Photovoltaic panel dust deposition detection and path planning method and device

Abstract

The invention provides a photovoltaic panel dust deposition detection and path planning method and device, and relates to the technical field of computer vision, and the method comprises the steps: obtaining a surface image of a target photovoltaic panel, inputting the surface image of the target photovoltaic panel into a pre-trained image recognition model, and outputting a target dust deposition detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust deposition data set; based on the target dust deposition detection result, a target cleaning path is generated through a pre-designed path planning algorithm, and a robot is controlled to execute cleaning operation according to the target cleaning path; wherein the path planning algorithm is obtained based on fusion of an ant colony algorithm and a genetic algorithm. According to the method provided by the invention, a mixed path planning method combining the ant colony algorithm and the genetic algorithm is adopted, the cleaning path is optimized, and the path length is reduced; stains and obstacles on the surface of the photovoltaic panel are recognized in real time, and a closed-loop feedback system is formed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, and in particular to a photovoltaic panel dust detection and path planning method and device. BACKGROUND

[0002] In the field of monitoring dust deposition on the surface of photovoltaic cells, traditional detection methods mainly rely on manual detection or simple image processing techniques, which are inefficient and difficult to meet the detection needs in complex environments, such as poor performance under conditions of reflection and shadow. Photovoltaic panel cleaning robots are important tools for improving the power generation efficiency of photovoltaic panels. Existing cleaning equipment mainly includes cleaning drones and cleaning trolleys. These devices clean through simple mechanical cleaning or preset paths, but lack intelligent path planning and image recognition technology.

[0003] How to achieve accurate photovoltaic panel dust detection and effective path planning is a technical problem that needs to be solved. SUMMARY

[0004] The present application provides a photovoltaic panel dust detection and path planning method and device to solve the defects in the prior art.

[0005] The present application provides a photovoltaic panel dust detection and path planning method, comprising the following steps: Obtain the target photovoltaic panel surface image, input the target photovoltaic panel surface image into the pre-trained image recognition model, and output the target dust detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on historical photovoltaic panel dust data sets; Based on the target dust detection result, a target cleaning path is generated through a pre-designed path planning algorithm, and a robot is controlled to perform cleaning operations according to the target cleaning path; wherein the path planning algorithm is obtained by fusing an ant colony algorithm and a genetic algorithm.

[0006] According to the photovoltaic panel dust detection and path planning method provided by the present application, the training process of the image recognition model comprises: Obtain a historical photovoltaic panel dust data set under multiple environmental conditions, convert the historical photovoltaic panel dust data set into a YOLO format and divide it into a training set and a validation set; The weights of the YOLOv10 detection model are used as initial weights, and the YOLOv10 detection model is trained based on the training set through cross-entropy and GIOU loss functions to obtain an initial image recognition model; The initial image recognition model is verified based on the validation set, the model parameters of the initial image recognition model are dynamically adjusted based on the verification result, and the image recognition model is obtained.

[0007] According to the photovoltaic panel dust detection and path planning method provided by the application, the target cleaning path is generated through a pre-designed path planning algorithm based on the target dust detection result, and the method comprises the following steps: An initial cleaning path is generated through the ant colony algorithm based on the target dust detection result, wherein the ant colony algorithm is used to determine a local optimal solution through a positive feedback mechanism of pheromone; The initial cleaning path is used as the initial population of the genetic algorithm, and the global optimization is performed through selection, crossover and mutation operations to obtain an optimized cleaning path; The pheromone matrix in the ant colony algorithm is updated based on the real-time detected obstacle coordinates, and the optimized cleaning path is optimized based on the updated pheromone matrix to obtain the target cleaning path.

[0008] According to the photovoltaic panel dust detection and path planning method provided by the application, the target cleaning path is generated through a pre-designed path planning algorithm based on the target dust detection result, and the method comprises the following steps: Real-time photovoltaic panel surface images of the target photovoltaic panel are acquired, and the real-time photovoltaic panel surface images are input into the image recognition model to output real-time dust detection results; An adjustment strategy is generated through the path planning algorithm based on the real-time dust detection results, and the target cleaning path is dynamically adjusted based on the adjustment strategy to obtain an adjusted cleaning path; The robot is controlled to perform a cleaning operation according to the adjusted cleaning path.

[0009] According to the photovoltaic panel dust detection and path planning method provided by the application, the target cleaning path is generated through a pre-designed path planning algorithm based on the target dust detection result, and the method comprises the following steps: After each single photovoltaic panel cleaning is completed, the image recognition model is automatically triggered for secondary detection; In the case that the secondary detection result indicates that the dust coverage rate is greater than a preset threshold, a cleaning path is regenerated through the path planning algorithm.

[0010] According to the photovoltaic panel dust detection and path planning method provided by the application, the target dust detection result comprises a bounding box coordinate, a confidence and a category; After the target photovoltaic panel surface images are acquired and input into the pre-trained image recognition model to output the target dust detection result, the method further comprises the following steps: The target dust detection result is displayed to a graphical user interface, and feedback information of the target dust detection result is received from a user; The feedback information comprises an IOU threshold and a confidence threshold, and the IOU threshold and the confidence threshold are used for optimizing detection accuracy.

[0011] The application further provides a photovoltaic panel dust detection and path planning device, comprising the following modules: A dust detection module is configured to acquire a target photovoltaic panel surface image, input the target photovoltaic panel surface image into a pre-trained image recognition model, and output a target dust detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust dataset. A path planning module is configured to generate a target cleaning path based on the target dust detection result by using a pre-designed path planning algorithm, and control a robot to perform a cleaning operation according to the target cleaning path; wherein the path planning algorithm is obtained by fusing an ant colony algorithm and a genetic algorithm.

[0012] According to the photovoltaic panel dust detection and path planning device provided by the application, the device further comprises a training module, which is specifically configured to: acquire a historical photovoltaic panel dust dataset under multiple environmental conditions, convert the historical photovoltaic panel dust dataset into a YOLO format, and divide the historical photovoltaic panel dust dataset into a training set and a verification set; use the weight of the YOLOv10 detection model as an initial weight, train the YOLOv10 detection model based on the training set by using a cross-entropy and GIOU loss function, and obtain an initial image recognition model; verify the initial image recognition model based on the verification set, dynamically adjust the model parameters of the initial image recognition model based on a verification result, and obtain the image recognition model.

[0013] The application further provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the photovoltaic panel dust detection and path planning method according to any one of the above when executing the computer program.

[0014] The application further provides a non-transitory computer-readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the photovoltaic panel dust detection and path planning method according to any one of the above.

[0015] The application further provides a computer program product, which comprises a computer program, and the computer program is executable on a processor to implement the photovoltaic panel dust detection and path planning method according to any one of the above.

[0016] The application provides a photovoltaic panel dust detection and path planning method and device, which comprises the following steps: obtaining a target photovoltaic panel surface image, inputting the target photovoltaic panel surface image into a pre-trained image recognition model, and outputting a target dust detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on historical photovoltaic panel dust data sets; based on the target dust detection result, a target cleaning path is generated through a pre-designed path planning algorithm, and a robot is controlled to perform a cleaning operation according to the target cleaning path; wherein the path planning algorithm is obtained based on the fusion of an ant colony algorithm and a genetic algorithm. Therefore, the application provides a hybrid path planning method combining an ant colony algorithm and a genetic algorithm, optimizes the cleaning path, and reduces the path length; the YOLO image recognition technology is introduced to identify the stains and obstacles on the surface of the photovoltaic panel in real time, and a closed-loop feedback system is formed. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0018] Figure 1 is a flowchart of the photovoltaic panel dust detection and path planning method provided by the application.

[0019] Figure 2 is a complete flowchart of the photovoltaic panel dust detection and path planning method provided by the application.

[0020] Figure 3 is an architecture diagram of the photovoltaic panel dust detection system based on deep learning provided by the application.

[0021] Figure 4 is a detection flowchart of the photovoltaic panel dust detection system based on deep learning provided by the application.

[0022] Figure 5 is a visual interface diagram of the photovoltaic panel dust detection system based on deep learning provided by the application.

[0023] Figure 6 is a hardware connection diagram of the Jetson Nano and the STM32F407ZG16 microcontroller provided by the application.

[0024] Figure 7 is a training set detection effect diagram provided by the application.

[0025] Figure 8is a verification set label display diagram provided by the present application.

[0026] Figure 9 is a verification set enhancement diagram provided by the present application.

[0027] Figure 10 is one of GUI interface detection result analysis diagrams provided by the present application.

[0028] Figure 11 is another GUI interface detection result analysis diagram provided by the present application.

[0029] Figure 12 is a structural diagram of a photovoltaic panel dust deposition detection and path planning device provided by the present application.

[0030] Figure 13 is a structural diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0031] To make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0032] It should be noted that in the field of photovoltaic cell surface dust deposition monitoring, traditional detection methods mainly rely on manual detection or simple image processing technology, which is inefficient and difficult to meet the detection needs in complex environments, such as poor performance under conditions of reflection and shadow. Photovoltaic panel cleaning robots are important tools for improving the efficiency of photovoltaic panel power generation. Existing cleaning equipment mainly includes cleaning drones and cleaning trolleys, etc. These devices clean through simple mechanical cleaning or preset paths, but lack intelligent path planning and image recognition technology. Related background technologies include: Numerical control machine tool control technology: controls mechanical movement through preset programs, but lacks dynamic environmental adaptability. Ant colony algorithm-based traveling salesman problem (TSP): used for path optimization, but single algorithm has limitations in complex environments. YOLO image recognition technology: used to identify stains and obstacles on the surface of photovoltaic panels, but existing technology has not been combined with path planning algorithms to form a closed-loop feedback system.

[0033] Current graphical user interface (GUI) systems lack interactivity and dynamic visualization features such as heat maps and historical trend analysis. The YOLO series of algorithms have limitations in lightweight processing and multi-scale feature fusion, which affects their deployment efficiency on mobile devices. Additionally, YOLO algorithms optimized for photovoltaic scenarios have limitations, particularly in handling glare and shadow interference.

[0034] The existing technology mainly has the following problems: 1. Imaging quality is easily disturbed by environmental factors.

[0035] The system relies on high-definition cameras or infrared cameras to capture images, but fluctuations in ambient light (such as strong sunlight or shadows) can negatively impact imaging quality, leading to dust identification errors. The reflective properties of the photovoltaic panel surface can obscure the dust area or cause overexposure or underexposure in the image, affecting the accuracy of the grayscale value calculation.

[0036] 2. Algorithm limitations in detecting small dust accumulations.

[0037] Current image segmentation and feature extraction algorithms (such as the histogram bimodal method and adaptive threshold method) may not effectively identify low-density or sparsely distributed dust accumulations. The algorithm relies on high-contrast images, and if the dust particles are small or evenly distributed, they may be misjudged as normal surface texture.

[0038] 3. Bottlenecks in data processing and encryption efficiency.

[0039] Asymmetric encryption technology ensures data security through binary conversion and dictionary mapping, but may increase processing complexity and affect the response speed of real-time monitoring systems. In large photovoltaic power stations, frequent image data collection may introduce delays due to encryption processes, affecting the immediate feedback of high real-time applications such as dynamic dust monitoring.

[0040] 4. Insufficient adaptability of models relying on historical data.

[0041] When faced with sudden environmental changes (such as sandstorms and seasonal pollution), the model may not adapt effectively, resulting in significant deviations between the corrected output power and the actual value. If historical data does not adequately cover extreme weather conditions, the model's corrected dust accumulation assessment may lack accuracy, adversely affecting operational decisions.

[0042] 5. Limitations of the de-griding algorithm.

[0043] In the surface image of the photovoltaic panel, the grid lines appear as high-frequency signals, and the application of a traditional low-pass filter can cause the image to be excessively blurred. Although a local low-pass filter aims to preserve image details, it can still interfere with the identification of dusty areas when dealing with complex grid line structures. Due to the overlap of grid lines and dust in the frequency spectrum, the algorithm may misjudge, so further optimization of filter parameters is needed, or deep learning techniques can be introduced to assist in identification.

[0044] 6、The cleaning device brush head path planning flexibility is insufficient.

[0045] Track-type equipment: Although it has high stability, it lacks intelligent path planning and image recognition technology, cannot dynamically adjust the cleaning path, and the cleaning effect is limited.

[0046] Wheeled and unmanned aerial vehicle equipment: Although mobile flexibility, existing path planning algorithms (such as ant colony algorithm, A* algorithm) are prone to local optimization in complex environments (such as dynamic obstacles, irregular photovoltaic panel layout), and cannot efficiently complete the cleaning task.

[0047] Lack of closed-loop feedback: existing equipment cannot monitor cleaning effect in real time and adjust path, resulting in low cleaning efficiency.

[0048] Therefore, the present application provides a photovoltaic panel dust detection and path planning method to solve at least one of the above problems.

[0049] The following Figures 1-13 A photovoltaic panel dust detection and path planning method and device are described.

[0050] Figure 1 is a flowchart of the photovoltaic panel dust detection and path planning method provided by the present application, as Figure 1 shown, the method comprises the following steps: Step 100, obtaining a target photovoltaic panel surface image, inputting the target photovoltaic panel surface image into a pre-trained image recognition model, and outputting a target dust detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on historical photovoltaic panel dust data set.

[0051] Figure 2 is a complete flowchart of the photovoltaic panel dust detection and path planning method provided by the present application, and the following Figure 2 , the photovoltaic panel dust detection and path planning method provided by the present application is described in detail.

[0052] It should be noted that in view of the key problems of insufficient algorithm accuracy, poor user interaction experience and low system integration of the current photovoltaic panel dust detection technology, the embodiment of the present application proposes a photovoltaic panel dust detection system based on deep learning and a graphical user interface (GUI) thereof.

[0053] Specifically, in the design stage of the present application, the hardware environment is carefully selected and configured to ensure the stable operation of the photovoltaic panel detection system. For the prototype system, at least a quad-core processor with a frequency not less than 2.5 GHz is required, and a more powerful processor such as Intel Core i7 or AMD Ryzen 7 is recommended to achieve faster processing speed. Since deep learning models are usually trained and inferred on GPUs, the present application recommends using NVIDIA series GPUs such as GeForce RTX 30 series or higher versions to fully utilize their powerful parallel processing performance. In the scenario where model training is required, at least 4GB of video memory is required. The system needs to have at least 8GB of RAM to ensure smooth operation. For tasks involving large models or complex data processing, it is recommended to use 32GB or more of RAM. Based on the above considerations, all data in this embodiment is operated on the hardware platform shown in Table 1.

[0054] Table 1

[0055] Specifically, the present application develops and deploys a photovoltaic panel detection system based on YOLOv10 and builds the corresponding software environment. Windows 10 is selected as the main operating system due to its wide hardware compatibility, stability and compatibility with most deep learning frameworks and libraries. A closed virtual environment is created using the Anaconda tool, designed specifically for the operation of this project. Python 3.8 is selected as the main programming language for the project, and its rich library resources and active development community make it the preferred language in the field of deep learning and machine learning. This embodiment uses PyTorch 2.5.1, a popular open-source machine learning library that is widely used in deep learning research and production environments, especially suitable for the construction of research prototypes and rapid iterative development. YOLOv10 is used to achieve this version, which has been optimized for efficient photovoltaic panel dust detection. In addition, this embodiment lists the third-party libraries used during model training and system development and their corresponding version numbers in detail, as shown in Table 2.

[0056] Table 2

[0057] The training process of the image recognition model provided by the embodiment of the present application is described below. The training process of the image recognition model comprises: In step 110, a historical photovoltaic panel soiling dataset under multiple environmental conditions is obtained, and the historical photovoltaic panel soiling dataset is converted into a YOLO format and divided into a training set and a validation set.

[0058] Specifically, the historical photovoltaic panel soiling dataset used in the embodiment is obtained from a network resource, and the labeling work is completed using a LabelImg tool. The tool is essential for constructing a high-quality training dataset through a simple operation process, such as creating a labeling file, selecting a category, and drawing a bounding box. Labeling the targets in the image and assigning a category label using the LabelImg tool helps to prepare the data required for machine learning model training, and improves the performance and accuracy of the model. The dataset is initially in XML format, and is converted into YOLO format through a Python script, and is randomly divided into a training set and a validation set to prevent model overfitting.

[0059] The dataset constructed in the embodiment is composed of 1463 images, of which 1058 images are used for model training and 405 images are used for verification. The dataset covers a variety of environments, thereby ensuring the diversity and complexity of the data. The dataset mainly focuses on the "Dirt" category and contains more than 6822 soiling target instances, providing rich training samples for developing an accurate photovoltaic panel soiling detection model. In addition, the embodiment records the relevant information of the dataset in detail, and the specific information is shown in Table 3.

[0060] Table 3

[0061] In step 120, the weights of the YOLOv10 detection model are used as initial weights, and the YOLOv10 detection model is trained based on the training set through cross-entropy and GIOU loss functions to obtain an initial image recognition model.

[0062] In step 130, the initial image recognition model is verified based on the validation set, and the model parameters of the initial image recognition model are dynamically adjusted based on the verification result to obtain the image recognition model.

[0063] Specifically, the model training process involves optimizing algorithm parameters through a dataset to achieve the best fit for the objective function. During this process, selecting the appropriate model architecture, loss function, optimization algorithm, and fine-tuning of hyperparameters are key steps to improve model performance. To avoid model overfitting, cross-validation techniques are widely used; regularization techniques are used to control model complexity; data augmentation strategies help improve the model's generalization ability. When analyzing performance indicators, appropriate evaluation criteria should be selected according to different task types: in classification tasks, usually focus on accuracy, precision, recall, F1 value, and AUC-ROC curve; regression tasks focus on MSE, MAE, R², etc. Clustering tasks may use indicators such as silhouette coefficient for evaluation.

[0064] This embodiment realizes the random division of the photovoltaic panel dataset by writing a Python script, simplifying the data analysis process. And set the training period to 100 rounds to ensure that the model fully learns the dataset. In training, combined with cross-entropy and GIOU loss function, using pre-trained model yolov10n.pt as initial weight, based on YOLOv10 architecture, to improve the convergence speed and optimization effect, and expect to achieve the best performance.

[0065] Further, the deep learning-based dust detection system for photovoltaic panels proposed in the embodiment of the present application mainly consists of four core modules: a detection module, a detection parameter setting module, a detection parameter selection module, and an operation module. The operation module allows users to select a detection source. Users can select static pictures, video files, or real-time video streams as detection objects according to specific needs. The detection module, as the core of the system, is responsible for loading a pre-trained deep learning model and performing inference analysis on preprocessed detection sources. After inference is completed, the system will draw the target inference frame on the detection source and display the results in the center area of the interface for user reference. The detection result module aims to handle multiple targets that may be included in the inference results. Users can quickly locate and select specific targets through this module to obtain detailed information such as target type, confidence, and location coordinates. In addition, this module is also responsible for recording the processing time of the model for the current detection source and the number of targets in the detection results. The detection result table module records the file path of each detection source and the confidence of each target in table form. Users can save the detection results to the Save_data folder under the project directory through the "Save" function in the operation module; while the "Exit" function is used to close the system. Figure 3 is the architecture diagram of the deep learning-based dust detection system for photovoltaic panels provided by the present application, and the specific operation program is introduced as shown in Figure 3 .

[0066] This embodiment uses the Qt Designer tool to build a user interface through an intuitive drag-and-drop interface design method, and uses the pyuic5 tool to convert the designed UiMain.ui file into Python code for use in PyQt5 applications. The interface design incorporates 11 different controls, mainly the Q Label and Q Line Edit controls of Qt Widgets, which account for 4 each. In the PyQt5 code implementation, control instances are created and managed by loading or converting .ui files, and the corresponding event handling logic is written. The design implements loading and preprocessing of images, videos, and camera inputs, provides object detection options, can display the confidence and coordinate information of the detection, and supports the saving function of the detection results. In addition, the YOLO model is loaded through the Open CV library for reasoning, the model output is parsed, the bounding box, confidence and category information are extracted, and the bounding box and label are drawn on the image to display the coordinate information.

[0067] The photovoltaic panel dust accumulation detection system constructed in this embodiment uses the PyQt5 framework to implement its interface logic. After the system starts, it first loads the user interface layout file Ui_Main Window and initializes the main window of the application. Subsequently, the system sets the signal and slot mechanism of the button and drop-down menu to ensure that user operations can trigger the program to execute the corresponding function. In addition, the application interface is beautified by introducing a CSS file. The system uses a pre-trained YOLOv10 model to perform target detection tasks. The detection results include detailed information such as the target's bounding box, category, and confidence level, and this information is displayed in the table and image preview area of ​​the interface.

[0068] The system user interface is designed with a "file import" function, which enables the selection of image detection and video detection through two icons. Figure 4 This is a schematic diagram of the detection process of the photovoltaic panel dust accumulation detection system provided by the present invention. The specific detection process is as follows Figure 4 As shown in the figure, after model inference is complete, the image results with detection boxes are displayed directly in the left area, and detailed detection results are output in the "Detection Results" area. Output information includes inference time, number of targets, target category, confidence level, and target coordinates. Users can use the "Target Selection" drop-down menu to select to view all or specific target objects. The detection results are clearly displayed in a table in the lower left corner of the interface. The "Save" button in the "Operation" area in the lower right corner of the interface allows users to save the detection results to local storage, and the "Exit" button is used to shut down the system. Figure 5 This is a schematic diagram of the visualization interface of the photovoltaic panel dust accumulation detection system based on deep learning provided by the present invention. Figure 5 Shown is the final visual operation interface of the system.

[0069] Step 200, based on the target dust detection result, generating a target cleaning path through a pre-designed path planning algorithm, and controlling the robot to perform cleaning operations according to the target cleaning path; wherein the path planning algorithm is obtained based on the fusion of ant colony algorithm and genetic algorithm.

[0070] Specifically, step 200 generates a target cleaning path based on the target dust detection result through a pre-designed path planning algorithm, including: Step 210, based on the target dust detection result, generating an initial cleaning path through the ant colony algorithm; wherein the ant colony algorithm is to determine a local optimal solution through a pheromone positive feedback mechanism.

[0071] Step 220, taking the initial cleaning path as the initial population of the genetic algorithm, and performing global optimization through selection, crossover and mutation operations to obtain an optimized cleaning path.

[0072] Step 230, updating the pheromone matrix in the ant colony algorithm based on the real-time detected obstacle coordinates, and optimizing the optimized cleaning path based on the updated pheromone matrix to obtain the target cleaning path.

[0073] Step 240, obtaining a real-time photovoltaic panel surface image of the target photovoltaic panel, inputting the real-time photovoltaic panel surface image into the image recognition model, and outputting a real-time dust detection result.

[0074] Step 250, based on the real-time dust detection result, generating an adjustment strategy through the path planning algorithm, and dynamically adjusting the target cleaning path based on the adjustment strategy to obtain an adjusted cleaning path.

[0075] Step 260, controlling the robot to perform cleaning operations according to the adjusted cleaning path.

[0076] Step 270, after completing cleaning of a single photovoltaic panel, automatically triggering the image recognition model for secondary detection.

[0077] Step 280, in the case where the secondary detection result indicates that the dust coverage rate is greater than a preset threshold, regenerating a cleaning path through the path planning algorithm.

[0078] Specifically, the path planning part mainly includes the following contents: 1. Hybrid algorithm design.

[0079] Ant colony algorithm: used for initial exploration of path, and a better path is found through a pheromone positive feedback mechanism.

[0080] Genetic algorithm: used for global optimization, and the path is further improved through selection, crossover and mutation operations.

[0081] Fusion strategy: first run the ant colony algorithm to generate the initial path, and then optimize the path through genetic algorithm, combining the advantages of the two algorithms.

[0082] 2、Image recognition module.

[0083] Use YOLO image recognition technology to monitor the stains and obstacles on the surface of photovoltaic panels in real time.

[0084] The recognition result is fed back to the path planning module to dynamically adjust the cleaning path.

[0085] 3、Hardware architecture.

[0086] Host computer: based on NVIDIA Jetson Nano, responsible for running hybrid algorithms and image recognition modules.

[0087] Lower computer: based on STM32F407ZGT6, responsible for receiving host computer instructions and controlling robot movement through GRBL firmware.

[0088] Communication protocol: UART serial communication is adopted to ensure the real-time and reliability of data transmission between upper and lower computers.

[0089] Motion mode: three-axis movable frame is adopted.

[0090] 4、Specific steps.

[0091] Initialization program.

[0092] Run the image recognition model to identify the coordinates of stains.

[0093] Run the ant colony algorithm to analyze the material coordinates and generate the initial cleaning path.

[0094] Use the path generated by the ant colony algorithm as the initial population of the genetic algorithm for further optimization.

[0095] Update the pheromone matrix and combine the global search results of the genetic algorithm to optimize the path.

[0096] Output the target cleaning path and control the robot to execute.

[0097] In summary, the embodiment of the invention takes YOLOv10 algorithm, graphical user interface (GUI) interaction innovation and system integration as the core, and proposes the following key technical innovations for the specific needs of photovoltaic panel dust detection. 1) YOLOv10 lightweight and precision balance. 2) GUI interaction design originality. 3) System-level integration and closed-loop management. 4) Differentiation comparison with existing technologies, please refer to Table 4.

[0098] Table 4

[0099] The path planning is as follows: The fusion strategy of the hybrid algorithm: Combining the local search ability of the ant colony algorithm and the global search ability of the genetic algorithm, the efficiency and accuracy of path planning are improved. The ant colony algorithm quickly finds the optimal path through the positive feedback mechanism of pheromone, and the genetic algorithm further optimizes the path through selection, crossover and mutation operations to avoid falling into local optimum.

[0100] Closed-loop feedback system: Real-time monitoring of stains and obstacles on the surface of photovoltaic panels through YOLO image recognition technology, and feeding back the recognition results to the path planning module. Dynamically adjust the cleaning path to ensure that the robot can adapt to complex environments and dynamic changes, and improve cleaning efficiency.

[0101] Collaborative design of hardware and algorithm: The host computer (NVIDIA Jetson Nano) is responsible for running the hybrid algorithm and image recognition module, providing powerful computing power. The lower computer (STM32F407ZGT6) is responsible for receiving instructions from the host computer and controlling the robot motion through the GRBL firmware, ensuring the accuracy and real-time performance of motion control. UART serial communication protocol is used to ensure stable and reliable data transmission between the upper and lower computers.

[0102] Adaptive design: The algorithm can adapt to photovoltaic panel matrices of different sizes and complex environments, with high robustness. By adjusting algorithm parameters such as pheromone evaporation rate in ant colony algorithm, heuristic information importance factor, and crossover probability and mutation probability in genetic algorithm, the algorithm's performance in different scenarios is improved.

[0103] Three-axis movable frame: Figure 6 The hardware connection diagram of Jetson Nano and STM32F407ZG16 microcontroller provided by the present application is shown in Figure 6 The microcontroller is mainly used to control the three-axis movable frame of the photovoltaic panel cleaning robot. The robot uses a three-axis movable frame that can flexibly adapt to photovoltaic panels of different sizes and layouts. Through precise motion control, the robot can maintain stability during cleaning and improve cleaning efficiency.

[0104] Integration of path planning and motion control: Convert the path planning results into G code, and parse and control the three-axis motion frame through the lower computer to realize seamless connection between path planning and motion control. This integrated approach improves the overall performance of the system, ensuring that the robot can efficiently and accurately complete the cleaning task.

[0105] The above is a step-by-step description of the photovoltaic panel dust detection and path planning method provided by the present application. From the description of the above steps, it can be seen that according to the photovoltaic panel dust detection and path planning method provided by the present application, the target photovoltaic panel surface image is obtained, the target photovoltaic panel surface image is input into the pre-trained image recognition model, and the target dust detection result is output. The image recognition model is obtained by training the YOLOv10 detection model based on historical photovoltaic panel dust data set; based on the target dust detection result, a target cleaning path is generated through a pre-designed path planning algorithm, and a robot is controlled to perform a cleaning operation according to the target cleaning path; wherein the path planning algorithm is obtained based on the fusion of ant colony algorithm and genetic algorithm. Therefore, the present application proposes a hybrid path planning method combining ant colony algorithm and genetic algorithm, which optimizes the cleaning path and reduces the path length; the YOLO image recognition technology is introduced to identify the stains and obstacles on the surface of the photovoltaic panel in real time, forming a closed-loop feedback system.

[0106] Based on the above embodiment, in this embodiment, the effect of the photovoltaic panel dust detection and path planning method provided by the present application is described.

[0107] 1. Confusion matrix. Based on the comparison between model prediction and actual category, it is determined that the model has high accuracy in predicting the Dirt category. The precision of the Dirt category is 0.974, the recall is 0.929, and the F1 score is 0.951.

[0108] 2. Normalized confusion matrix. The prediction accuracy of the model for the Dirt category (true positive rate, TPR) is 0.97, the prediction error rate of the model for the background category (false positive rate, FPR) is 1.00, the prediction error rate of the model for the Dirt category (false negative rate, FNR) is 0.03, and in addition, the prediction accuracy of the model for the background category (true negative rate, TNR) is also included. Through the normalized confusion matrix, the excellent performance of the model in the classification task can be clearly observed, especially in identifying the Dirt category, the precision, recall and F1 score all perform well, which fully proves the superiority of the overall performance of the model.

[0109] 3. Curve analysis.

[0110] 1) F1 score and confidence. When the confidence threshold is 0.341, the model exhibits optimal performance with an F1 score of 0.95, indicating that at this confidence threshold, the model can effectively balance precision and recall.

[0111] 2) Precision-Confidence. The model achieved optimal performance at a confidence threshold of 0.973, with a peak precision of 1.00, indicating that at this confidence threshold, the model can achieve complete accurate prediction of positive class samples.

[0112] 3) Precision-Recall. The model performed very well on the "Dirt" category and all categories, with an average precision (AP) of 0.977 at a threshold of 0.5.

[0113] 4) Recall-Confidence. At lower thresholds, the model can achieve higher recall rates; however, as the threshold increases, the recall rate decreases, indicating that the reduction of false positives may be accompanied by an increase in false negatives.

[0114] 4. Model result analysis. The model's performance improved continuously during training, with key indicators such as precision, recall, and mAP showing an upward trend.

[0115] 5. Model detection results. Figure 7 is a training set detection effect diagram provided by the present application, Figure 8 is a validation set label indication diagram provided by the present application, Figure 9 is a validation set enhancement diagram provided by the present application, see Figures 7-9 , which illustrates the detection effect of the image recognition model provided by the present application.

[0116] 6. GUI interface detection result analysis. Figure 10 is one of the GUI interface detection result analysis diagrams provided by the present application, Figure 11 is the second GUI interface detection result analysis diagram provided by the present application, which presents the interface and effect of the photovoltaic panel dust detection system. The system identifies the dust accumulation area through deep learning algorithm and displays the detailed detection results. Users can adjust parameters to optimize detection and save or stop the system. After model training, it can accurately identify the dust points in the image, even in a blurred background, and can detect small targets with high recall rate, ensuring that all targets are identified.

[0117] The present application realizes high precision, high efficiency and intelligence of photovoltaic panel dust detection through YOLOv10 algorithm, optimization of graphical user interface (GUI) interaction and system integration. Through the collaborative design of hardware and algorithm, the cleaning efficiency is improved, the energy consumption of the robot is reduced, and the service life is prolonged, filling the gap in the prior art and providing a comprehensive solution for the operation and management of photovoltaic power stations. The specific effects can be seen in Table 5.

[0118] Table 5

[0119] The photovoltaic panel dust detection and path planning device provided by the present application is described below, and the photovoltaic panel dust detection and path planning device described below can be referred to in correspondence with the photovoltaic panel dust detection and path planning method described above.

[0120] Figure 12 The photovoltaic panel dust detection and path planning device provided by the present application is described below, and the photovoltaic panel dust detection and path planning device described below can be referred to in correspondence with the photovoltaic panel dust detection and path planning method described above. Figure 12 As shown in FIG. 1, the photovoltaic panel dust detection and path planning device provided by the present application comprises: The dust detection module 1201 is configured to acquire a target photovoltaic panel surface image, input the target photovoltaic panel surface image into a pre-trained image recognition model, and output a target dust detection result. The image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust data set. The path planning module 1202 is configured to generate a target cleaning path based on the target dust detection result by using a pre-designed path planning algorithm, and control a robot to perform a cleaning operation according to the target cleaning path. The path planning algorithm is obtained by fusing an ant colony algorithm and a genetic algorithm.

[0121] The photovoltaic panel dust detection and path planning device provided by the present application is described below, and the photovoltaic panel dust detection and path planning device described below can be referred to in correspondence with the photovoltaic panel dust detection and path planning method described above.

[0122] Based on the above embodiment, in the present embodiment, the device further comprises a training module, which is specifically configured to: acquire a historical photovoltaic panel dust data set under multiple environmental conditions, convert the historical photovoltaic panel dust data set into a YOLO format, and divide the historical photovoltaic panel dust data set into a training set and a validation set; use the weights of the YOLOv10 detection model as initial weights, train the YOLOv10 detection model based on the training set by using a cross-entropy and GIOU loss function, and obtain an initial image recognition model. verify the initial image recognition model based on the verification set, dynamically adjust model parameters of the initial image recognition model based on a verification result, and obtain the image recognition model.

[0123] Based on the above embodiment, in the present embodiment, the path planning module 1202 is specifically configured to: generate an initial cleaning path based on the target dust accumulation detection result through the ant colony algorithm; wherein the ant colony algorithm is to determine a local optimal solution through a pheromone positive feedback mechanism; take the initial cleaning path as an initial population of the genetic algorithm, and perform global optimization through selection, crossover and mutation operations to obtain an optimized cleaning path; update a pheromone matrix in the ant colony algorithm based on real-time detected obstacle coordinates, and optimize the optimized cleaning path based on the updated pheromone matrix to obtain the target cleaning path.

[0124] Based on the above embodiment, in the present embodiment, the device further comprises a dynamic adjustment module, which is specifically configured to: generate a target cleaning path through a pre-designed path planning algorithm based on the target dust accumulation detection result, and control the robot to perform a cleaning operation according to the target cleaning path, then acquire a real-time photovoltaic panel surface image of the target photovoltaic panel, input the real-time photovoltaic panel surface image into the image recognition model, and output a real-time dust accumulation detection result; generate an adjustment strategy through the path planning algorithm based on the real-time dust accumulation detection result, and dynamically adjust the target cleaning path based on the adjustment strategy to obtain an adjusted cleaning path; control the robot to perform a cleaning operation according to the adjusted cleaning path.

[0125] Based on the above embodiment, in the present embodiment, the device further comprises a secondary detection module, which is specifically configured to: after the robot performs a cleaning operation according to the adjusted cleaning path, automatically trigger the image recognition model to perform secondary detection after each single photovoltaic panel cleaning is completed; in a case where the secondary detection result indicates that the dust coverage rate is greater than a preset threshold, regenerate a cleaning path through the path planning algorithm.

[0126] Based on the above embodiment, in the present embodiment, the target dust accumulation detection result comprises a bounding box coordinate, a confidence and a category; the device further comprises an interaction module, which is specifically configured to: The target photovoltaic panel surface image is obtained, the target photovoltaic panel surface image is input into a pre-trained image recognition model, and after outputting a target dust accumulation detection result, the target dust accumulation detection result is displayed on a graphical user interface, and user feedback on the target dust accumulation detection result is received; The feedback information includes an IOU threshold and a confidence threshold, and the IOU threshold and the confidence threshold are used to optimize detection accuracy.

[0127] Figure 13 An example of a physical structure diagram of an electronic device is shown below. Figure 13 As shown, the electronic device may be a robot or other electronic device, and may include: a processor 1310, a communications interface 1320, a memory 1330, and a communication bus 1340. The processor 1310, the communications interface 1320, and the memory 1330 communicate with each other via the communication bus 1340. The processor 1310 may call the logic instructions in the memory 1330 to execute the photovoltaic panel dust accumulation detection and path planning method, including: Obtain a target photovoltaic panel surface image, input the target photovoltaic panel surface image into a pre-trained image recognition model, and output a target dust accumulation detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust accumulation dataset; Based on the target dust accumulation detection result, a target cleaning path is generated through a pre-designed path planning algorithm, and the robot is controlled to perform cleaning operations according to the target cleaning path; wherein, the path planning algorithm is obtained based on the fusion of ant colony algorithm and genetic algorithm.

[0128] Furthermore, the logic instructions in the aforementioned memory 1330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0129] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program being stored on a non-transitory computer-readable storage medium, and the computer program being executable by a processor to enable a computer to perform the photovoltaic panel dust detection and path planning method provided by the above-mentioned methods, comprising: obtaining a target photovoltaic panel surface image, inputting the target photovoltaic panel surface image into a pre-trained image recognition model, and outputting a target dust detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust data set; based on the target dust detection result, generating a target cleaning path through a pre-designed path planning algorithm, and controlling a robot to perform a cleaning operation according to the target cleaning path; wherein the path planning algorithm is obtained by fusing an ant colony algorithm and a genetic algorithm.

[0130] In another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being executable by a processor to implement the photovoltaic panel dust detection and path planning method provided by the above-mentioned methods, comprising: obtaining a target photovoltaic panel surface image, inputting the target photovoltaic panel surface image into a pre-trained image recognition model, and outputting a target dust detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust data set; based on the target dust detection result, generating a target cleaning path through a pre-designed path planning algorithm, and controlling a robot to perform a cleaning operation according to the target cleaning path; wherein the path planning algorithm is obtained by fusing an ant colony algorithm and a genetic algorithm.

[0131] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place or distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0132] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0133] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A photovoltaic panel dust accumulation detection and path planning method, characterized in that: include: Obtain a target photovoltaic panel surface image, input the target photovoltaic panel surface image into a pre-trained image recognition model, and output a target dust accumulation detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust accumulation dataset; Based on the target dust accumulation detection result, a target cleaning path is generated through a pre-designed path planning algorithm, and the robot is controlled to perform cleaning operations according to the target cleaning path; wherein, the path planning algorithm is obtained based on the fusion of ant colony algorithm and genetic algorithm.

2. The photovoltaic panel dust accumulation detection and path planning method according to claim 1, characterized in that: The training process of the image recognition model includes: Obtain a historical photovoltaic panel dust accumulation dataset under various environmental conditions, convert the historical photovoltaic panel dust accumulation dataset into YOLO format and divide it into a training set and a validation set; Using the weight of the YOLOv10 detection model as the initial weight, training the YOLOv10 detection model based on the training set using the cross entropy and GIOU loss function to obtain an initial image recognition model; The initial image recognition model is verified based on the verification set, and the model parameters of the initial image recognition model are dynamically adjusted based on the verification result to obtain the image recognition model.

3. The photovoltaic panel dust accumulation detection and path planning method according to claim 1, characterized in that: The target cleaning path is generated by a pre-designed path planning algorithm based on the target dust accumulation detection result, including: Based on the target dust accumulation detection result, an initial cleaning path is generated by the ant colony algorithm; wherein the ant colony algorithm is: determining a local optimal solution through a pheromone positive feedback mechanism; The initial cleaning path is used as the initial population of the genetic algorithm, and global optimization is performed through selection, crossover and mutation operations to obtain an optimized cleaning path; The pheromone matrix in the ant colony algorithm is updated based on the obstacle coordinates detected in real time, and the optimized cleaning path is optimized based on the updated pheromone matrix to obtain the target cleaning path.

4. The photovoltaic panel dust accumulation detection and path planning method according to claim 1, characterized in that: After generating a target cleaning path based on the target dust accumulation detection result using a pre-designed path planning algorithm and controlling the robot to perform a cleaning operation according to the target cleaning path, the method further includes: Acquire a real-time photovoltaic panel surface image of a target photovoltaic panel, input the real-time photovoltaic panel surface image into the image recognition model, and output a real-time dust accumulation detection result; Based on the real-time dust accumulation detection result, an adjustment strategy is generated by the path planning algorithm, and the target cleaning path is dynamically adjusted based on the adjustment strategy to obtain an adjusted cleaning path; The robot is controlled to perform a cleaning operation according to the adjusted cleaning path.

5. The photovoltaic panel dust accumulation detection and path planning method according to claim 4, characterized in that: After controlling the robot to perform the cleaning operation according to the adjusted cleaning path, the method further includes: After each photovoltaic panel is cleaned, the image recognition model is automatically triggered to perform a secondary inspection; When the secondary detection result indicates that the dust accumulation coverage is greater than a preset threshold, the cleaning path is regenerated using the path planning algorithm.

6. The photovoltaic panel dust accumulation detection and path planning method according to claim 1, characterized in that: The target dust accumulation detection result includes: bounding box coordinates, confidence level and category; After acquiring the target photovoltaic panel surface image, inputting the target photovoltaic panel surface image into a pre-trained image recognition model, and outputting the target dust accumulation detection result, the method further includes: Displaying the target dust accumulation detection result on a graphical user interface, and receiving user feedback on the target dust accumulation detection result; The feedback information includes an IOU threshold and a confidence threshold, and the IOU threshold and the confidence threshold are used to optimize detection accuracy.

7. A photovoltaic panel dust accumulation detection and path planning device, characterized in that: include: A dust accumulation detection module is configured to obtain an image of the target photovoltaic panel surface, input the image into a pre-trained image recognition model, and output a target dust accumulation detection result; wherein the image recognition model is obtained by training a YOLOv10 detection model based on a historical photovoltaic panel dust accumulation dataset; A path planning module is used to generate a target cleaning path based on the target dust accumulation detection result through a pre-designed path planning algorithm, and control the robot to perform cleaning operations according to the target cleaning path; wherein, the path planning algorithm is obtained by integrating the ant colony algorithm and the genetic algorithm.

8. The photovoltaic panel dust accumulation detection and path planning device according to claim 7, characterized in that: The device also includes a training module, specifically configured to: Obtain a historical photovoltaic panel dust accumulation dataset under various environmental conditions, convert the historical photovoltaic panel dust accumulation dataset into YOLO format and divide it into a training set and a validation set; Using the weight of the YOLOv10 detection model as the initial weight, training the YOLOv10 detection model based on the training set using the cross entropy and GIOU loss function to obtain an initial image recognition model; The initial image recognition model is verified based on the verification set, and the model parameters of the initial image recognition model are dynamically adjusted based on the verification result to obtain the image recognition model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the photovoltaic panel dust accumulation detection and path planning method according to any one of claims 1 to 6 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the photovoltaic panel dust accumulation detection and path planning method according to any one of claims 1 to 6 is implemented.

Images