification Layer 1 classifies the input images according to the following rules: images taken by a drone flying at the target photovoltaic panel's tilt angle and at a vertical distance of 5 meters; images taken by a drone flying at a different tilt angle than the target photovoltaic panel or at a vertical distance not exceeding 5 meters (i.e., other images). Classification Layer 2 classifies the input images according to the following rules: images taken by a drone flying at the target photovoltaic panel's tilt angle and at a vertical distance of 6 meters; images taken by a drone flying at a different tilt angle than the target photovoltaic panel or at a vertical distance not exceeding 6 meters (i.e., other images). Classification layer 3 is used to classify the input images according to the following rules: images taken by the drone when flying with the target photovoltaic panel tilted at an angle of 7 meters and a vertical distance of 7 meters; images taken by the drone when flying without the target photovoltaic panel tilted at an angle of 7 meters or a vertical distance of 7 meters (i.e., other images).
[0067] The training process for this classification model is as follows:
[0068] 1. Collect training samples to train the classification model.
[0069] The training samples include positive and negative samples. Positive samples include images of the photovoltaic panel sample taken by a UAV at the known tilt angle of the photovoltaic panel sample and at a fixed vertical distance from the sample. These positive samples are grouped according to different vertical distances (e.g., the aforementioned 5 meters, 6 meters, and 7 meters) to obtain multiple positive sample groups. The negative sample groups include images of the photovoltaic panel sample taken by a UAV at a different tilt angle or at a different vertical distance (e.g., not the aforementioned 5 meters, 6 meters, or 7 meters) when the tilt angle of the photovoltaic panel sample is known.
[0070] 2. Construct the loss function l:
[0071]
[0072] Where i is the sample ID, n is the number of samples, m is the number of classification layers, and j is the classification layer ID; y ij It is the true value of the j-th classification layer for the i-th sample, which indicates whether it is a negative or positive sample classification. x is the predicted value of the j-th classification layer for the i-th sample, S is the positive sample group, and x is the predicted value of the j-th classification layer for the i-th sample. i Let x be the i-th positive sample, S′ be the negative sample group, and x be the i-th positive sample. i ′ is the i-th negative sample. Note that the negative sample group S′ here is an image of the photovoltaic panel sample taken by the drone without tilting it at the photovoltaic panel sample angle. It should not include images of the photovoltaic panel sample taken by the drone when it is tilting at the photovoltaic panel sample angle and not at a fixed vertical distance from the photovoltaic panel sample.
[0073] θ is the tilt angle of the photovoltaic panel sample. It is the pitch angle of the drone when the i-th positive sample is captured. It is the pitch angle of the drone when shooting the i-th negative sample.
[0074] The loss function is minimized to iterate through the input layer, feature extraction layer, edge extraction layer, fusion layer, and the classification layer corresponding to the positive and negative sample groups in the classification model. This embodiment does not limit the specific iterative algorithm; gradient descent can be used.
[0075] For each classification layer j, the output will be... pass Calculation output With truth value y ij The distance is then calculated. Then, the distances of each classification layer j are summed. This is done by... A loss function is constructed by incorporating the true pitch angle of the drone images. When capturing positive samples, the drone's pitch angle should be equal to or approximately equal to the tilt angle of the photovoltaic panel sample; when capturing negative samples, the drone's pitch angle should differ from the tilt angle of the photovoltaic panel sample. This loss function is achieved by minimizing... By using the drone's pitch angle as a supervisory condition, the classification model is trained to further improve the model's accuracy.
[0076] Optionally, selecting multiple target image blocks from multiple image blocks of the target photovoltaic panel according to a classification model includes: selecting multiple groups of target image blocks from multiple image blocks of the target photovoltaic panel according to a classification model, each group of target image blocks including multiple target image blocks; the multiple target image blocks included in each group of target image blocks are captured when the pitch angle of the UAV is the same as the current tilt angle of the target photovoltaic panel, and the vertical distance between the UAV and the target photovoltaic panel is the same; and the vertical distance (e.g., the aforementioned 5 meters, 6 meters and 7 meters) is different for different groups of target image blocks at the time of capture.
[0077] Optionally, linear fitting is performed on the multiple spatial locations, and the tilt angle of the fitted line is used as the current tilt angle of the target photovoltaic panel. This includes: performing linear fitting on multiple spatial locations when capturing the same set of target image blocks to obtain the tilt angle of the fitted line; performing linear fitting on the UAV spatial locations corresponding to vertical distances of, for example, 5 meters, 6 meters, and 7 meters; and averaging the tilt angles of the multiple lines to obtain the current tilt angle of the target photovoltaic panel. By fitting multiple lines at different vertical distances and averaging the tilt angles of the fitted lines, the error in tilt angle monitoring can be reduced.
[0078] It should be noted that the terminology used in this invention is for describing specific embodiments only and is not intended to limit the scope of this application. As shown in this specification, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may include the plural. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising an..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element.
[0079] It should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Unless otherwise expressly specified and limited, the terms "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood according to the specific circumstances.
[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.
Claims
1. A method for intelligent monitoring of the tilt of a photovoltaic array from an unmanned aerial vehicle (UAV), characterized in that, include: The drone is controlled to fly above the photovoltaic array, and images of the photovoltaic array are captured in real time during the flight. The drone dynamically adjusts its pitch angle and altitude during the flight, so that the adjustment range of the pitch angle covers the maximum tilt angle of the photovoltaic array. The drone acquires multiple images captured in real time, along with the drone's pitch angle and spatial position when each image is captured. A target photovoltaic panel is selected in the photovoltaic array, and the target photovoltaic panel is identified and segmented from multiple images captured in real time to obtain multiple image blocks of the target photovoltaic panel; Multiple target image blocks are selected from multiple image blocks of the target photovoltaic panel according to the classification model; wherein, the multiple target image blocks are captured when the pitch angle of the UAV is the same as the current tilt angle of the target photovoltaic panel and the vertical distance between the UAV and the target photovoltaic panel is the same; When capturing multiple target image blocks, the drone acquires multiple spatial positions. Linear fitting is performed on the multiple spatial locations, and the tilt angle of the fitted line is taken as the current tilt angle of the target photovoltaic panel.
2. The intelligent monitoring method for tilting photovoltaic arrays of unmanned aerial vehicles according to claim 1, characterized in that, Before selecting multiple target image patches from multiple image patches of the target photovoltaic panel based on the classification model, the following steps are also included: Training samples are collected. The positive samples in the training samples include: images of photovoltaic panel samples taken by a UAV at the tilt angle of the photovoltaic panel sample and at a fixed vertical distance from the photovoltaic panel sample, given that the tilt angle of the photovoltaic panel sample is known; the positive samples are grouped according to different vertical distances to obtain multiple positive sample groups; the negative sample groups in the training samples include: images of photovoltaic panel samples taken by a UAV at a different tilt angle of the photovoltaic panel sample or at a different vertical distance from the photovoltaic panel sample, given that the tilt angle of the photovoltaic panel sample is known. The classification model is trained based on the training samples.
3. The intelligent monitoring method for tilting photovoltaic arrays of unmanned aerial vehicles according to claim 2, characterized in that, The classification model includes an input layer, a feature extraction layer, an edge extraction layer, a fusion layer, and a classification layer.
4. The intelligent monitoring method for tilting photovoltaic arrays of unmanned aerial vehicles according to claim 3, characterized in that, Training the classification model based on the training samples includes: Construct the loss function l: Where i is the sample ID, n is the number of samples, m is the number of classification layers, and j is the classification layer ID; y ij It is the true value of the j-th classification layer for the i-th sample, which indicates whether it is a negative or positive sample classification. x is the predicted value of the j-th classification layer for the i-th sample, S is the positive sample group, and x is the predicted value of the j-th classification layer for the i-th sample. i Let x' be the i-th positive sample, S′ be the negative sample group, and x′ be the negative sample group. i It is the i-th negative sample, and θ is the tilt angle of the photovoltaic panel sample. It is the pitch angle of the drone when the i-th positive sample is captured. It is the pitch angle of the drone when the i-th negative sample is captured; The loss function is minimized to iterate over the input layer, feature extraction layer, edge extraction layer, fusion layer, and classification layer corresponding to the positive and negative sample groups in the classification model.
5. The intelligent monitoring method for tilting photovoltaic arrays of unmanned aerial vehicles according to claim 1, characterized in that, Multiple target image patches are selected from multiple image patches of the target photovoltaic panel based on a classification model, including: Multiple sets of target image blocks are selected from multiple image blocks of the target photovoltaic panel based on the classification model, and each set of target image blocks includes multiple target image blocks; Each group of target image blocks includes multiple target image blocks that were captured when the drone's pitch angle was the same as the target photovoltaic panel's current tilt angle, and the vertical distance between the drone and the target photovoltaic panel was the same; and the vertical distances of the target image blocks in different groups were different when they were captured.
6. The intelligent monitoring method for tilting photovoltaic arrays of unmanned aerial vehicles according to claim 5, characterized in that, Performing linear fitting on the multiple spatial locations, and using the tilt angle of the fitted line as the current tilt angle of the target photovoltaic panel, includes: Linear fitting is performed on multiple spatial positions when capturing the same set of target image blocks to obtain the inclination angle of the fitted line; The current tilt angle of the target photovoltaic panel is obtained by averaging the tilt angles of multiple straight lines.
7. The method according to claim 3, characterized in that, The input layer is used to convert the input image patches into embedding vectors; The feature extraction layer is used to perform multi-level feature extraction on the embedding vector to obtain high-dimensional features; The edge extraction layer is used to extract low-dimensional features from the embedding vector to obtain low-dimensional features; The fusion layer is used to concatenate the high-dimensional features and low-dimensional features, and then perform a convolution operation, a nonlinear transformation, a bilinear interpolation, and a second convolution operation on the concatenated features in sequence to obtain the fused features. The system has multiple classification layers, which are used to classify images taken by a UAV at the tilt angle of the target photovoltaic panel according to the vertical distance based on the fusion features; and to classify images taken by a UAV at a different tilt angle than the target photovoltaic panel sample, or at a different vertical distance from the target photovoltaic panel.