A flight control method, device and equipment of a photovoltaic cleaning unmanned aerial vehicle and a medium

Abstract

The application discloses a flight control method, device and equipment of a photovoltaic cleaning unmanned aerial vehicle and a medium. The method comprises the following steps: generating a flight basic route of the photovoltaic cleaning unmanned aerial vehicle; determining a terrain scene of a photovoltaic power station according to a maximum slope and a maximum component interval of the photovoltaic power station; if the terrain scene is flat ground, adjusting the height of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of a spray head according to the detection result of a ground radar; if the terrain scene is conventional mountainous land, adjusting the height of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of the spray head according to the detection result of the ground radar and a front radar; if the terrain scene is complex mountainous land, determining the lifting speed or the descending speed of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of the spray head according to the recognition result of current image data and the recognition result of last image data by using a photovoltaic gap image recognition model. The application improves the safety and cleaning efficiency of the photovoltaic cleaning unmanned aerial vehicle and improves the accuracy of the spray head control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power stations, and more particularly to a flight control method, device and equipment of a photovoltaic cleaning unmanned aerial vehicle and a medium. BACKGROUND

[0002] With the continuous development of renewable energy, the construction and operation of photovoltaic power stations become more and more important. The accumulation of pollutants on the surface of photovoltaic components can reduce their light transmittance and affect the power generation efficiency, so regular cleaning is needed. Currently, photovoltaic cleaning mainly relies on manual or automated equipment. Unmanned aerial vehicles (UAVs) are gradually being applied in the field of photovoltaic cleaning due to their flexibility and efficiency. In the prior art, photovoltaic component cleaning by UAVs is mainly achieved through preset flight routes and ground-following strategies.

[0003] Under the same cleaning conditions, the distance of the photovoltaic cleaning UAV from the components during flight is negatively correlated with the cleaning effect. To ensure the cleaning effect, the UAV needs to fly at a low height of 0.5-1m above the photovoltaic components. However, there are about 1m channel gaps between some components of the photovoltaic power station. When the cleaning UAV executes the task according to the flight route and ground-following flight is enabled, the ground radar detects a sudden increase in the detection distance of the channel gap. If the cleaning UAV detects a decrease in the height of the gap during ground-following flight, the cleaning UAV will follow the decrease in the height of the gap, causing a safety hazard of the cleaning UAV colliding with the components behind the gap. Frequent adjustment of the height of the UAV at the gap can cause the cleaning water tank to shake, leading to unstable flight. Therefore, in the existing ground-following technology, the UAV needs to frequently adjust the height to adapt to the height of the components and the ground during flight, which not only poses a safety hazard, but also leads to increased energy consumption and reduced cleaning efficiency. In addition, the start-stop control of the spray head in the existing technology is often based on a simple height threshold, which cannot accurately identify the gap between the components, leading to waste of water resources and poor cleaning effect.

[0004] Therefore, how to improve the safety and cleaning efficiency of the photovoltaic cleaning UAV and improve the accuracy of the spray head control is a technical problem that needs to be solved by those skilled in the art. SUMMARY

[0005] The purpose of the present application is to provide a flight control method, device and equipment of a photovoltaic cleaning UAV and a computer-readable storage medium, which improves the safety and cleaning efficiency of the photovoltaic cleaning UAV and improves the accuracy of the spray head control.

[0006] To achieve the above purpose, the present application provides a flight control method of a photovoltaic cleaning UAV, comprising:

[0007] generating a flight base route of the photovoltaic cleaning UAV based on image data of the photovoltaic power station;

[0008] obtaining a maximum slope and a maximum component interval of the photovoltaic power station, and determining a terrain scene of the photovoltaic power station according to the maximum slope and the maximum component interval; wherein the terrain scene comprises flat land, regular mountainous land and complex mountainous land;

[0009] if the terrain scene of the photovoltaic power station is flat land, adjusting the height of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of the spray head according to the detection result of the ground radar; wherein the detection result of the ground radar comprises a component or a gap, and the state of the spray head comprises opening and closing;

[0010] if the terrain scene of the photovoltaic power station is regular mountainous land, adjusting the height of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of the spray head according to the detection results of the ground radar and the front radar; wherein the detection result of the front radar comprises a component or a gap;

[0011] if the terrain scene of the photovoltaic power station is complex mountainous land, inputting image data collected by the photovoltaic cleaning unmanned aerial vehicle into a photovoltaic gap image recognition model, and determining the lifting rate or the descending rate of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of the spray head according to the recognition result of the current time image data and the recognition result of the last time image data by using the photovoltaic gap image recognition model; wherein the recognition result of the photovoltaic gap image recognition model comprises a component or a gap.

[0012] wherein the flight basic route of the photovoltaic cleaning unmanned aerial vehicle is generated based on image data of the photovoltaic power station, comprising:

[0013] collecting image data of the photovoltaic power station, and generating a digital surface model and a digital orthographic image map with different ground resolutions according to the image data;

[0014] generating the latitude and longitude coordinates of the flight route of the photovoltaic cleaning unmanned aerial vehicle based on the digital surface model and the digital orthographic image map with different ground resolutions;

[0015] determining the absolute elevation of each waypoint of the photovoltaic cleaning unmanned aerial vehicle, and generating the flight basic route of the photovoltaic cleaning unmanned aerial vehicle based on the latitude and longitude coordinates of the flight route and the absolute elevation of each waypoint.

[0016] wherein the terrain scene of the photovoltaic power station is determined according to the maximum slope and the maximum component interval, comprising:

[0017] determining whether the absolute value of the maximum slope is less than or equal to a first slope threshold;

[0018] if the absolute value of the maximum slope is less than or equal to the first slope threshold, it is determined that the terrain scene of the photovoltaic power station is flat land;

[0019] if the absolute value of the maximum slope is greater than the first slope threshold, determining whether the absolute value of the maximum slope is greater than a second slope threshold; wherein the second slope threshold is greater than the first slope threshold;

[0020] if the absolute value of the maximum slope is greater than the second slope threshold, determining that the terrain scene of the photovoltaic power station is complex mountainous area;

[0021] if the absolute value of the maximum slope is less than or equal to the second slope threshold, determining whether the maximum component interval is less than or equal to a distance threshold;

[0022] if the maximum component interval is less than or equal to the distance threshold, determining that the terrain scene of the photovoltaic power station is regular mountainous area;

[0023] if the maximum component interval is greater than the distance threshold, determining that the terrain scene of the photovoltaic power station is complex mountainous area.

[0024] wherein the adjusting the height of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of the spray head according to the detection result of the ground penetrating radar comprises:

[0025] determining the height h of the component center from the ground, and determining the height fluctuation threshold when the absolute value of the slope is less than or equal to the first slope threshold according to the update time interval of the ground penetrating radar dotting and the horizontal direction speed of the photovoltaic cleaning unmanned aerial vehicle ;

[0026] calculating the height difference between the height detected by the ground penetrating radar at the current time i and the height detected at the last time i-1 ; ;

[0027] if , the detection result of the ground penetrating radar is a component or an interval, maintaining the height and the state of the spray head of the last time;

[0028] if , the detection result of the ground penetrating radar is an interval, adjusting the height of the photovoltaic cleaning unmanned aerial vehicle to , and controlling the state of the spray head to be closed;

[0029] if , the detection result of the ground penetrating radar is a component, adjusting the height of the photovoltaic cleaning unmanned aerial vehicle to , and controlling the state of the spray head to be opened.

[0030] wherein the adjusting the height of the photovoltaic cleaning unmanned aerial vehicle and controlling the state of the spray head according to the detection result of the ground penetrating radar and the front radar comprises:

[0031] ​In the flight process of the photovoltaic cleaning unmanned aerial vehicle, the slope is detected in real time by using the ground radar and the front radar;

[0032] If the slope at the last moment and the slope at the current moment are both positive values, and the difference between the slope at the current moment and the slope at the last moment is less than or equal to the third slope threshold, the detection result of the front radar and the detection result of the ground radar are both the front gap component or the rear gap component, the height of the photovoltaic cleaning unmanned aerial vehicle is adjusted according to the slope at the current moment, and the state of the spray head is controlled to be opened;

[0033] If the slope at the current moment is counterclockwise, the slope at the current moment is negative, and the slope at the last moment is positive, the detection result of the front radar is a gap, and the detection result of the ground radar is a component, the height of the photovoltaic cleaning unmanned aerial vehicle adjusted according to the slope at the last moment is maintained, and the state of the spray head at the last moment is maintained;

[0034] If the slope at the current moment is clockwise, the slope at the current moment is positive, and the slope at the last moment is negative, the detection result of the front radar is a rear gap component, and the detection result of the ground radar is a front gap component, the height of the photovoltaic cleaning unmanned aerial vehicle is adjusted according to the slope at the current moment, and the state of the spray head at the last moment is maintained;

[0035] If the slope at the last moment and the slope at the current moment are both positive values, and the difference between the slope at the current moment and the slope at the last moment is greater than the third slope threshold, the detection result of the front radar is a rear gap component, and the detection result of the ground radar is a gap, the height of the photovoltaic cleaning unmanned aerial vehicle adjusted according to the slope at the last moment is maintained, and the state of the spray head is controlled to be closed.

[0036] The image data collected by the photovoltaic cleaning unmanned aerial vehicle is input into a photovoltaic gap image recognition model, and the recognition result of the current moment image data and the recognition result of the last moment image data are used to determine the lifting rate or the descending rate of the photovoltaic cleaning unmanned aerial vehicle and control the state of the spray head, including:

[0037] In the flight process of the photovoltaic cleaning unmanned aerial vehicle, the slope is detected in real time by using the ground radar and the front radar;

[0038] The image data collected by the photovoltaic cleaning unmanned aerial vehicle is input into a photovoltaic gap image recognition model, and the recognition result of the current moment image data and the recognition result of the last moment image data are used to determine the lifting rate or the descending rate of the photovoltaic cleaning unmanned aerial vehicle and control the state of the spray head, including:

[0039] If the recognition result of the image data at the current moment and the recognition result of the image data at the last moment are both gaps, the lifting speed or the descending speed of the photovoltaic cleaning unmanned aerial vehicle is adjusted according to the slope at the current moment, and the nozzle state at the last moment is maintained;

[0040] If the recognition result of the image data at the current moment is a gap, and the recognition result of the image data at the last moment is a component, the lifting speed or the descending speed is maintained according to the slope at the last moment, and the nozzle state is controlled to be closed;

[0041] If the recognition result of the image data at the current moment and the recognition result of the image data at the last moment are both components, the lifting speed or the descending speed of the photovoltaic cleaning unmanned aerial vehicle is adjusted according to the slope at the current moment, and the nozzle state at the last moment is maintained;

[0042] If the recognition result of the image data at the current moment is a component, and the recognition result of the image data at the last moment is a gap, the lifting speed or the descending speed of the photovoltaic cleaning unmanned aerial vehicle is adjusted according to the slope at the current moment, and the nozzle state is controlled to be opened.

[0043] Further comprising:

[0044] Obtain a training image set; wherein the training image set includes multiple groups of training images, and each group of training images includes a visible light image and an infrared image;

[0045] Label the training images; wherein the labeling result includes a component or a gap;

[0046] Perform preprocessing operations on the visible light image and the infrared image; wherein the preprocessing operation of the visible light image includes any one or a combination of suppressing the feature response of the light reflection area and exposure correction, and the preprocessing operation of the infrared image includes any one or a combination of suppressing the feature response of the light reflection area, non-uniformity correction, and temperature calibration;

[0047] Register the visible light image and the infrared image in each group of training images, and fuse to obtain a multispectral fusion image based on the registration result; wherein in the fusion process, the visible light image and the infrared image are fused by pixel-level weighting, and the features of the visible light image and the infrared image are fused by a cross-modal attention mechanism;

[0048] Train a YOLOv8n model using the multispectral fusion image and the corresponding label to obtain the photovoltaic gap image recognition model; wherein the YOLOv8n model includes a photovoltaic global attention module, the photovoltaic global attention module is used to process image features based on a channel attention mechanism and a spatial attention mechanism, and a fixed weight grid mask is used to perform priori on a battery sheet grid to enhance the feature response of a component area.

[0049] To achieve the above object, the application provides a flight control device of a photovoltaic cleaning unmanned aerial vehicle, which comprises

[0050] A generating module is configured to generate a flight base route of the photovoltaic cleaning unmanned aerial vehicle based on image data of a photovoltaic power station.

[0051] A determining module is configured to acquire a maximum slope and a maximum component interval of the photovoltaic power station, and determine a terrain scene of the photovoltaic power station according to the maximum slope and the maximum component interval, wherein the terrain scene comprises flat land, regular mountainous land and complex mountainous land.

[0052] A first control module is configured to adjust a height of the photovoltaic cleaning unmanned aerial vehicle and control a state of a spray head according to a detection result of a ground radar when the terrain scene is flat land, wherein the detection result of the ground radar comprises a component or a gap, and the state of the spray head comprises opening and closing.

[0053] A second control module is configured to adjust the height of the photovoltaic cleaning unmanned aerial vehicle and control the state of the spray head according to detection results of the ground radar and a front radar when the terrain scene is regular mountainous land, wherein the detection result of the front radar comprises the component or the gap.

[0054] A third control module is configured to input image data collected by the photovoltaic cleaning unmanned aerial vehicle into a photovoltaic gap image recognition model, and determine a lifting rate or a descending rate of the photovoltaic cleaning unmanned aerial vehicle and control the state of the spray head according to a recognition result of current image data and a recognition result of previous image data of the photovoltaic gap image recognition model, wherein the recognition result of the photovoltaic gap image recognition model comprises the component or the gap.

[0055] To achieve the above object, the application provides an electronic device, which comprises

[0056] A memory is configured to store a computer program.

[0057] A processor is configured to execute the computer program to realize steps of the flight control method of the photovoltaic cleaning unmanned aerial vehicle.

[0058] To achieve the above object, the application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize steps of the flight control method of the photovoltaic cleaning unmanned aerial vehicle.

[0059] The flight control method of the photovoltaic cleaning unmanned aerial vehicle provided in the application can make the unmanned aerial vehicle better adapt to diversified terrains, ensure the smooth progress of the cleaning operation, significantly reduce the energy consumption of the unmanned aerial vehicle, and improve the efficiency of the cleaning operation, by identifying the terrain scene and adopting differentiated flight strategies according to the characteristics of different terrain scenes. In addition, the application adjusts the flight height by adopting different strategies according to the characteristics of different terrain scenes, reduces unnecessary height adjustment, thereby avoiding the risk of collision with the front or rear component, and improves the safety of the photovoltaic cleaning unmanned aerial vehicle. Further, the application accurately identifies the component gap and accurately controls the state of the spray head according to the characteristics of different terrain scenes, which not only saves water resources but also improves the uniformity of cleaning. The application also discloses a flight control device of a photovoltaic cleaning unmanned aerial vehicle, an electronic device and a computer readable storage medium, which can also achieve the above technical effects.

[0060] It should be understood that the above general description and the following detailed description are only exemplary and cannot limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor. The drawings are used to provide further understanding of the disclosure and constitute a part of the specification, and are used to explain the disclosure together with the following specific embodiments, but do not constitute a limitation on the disclosure. In the drawings:

[0062] Figure 1 A flow chart of a flight control method of a photovoltaic cleaning unmanned aerial vehicle according to an exemplary embodiment is shown;

[0063] Figure 2 A schematic diagram of selecting a distance threshold according to an exemplary embodiment is shown;

[0064] Figure 3 A judgment flow chart of the photovoltaic component gap on the flat ground according to an exemplary embodiment is shown;

[0065] Figure 4 A judgment flow chart of the photovoltaic component gap on the general mountainous terrain according to an exemplary embodiment is shown;

[0066] Figure 5 A schematic diagram of the front radar and the ground radar both detecting the front or rear component according to an exemplary embodiment is shown;

[0067] Figure 6 A schematic diagram of a gap detected by a front radar and a gap detected by a ground radar for a complex mountain photovoltaic module according to an example embodiment;

[0068] Figure 7 A schematic diagram of a gap detected by a front radar and a gap detected by a ground radar for a complex mountain photovoltaic module according to an example embodiment;

[0069] Figure 8 A schematic diagram of a gap detected by a front radar and a gap detected by a ground radar for a complex mountain photovoltaic module according to an example embodiment;

[0070] Figure 9 A flow chart of a judgment process for a gap of a complex mountain photovoltaic module according to an example embodiment;

[0071] Figure 10 A schematic diagram of a photovoltaic gap image recognition model according to an example embodiment;

[0072] Figure 11 A structural diagram of a flight control device of a photovoltaic cleaning unmanned aerial vehicle according to an example embodiment;

[0073] Figure 12 A structural diagram of an electronic device according to an example embodiment. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application. In addition, in the embodiments of the present application, “first”, “second”, etc. are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0075] The embodiments of the present application disclose a flight control method of a photovoltaic cleaning unmanned aerial vehicle, which improves the cleaning efficiency of the photovoltaic cleaning unmanned aerial vehicle and improves the accuracy of nozzle control.

[0076] Referring to Figure 1 , a flow chart of a flight control method of a photovoltaic cleaning unmanned aerial vehicle according to an example embodiment, as shown in Figure 1 , includes:

[0077] S101: generating a flight base route of a photovoltaic cleaning unmanned aerial vehicle based on image data of a photovoltaic power station;

[0078] The flight base route refers to a rough flight path of the photovoltaic cleaning unmanned aerial vehicle according to a certain rule before performing the cleaning task.

[0079] In this step, according to the layout of the photovoltaic power station, the arrangement of the photovoltaic components and the requirements of the cleaning task provided by the image data of the photovoltaic power station and other factors, through professional flight planning software or algorithm, the safety, efficiency and cleaning coverage of flight are comprehensively considered, the component area to be cleaned is selected and a suitable flight base route of the photovoltaic cleaning unmanned aerial vehicle is calculated. This route will try to avoid obvious obstacles to ensure that the unmanned aerial vehicle can fly along a relatively safe path while ensuring that most of the photovoltaic component areas that need to be cleaned can be covered.

[0080] As a feasible implementation manner, the image data of the photovoltaic power station generates the flight base route of the photovoltaic cleaning unmanned aerial vehicle, comprising: collecting image data of the photovoltaic power station, generating digital surface model and digital orthophoto map of different ground resolutions according to the image data; generating the latitude and longitude coordinates of the route of the photovoltaic cleaning unmanned aerial vehicle based on the digital surface model and the digital orthophoto map of different ground resolutions; determining the absolute elevation of each waypoint of the photovoltaic cleaning unmanned aerial vehicle, and generating the flight base route of the photovoltaic cleaning unmanned aerial vehicle based on the latitude and longitude coordinates of the route and the absolute elevation of each waypoint.

[0081] The image data of the photovoltaic power station refers to high-resolution images of the photovoltaic power station area obtained by unmanned aerial vehicle aerial photography and other methods. These images can record the terrain, component layout and other related facilities of the photovoltaic power station in detail. Digital surface model (Digital Surface Model, DSM) is a kind of three-dimensional terrain model, which records the surface height information of all objects on the ground, while digital orthophoto map (Digital Orthophoto Map, DOM) is a kind of high-resolution image after geometric correction, which can truly reflect the planar position of the ground object. The absolute elevation refers to the vertical height of the waypoint relative to the sea level, including the ground elevation, component height and relative flight height of the unmanned aerial vehicle and the component.

[0082] In a specific implementation, first, high-resolution image data of the photovoltaic power station is collected by using a UAV equipped with an RTK (Real-Time Kinematic) positioning system for aerial photography. The UAV requires an RTK positioning accuracy of 3 cm or more, a fixed-focus lens with a pixel of 20 million or more, and a shutter speed of no less than 1 / 1000 s. The flight heading overlap is 70%-80%, the lateral overlap is 60%-70%, and the flight height is 30-50 m. These image data are processed by professional image processing software, such as Pix4D, ContextCapture, etc., for image stitching and aerial triangulation encryption to generate sparse point clouds, DSM and DOM with different ground resolutions (GSD). The ground resolution is the spatial distance represented by one pixel point of the aerial photograph or remote sensing image, such as 0.1 m, 0.5 m, 1 m, etc.

[0083] ;

[0084] wherein, is the ground resolution, is the physical width of the camera sensor, is the vertical height of the aerial photography UAV from the ground, is the focal length of the camera lens, is the pixel width of the image.

[0085] Secondly, the DSM data is down-sampled using the data conversion library GDAL (Geospatial Data Abstraction Library), and the DOM image is down-sampled using OpenCV (Open Source Computer Vision Library) to adapt to different precision requirements. Then, the DSM and DOM data are organized in the form of image pyramid structure, and different resolution data is dynamically switched according to the distance between the viewpoint and the terrain to ensure real-time loading of high-precision data. The area range of the photovoltaic power station is obtained on the loaded DOM image, and the area of the photovoltaic module to be cleaned is extracted by using a geographic information system (Geographic Information System, GIS) tool such as ArcGIS, etc. The flight path of the unmanned aerial vehicle is planned, and the latitude and longitude coordinates of the route are generated. Then, the elevation data corresponding to each latitude and longitude waypoint on the DSM is extracted using ArcGIS, the ground height is verified, and the absolute elevation of the waypoint is calculated. Finally, the waypoint file is generated in combination with the latitude and longitude coordinates of the route and the absolute elevation of each waypoint, and the waypoint file is superimposed with the original DSM to perform visibility analysis between waypoints to check the safety of the flight path. After the visibility analysis, the flight basic route file is generated, and the unmanned aerial vehicle will fly according to this route file, and the route will be adjusted in real time during the flight to adapt to the terrain changes.

[0086] As can be seen, through high-precision aerial photography and professional image processing, accurate DSM and DOM are generated, which can ensure the accuracy of the route planning and make the route cover the components to be cleaned. Dynamic switching of different resolution data optimizes the data processing efficiency and ensures that high-precision data can be quickly loaded and processed in real-time flight. Accurate determination of the absolute elevation of each waypoint can ensure that the unmanned aerial vehicle always maintains a safe height during flight, avoiding collision with the ground or photovoltaic modules, while ensuring cleaning effect. Visibility analysis ensures the safety of the flight path, avoiding flight risks caused by complex terrain. The finally generated flight basic route can ensure that the unmanned aerial vehicle safely and efficiently completes the cleaning task during flight, improving the overall efficiency and safety of the cleaning operation.

[0087] S102: Obtain the maximum slope and the maximum component interval of the photovoltaic power station, and determine the terrain scene of the photovoltaic power station according to the maximum slope and the maximum component interval; wherein the terrain scene includes flat land, regular mountainous land and complex mountainous land;

[0088] wherein the ground penetrating radar is a device installed on the unmanned aerial vehicle for measuring the distance between the unmanned aerial vehicle and the ground. The slope refers to the degree of inclination of the ground, and the component gap refers to the spatial distance between adjacent photovoltaic components. The terrain scene is a classification description of the ground conditions of the area where the unmanned aerial vehicle is located. The flat ground refers to a region with a flat ground, the conventional mountainous region refers to a mountainous region with a certain slope but relatively regular terrain, and the complex mountainous region refers to a mountainous region with large terrain undulations, complex slope changes and possibly many obstacles.

[0089] In this step, the project land DEM (Digital Elevation Model) file before the construction of the target photovoltaic power station is imported into ArcGIS, the 3D analyst in the extension module is opened, the slope grid and the slope direction grid are obtained respectively using ARctoolbox-3D analyst-raster surface-slope / aspect, the slope and slope direction of the power station are viewed, and the maximum slope is determined. The base route file is overlaid with the original DSM, and the maximum component interval covered by the base route is measured. Then, according to the specific values of the maximum slope and the maximum component interval, combined with the pre-set threshold or classification standard, it is determined whether the terrain scene of the photovoltaic power station is flat, conventional mountainous region or complex mountainous region.

[0090] As a feasible implementation manner, the method for determining the terrain scene of the photovoltaic power station according to the maximum slope and the maximum component interval comprises: judging whether the absolute value of the maximum slope is less than or equal to a first slope threshold value; if the absolute value of the maximum slope is less than or equal to the first slope threshold value, determining that the terrain scene of the photovoltaic power station is flat; if the absolute value of the maximum slope is greater than the first slope threshold value, judging whether the absolute value of the maximum slope is greater than a second slope threshold value; wherein the second slope threshold value is greater than the first slope threshold value; if the absolute value of the maximum slope is greater than the second slope threshold value, determining that the terrain scene of the photovoltaic power station is complex mountainous region; if the absolute value of the maximum slope is less than or equal to the second slope threshold value, judging whether the maximum component interval is less than or equal to a distance threshold value; if the maximum component interval is less than or equal to the distance threshold value, determining that the terrain scene of the photovoltaic power station is conventional mountainous region; if the maximum component interval is greater than the distance threshold value, determining that the terrain scene of the photovoltaic power station is complex mountainous region.

[0091] In a specific implementation, if the absolute value of the maximum slope is less than or equal to the first slope threshold, the terrain scene of the photovoltaic power station is determined as flat land. If the absolute value of the maximum slope is greater than the second slope threshold, the terrain scene of the photovoltaic power station is determined as complex mountainous area. If the absolute value of the maximum slope is greater than the first slope threshold and less than or equal to the second slope threshold, the maximum component interval covered by the basic flight route file needs to be further determined. If the absolute value of the maximum slope is greater than the first slope threshold and less than or equal to the second slope threshold, and the maximum component interval covered by the basic flight route is greater than the distance threshold, the terrain scene of the photovoltaic power station is determined as complex mountainous area. If the absolute value of the power station slope is greater than the first slope threshold and less than or equal to the second slope threshold, and the maximum component interval covered by the basic flight route is less than or equal to the distance threshold, the terrain scene of the photovoltaic power station is determined as regular mountainous area.

[0092] For the selection of the slope threshold, according to the flight process of the unmanned aerial vehicle, the following can be obtained from the trigonometric function:

[0093] .

[0094] Then: .

[0095] wherein, is the slope, is flat land, the slope is in the clockwise direction based on the horizontal plane, is uphill; the slope is in the counterclockwise direction, is downhill. is the longitudinal adjustment distance of the unmanned aerial vehicle, is the lateral flight distance of the unmanned aerial vehicle, is the vertical direction speed of the unmanned aerial vehicle, is the horizontal flight speed of the photovoltaic unmanned aerial vehicle, and t is the flight time of the unmanned aerial vehicle.

[0096] For the selection of the first slope threshold, the first slope threshold can be pre-set to be not more than 15°, and adaptive adjustment can be performed in the application process. At the same time, the first slope threshold needs to satisfy , that is, the sum of the height fluctuations caused by the slopes between two radar dotting times is not more than half of the installation height of the component relative to the ground, so as to identify the component gap, and therefore the first slope threshold can be calculated according to the following formula:

[0097] ;

[0098] ;

[0099] ;

[0100] wherein, is the first slope threshold, The update interval for ground-based radar markers is determined by the selected UAV model and radar parameters, and is typically 10-50ms. The horizontal flight speed of the drone is set according to the cleaning requirements. To ensure the cleaning effect, it is usually below 5m / s. The installation height of the module relative to the ground is a known parameter of the photovoltaic power station, typically not exceeding 1m. For the slope at The range of height fluctuation values.

[0101] For the second slope threshold The selection of the second slope threshold can be preset to not exceed 35°, and can be adaptively adjusted during application.

[0102] Regarding the selection of the distance threshold, assuming that both the ground-detection radar and the forward-looking radar detect the gap simultaneously, such as... Figure 2 As shown, in this situation, the data filtering components of the forward-facing radar and ground-based radar cannot be used at intervals. Image recognition from the drone's mounted camera is required for judgment. To avoid this, a distance threshold needs to be calculated as one of the bases for terrain classification. In this case, according to the sine theorem:

[0103] ;

[0104] ;

[0105] but: , .

[0106] in, The angle between the forward-facing radar and the ground-based radar is typically 45°. The distance measured by the ground-based radar is the drone's flight altitude relative to the lower plane (in this case, the ground). The distance measured by the forward radar is the diagonal distance between the drone and the front (in this case, the ground), and S is the distance threshold.

[0107] In distance threshold calculation, The value is the sum of the cleaning height set according to cleaning requirements and the installation height of the component relative to the ground, that is: , The installation height of the module relative to the ground is a known parameter of the photovoltaic power station, typically not exceeding 1m. The cleaning height is set according to the cleaning requirements, that is, the flight altitude of the drone relative to the components to be maintained during flight. To ensure the cleaning effect, it is usually no more than 2m. Based on the above description, the distance threshold S can be calculated.

[0108] S103: If the terrain of the photovoltaic power station is flat, the height of the photovoltaic cleaning drone and the nozzle status are adjusted according to the detection results of the ground radar; wherein, the detection results of the ground radar include components or gaps, and the nozzle status includes on and off.

[0109] The detection results from the ground-based radar are mainly categorized into components and gaps. If a component is detected, the nozzle typically needs to be turned on for cleaning. If a gap is detected, it indicates that the drone is located between adjacent photovoltaic modules. To avoid wasting cleaning fluid and to prevent unnecessary impact on the modules, the nozzle generally needs to be turned off.

[0110] In this step, once the terrain is determined to be flat, only the ground-based radar is activated, which continues to monitor altitude changes in real time. By analyzing the altitude change data, it is determined whether the current location belongs to a component or a gap. Based on the radar's detection results, the drone's altitude is adjusted accordingly to maintain it within a suitable distance range, ensuring cleaning effectiveness while avoiding collisions with components. Simultaneously, the nozzle's on / off state is controlled based on the detection results; the nozzle is activated for cleaning when a component is detected, and deactivated when a gap is detected.

[0111] As a feasible implementation, adjusting the altitude of the photovoltaic cleaning drone and controlling the nozzle state based on the detection results of the ground radar includes: determining the height h of the component center from the ground, and determining the height fluctuation threshold when the absolute value of the slope is less than or equal to the first slope threshold based on the update interval of the ground radar and the horizontal speed of the photovoltaic cleaning drone. ; Calculate the altitude detected by the ground-detecting radar at the current time i. The height detected at the previous time i-1 The height difference between ;like If the detection result of the ground radar is a component or gap, maintaining the height and nozzle status of the previous moment; if If the detection result of the ground radar is a gap, then the altitude of the photovoltaic cleaning drone is adjusted to... The nozzle status is set to off; if The detection result of the ground radar is the component, and the altitude of the photovoltaic cleaning drone is adjusted to... Set the nozzle status to "on".

[0112] In specific implementation, for flat land ( The flowchart for determining the gap between photovoltaic modules is as follows: Figure 3As shown. The heights of the components and gaps are sampled, and the heights of the components and gaps are extracted. The height h of the component center from the ground is calculated. The height of the ground-based radar is obtained, and the height of the ground-based radar at time i is calculated. And the current and previous ground radar altitude The difference ,Depend on Determine if the drone is in a gap. If This indicates that the drone detected above the component or gap, maintaining the previous altitude and nozzle status. If This indicates that the UAV's ground-point radar left the assembly, detected the gap, and used the ground-point radar altitude... Set height and close the nozzle. If This indicates that the drone left the gap, the component was detected, and the ground radar altitude was used. Set height and turn on the nozzle.

[0113] As can be seen, under relatively simple terrain conditions such as flat ground, this step can precisely control the drone's altitude and nozzle status, ensuring that the cleaning operation can be carried out efficiently and accurately, avoiding waste of cleaning fluid and potential damage to photovoltaic modules, improving cleaning quality and resource utilization efficiency, and also ensuring the flight safety of the drone.

[0114] S104: If the terrain of the photovoltaic power station is a conventional mountainous area, the altitude of the photovoltaic cleaning drone and the state of the nozzles are adjusted according to the detection results of the ground radar and the front radar; wherein, the detection results of the front radar include components or gaps.

[0115] Among them, the forward radar is used to detect changes in altitude within a certain distance in front of the drone, enabling it to perceive changes in the terrain ahead in advance and provide more comprehensive information for the drone's flight control. The detection results of the forward radar are mainly divided into components and gaps.

[0116] In this step, when the terrain is determined to be typical mountainous terrain, both the ground-based radar and the forward-facing radar operate simultaneously. The forward-facing radar detects the terrain in advance, extracts the component slope based on the data relationship between the forward-facing and ground-based radars, and determines the drone's position. During slope determination, the drone's flight control system uses a multi-radar altitude information fusion algorithm to ensure that the drone's ascent or descent distance matches the height of the terrain undulations, thus maintaining a constant drone altitude relative to the component and ignoring disturbances caused by changes in gap height. This allows for timely control of nozzle activation and deactivation, ensuring the nozzles are in the appropriate state at different locations, thereby guaranteeing the continuity and effectiveness of the cleaning operation.

[0117] As a feasible implementation, adjusting the altitude of the photovoltaic cleaning drone and controlling the nozzle status based on the detection results of the ground-based radar and the front-facing radar includes: during the flight of the photovoltaic cleaning drone, using the ground-based radar and the front-facing radar to detect the slope in real time; if the slope at the previous moment and the slope at the current moment are both positive, and the difference between the slope at the current moment and the slope at the previous moment is less than or equal to a third slope threshold, then the detection results of the front-facing radar and the ground-based radar are both for the gap before component or the gap after component, and the altitude of the photovoltaic cleaning drone is adjusted according to the slope at the current moment, and the nozzle status is controlled to be on; if the slope at the current moment is counterclockwise, the slope at the current moment is negative, and the slope at the previous moment is positive, then the detection result of the front-facing radar is a gap, and the detection result of the ground-based radar is... The detection result is the component, maintaining the height of the photovoltaic cleaning drone adjusted according to the slope at the previous moment, and maintaining the nozzle state at the previous moment; if the slope at the current moment is clockwise, the slope at the current moment is positive, and the slope at the previous moment is negative, then the detection result of the front radar is the component after the gap, and the detection result of the ground radar is the component before the gap, adjusting the height of the photovoltaic cleaning drone according to the slope at the current moment, and maintaining the nozzle state at the previous moment; if the slope at the previous moment and the slope at the current moment are both positive, and the difference between the slope at the current moment and the slope at the previous moment is greater than the third slope threshold, then the detection result of the front radar is the component after the gap, and the detection result of the ground radar is the gap, maintaining the height of the photovoltaic cleaning drone adjusted according to the slope at the previous moment, and controlling the nozzle state to be closed.

[0118] In specific implementation, for conventional mountainous areas ( And the flowchart for determining the photovoltaic module gap (with the maximum module spacing ≤ S) is as follows: Figure 4 As shown. Calculate the third slope threshold. The angle between the line connecting the two radar monitoring points and the horizontal plane when the ground-based radar detects the ground and the front-end radar detects the component can be obtained using trigonometric functions: At the third slope threshold During the calculation, The value is the sum of the cleaning height set according to cleaning requirements and the installation height of the component relative to the ground, that is: , The installation height of the module relative to the ground is a known parameter of the photovoltaic power station, typically not exceeding 1m. The cleaning height is set according to the cleaning requirements, which is the flight height of the drone relative to the components during flight. To ensure the cleaning effect, it usually does not exceed 2m.

[0119] During its flight over a conventional mountain power station, the UAV employs a slope-stabilized flight path with altitude hold to acquire data from both forward-facing radar and ground-based radar. The UAV begins its flight above the components, calculating the initial slope. The ground-based radar altitude at time i-1 is then extracted. Forward radar altitude Calculate the slope at time i-1. Extract the current radar altitude of i at the ground. Forward radar altitude The difference in ground radar altitude between the current and previous moments Calculate the slope at the current time i. The formula for calculating the slope is:

[0120] .

[0121] according to and Determine the current positions of the forward radar and ground radar, and then determine whether to adjust the drone's ascent or descent rate and nozzle status based on the current slope. If For positive and It is positive and satisfies At this point, both the forward radar and the ground radar detect the pre-spaced or post-spaced component, such as... Figure 5 As shown, the slope at the current moment is used. Adjust the drone's altitude and turn on the nozzles. Using the horizontal plane as a reference, if the slope is counter-clockwise at this point... negative and If the value is positive, then the front-end radar detects the gap, and the ground-based radar detects the component, such as... Figure 6 As shown, it does not respond to changes in the height of the gap, and at this time it maintains according to Adjust the drone's altitude to maintain the nozzle's position from the previous moment. Using the horizontal plane as a reference, if the slope is currently clockwise... For positive and If the value is negative, then the forward radar detects the component after the interval, while the ground radar is still on the component before the interval. Figure 7 As shown, the slope at the current moment is used. Adjust the drone's altitude to maintain the nozzle's position from the previous moment. If For positive and It is positive and satisfies At this point, the forward radar detects the component after the interval, and the ground radar detects the gap, such as... Figure 8 As shown, maintain according to When the drone's altitude is adjusted, the nozzles shut off and do not respond to changes in the gap's altitude.

[0122] It is evident that, in typical mountainous terrain with varying slopes but relatively regular patterns, this step, by combining detection data from ground-based radar and forward-facing radar, can predict terrain changes in advance. This allows the drone to adjust its altitude more smoothly during flight, reducing flight turbulence caused by terrain undulations and the risk of collisions with later components. This improves the stability and continuity of cleaning operations in complex terrain, further enhancing cleaning efficiency and quality.

[0123] S105: If the terrain of the photovoltaic power station is a complex mountainous area, the image data collected by the photovoltaic cleaning drone is input into the photovoltaic gap image recognition model. The photovoltaic gap image recognition model is used to determine the lifting rate or descent rate of the photovoltaic cleaning drone and control the nozzle state based on the recognition results of the image data at the current moment and the recognition results of the image data at the previous moment. The recognition results of the photovoltaic gap image recognition model include components or gaps.

[0124] The photovoltaic gap image recognition model is a model based on image recognition technology used to classify components and gaps. The lift rate refers to the speed at which the drone ascends per unit time, and the descent rate refers to the speed at which the drone descends per unit time.

[0125] In this step, when the terrain is complex and mountainous, radar detection alone may be insufficient to accurately determine the drone's position and terrain changes due to the undulating terrain and numerous obstacles. Therefore, a YOLOv8n photovoltaic gap image recognition model, accelerated by multispectral fusion and edge computing, is established based on component and gap feature analysis. This model is then used by the drone's onboard camera and computer for image recognition. Image data is collected using the camera mounted on the photovoltaic cleaning drone and input into the pre-trained photovoltaic gap image recognition model. This model accurately classifies components and gaps through image analysis and recognition; the model's recognition result includes either components or gaps. Then, the drone's ascent or descent rate is calculated based on the photovoltaic gap image recognition model's results for the current and previous image data. The drone's altitude is adjusted accordingly based on this rate and its current position to safely adapt to the complex mountainous terrain. Simultaneously, the nozzle's opening and closing status is controlled based on the photovoltaic gap image recognition model's results for the current and previous image data to ensure cleaning operations are performed at the appropriate time.

[0126] As a feasible implementation, the image data collected by the photovoltaic cleaning drone is input into a photovoltaic gap image recognition model. The model is used to determine the lift-up or descent rate of the photovoltaic cleaning drone and control the nozzle state based on the recognition results of the current image data and the previous image data. This includes: during the flight of the photovoltaic cleaning drone, real-time slope detection using ground-based radar and forward-facing radar; inputting the image data collected by the photovoltaic cleaning drone into the photovoltaic gap image recognition model; and outputting the recognition results of the image data using the photovoltaic gap image recognition model. The image data includes visible light images and infrared images. If the recognition results of the current image data and the previous image data are both... If the image data shows a gap, the lifting or lowering rate of the photovoltaic cleaning drone is adjusted according to the current slope to maintain the nozzle state of the previous moment. If the recognition result of the current image data is a gap and the recognition result of the previous image data is a component, the lifting or lowering rate is maintained according to the slope of the previous moment, and the nozzle state is controlled to be closed. If the recognition result of the current image data and the recognition result of the previous image data are both components, the lifting or lowering rate of the photovoltaic cleaning drone is adjusted according to the current slope to maintain the nozzle state of the previous moment. If the recognition result of the current image data is a component and the recognition result of the previous image data is a gap, the lifting or lowering rate of the photovoltaic cleaning drone is adjusted according to the slope of the current moment, and the nozzle state is controlled to be open.

[0127] In practical implementation, for complex mountainous areas ( The flowchart for determining the photovoltaic module gap (or maximum module spacing > S) is as follows: Figure 9 As shown. During its flight over a complex mountainous power station, the UAV acquires data from both forward-facing radar and ground-based radar. The UAV begins its flight from above the components, calculating the initial slope. The ground-based radar altitude at time i-1 is extracted. Forward radar altitude Calculate the slope at time i-1. Extract the current radar altitude of i at the ground. Forward radar altitude The difference in ground radar altitude between the current and previous moments Calculate the slope at the current time i. Visible light and infrared images captured by a drone are input into a photovoltaic gap image recognition model. This model can classify components and gaps based on image features. If the current image data is identified as a gap, and the previous image data also identified it as a gap, the uplift or descent rate is calculated using the current slope, maintaining the nozzle state from the previous moment. If the current image data is identified as a gap, and the previous image data identified it as a component, the uplift or descent rate is maintained using the slope from the previous state, and the nozzle is closed. If the current image data is identified as a component, and the previous image data also identified it as a component, the uplift or descent rate is calculated using the current slope, maintaining the nozzle state from the previous moment. If the current image data is identified as a component, and the previous image data identified it as a gap, the uplift or descent rate is calculated using the current slope, and the nozzle is opened.

[0128] As a feasible implementation method, the training process of the photovoltaic gap image recognition model includes: acquiring a training image set; wherein the training image set includes multiple sets of training images, each set of training images including a visible light image and an infrared image; labeling the training images; wherein the labeling results include components or gaps; performing preprocessing operations on the visible light images and the infrared images; wherein the preprocessing operations on the visible light images include suppressing the feature response of reflective areas and / or exposure correction, and the preprocessing operations on the infrared images include any one or a combination of any of the following: suppressing the feature response of reflective areas, non-uniformity correction, and temperature calibration; and matching the visible light images and infrared images in each set of training images. The images are registered and then fused to obtain a multispectral fused image. During the fusion process, pixel-level weighted fusion of the visible light image and the infrared image is used, and a cross-modal attention mechanism is employed to fuse the features of the visible light image and the infrared image. The multispectral fused image and its corresponding annotations are used to train a YOLOv8n model to obtain the photovoltaic gap image recognition model. The YOLOv8n model includes a photovoltaic global attention module, which processes image features based on channel attention and spatial attention mechanisms, and performs prior analysis on the cell grid using a fixed-weight grid mask to enhance the feature response of the module region.

[0129] In practice, visible light and infrared images were collected by drones at a distance of 2m above the photovoltaic modules under different environmental types (bare land, concrete surface, vegetated land), color types (dark blue, black, brownish-red), and time periods (morning, noon, afternoon). These images were used to construct a training image set, providing a rich and diverse set of image samples for model training. The characteristics of the photovoltaic modules and the surrounding ground were clearly defined. The photovoltaic modules are characterized as follows: crystalline silicon modules are dark blue / black, thin-film modules are brownish-red, with regularly arranged rectangular cell textures, potentially showing bright reflective areas, and exhibiting dust pollution and shading. The surrounding ground features are: vegetation-covered areas are green, bare soil is brown, and concrete and cement roof photovoltaic gaps are gray with irregular textures. A schematic diagram of the photovoltaic gap image recognition model is shown below. Figure 10As shown, to improve the model's generalization ability and simulate image conditions in complex real-world environments, the following image data augmentation methods can be used: The RandomSunFlare illumination transformation optimization algorithm is used to simulate component reflections under strong midday sunlight; the RandomShadow shadow transformation algorithm is used to simulate component shadows in cloudy or occluded environments; and the MultiSpectralDropout optimization algorithm is used to simulate multispectral random occlusion in complex environments such as foliage or birds. Next, these training images are labeled with their locations, explicitly marking the positions above the components and gaps in the images so that the model can learn to distinguish between these two different scenes. After labeling, preprocessing operations are performed on the visible light and infrared images respectively. For visible light image data, a reflectivity sensing module is added to detect and suppress the feature response of reflective areas. Automatic exposure correction is then performed, adjusting the median brightness of the target to 100-120, with the brightness percentile set to 50%. For infrared thermal imaging image data, a similar reflectivity sensing module is added to detect and suppress the feature response of reflective areas. Due to the fixed-mode noise inherent in infrared thermal imagers, dynamic statistical compensation is used for non-uniformity correction, and a radiation-temperature conversion model is used for temperature calibration. For the preprocessed visible light and infrared images, SIFT (Scale-Invariant Feature Transform) feature extraction, FLANN (Fast Library for Approximate Nearest Neighbors) feature matching, and affine transformation matrix calculation are employed. CUDA (Compute Unified Device Architecture) is used to accelerate resampling, achieving real-time spectral registration and ensuring spatial consistency between the two images. Pixel-level weighted fusion is used as the basic fusion method, while a cross-modal attention mechanism is employed to fuse features from visible light and infrared images. This fully leverages the advantages of both image types to enhance the model's ability to identify photovoltaic modules and their gaps. Finally, these multispectral fused images and their corresponding location annotations are used to train the YOLOv8n model, resulting in the final photovoltaic gap image recognition model.It is worth noting that the YOLOv8n model includes a Global Attention Module (GAM) for photovoltaics. This module processes image features through channel attention and spatial attention mechanisms, and uses a fixed-weight grid mask to perform prior analysis on the cell grid, thereby enhancing the feature response of the component region. This allows the model to more accurately identify the location of photovoltaic modules and gaps, further improving the model's recognition accuracy for photovoltaic modules. This provides precise image recognition support for photovoltaic cleaning drones flying and cleaning in complex mountainous environments, ensuring the efficient and accurate completion of cleaning tasks. The trained photovoltaic gap image recognition model can be ported to the drone's onboard computer for edge computing deployment. The hardware parameters can be NVIDIA Jetson Orin NX, 16GB of memory, and support for 4-channel input (visible light image + infrared image).

[0130] As can be seen, in complex mountainous terrain, where conditions are most intricate and unpredictable, this step, utilizing a photovoltaic gap image recognition model, can more accurately classify components and gaps. By comparing positions at different times, it determines the lift or descent rate, thereby enabling precise adjustments to the drone's altitude and reasonable control of the nozzle status. This allows the drone to fly and clean more safely and flexibly in complex mountainous environments, effectively avoiding collisions with obstacles and wasting cleaning fluid. It significantly improves the feasibility and reliability of cleaning operations in complex terrain, ensuring the successful completion of the entire cleaning task.

[0131] The flight control method for a photovoltaic cleaning drone provided in this application identifies terrain scenarios and adopts differentiated flight strategies based on the characteristics of different terrain scenarios. This allows the drone to better adapt to diverse terrains, ensuring smooth cleaning operations while significantly reducing drone energy consumption and improving cleaning efficiency. Furthermore, this application uses different strategies to adjust flight altitude according to the characteristics of different terrain scenarios, reducing unnecessary altitude adjustments and thus avoiding the risk of collisions with spaced components, improving the safety of the photovoltaic cleaning drone. Moreover, this application uses different strategies to accurately identify component gaps and precisely control nozzle status based on the characteristics of different terrain scenarios, saving water resources and improving cleaning uniformity.

[0132] The following describes a flight control device for a photovoltaic cleaning drone provided in an embodiment of this application. The flight control device for a photovoltaic cleaning drone described below and the flight control method for a photovoltaic cleaning drone described above can be referred to each other.

[0133] See Figure 11A structural diagram of a flight control device for a photovoltaic cleaning drone, as shown in an exemplary embodiment, is provided. Figure 11 As shown, it includes:

[0134] The generation module 100 is used to generate the basic flight path of the photovoltaic cleaning drone based on the image data of the photovoltaic power station;

[0135] The determination module 200 is used to obtain the maximum slope and maximum component spacing of the photovoltaic power station, and determine the terrain scene of the photovoltaic power station based on the maximum slope and the maximum component spacing; wherein, the terrain scene includes flat land, conventional mountainous land and complex mountainous land;

[0136] The first control module 300 is used to adjust the altitude of the photovoltaic cleaning drone and control the nozzle status based on the detection results of the ground radar when the terrain scene is flat; wherein, the detection results of the ground radar include components or gaps, and the nozzle status includes on and off.

[0137] The second control module 400 is used to adjust the altitude of the photovoltaic cleaning drone and control the nozzle status based on the detection results of the ground radar and the front radar when the terrain scene is a conventional mountainous area; wherein, the detection results of the front radar include components or gaps.

[0138] The third control module 500 is used to input the image data collected by the photovoltaic cleaning drone into the photovoltaic gap image recognition model, and use the photovoltaic gap image recognition model to determine the lifting rate or descent rate of the photovoltaic cleaning drone and control the nozzle state based on the recognition results of the image data at the current moment and the recognition results of the image data at the previous moment; wherein, the recognition results of the photovoltaic gap image recognition model include components or gaps.

[0139] The flight control device for the photovoltaic cleaning drone provided in this application identifies terrain scenarios and adopts differentiated flight strategies based on the characteristics of different terrain scenarios. This allows the drone to better adapt to diverse terrains, ensuring smooth cleaning operations while significantly reducing drone energy consumption and improving cleaning efficiency. Furthermore, this application adjusts the flight altitude using different strategies based on the characteristics of different terrain scenarios, reducing unnecessary altitude adjustments and thus avoiding the risk of collisions with spaced components, improving the safety of the photovoltaic cleaning drone. Moreover, this application accurately identifies component gaps and precisely controls the nozzle state using different strategies based on the characteristics of different terrain scenarios, saving water resources and improving cleaning uniformity.

[0140] Based on the above embodiments, as a preferred implementation, the generation module 100 is specifically used for: collecting image data of photovoltaic power stations; generating digital surface models and digital orthophoto maps with different ground resolutions based on the image data; generating the latitude and longitude coordinates of the flight path of the photovoltaic cleaning drone based on the digital surface models and digital orthophoto maps with different ground resolutions; determining the absolute elevation of each waypoint of the photovoltaic cleaning drone; and generating the basic flight path of the photovoltaic cleaning drone based on the latitude and longitude coordinates of the flight path and the absolute elevation of each waypoint.

[0141] Based on the above embodiments, as a preferred implementation, the determining module 200 is specifically used to: determine whether the absolute value of the maximum slope is less than or equal to a first slope threshold; if the absolute value of the maximum slope is less than or equal to the first slope threshold, then determine that the terrain scene of the photovoltaic power station is flat land; if the absolute value of the maximum slope is greater than the first slope threshold, then determine whether the absolute value of the maximum slope is greater than a second slope threshold; wherein, the second slope threshold is greater than the first slope threshold; if the absolute value of the maximum slope is greater than the second slope threshold, then determine that the terrain scene of the photovoltaic power station is complex mountainous terrain; if the absolute value of the maximum slope is less than or equal to the second slope threshold, then determine whether the maximum component spacing is less than or equal to a distance threshold; if the maximum component spacing is less than or equal to the distance threshold, then determine that the terrain scene of the photovoltaic power station is conventional mountainous terrain; if the maximum component spacing is greater than the distance threshold, then determine that the terrain scene of the photovoltaic power station is complex mountainous terrain.

[0142] Based on the above embodiments, as a preferred implementation, the first control module 300 is specifically used to: determine the height h of the component center from the ground, and determine the height fluctuation threshold when the absolute value of the slope is less than or equal to the first slope threshold based on the update interval of the ground radar and the horizontal speed of the photovoltaic cleaning drone. ; Calculate the altitude detected by the ground-detecting radar at the current time i. The height detected at the previous time i-1 The height difference between ;like If the detection result of the ground radar is a component or gap, maintaining the height and nozzle status of the previous moment; if If the detection result of the ground radar is a gap, then the altitude of the photovoltaic cleaning drone is adjusted to... The nozzle status is set to off; if The detection result of the ground radar is the component, and the altitude of the photovoltaic cleaning drone is adjusted to... Set the nozzle status to "on".

[0143] Based on the above embodiments, as a preferred implementation, the second control module 400 is specifically used for: during the flight of the photovoltaic cleaning drone, using ground-based radar and front-facing radar to detect the slope in real time; if the slope at the previous moment and the slope at the current moment are both positive, and the difference between the slope at the current moment and the slope at the previous moment is less than or equal to a third slope threshold, then the detection results of the front-facing radar and the ground-based radar are both gap-before component or gap-after component, and the height of the photovoltaic cleaning drone is adjusted according to the slope at the current moment, controlling the nozzle state to be on; if the slope at the current moment is counterclockwise, the slope at the current moment is negative, and the slope at the previous moment is positive, then the detection result of the front-facing radar is gap, and the detection result of the ground-based radar is component, maintaining... The height of the photovoltaic cleaning drone is adjusted according to the slope at the previous moment, maintaining the nozzle state of the previous moment. If the current slope is clockwise, positive, and the previous slope was negative, then the detection result of the front radar is the gap rear component, and the detection result of the ground radar is the gap front component. The height of the photovoltaic cleaning drone is adjusted according to the current slope, maintaining the nozzle state of the previous moment. If both the previous and current slopes are positive, and the difference between the current and previous slopes is greater than the third slope threshold, then the detection result of the front radar is the gap rear component, and the detection result of the ground radar is the gap. The height of the photovoltaic cleaning drone adjusted according to the slope at the previous moment is maintained, and the nozzle state is controlled to be closed.

[0144] Based on the above embodiments, as a preferred implementation, the third control module 500 is specifically used for: detecting the slope in real time using ground-based radar and forward-facing radar during the flight of the photovoltaic cleaning drone; inputting the image data collected by the photovoltaic cleaning drone into a photovoltaic gap image recognition model, and outputting the recognition result of the image data using the photovoltaic gap image recognition model; wherein, the image data includes visible light images and infrared images; if the recognition result of the image data at the current moment and the recognition result of the image data at the previous moment are both gaps, then adjusting the lift rate or descent rate of the photovoltaic cleaning drone according to the slope at the current moment, maintaining the slope at the previous moment. The nozzle status is determined as follows: If the recognition result of the current image data is a gap, and the recognition result of the previous image data is a component, then the lifting rate or descent rate is maintained according to the slope of the previous moment, and the nozzle status is controlled to be closed; if the recognition result of the current image data and the recognition result of the previous image data are both components, then the lifting rate or descent rate of the photovoltaic cleaning drone is adjusted according to the slope of the current moment to maintain the nozzle status of the previous moment; if the recognition result of the current image data is a component, and the recognition result of the previous image data is a gap, then the lifting rate or descent rate of the photovoltaic cleaning drone is adjusted according to the slope of the current moment, and the nozzle status is controlled to be open.

[0145] Based on the above embodiments, as a preferred embodiment, it further includes:

[0146] A training module is used to acquire a training image set, wherein the training image set includes multiple sets of training images, each set including a visible light image and an infrared image; to annotate the training images, wherein the annotation results include components or gaps; to perform preprocessing operations on the visible light images and the infrared images, wherein the preprocessing operations on the visible light images include suppressing the feature response of reflective areas and / or exposure correction, and the preprocessing operations on the infrared images include any one or a combination of suppressing the feature response of reflective areas, non-uniformity correction, and temperature calibration; to register the visible light image and the infrared image in each set of training images, and to perform [further processing] based on the registration results. A multispectral fused image is obtained by fusing the visible light image and the infrared image. During the fusion process, pixel-level weighted fusion is used to fuse the visible light image and the infrared image, and a cross-modal attention mechanism is used to fuse the features of the visible light image and the infrared image. The YOLOv8n model is trained using the multispectral fused image and the corresponding annotations to obtain the photovoltaic gap image recognition model. The YOLOv8n model includes a photovoltaic global attention module, which is used to process image features based on channel attention mechanism and spatial attention mechanism, and to perform prior arithmetic on the cell grid through a fixed-weight grid mask to enhance the feature response of the module region.

[0147] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0148] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiments of this application, the embodiments of this application also provide an electronic device. Figure 12 This is a structural diagram of an electronic device according to an exemplary embodiment, such as... Figure 12 As shown, the electronic device includes:

[0149] Communication interface 1 enables information exchange with other devices, such as network devices;

[0150] Processor 2 is connected to communication interface 1 to enable information interaction with other devices. When running a computer program, it executes the flight control method of the photovoltaic cleaning drone provided by one or more of the above-mentioned technical solutions. The computer program is stored in memory 3.

[0151] Of course, in practical applications, the various components in an electronic device are coupled together through bus system 4. It can be understood that bus system 4 is used to achieve communication and connection between these components. In addition to the data bus, bus system 4 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 12 The general will label all buses as Bus System 4.

[0152] The memory 3 in this embodiment is used to store various types of data to support the operation of the electronic device. Examples of such data include any computer program used to operate on the electronic device.

[0153] It is understood that memory 3 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memory 3 described in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0154] The methods disclosed in the embodiments of this application can be applied to processor 2, or implemented by processor 2. Processor 2 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 2 or by instructions in the form of software. The processor 2 may be a general-purpose processor, DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 2 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in memory 3. Processor 2 reads the program in memory 3 and completes the steps of the aforementioned method in combination with its hardware.

[0155] When processor 2 executes the program, it implements the corresponding processes in the various methods of the embodiments of this application. For the sake of brevity, these will not be described in detail here.

[0156] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory 3 that stores a computer program, which can be executed by a processor 2 to complete the steps described in the aforementioned method. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0157] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0158] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0159] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A flight control method for a photovoltaic cleaning drone, characterized in that, include: Generate the basic flight path of photovoltaic cleaning drones based on image data of photovoltaic power plants; The maximum slope and maximum module spacing of the photovoltaic power station are obtained, and the terrain scenario of the photovoltaic power station is determined based on the maximum slope and the maximum module spacing; wherein, the terrain scenario includes flat land, conventional mountainous land and complex mountainous land; If the terrain of the photovoltaic power station is flat, the altitude of the photovoltaic cleaning drone and the status of the nozzles are adjusted according to the detection results of the ground radar; wherein, the detection results of the ground radar include components or gaps, and the status of the nozzles includes on and off. If the terrain of the photovoltaic power station is a typical mountainous area, the altitude of the photovoltaic cleaning drone and the state of the nozzles are adjusted according to the detection results of the ground radar and the front radar; wherein, the detection results of the front radar include components or gaps. If the terrain of the photovoltaic power station is a complex mountainous area, the image data collected by the photovoltaic cleaning drone is input into the photovoltaic gap image recognition model. The photovoltaic gap image recognition model is used to determine the lifting or lowering rate of the photovoltaic cleaning drone and control the nozzle state based on the recognition results of the image data at the current moment and the recognition results of the image data at the previous moment. The recognition results of the photovoltaic gap image recognition model include components or gaps. The image data collected by the photovoltaic cleaning drone is input into a photovoltaic gap image recognition model. The model uses the recognition results of the current image data and the previous image data to determine the lifting or lowering rate of the photovoltaic cleaning drone and control the nozzle state, including: During the flight of the photovoltaic cleaning drone, the slope is detected in real time using ground radar and forward radar; The image data collected by the photovoltaic cleaning drone is input into the photovoltaic gap image recognition model, and the photovoltaic gap image recognition model is used to output the recognition result of the image data; wherein, the image data includes visible light images and infrared images; If the recognition result of the image data at the current moment is a gap with the recognition result of the image data at the previous moment, then the lifting rate or descent rate of the photovoltaic cleaning drone is adjusted according to the slope at the current moment to maintain the nozzle state at the previous moment. If the recognition result of the image data at the current moment is a component of the gap and the recognition result of the image data at the previous moment, then the lifting rate or the lowering rate is maintained according to the slope at the previous moment, and the nozzle state is controlled to be closed. If the recognition result of the image data at the current moment and the recognition result of the image data at the previous moment are both components, then the lifting rate or descent rate of the photovoltaic cleaning drone is adjusted according to the slope at the current moment to maintain the nozzle state at the previous moment. If the recognition result of the image data at the current moment is a component and the recognition result of the image data at the previous moment is a gap, then the lifting rate or descent rate of the photovoltaic cleaning drone is adjusted according to the slope at the current moment, and the nozzle status is controlled to be on. The method further includes: Obtain a training image set; wherein the training image set includes multiple sets of training images, and each set of training images includes visible light images and infrared images; The training images are labeled; wherein the labeling results include components or gaps; The visible light image and the infrared image are preprocessed; wherein the preprocessing operation of the visible light image includes suppressing the feature response of the reflective area and / or exposure correction, and the preprocessing operation of the infrared image includes any one or a combination of any of the following: suppressing the feature response of the reflective area, non-uniformity correction, and temperature calibration. The visible light image and infrared image in each group of training images are registered, and the multispectral fused image is obtained based on the registration results. In the fusion process, the visible light image and the infrared image are fused using pixel-level weighting, and the features of the visible light image and the infrared image are fused using a cross-modal attention mechanism. The photovoltaic gap image recognition model is obtained by training the YOLOv8n model using the multispectral fusion image and the corresponding annotations; wherein, the YOLOv8n model includes a photovoltaic global attention module, which is used to process image features based on channel attention mechanism and spatial attention mechanism, and to perform prior on cell grid through fixed weight grid mask to enhance the feature response of the module region.

2. The flight control method for the photovoltaic cleaning drone according to claim 1, characterized in that, The basic flight path for the photovoltaic cleaning drone, generated based on image data from the photovoltaic power station, includes: Collect image data of photovoltaic power plants, and generate digital surface models and digital orthophoto maps with different ground resolutions based on the image data; Generate the latitude and longitude coordinates of the flight path of the photovoltaic cleaning drone based on digital surface models and digital orthophoto maps with different ground resolutions; The absolute elevation of each waypoint of the photovoltaic cleaning drone is determined, and the basic flight route of the photovoltaic cleaning drone is generated based on the latitude and longitude coordinates of the route and the absolute elevation of each waypoint.

3. The flight control method for the photovoltaic cleaning drone according to claim 1, characterized in that, The step of determining the terrain scenario of the photovoltaic power station based on the maximum slope and the maximum component spacing includes: Determine whether the absolute value of the maximum slope is less than or equal to the first slope threshold; If the absolute value of the maximum slope is less than or equal to the first slope threshold, then the terrain of the photovoltaic power station is determined to be flat. If the absolute value of the maximum slope is greater than the first slope threshold, then it is determined whether the absolute value of the maximum slope is greater than the second slope threshold; wherein the second slope threshold is greater than the first slope threshold. If the absolute value of the maximum slope is greater than the second slope threshold, the terrain of the photovoltaic power station is determined to be complex mountainous terrain. If the absolute value of the maximum slope is less than or equal to the second slope threshold, then determine whether the maximum component interval is less than or equal to the distance threshold. If the maximum component spacing is less than or equal to the distance threshold, the terrain of the photovoltaic power station is determined to be conventional mountainous terrain. If the maximum component spacing is greater than the distance threshold, the terrain of the photovoltaic power station is determined to be complex mountainous terrain.

4. The flight control method for the photovoltaic cleaning drone according to claim 3, characterized in that, The step of adjusting the altitude of the photovoltaic cleaning drone and controlling the nozzle status based on the detection results of the ground radar includes: Determine the height h of the component center above the ground, and based on the update interval of the ground-based radar and the horizontal speed of the photovoltaic cleaning drone, determine the height fluctuation threshold when the absolute value of the slope is less than or equal to the first slope threshold. ; Calculate the altitude detected by the ground-detecting radar at the current time i. The height detected at the previous time i-1 The height difference between ; like If the detection result of the ground radar is a component or gap, the height and nozzle status of the previous moment are maintained. like If the detection result of the ground radar is a gap, then the altitude of the photovoltaic cleaning drone is adjusted to... Set the nozzle status to off; like The detection result of the ground radar is the component, and the altitude of the photovoltaic cleaning drone is adjusted to... Set the nozzle status to "on".

5. The flight control method for the photovoltaic cleaning drone according to claim 1, characterized in that, The step of adjusting the altitude of the photovoltaic cleaning drone and controlling the nozzle status based on the detection results of the ground radar and the forward radar includes: During the flight of the photovoltaic cleaning drone, the slope is detected in real time using ground radar and forward radar; If the slope at the previous moment and the slope at the current moment are both positive, and the difference between the slope at the current moment and the slope at the previous moment is less than or equal to the third slope threshold, then the detection results of the front radar and the detection results of the ground radar are both the gap front component or the gap rear component. The height of the photovoltaic cleaning drone is adjusted according to the slope at the current moment, and the nozzle status is controlled to be on. If the current slope is counterclockwise, the current slope is negative, and the previous slope was positive, then the detection result of the front radar is a gap, and the detection result of the ground radar is a component. The height of the photovoltaic cleaning drone is maintained according to the slope of the previous moment, and the nozzle state of the previous moment is maintained. If the current slope is clockwise, the current slope is positive, and the previous slope was negative, then the detection result of the front radar is the gap rear component, and the detection result of the ground radar is the gap front component. The height of the photovoltaic cleaning drone is adjusted according to the current slope to maintain the nozzle state of the previous moment. If the slope at the previous moment and the slope at the current moment are both positive, and the difference between the slope at the current moment and the slope at the previous moment is greater than the third slope threshold, then the detection result of the front radar is the gap behind the component, the detection result of the ground radar is the gap, the height of the photovoltaic cleaning drone is maintained according to the slope at the previous moment, and the nozzle status is controlled to be closed.

6. A flight control device for a photovoltaic cleaning drone, characterized in that, include: The generation module is used to generate the basic flight path of the photovoltaic cleaning drone based on the image data of the photovoltaic power station; A determination module is used to obtain the maximum slope and maximum module spacing of the photovoltaic power station, and determine the terrain scenario of the photovoltaic power station based on the maximum slope and maximum module spacing; wherein, the terrain scenario includes flat land, conventional mountainous land and complex mountainous land; The first control module is used to adjust the altitude of the photovoltaic cleaning drone and control the nozzle status based on the detection results of the ground radar when the terrain scene is flat; wherein, the detection results of the ground radar include components or gaps, and the nozzle status includes on and off. The second control module is used to adjust the altitude of the photovoltaic cleaning drone and control the nozzle status based on the detection results of the ground radar and the front radar when the terrain scene is a conventional mountainous area; wherein, the detection results of the front radar include components or gaps. The third control module is used to input the image data collected by the photovoltaic cleaning drone into the photovoltaic gap image recognition model, and use the recognition results of the photovoltaic gap image recognition model for the current image data and the recognition results for the previous image data to determine the lifting rate or descent rate of the photovoltaic cleaning drone and control the nozzle state; wherein, the recognition results of the photovoltaic gap image recognition model include components or gaps; Specifically, the third control module is used to: detect the slope in real time using ground-based radar and forward-facing radar during the flight of the photovoltaic cleaning drone; input the image data collected by the photovoltaic cleaning drone into a photovoltaic gap image recognition model, and output the recognition result of the image data using the photovoltaic gap image recognition model; wherein the image data includes visible light images and infrared images; if the recognition result of the image data at the current moment and the recognition result of the image data at the previous moment are both gaps, then adjust the lifting rate or descent rate of the photovoltaic cleaning drone according to the slope at the current moment to maintain the nozzle state at the previous moment; if the current moment... If the image data recognition result is a gap or a component of the image data recognition result from the previous moment, then the lifting or lowering rate is maintained according to the slope of the previous moment, and the nozzle state is controlled to be closed; if the image data recognition result at the current moment and the image data recognition result at the previous moment are both components, then the lifting or lowering rate of the photovoltaic cleaning drone is adjusted according to the slope of the current moment, maintaining the nozzle state of the previous moment; if the image data recognition result at the current moment is a component and the image data recognition result at the previous moment is a gap, then the lifting or lowering rate of the photovoltaic cleaning drone is adjusted according to the slope of the current moment, and the nozzle state is controlled to be open. The device further includes: A training module is used to acquire a training image set, wherein the training image set includes multiple sets of training images, each set including a visible light image and an infrared image; to annotate the training images, wherein the annotation results include components or gaps; to perform preprocessing operations on the visible light images and the infrared images, wherein the preprocessing operations on the visible light images include suppressing the feature response of reflective areas and / or exposure correction, and the preprocessing operations on the infrared images include any one or a combination of suppressing the feature response of reflective areas, non-uniformity correction, and temperature calibration; to register the visible light image and the infrared image in each set of training images, and to perform [further processing] based on the registration results. A multispectral fused image is obtained by fusing the visible light image and the infrared image. During the fusion process, pixel-level weighted fusion is used to fuse the visible light image and the infrared image, and a cross-modal attention mechanism is used to fuse the features of the visible light image and the infrared image. The YOLOv8n model is trained using the multispectral fused image and the corresponding annotations to obtain the photovoltaic gap image recognition model. The YOLOv8n model includes a photovoltaic global attention module, which is used to process image features based on channel attention mechanism and spatial attention mechanism, and to perform prior arithmetic on the cell grid through a fixed-weight grid mask to enhance the feature response of the module region.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the flight control method for the photovoltaic cleaning drone as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the flight control method for the photovoltaic cleaning drone as described in any one of claims 1 to 5.

Images