Navigation deviation correction method and device for photovoltaic cleaning

Abstract

The application provides a navigation deviation correction method and device for photovoltaic cleaning. The method comprises the following steps: continuously acquiring image data on a driving path of a cleaning vehicle during cleaning, and extracting an upper edge feature of a current component row where the cleaning vehicle is located according to the image data; acquiring an upper edge reference path of the current component row from a pre-stored component layout model; and correcting a driving direction of the cleaning vehicle according to a comparison result of the upper edge feature and the upper edge reference path, so as to maintain parallel driving with the upper edge of the component. The application realizes accurate dynamic deviation correction of the driving path of the cleaning vehicle under the condition of no pre-set parameters by collecting the upper edge image of the photovoltaic component in real time and adaptively establishing the component layout model.

Description

Technical Field

[0001] This application relates to the field of photovoltaic cleaning, and more particularly to a navigation and correction method and device for photovoltaic cleaning. Background Technology

[0002] The surfaces of photovoltaic modules in photovoltaic power plants are prone to accumulating pollutants such as dust, sand, and bird droppings, leading to a significant decrease in photoelectric conversion efficiency. Therefore, regular cleaning is necessary. Currently, photovoltaic power plant cleaning vehicles on the market mainly employ two navigation and correction methods: one is to pre-set a fixed path using satellite maps, laser modeling, etc., and the cleaning vehicle travels along the preset path; the other relies on manual remote control, where the operator controls the cleaning vehicle's direction and cleaning trajectory in real time.

[0003] However, both methods have obvious drawbacks: the preset fixed path method has poor adaptability. When the installation angle of the photovoltaic modules is slightly adjusted, the distribution of the module array is locally changed, or temporary obstacles appear in the power station, the preset path cannot match the actual working conditions, and it is easy to deviate from the effective area of ​​the modules due to driving deviation, resulting in missed cleaning. The manual remote control method requires a lot of manpower, and the cleaning efficiency is significantly affected by human factors. It is difficult to accurately maintain alignment with the edge of the modules, and it also has the problem of incomplete coverage. Summary of the Invention

[0004] One of the objectives of this application is to address at least some of the problems existing in the prior art by providing a navigation and correction method and apparatus for photovoltaic cleaning. Specifically, the technical solution of this application is as follows: Firstly, this application outlines a navigation and correction method for photovoltaic cleaning, including: During the cleaning process, the sweeper continuously acquires image data along its travel path; Extract the upper edge features of the current component row where the sweeper is located based on the image data; Obtain the top edge reference path of the current component row from the pre-stored component layout model; Based on the comparison results between the upper edge features and the upper edge reference path, the driving direction of the sweeper is corrected to maintain the sweeper driving parallel to the upper edge of the current component row.

[0005] In some implementations, the continuous acquisition of image data along the driving path specifically includes: The image data is acquired by an image acquisition device installed on the sweeper body. The image acquisition device includes three cameras: a front-view camera, a side-view camera, and a rear-view camera. The front-view camera is used to capture the area to be cleaned directly in front, the side-view camera is used to capture the areas on both sides, and the rear-view camera is used to capture the area that has been cleaned behind.

[0006] In some implementations, correcting the sweeper's travel direction based on the comparison result between the upper edge feature and the upper edge reference path includes: If a break or obstruction is detected on the upper edge of the component row, the path of the broken upper edge is predicted based on the extension trend of the upper edge. The vehicle temporarily travels along the predicted upper edge path, while the side-view camera helps to capture the upper edges of adjacent components. Once the complete upper edge is detected again, the vehicle immediately corrects to the upper edge reference path.

[0007] In other embodiments, when the sweeper reaches the end of the current component row, a row switching judgment is performed based on the row distribution information of the component layout model.

[0008] In other embodiments, the navigation and correction method for photovoltaic cleaning further includes pre-constructing the component layout model, specifically including: By conducting a complete cleaning process of the photovoltaic power station, the cleaning vehicle travels along the upper edge of any row of components, simultaneously collecting image data. Extract the top edge features, angles, spacing, distribution features, and array features of the component rows from the image data, and cluster these features to form feature clusters; As the sweeper moves, it continuously collects image data and simultaneously updates feature clusters. Repeat the above process until a complete sweep is completed, and establish a component layout model that saves the feature clusters of each of the component rows.

[0009] In other embodiments, the inter-row switching determination includes: The side-view camera captures component images on both sides of the current component row, segments and extracts the upper edge features of adjacent uncleaned component rows, and determines their starting position, extension direction, parallelism with the upper edge of the current component row, and spacing between rows. The sweeper first adjusts its direction to ensure that the end of the current component row is aligned with the upper edge of the starting position of the adjacent row, then travels a distance in a direction perpendicular to the upper edge of the current component row, and finally adjusts the direction of the sweeper to be parallel to the upper edge of the adjacent row. After switching to the starting position of the adjacent row, the upper edge image of the components in the adjacent row is acquired, and the upper edge parallelism deviation is calculated; If the switching is successful based on the upper edge parallelism deviation, the sweeper will begin sweeping along the upper edge extension direction of the adjacent row.

[0010] In some implementations, when the sweeper reaches a fork in the road, the component features of the branches contained in the current fork in the road are obtained from the component layout model; The component features of each branch are matched with the component features of the current component row, and the branch with the highest matching degree is selected as the candidate direction.

[0011] In other embodiments, matching the component features of each branch with the component features of the current component row includes: Calculate the comprehensive feature similarity between the component angle features, component spacing features, and upper edge parallelism deviation of each branch and the current path feature cluster, and select the branch with the highest comprehensive feature similarity as the candidate direction; If the similarity of the comprehensive features of each branch is small, then the component distribution density of each branch is calculated, and the branch with the highest density is selected as the candidate direction. If the density difference between the branches is small, the extension length is calculated, and the branch with the longest extension length is selected as the candidate direction.

[0012] In some embodiments, the navigation and correction method for photovoltaic cleaning further includes calculating the extension length using the following formula: ; Where L is the extension length. Extend the pixel length of the upper edge of the detected components in the image. The conversion factor is the physical length of the image pixel to the upper edge of the component. This is the predicted value of the trend extension length based on the feature cluster.

[0013] Secondly, this application also discloses a navigation and correction device for photovoltaic cleaning, comprising: Memory, used to store computer programs; A processor is used to implement the navigation and correction method for photovoltaic cleaning as described in any of the above embodiments when running the computer program.

[0014] The navigation and correction method and device for photovoltaic cleaning provided in this application can bring at least the following beneficial effects: 1. This application acquires real-time images of the upper edge of the component and dynamically fits a baseline path, combined with a closed-loop correction mechanism, ensuring that the sweeper always travels parallel to the upper edge of the component. This effectively addresses actual working conditions such as component angle adjustments, local layout changes, and temporary obstacles, fundamentally eliminating sweeping omissions caused by driving deviation and significantly improving coverage accuracy and operational reliability.

[0015] 2. This application optimizes the direction judgment logic at intersections by introducing the angle, spacing features, and upper edge parallelism deviation of the multi-branch component row. These features are compared with the comprehensive feature similarity of the current path feature cluster. The branch with the highest comprehensive feature similarity is selected as the candidate direction, which not only matches the component layout pattern but also ensures the feasibility of upper edge alignment for subsequent cleaning, thereby improving the rationality of path decision-making.

[0016] 3. The proposed solution uses image data as the sole core input. The deviation threshold and similarity weight of the component's upper edge are all adaptively generated by the model, which can adapt to the differences in component size, installation angle and layout of different photovoltaic power plants, making it highly versatile. Attached Figure Description

[0017] The preferred embodiments will be described below in a clear and easy-to-understand manner, with reference to the accompanying drawings, to further explain the above-mentioned characteristics, technical features, advantages, and implementation methods of a navigation and correction method and device for photovoltaic cleaning.

[0018] Figure 1 This is a flowchart of an embodiment of a navigation and correction method for photovoltaic cleaning according to this application; Figure 2 This is a flowchart of another embodiment of a navigation and correction method for photovoltaic cleaning according to this application; Figure 3 This is a schematic diagram of one embodiment of a navigation and correction device for photovoltaic cleaning according to this application. Detailed Implementation

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the specific implementation methods of this application will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without creative effort.

[0020] To keep the drawings concise, each drawing only schematically shows the parts relevant to this application, and they do not represent the actual structure of the product. Furthermore, for ease of understanding, in some drawings, components with the same structure or function are only schematically drawn or labeled as one. In this document, "one" not only means "only one," but can also mean "more than one."

[0021] Reference manual attached Figure 1 As shown, one embodiment of the navigation and correction method for photovoltaic cleaning according to this application specifically includes the following steps: S1, the sweeper continuously acquires image data along its travel path during the sweeping process.

[0022] In some optional implementations, step S1 involves acquiring image data using an image acquisition device mounted on the sweeper body. The image acquisition device includes three cameras: a front-view camera, a side-view camera, and a rear-view camera. The front-view camera is used to acquire the area to be cleaned directly in front, the side-view camera is used to acquire the areas on both sides, and the rear-view camera is used to acquire the cleaned area behind.

[0023] Specifically, during the cleaning process within the photovoltaic power station, the sweeper's front-view, side-view, and rear-view cameras, all mounted on the vehicle, simultaneously and continuously collect image data within the driving path. The frame rate is no less than 30fps, and the resolution is no less than 1080P.

[0024] Optionally, the image acquisition device features an ambient light adaptive adjustment function, enabling stable imaging under complex lighting conditions such as strong light and shadow. The front-view camera is installed at the same height as the upper edge of the component, with an error not exceeding ±5 cm, and the lens is oriented in the same direction as the sweeper's travel to ensure real-time capture of images of the upper edge of the component in front of the current travel direction. The acquired image data is transmitted in real-time to the edge computing host, serving as the basic input for subsequent path recognition and navigation control.

[0025] S2, extract the upper edge features of the current component row where the sweeper is located based on the image data.

[0026] In some implementations, after receiving image data, the edge computing host first preprocesses the image, including denoising, distortion correction, and grayscale enhancement, to improve image quality. Then, by introducing an attention-enhanced U-Net model, the photovoltaic module region in the image is segmented from the background region, thereby automatically extracting the upper edge contour features of the module.

[0027] Optionally, within the segmented component region, the Canny edge detection algorithm combined with morphological processing is used to accurately extract the upper edge contour of the current component row and fit it into the upper edge straight line equation y=kx+b, where k is the slope and b is the intercept. Simultaneously, the parallelism deviation δ between the upper edge and the current driving direction of the sweeper is calculated. This deviation directly reflects whether the sweeper is skewed relative to the upper edge of the component, without requiring any preset deviation threshold, relying entirely on the adaptive extraction of real-time image features.

[0028] In other embodiments, step S2 further includes automatically extracting other core features of the component, including the upper edge contour features, based on the segmented component region, including but not limited to the following: Component angular characteristics: The equation of the straight line at the edge of the component is detected by Hough transform, and the angle between the straight line and the horizontal direction is calculated to obtain the installation angle α of each component. i .

[0029] Component spacing characteristics: Based on the image pixel scale (obtained by the visual sensor), the pixel distance between the edges of adjacent components is calculated and converted into the actual physical spacing d. i .

[0030] Component distribution characteristics: Through connected component analysis, the number of components and their distribution compactness within a unit image region are statistically analyzed to form an image representation of component distribution density.

[0031] Intersection features: When at least two independent component connected regions are detected in the image, it is determined to be an intersection. At the same time, the number of branches and the component region boundary and upper edge features of each branch are identified.

[0032] In addition, the array features: by the row and column distribution pattern of the component connected regions, the extension direction (horizontal / vertical), the spacing between rows, and the total number of arrays of the components are identified, and the overall extension trend of the upper edge of each row of components is extracted.

[0033] S3, obtain the top edge reference path of the current component row from the pre-stored component layout model.

[0034] Specifically, during the initial driving phase of the sweeper, the edge computing host has continuously collected multiple frames of image data, extracted edge features, angle features, spacing features, distribution features, intersection features, and array features of the components, and used the DBSCAN clustering algorithm with spatial weight factors to automatically cluster them into component layout feature clusters, thereby establishing an adaptive component layout model that does not require preset fixed parameters.

[0035] The component layout adaptive model stores the upper edge reference path of each component row. The reference path is generated by fitting the upper edge straight line equation and represents the ideal extension direction of the upper edge of the component row. When the sweeper enters the current component row, the edge computing host retrieves the corresponding upper edge reference path from this model as a reference for subsequent driving direction comparison and correction.

[0036] S4. Based on the comparison result between the upper edge feature and the upper edge reference path, the driving direction of the sweeper is corrected to maintain the sweeper driving parallel to the upper edge of the current component row.

[0037] Specifically, as the sweeper travels along the current row of components, the forward-facing camera continuously captures images of the upper edges of the components. The edge computing host compares the currently extracted upper edge features with the upper edge reference path obtained from the model in real time, calculating the parallelism deviation δ between the two. When δ is detected to exceed the adaptively determined deviation tolerance range, an alignment correction operation is automatically triggered. Specifically, the sweeper's steering mechanism is adjusted through a PID control algorithm to gradually adjust the travel direction until δ returns to within the tolerance range, thereby maintaining the parallel travel of the sweeper with the upper edge of the components.

[0038] This application discloses another embodiment of a navigation and correction method for photovoltaic cleaning. Based on any of the above method embodiments, before executing step S3 to obtain the upper edge reference path of the current component row from the pre-stored component layout model, it further includes pre-constructing the component layout model, specifically including: S01, through a complete cleaning process of the photovoltaic power station, the cleaning vehicle travels along the upper edge of any component row, and simultaneously collects image data.

[0039] Specifically, during a complete cleaning operation within the photovoltaic power station, the sweeper starts driving along the upper edge of any selected starting module row. At this time, the front-view camera and the side-view camera are turned on simultaneously to continuously collect image data of the modules in front of and to the side of the driving path, which serves as the raw input for subsequent feature extraction and modeling.

[0040] S02, extract the upper edge features, angles, spacing, distribution features, and array features of the component rows from the image data, and cluster these features to form feature clusters.

[0041] Specifically, after preprocessing and segmenting the component regions of each frame of image data, the edge computing host extracts features such as the equation of the upper edge line of the component row, the component installation angle, the spacing between adjacent components, the component distribution density, and the extension direction of the row. These features are then automatically aggregated by the DBSCAN clustering algorithm, which introduces spatial weight factors, to form a set of feature clusters that characterize the layout attributes of the row of components.

[0042] S03, as the sweeper moves, it continuously collects image data and simultaneously updates the feature clusters.

[0043] Specifically, as the sweeper continues to move forward along the current component row, the forward-looking camera continuously collects new image data. The edge computing host extracts the above-mentioned features frame by frame and dynamically incorporates the newly extracted features into the existing feature clusters. This allows the statistical characteristics of the feature clusters to converge continuously as the amount of collected data increases, thereby improving the representativeness and stability of the feature representation.

[0044] S04. Repeat the above process until a complete cleaning is completed, and establish a component layout model that saves the feature clusters of each component row.

[0045] Specifically, the above collection, extraction, and clustering update process is executed repeatedly until the sweeper completes a full cleaning operation on all component rows. At this point, the edge computing host will collect and store all feature clusters generated during the traversal process according to the corresponding component row positions, forming a component layout model that includes the upper edge reference path of each component row, the spacing between rows, the arrangement order, and the overall layout relationship, for subsequent cleaning tasks at the same power station to directly call.

[0046] Based on the above embodiments, this application provides another embodiment of a navigation correction method for photovoltaic cleaning, wherein step S4, correcting the driving direction of the sweeper based on the comparison result of the upper edge feature and the upper edge reference path, specifically includes: If a break or obstruction is detected on the upper edge of the component row, the path of the broken upper edge is predicted based on the extension trend of the upper edge. The vehicle temporarily travels along the predicted upper edge path, while the side-view camera helps to capture the upper edges of adjacent components. Once the complete upper edge is detected again, the vehicle immediately corrects to the upper edge reference path.

[0047] Specifically, the predicted upper edge path refers to the predicted extension trajectory of the upper edge after the breakage area when the upper edge of the front component of the sweeper cannot be fully captured by the forward-looking camera due to reasons such as breakage (e.g., the proportion of continuous features is <80%) or obstruction by obstacles. The edge computing host infers the predicted extension trajectory of the upper edge after the breakage area based on the continuous multi-frame upper edge image data collected by the component before the breakage occurred, by fitting the upper edge straight line equation, i.e., the slope k and intercept b reflected by y=kx+b.

[0048] Based on the predicted trajectory, the sweeper temporarily continues to move forward along the predicted path to maintain the continuity of the sweeping operation and the basic stability of the direction of travel, so as to avoid missing any sweeping due to stopping or deviating from the course caused by interruption of visual information.

[0049] Optionally, after the sweeper continues forward and passes through the broken or obstructed area, the sweeper's camera is simultaneously activated and resumes auxiliary operation. At this time, if the forward-looking camera recaptures a complete and continuous image of the upper edge of the current component row, the edge computing host immediately compares the real-time extracted upper edge features with the upper edge reference path stored in the model to calculate the current parallelism deviation δ. Subsequently, the steering mechanism is adjusted through a PID control algorithm to gradually correct the sweeper's driving direction back to the reference path, ensuring that the sweeper resumes precise parallel driving with the upper edge of the component, thereby completing autonomous passage through the broken or obstructed area and path correction.

[0050] This application provides another embodiment of a navigation and correction method for photovoltaic cleaning, based on any of the embodiments described above, with reference to the appendix to the specification. Figure 2 As shown, it also includes: S5, when the sweeper reaches the end of the current component row, it performs a row switching judgment based on the row distribution information of the component layout model.

[0051] Specifically, when the sweeper travels to the end of the current component row and both the front and rear views confirm that there are no more continuous component areas to sweep in the current row, the edge computing host initiates the row switching judgment process based on the row distribution information in the established component layout model to determine the position and direction of the adjacent undisturbed component rows.

[0052] In other embodiments, the inter-row switching determination includes: S51, the side-view camera captures component images on both sides of the current component row, segments and extracts the upper edge features of adjacent uncleaned component rows, and determines their starting position, extension direction, parallelism with the upper edge of the current component row, and spacing between rows.

[0053] S52, the sweeper first adjusts its direction to ensure that the end of the current component row is aligned with the upper edge of the starting position of the adjacent row, then travels a distance in a direction perpendicular to the upper edge of the current component row, and finally adjusts the direction of the sweeper to be parallel to the upper edge of the adjacent row.

[0054] S53, after switching to the starting position of the adjacent row, acquire the upper edge image of the components in the adjacent row and calculate the upper edge parallelism deviation.

[0055] S54, if the switching is successful based on the upper edge parallelism deviation, the sweeper will start sweeping along the upper edge extension direction of the adjacent row.

[0056] Specifically, the row-to-row switching determination first uses a side-view camera to collect image data from the left and right sides of the current row of components. After image segmentation and feature extraction, the starting position and extension direction of the upper edge of the adjacent uncleaned row are identified, and the parallelism deviation between the upper edge of this upper edge and the upper edge of the current row, as well as the vertical distance between the two rows, are calculated. Subsequently, the sweeper executes the following switching path: First, the driving direction is slightly adjusted so that the upper edge of the end position of the current row is spatially aligned with the upper edge of the starting position of the adjacent row. Then, it moves horizontally along a direction perpendicular to the upper edge of the current row, with the horizontal distance equal to the previously measured row-to-row distance. Finally, the driving direction is adjusted again to keep the sweeper parallel to the upper edge of the adjacent row.

[0057] After switching to the starting position of the adjacent row, the sweeper uses a forward-facing camera to capture an image of the upper edge of the components in the starting area of ​​that row. The edge computing host extracts the equation of the upper edge line and calculates its parallelism deviation with the current driving direction. If the deviation is within the preset tolerance range, the switch is considered successful, and the sweeper continues to drive along the extension direction of the upper edge of that row and begins sweeping.

[0058] In another embodiment of this example, step S5 further includes: S55, switching validity verification. After switching to the starting position of the adjacent row, the upper edge image of the component in that row is acquired, and the upper edge parallelism deviation is calculated. and comprehensive feature similarity ,like If the switch is successful, begin cleaning along the upper edge of the row. Otherwise, reposition and adjust the switch path.

[0059] S56, Statistics of Unswept Component Rows. After each row is swept and switched, update the number of unswept component rows. If the number > 0, repeat the row switching logic. If the number = 0, perform a coverage integrity check.

[0060] S57, after all the array cleaning is completed, the edge computing host, based on the pre-stored component layout model, counts the number of component feature clusters in the cleaned area and the number of uncovered feature clusters. Combining this with the cleaning trajectory records on the upper edge of the components, the cleaning coverage rate is calculated using the following formula: .

[0061] in: To improve cleaning coverage, This represents the number of component feature clusters that have been cleaned. The total number of feature clusters identified by the model. The top edge alignment coverage factor ( Indicates perfect alignment. The alignment deviation decreases linearly as it increases, ensuring that misaligned areas are included in the uncovered data.

[0062] Optionally, the forward-facing camera can employ a binocular vision sensor to provide more accurate distance feedback for path alignment by calculating the actual distance between the upper edge of the stereo vision component and the sweeper.

[0063] In this embodiment, by accurately extracting the upper edge reference path and combining it with a real-time alignment correction mechanism, the sweeper is ensured to always travel parallel to the upper edge of the component, which fundamentally solves the problem of sweeping omissions caused by driving deviation in traditional methods and greatly improves the sweeping coverage.

[0064] This application provides another embodiment of a navigation and correction method for photovoltaic cleaning, based on any of the embodiments described above, with reference to the appendix to the specification. Figure 2 As shown, it also includes: S6, when the sweeper reaches a fork in the road, the component features of the branches included in the current fork are obtained from the component layout model. The component features of each branch are matched with the component features of the current component row, and the branch with the highest matching degree is selected as the candidate direction.

[0065] In some implementations, step S6 involves matching the component features of each branch with the component features of the current component row, specifically including the following sub-steps: S61, calculate the comprehensive feature similarity between the component angle features, component spacing features, and upper edge parallelism deviation of each branch and the current path feature cluster, and select the branch with the highest comprehensive feature similarity as the candidate direction.

[0066] The edge computing host first calculates the comprehensive cosine similarity between the component angle features, component spacing features, and upper edge parallelism deviation of each branch and the current path feature cluster. Here, component angle features refer to the angle between the component edge and the horizontal direction, component spacing features refer to the actual physical distance between the edges of adjacent components, and upper edge parallelism deviation reflects the degree of deviation of the branch's upper edge from the current driving direction. A higher comprehensive similarity indicates that the layout pattern of the branch is more consistent with the current component arrangement; therefore, the branch with the highest comprehensive feature similarity is selected as the preferred candidate direction.

[0067] In some implementations, the overall feature similarity is considered. The calculation formula is as follows: .

[0068] in: The value range is 0~1. , This is a weighting coefficient (which can be optimized based on component layout adaptability). The cosine similarity is between the angle and spacing features. The similarity of the upper edge parallelism deviation is calculated using the following formula.

[0069] .

[0070] in: The attenuation coefficient is... This refers to the parallelism deviation of the upper edge of the branch component. This represents the average edge parallelism deviation along the current path.

[0071] S62, if the similarity of the comprehensive features of each branch is small, then calculate the component distribution density of each branch and select the branch with the highest density as the candidate direction.

[0072] In some implementations, if multiple branches If the difference is ≤5%, calculate the component distribution density ρ using the following formula and select the branch with the highest density: .

[0073] in: This represents the number of components within the connected region of the branch component. This represents the actual area of ​​the connected region.

[0074] In this embodiment, by introducing a comprehensive similarity calculation of the parallelism deviation similarity of the upper edge and the features of angle and component spacing, the branch selection not only matches the component layout pattern, but also ensures the feasibility of upper edge alignment in subsequent cleaning, thereby improving the rationality of path decision-making.

[0075] S63, if the density difference of each branch is small, calculate the extension length and select the branch with the longest length as the candidate direction.

[0076] For example, if the density difference between branches is ≤5%, the extension length L is calculated using the following formula, and the branch with the longest length is selected: .

[0077] Where: L is the extension length. Extend the pixel length of the upper edge of the detected components in the image. The pixel-to-physical length conversion factor. This is the predicted extension length based on the feature cluster trend.

[0078] This application optimizes the direction judgment logic at intersections by introducing the angle, spacing features, and upper edge parallelism deviation of multiple branches of the component row. These features are compared with the comprehensive feature similarity of the current path feature cluster. The branch with the highest comprehensive feature similarity is selected as the candidate direction, which not only matches the component layout pattern but also ensures the feasibility of upper edge alignment for subsequent cleaning, thereby improving the rationality of path decision-making.

[0079] Based on the same technical concept, this application also discloses a navigation and correction device for photovoltaic cleaning, see attached specification. Figure 3 As shown, it includes: Memory 100 is used to store computer program 200.

[0080] The processor 300 is used to implement the navigation and correction method for photovoltaic cleaning in any of the foregoing embodiments when running the computer program 200.

[0081] The memory 100 can be any internal storage unit and / or external storage device capable of storing data and programs. For example, the memory 100 can be a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD card), or a flash memory card.

[0082] Depending on the requirements, the processor 300 may be a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a general-purpose processor, or other logic devices.

[0083] It should be noted that the above embodiments can be freely combined as needed. The above are merely preferred embodiments of this application. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A navigation and correction method for photovoltaic cleaning, characterized in that, include: During the cleaning process, the sweeper continuously acquires image data along its travel path; Extract the upper edge features of the current component row where the sweeper is located based on the image data; Obtain the top edge reference path of the current component row from the pre-stored component layout model; Based on the comparison results between the upper edge features and the upper edge reference path, the driving direction of the sweeper is corrected to maintain the sweeper driving parallel to the upper edge of the current component row.

2. The navigation and correction method for photovoltaic cleaning according to claim 1, characterized in that, The continuous acquisition of image data along the driving path specifically includes: The image data is acquired by an image acquisition device installed on the sweeper body. The image acquisition device includes three cameras: a front-view camera, a side-view camera, and a rear-view camera. The front-view camera is used to capture the area to be cleaned directly in front, the side-view camera is used to capture the areas on both sides, and the rear-view camera is used to capture the area that has been cleaned behind.

3. The navigation and correction method for photovoltaic cleaning according to claim 2, characterized in that, The step of correcting the sweeper's driving direction based on the comparison result between the upper edge feature and the upper edge reference path includes: If a break or obstruction is detected on the upper edge of the component row, the path of the broken upper edge is predicted based on the extension trend of the upper edge. The vehicle temporarily travels along the predicted upper edge path, while the side-view camera helps to capture the upper edges of adjacent components. Once the complete upper edge is detected again, the vehicle immediately corrects to the upper edge reference path.

4. The navigation and correction method for photovoltaic cleaning according to claim 1, characterized in that, Also includes: When the sweeper reaches the end of the current component row, it performs a row switching judgment based on the row distribution information of the component layout model.

5. The navigation and correction method for photovoltaic cleaning according to claim 1, characterized in that, It also includes pre-building the component layout model, specifically including: By conducting a complete cleaning process of the photovoltaic power station, the cleaning vehicle travels along the upper edge of any row of components, simultaneously collecting image data. Extract the top edge features, angles, spacing, distribution features, and array features of the component rows from the image data, and cluster these features to form feature clusters; As the sweeper moves, it continuously collects image data and simultaneously updates feature clusters. Repeat the above process until a complete sweep is completed, and establish a component layout model that saves the feature clusters of each of the component rows.

6. The navigation and correction method for photovoltaic cleaning according to claim 4, characterized in that, The inter-row switching determination includes: Images of the components on both sides of the current component row are captured by a side-view camera. The upper edge features of adjacent uncleaned component rows are segmented and extracted to determine their starting position, extension direction, parallelism with the upper edge of the current component row, and spacing between rows. The sweeper first adjusts its direction to ensure that the end of the current component row is aligned with the upper edge of the starting position of the adjacent row, then travels a distance in a direction perpendicular to the upper edge of the current component row, and finally adjusts the direction of the sweeper to be parallel to the upper edge of the adjacent row. After switching to the starting position of the adjacent row, the upper edge image of the components in the adjacent row is acquired, and the upper edge parallelism deviation is calculated; If the switching is successful based on the upper edge parallelism deviation, the sweeper will begin sweeping along the upper edge extension direction of the adjacent row.

7. The navigation and correction method for photovoltaic cleaning according to claim 1, characterized in that, Also includes: When the sweeper reaches a fork in the road, the component features of the branches contained in the current fork in the road are obtained from the component layout model; The component features of each branch are matched with the component features of the current component row, and the branch with the highest matching degree is selected as the candidate direction.

8. The navigation and correction method for photovoltaic cleaning according to claim 7, characterized in that, The process of matching the component features of each branch with the component features of the current component row includes: Calculate the comprehensive feature similarity between the component angle features, component spacing features, and upper edge parallelism deviation of each branch and the current path feature cluster, and select the branch with the highest comprehensive feature similarity as the candidate direction; If the similarity of the comprehensive features of each branch is small, then the component distribution density of each branch is calculated, and the branch with the highest density is selected as the candidate direction. If the density difference between the branches is small, the extension length is calculated, and the branch with the longest extension length is selected as the candidate direction.

9. The navigation and correction method for photovoltaic cleaning according to claim 8, characterized in that, Also includes: The extension length is calculated using the following formula: ； Where L is the extension length. Extend the pixel length of the upper edge of the detected components in the image. The conversion factor is the physical length of the image pixel to the upper edge of the component. This is the predicted value of the trend extension length based on the feature cluster.

10. A navigation and correction device for photovoltaic cleaning, characterized in that, include: Memory, used to store computer programs; A processor for implementing the navigation and correction method for photovoltaic cleaning as described in any one of claims 1 to 9 when running the computer program.

Images