A method for stain identification and precise cleaning of a photovoltaic panel intelligent cleaning trolley

Abstract

The application discloses a kind of photovoltaic panel intelligent cleaning trolley stain identification and accurate cleaning method, belong to photovoltaic operation and maintenance technical field, including cleaning trolley initialization and moves to photovoltaic panel starting detection position, completes attitude calibration and detection system preheating, the present application is by fusing multispectral image acquisition, deep learning identification model and self-adapting cleaning control strategy, constructs a set of set stain identification, classification determination, path planning and accurate cleaning in one integrated intelligent cleaning method, realizes the high-precision identification of photovoltaic panel surface stain and differentiates accurate processing, significantly improves cleaning effect and energy utilization efficiency;Stain is divided into mild, moderate, severe three grades, which provides a scientific basis for subsequent differentiated cleaning;Overall identification process is fast, stable, supports real-time processing, which greatly improves the intelligent decision-making ability of cleaning system.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic operation and maintenance technology, specifically referring to a method for identifying and precisely cleaning stains on photovoltaic panels using an intelligent cleaning vehicle. Background Technology

[0002] With the transformation of the global energy structure, solar photovoltaic power generation has been widely used due to its clean and renewable characteristics. As the core component of the power generation system, the surface cleanliness of photovoltaic panels directly affects the light energy absorption efficiency and power generation.

[0003] However, existing methods for stain identification and precise cleaning of photovoltaic panels using intelligent cleaning carts still have certain shortcomings. Existing technologies use a single visible light image recognition and a fixed cleaning mode, which makes it difficult to accurately distinguish the types and boundaries of stains under complex lighting conditions. They also lack differentiated cleaning capabilities for stains with different adhesion, which can easily lead to incomplete cleaning or over-cleaning that damages the coating on the surface of the photovoltaic panels. Furthermore, they do not fully consider the influence of environmental parameters on the characteristics of stains, resulting in unstable cleaning effects and low resource utilization. They cannot meet the actual needs of efficient, precise, and non-destructive cleaning. Therefore, a method for stain identification and precise cleaning of photovoltaic panels using intelligent cleaning carts is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying and precisely cleaning stains on a photovoltaic panel intelligent cleaning vehicle, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for stain identification and precise cleaning of photovoltaic panels using an intelligent cleaning vehicle, comprising the following steps:

[0006] S1. The cleaning vehicle initializes and moves to the photovoltaic panel initial detection position, completing attitude calibration and detection system preheating;

[0007] S2. Acquire a full-area image of the photovoltaic panel surface through a multispectral acquisition module, and simultaneously acquire ambient light intensity and panel temperature parameters.

[0008] S3. Preprocess the acquired image, segment out the suspected stain area and extract the regional feature parameters;

[0009] S4. Construct a stain recognition model based on feature parameters, classify and identify suspected stain areas, and determine the stain type and level;

[0010] S5. Based on the type and grade of stains and the surface shape of the photovoltaic panel, plan a precise cleaning path and cleaning parameters;

[0011] S6. According to the planned path and parameters, control the cleaning execution mechanism to perform differentiated and precise cleaning on the stained areas;

[0012] S7. After cleaning is completed, a second inspection is conducted on the cleaned area to determine whether the cleaning effect meets the standard. If it does not meet the standard, the cleaning parameters are recalculated.

[0013] Preferably, in step S1, the initialization of the cleaning vehicle includes battery power detection, cleaning actuator reset, and positioning module calibration. The positioning module uses a combination of BeiDou positioning and visual positioning. Visual positioning completes the relative position calibration between the vehicle and the photovoltaic panel by collecting feature points on the edge of the photovoltaic panel. Attitude calibration uses an inclination sensor installed at the bottom of the vehicle to collect the tilt angle of the vehicle. Combined with the preset installation angle of the photovoltaic panel, the attitude of the vehicle's walking mechanism is adjusted so that the vehicle fits the surface of the photovoltaic panel. The preheating of the detection system includes the power-on preheating of the multispectral acquisition module, image processing module, and control module.

[0014] Preferably, in step S2, the multispectral acquisition module includes a visible light camera, a near-infrared camera, and an ultraviolet camera, which simultaneously acquire visible light images, near-infrared images, and ultraviolet images of the photovoltaic panel surface. During the acquisition process, the trolley moves at a constant speed along the length of the photovoltaic panel, maintaining a constant distance between the multispectral acquisition module and the photovoltaic panel surface. Ambient light intensity is acquired in real time by a light intensity sensor installed on the top of the trolley, and panel temperature is acquired by a contact temperature sensor that is in flexible contact with the photovoltaic panel surface. The acquired image data, ambient light intensity data, and panel temperature data are simultaneously transmitted to the image processing module for data caching and preliminary processing, removing obvious abnormal data.

[0015] Preferably, in step S2, when the multispectral acquisition module acquires images, it adopts an interleaved acquisition method, ensuring that the acquisition frequencies of the visible light camera, near-infrared camera, and ultraviolet camera are consistent, and that there is a preset time difference in acquisition time; the resolution of the acquired images is adaptively adjusted according to the size of the photovoltaic panel; ambient light intensity acquisition is performed synchronously with image acquisition, and when the ambient light intensity is lower than a preset threshold, the supplementary lighting device of the multispectral acquisition module is activated, and the brightness of the supplementary lighting device is adaptively adjusted according to the ambient light intensity.

[0016] Preferably, in step S3, image preprocessing includes three sub-steps: image denoising, image enhancement, and image segmentation. Image denoising uses adaptive median filtering to remove environmental and sensor noise from the acquired image. Image enhancement adjusts the contrast and brightness of the image to highlight the grayscale difference between the stained area and the photovoltaic panel surface. Brightness adjustment is based on the acquired ambient light intensity data for adaptive adjustment. Image segmentation combines threshold segmentation and edge detection. Threshold segmentation initially separates suspected stained areas whose grayscale values ​​differ from those of the normal photovoltaic panel surface. Then, edge detection algorithms extract the contours of the suspected stained areas to obtain their boundary coordinates. The extracted regional feature parameters include the area of ​​the stained area, grayscale mean, grayscale variance, edge roughness, and near-infrared reflectivity. All feature parameters are integrated into a feature vector.

[0017] Preferably, in step S4, the stain recognition model is built based on deep learning. The input of the model is the extracted feature vector, and the output is the stain type and stain level. The stain level is divided into three levels: light, moderate, and heavy, based on the stain area and its impact on the light transmittance of the photovoltaic panel. In the stain recognition process, a stain feature correction coefficient K is introduced, which is implemented as follows:

[0018] ,

[0019] In the formula, K is the stain characteristic correction coefficient. This is the weighting coefficient for near-infrared reflectance. For the extracted near-infrared reflectance, This is the plate temperature weighting coefficient. The temperature of the plate is collected; the feature vector is corrected by a correction coefficient K to improve the accuracy of stain identification.

[0020] Preferably, in step S5, the cleaning path planning adopts a grid map method, dividing the photovoltaic panel surface into several uniform grids. Based on the identified boundary coordinates of the stained areas, the grids requiring cleaning are marked. Cleaning paths for stained areas are planned first, followed by auxiliary cleaning paths for non-stained areas. Cleaning parameters include cleaning pressure, cleaning speed, cleaning medium dosage, and cleaning frequency. These parameters are determined based on the stain type and stain level, and adjusted using a correction coefficient K. Specifically, the adjustment is implemented as follows:

[0021] ,

[0022] In the formula, P represents the final cleaning pressure. S represents the basic cleaning pressure, and S represents the stain level coefficient.

[0023] Preferably, in step S6, the cleaning actuator includes a liftable cleaning brush, a high-pressure spray head, and a negative pressure adsorption device. The height of the liftable cleaning brush is adjusted in real time according to the flatness of the photovoltaic panel surface. The height adjustment is achieved as follows:

[0024] ,

[0025] In the formula, H represents the final height of the cleaning brush. For the basic height of the cleaning brush, The value represents the flatness deviation of the photovoltaic panel surface; the spray angle and spray pressure of the high-pressure spray head are synchronously matched with the cleaning path and cleaning pressure; the negative pressure adsorption device works behind the cleaning brush to adsorb the sewage and residual stains after cleaning in real time; during the cleaning process, the control module collects the working parameters of the cleaning actuator in real time, compares them with the planned cleaning parameters, and adjusts them in time when deviations occur.

[0026] Preferably, in step S7, the secondary detection uses a multispectral acquisition module to acquire image data of the cleaned area, and through image preprocessing and a stain recognition model, detects whether there are still unremoved stains in the cleaned area.

[0027] Preferably, in step S7, the criteria for determining whether the cleaning effect meets the standard are: there are no heavy stains in the cleaning area, the area ratio of light and moderate stains is lower than a preset threshold, and the light transmittance of the cleaning area is restored to a preset ratio of the normal photovoltaic panel light transmittance. If the cleaning effect does not meet the standard, the characteristic parameters of the unremoved stains are extracted, and the process returns to step S5 to replan the cleaning path and cleaning parameters, adjust key parameters such as cleaning pressure and cleaning times, and perform the cleaning operation again. If the cleaning effect meets the standard, the relevant data of this cleaning, including stain type, cleaning parameters, cleaning time, etc., are recorded and stored in the storage module of the vehicle.

[0028] Compared with the prior art, the beneficial effects of the present invention are:

[0029] 1. In order to solve the problems of low cleaning efficiency, insufficient identification accuracy and waste of cleaning resources in traditional photovoltaic panel cleaning, and to improve the power generation efficiency and intelligent level of photovoltaic panel cleaning operations, this invention integrates multispectral image acquisition, deep learning recognition model and adaptive cleaning control strategy to build an intelligent cleaning method that integrates stain identification, classification judgment, path planning and precise cleaning. This method achieves high-precision identification and differentiated precise processing of stains on the surface of photovoltaic panels, significantly improving cleaning effect and energy utilization efficiency.

[0030] 2. This invention simultaneously acquires visible light, near-infrared, and ultraviolet images through a multispectral acquisition module, and combines ambient light intensity and panel temperature parameters to achieve multi-dimensional data fusion of the photovoltaic panel surface condition, significantly improving the information richness and accuracy of stain identification; the staggered acquisition method avoids mutual interference between cameras, ensuring image quality; constant distance and uniform speed acquisition ensures consistent image resolution, and the introduction of an adaptive adjustment mechanism for the supplementary lighting device enables clear images to be acquired even under low light conditions, enhancing the system's adaptability in complex environments;

[0031] 3. This invention uses an improved YOLO deep learning model to automatically classify and grade stains, achieving efficient identification of various stain types such as dust, oil, bird droppings, and corrosion. The model incorporates near-infrared reflectivity and board temperature for feature correction, and dynamically adjusts the recognition weights through stain feature correction coefficients, significantly improving the recognition accuracy under complex backgrounds and similar textures. Stains are classified into three levels: light, moderate, and heavy, providing a scientific basis for subsequent differentiated cleaning. The overall recognition process is fast and stable, supports real-time processing, and greatly enhances the intelligent decision-making capabilities of the cleaning system.

[0032] 4. This invention plans the cleaning path using a grid map method, dividing the photovoltaic panel surface into uniform grids and marking the areas to be cleaned, achieving optimal coverage of the cleaning path and avoiding repeated or missed cleaning; it prioritizes the treatment of stained areas before performing auxiliary cleaning, improving work efficiency; the cleaning parameters are dynamically adjusted according to the type, grade, and correction coefficient of the stain, achieving adaptive configuration of pressure, speed, media usage, and number of times. The overall path and parameter collaborative planning strategy makes the cleaning process more scientific, efficient, and energy-saving, significantly improving the precision level of cleaning operations. Attached Figure Description

[0033] Figure 1 The present invention describes the operation flow of a method for stain identification and precise cleaning of photovoltaic panels using an intelligent cleaning vehicle. Figure 1 ;

[0034] Figure 2 The present invention describes the operation flow of a method for stain identification and precise cleaning of photovoltaic panels using an intelligent cleaning vehicle. Figure 2 ;

[0035] Figure 3 The present invention describes the operation flow of a method for stain identification and precise cleaning of photovoltaic panels using an intelligent cleaning vehicle. Figure 3 ;

[0036] Figure 4 The present invention describes the operation flow of a method for stain identification and precise cleaning of photovoltaic panels using an intelligent cleaning vehicle. Figure 4 . Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Example

[0039] Please see Figures 1-4 As shown, the present invention provides a technical solution comprising the following steps:

[0040] S1. The cleaning vehicle initializes and moves to the photovoltaic panel initial detection position, completing attitude calibration and detection system preheating;

[0041] S2. Acquire a full-area image of the photovoltaic panel surface through a multispectral acquisition module, and simultaneously acquire ambient light intensity and panel temperature parameters.

[0042] S3. Preprocess the acquired image, segment out the suspected stain area and extract the regional feature parameters;

[0043] S4. Construct a stain recognition model based on feature parameters, classify and identify suspected stain areas, and determine the stain type and level;

[0044] S5. Based on the type and grade of stains and the surface shape of the photovoltaic panel, plan a precise cleaning path and cleaning parameters;

[0045] S6. According to the planned path and parameters, control the cleaning execution mechanism to perform differentiated and precise cleaning on the stained areas;

[0046] S7. After cleaning is completed, a second inspection is conducted on the cleaned area to determine whether the cleaning effect meets the standard. If it does not meet the standard, the cleaning parameters are recalculated.

[0047] In this embodiment, the initialization of the cleaning vehicle in S1 includes battery power detection, cleaning actuator reset, and positioning module calibration. The positioning module uses a combination of Beidou positioning and visual positioning. Visual positioning completes the relative position calibration between the vehicle and the photovoltaic panel by collecting feature points on the edge of the photovoltaic panel. The attitude calibration uses an inclination sensor set at the bottom of the vehicle to collect the tilt angle of the vehicle. Combined with the preset installation angle of the photovoltaic panel, the attitude of the vehicle's walking mechanism is adjusted so that the vehicle fits the surface of the photovoltaic panel.

[0048] The preheating of the detection system includes powering on and preheating the multispectral acquisition module, image processing module, and control module to ensure that the working parameters of each module are stable. After this, the trolley moves along the edge of the photovoltaic panel to the preset starting detection position via the walking mechanism. The starting detection position is a corner of the photovoltaic panel, avoiding vulnerable parts such as the photovoltaic panel junction box.

[0049] In this embodiment, in S2, the multispectral acquisition module includes a visible light camera, a near-infrared camera, and an ultraviolet camera, which simultaneously acquire visible light images, near-infrared images, and ultraviolet images of the photovoltaic panel surface. During the acquisition process, the trolley moves at a constant speed along the length of the photovoltaic panel, and the distance between the multispectral acquisition module and the photovoltaic panel surface remains constant during the movement.

[0050] Ambient light intensity is collected in real time by a light intensity sensor installed on the top of the vehicle, and the panel temperature is collected by a contact temperature sensor. The temperature sensor is in flexible contact with the surface of the photovoltaic panel to avoid scratching the coating of the photovoltaic panel.

[0051] The collected image data, ambient light intensity data, and plate temperature data are synchronously transmitted to the image processing module for data caching and preliminary processing, and obvious abnormal data is removed.

[0052] In this embodiment, during step S2, the multispectral acquisition module uses an interleaved acquisition method when acquiring images. The acquisition frequencies of the visible light camera, near-infrared camera, and ultraviolet camera are kept consistent, and there is a preset time difference in the acquisition time to avoid mutual interference when different cameras acquire images.

[0053] The resolution of the acquired images is adaptively adjusted according to the size of the photovoltaic panel to ensure that even small stains can be clearly captured. Ambient light intensity acquisition is carried out simultaneously with image acquisition. When the ambient light intensity is lower than the preset threshold, the supplementary lighting device of the multispectral acquisition module is activated. The brightness of the supplementary lighting device is adaptively adjusted according to the ambient light intensity to avoid overexposure or blurring caused by insufficient lighting.

[0054] In this embodiment, in step S3, image preprocessing includes three sub-steps: image denoising, image enhancement, and image segmentation. Image denoising uses adaptive median filtering to remove environmental noise and sensor noise from the acquired image.

[0055] Image enhancement highlights the grayscale difference between the stained area and the photovoltaic panel surface by adjusting the contrast and brightness of the image. The brightness adjustment is adaptively adjusted based on the collected ambient light intensity data.

[0056] Image segmentation combines threshold segmentation with edge detection. Threshold segmentation initially separates suspected stain areas whose gray values ​​differ from those on the normal photovoltaic panel surface. Then, edge detection algorithm is used to extract the contours of the suspected stain areas and obtain their boundary coordinates.

[0057] The extracted regional feature parameters include the area of ​​the stained area, the mean gray level, the gray level variance, the edge roughness, and the near-infrared reflectance. All feature parameters are integrated into a feature vector.

[0058] In this embodiment, in step S4, the stain recognition model is built based on deep learning, using an improved YOLO model. The model's input is the extracted feature vector, and the output is the stain type and stain level. The stain types include dust stains, oil stains, bird droppings stains, and corrosive stains. The stain level is divided into three grades: light, moderate, and heavy, based on the stain area and its impact on the photovoltaic panel's light transmittance. During the stain recognition process, a stain feature correction coefficient K is introduced, which is implemented as follows:

[0059] ,

[0060] In the formula, K is the stain characteristic correction coefficient. This is the weighting coefficient for near-infrared reflectance. For the extracted near-infrared reflectance, This is the plate temperature weighting coefficient. The temperature of the plate is collected; the feature vector is corrected by a correction coefficient K to improve the accuracy of stain identification.

[0061] In this embodiment, in step S5, the cleaning path planning adopts the grid map method, which divides the surface of the photovoltaic panel into several uniform grids. Combined with the identified boundary coordinates of the stained areas, the grids that need to be cleaned are marked. The cleaning path for the stained areas is planned first, and then the auxiliary cleaning path for the non-stained areas is planned to ensure that the cleaning path is complete and without repetition.

[0062] Specifically, the cleaning parameters include cleaning pressure, cleaning speed, cleaning medium dosage, and number of cleaning cycles. These parameters are determined based on the type and severity of the stain, and adjusted using a correction factor K. The specific adjustments are as follows:

[0063] ,

[0064] In the formula, P represents the final cleaning pressure. The basic cleaning pressure is S, which is the stain grade coefficient. For example, light stains S=1, medium stains S>1 and <2, and heavy stains S≥2. The cleaning medium is selected according to the type of stain. Dust stains are cleaned with water, oil stains are cleaned with a mixture of neutral cleaning solution and water, and bird droppings and corrosive stains are cleaned with special cleaning media.

[0065] In this embodiment, step S6, the cleaning actuator includes a liftable cleaning brush, a high-pressure spray head, and a negative pressure adsorption device. The height of the liftable cleaning brush is adjusted in real time according to the flatness of the photovoltaic panel surface. The height adjustment is achieved as follows:

[0066] ,

[0067] In the formula, H represents the final height of the cleaning brush. For the basic height of the cleaning brush, The flatness deviation value of the photovoltaic panel surface; the spray angle and spray pressure of the high-pressure spray head are synchronously matched with the cleaning path and cleaning pressure; the negative pressure adsorption device works behind the cleaning brush to adsorb the sewage and residual stains after cleaning in real time, avoiding sewage from flowing on the photovoltaic panel surface and causing secondary pollution; during the cleaning process, the control module collects the working parameters of the cleaning actuator in real time, compares them with the planned cleaning parameters, and adjusts them in time when deviations occur;

[0068] Specifically, the lifting cleaning brush of the cleaning actuator uses flexible brushes with a scratch-resistant coating to prevent damage to the nano-coating or anti-reflective coating on the photovoltaic panel surface; the high-pressure spray head has an adjustable spray nozzle, which can adjust the spray range according to the size of the cleaning area, reducing the waste of cleaning media; the negative pressure adsorption device has a filter structure at the adsorption port to filter the adsorbed dirt particles, preventing dirt particles from clogging the adsorption pipes. The adsorbed wastewater and dirt particles are collected centrally after filtration for subsequent treatment, avoiding environmental pollution.

[0069] In this embodiment, in step S7, the secondary detection uses a multispectral acquisition module to acquire image data of the cleaned area. Through image preprocessing and a stain recognition model, it detects whether there are still unremoved stains in the cleaned area.

[0070] In this embodiment, in step S7, the criteria for determining whether the cleaning effect meets the standard are: there are no heavy stains in the cleaning area, the area ratio of light and moderate stains is lower than a preset threshold, and the light transmittance of the cleaning area is restored to a preset ratio of the normal photovoltaic panel light transmittance. If the cleaning effect does not meet the standard, the characteristic parameters of the unremoved stains are extracted, and the process returns to step S5 to replan the cleaning path and cleaning parameters, adjust key parameters such as cleaning pressure and cleaning times, and perform the cleaning operation again. If the cleaning effect meets the standard, the relevant data of this cleaning, including stain type, cleaning parameters, cleaning time, etc., are recorded and stored in the storage module of the vehicle.

[0071] Working Principle: The initialization and attitude calibration of the cleaning trolley lay the foundation for subsequent detection and cleaning operations. The trolley first checks battery power, resets actuators, and calibrates the positioning module to ensure the system is ready for operation. Using a combination of BeiDou and visual positioning, relative position calibration is achieved using feature points on the edge of the photovoltaic panel, precisely moving to the initial detection position. Simultaneously, a tilt sensor collects the trolley's tilt angle, adjusting the walking mechanism's posture to fit the photovoltaic panel surface. The detection system preheats to ensure stable parameters for each module. A multispectral acquisition module simultaneously acquires images of the photovoltaic panel surface and environmental parameters, achieving multi-dimensional data fusion. Visible light, near-infrared, and ultraviolet cameras use an interleaved acquisition method to avoid mutual interference and ensure image clarity. The trolley moves at a constant speed while maintaining a constant distance between the acquisition module and the photovoltaic panel surface, ensuring consistent image resolution. An ambient light intensity sensor and a contact temperature sensor simultaneously collect environmental parameters, and the data is transmitted to the image processing module for caching and processing, eliminating abnormal data.

[0072] The acquired images were denoised, enhanced, and segmented to extract stain feature parameters. Adaptive median filtering removed environmental and sensor noise while preserving stain details. Image contrast and brightness were adjusted based on ambient light intensity data to highlight the grayscale difference between the stain area and the background. Threshold segmentation and edge detection algorithms were combined to initially separate suspected stain areas and extract their contour boundary coordinates. Finally, feature parameters such as area, mean grayscale, and edge roughness were extracted and integrated into a feature vector. An improved YOLO deep learning model was used to classify and grade stains. The model automatically identified stain types, including grayscale, using the extracted feature vector as input. Dust, oil, bird droppings, and corrosive stains are categorized into light, moderate, and heavy levels based on stain area and impact on light transmittance. Near-infrared reflectance and panel temperature are incorporated into feature correction, and the recognition weight is dynamically adjusted using stain feature correction coefficients to improve accuracy. A grid map method is used to plan cleaning paths and parameters for precise cleaning. The photovoltaic panel surface is divided into uniform grids, marking areas requiring cleaning. Priority is given to planning paths for stained areas, followed by auxiliary cleaning paths for non-stained areas, ensuring no omissions or repetitions. Cleaning pressure, speed, media usage, and number of cleaning cycles are dynamically adjusted based on stain type, level, and correction coefficients.

[0073] The system utilizes a height-adjustable cleaning brush, a high-pressure spray head, and a negative pressure adsorption device to perform differentiated and precise cleaning. The flexible brush adjusts its height in real time according to the flatness of the photovoltaic panel surface to avoid scratching the coating. The high-pressure spray head matches the cleaning path and pressure to precisely apply the cleaning medium, selecting different cleaning solutions for different types of stains. The negative pressure adsorption device immediately recovers wastewater and residual stains to prevent secondary pollution. The control module monitors the actuator parameters in real time. The cleaning effect is verified through secondary testing to achieve closed-loop control of cleaning quality. The multispectral acquisition module re-acquires images of the cleaned area, which are pre-processed and used to determine the presence of residual stains using a stain recognition model. The cleaning effect is judged based on standards such as the proportion of areas with no heavy stains, light to moderate stains, and the light transmittance recovery ratio. If the standards are not met, the system feeds back to the path planning module to optimize the parameters and clean again until the standards are met. Simultaneously, the cleaning data is recorded and stored.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

[0075] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for stain identification and precise cleaning of photovoltaic panel intelligent cleaning carts, characterized in that, Includes the following steps: S1. The cleaning vehicle initializes and moves to the photovoltaic panel initial detection position, completing attitude calibration and detection system preheating; S2. Acquire a full-area image of the photovoltaic panel surface through a multispectral acquisition module, and simultaneously acquire ambient light intensity and panel temperature parameters. S3. Preprocess the acquired image, segment out the suspected stain area and extract the regional feature parameters; S4. Construct a stain recognition model based on feature parameters, classify and identify suspected stain areas, and determine the stain type and level; S5. Based on the type and grade of stains and the surface shape of the photovoltaic panel, plan a precise cleaning path and cleaning parameters; S6. According to the planned path and parameters, control the cleaning execution mechanism to perform differentiated and precise cleaning on the stained areas; S7. After cleaning is completed, a second inspection is conducted on the cleaned area to determine whether the cleaning effect meets the standard. If it does not meet the standard, the cleaning parameters are recalculated.

2. The method for stain identification and precise cleaning of photovoltaic panel intelligent cleaning carts according to claim 1, characterized in that: In step S1, the initialization of the cleaning vehicle includes battery power detection, cleaning actuator reset, and positioning module calibration. The positioning module uses a combination of BeiDou positioning and visual positioning. Visual positioning completes the relative position calibration between the vehicle and the photovoltaic panel by collecting feature points on the edge of the photovoltaic panel. Attitude calibration uses an inclination sensor set at the bottom of the vehicle to collect the tilt angle of the vehicle. Combined with the preset installation angle of the photovoltaic panel, the attitude of the vehicle's walking mechanism is adjusted so that the vehicle fits the surface of the photovoltaic panel. The preheating of the detection system includes the power-on preheating of the multispectral acquisition module, image processing module, and control module.

3. The method for stain identification and precise cleaning of a photovoltaic panel intelligent cleaning vehicle according to claim 1, characterized in that: In step S2, the multispectral acquisition module includes a visible light camera, a near-infrared camera, and an ultraviolet camera, which simultaneously acquire visible light, near-infrared, and ultraviolet images of the photovoltaic panel surface. During the acquisition process, the trolley moves at a constant speed along the length of the photovoltaic panel, maintaining a constant distance between the multispectral acquisition module and the photovoltaic panel surface. Ambient light intensity is acquired in real time by a light intensity sensor installed on the top of the trolley, and panel temperature is acquired by a contact temperature sensor that flexibly contacts the photovoltaic panel surface. The acquired image data, ambient light intensity data, and panel temperature data are simultaneously transmitted to the image processing module for data caching and preliminary processing, removing obvious abnormal data.

4. The method for stain identification and precise cleaning of a photovoltaic panel intelligent cleaning vehicle according to claim 1, characterized in that: In step S2, when the multispectral acquisition module acquires images, it adopts an interleaved acquisition method, ensuring that the acquisition frequencies of the visible light camera, near-infrared camera, and ultraviolet camera are consistent, and that there is a preset time difference in the acquisition time; the resolution of the acquired images is adaptively adjusted according to the size of the photovoltaic panel. Ambient light intensity acquisition and image acquisition are performed simultaneously. When the ambient light intensity is lower than the preset threshold, the supplementary lighting device of the multispectral acquisition module is activated, and the brightness of the supplementary lighting device is adaptively adjusted according to the ambient light intensity.

5. The method for stain identification and precise cleaning of a photovoltaic panel intelligent cleaning vehicle according to claim 1, characterized in that: In step S3, image preprocessing includes three sub-steps: image denoising, image enhancement, and image segmentation. Image denoising uses adaptive median filtering to remove environmental noise and sensor noise from the acquired image. Image enhancement highlights the grayscale difference between the stained area and the photovoltaic panel surface by adjusting the contrast and brightness of the image. The brightness adjustment is adaptively adjusted based on the collected ambient light intensity data. Image segmentation employs a combination of threshold segmentation and edge detection. Threshold segmentation initially separates suspected stain areas whose grayscale values ​​differ from those of normal photovoltaic panel surfaces. Edge detection then extracts the contours of these suspected stain areas, yielding their boundary coordinates. Extracted regional feature parameters include the area of ​​the stain area, mean grayscale value, grayscale variance, edge roughness, and near-infrared reflectance. All feature parameters are integrated into a feature vector.

6. The method for stain identification and precise cleaning of a photovoltaic panel intelligent cleaning vehicle according to claim 1, characterized in that: In S4, the stain recognition model is built based on deep learning. The input of the model is the extracted feature vector, and the output is the stain type and stain level. The stain level is divided into three levels: light, moderate and heavy, according to the stain area and the degree of impact on the light transmittance of the photovoltaic panel. In the stain recognition process, a stain feature correction coefficient K is introduced. The feature vector is corrected by the correction coefficient K.

7. The method for stain identification and precise cleaning of photovoltaic panel intelligent cleaning carts according to claim 1, characterized in that: In S5, the cleaning path planning adopts the grid map method, which divides the surface of the photovoltaic panel into several uniform grids. Combined with the identified boundary coordinates of the stained areas, the grids that need to be cleaned are marked. The cleaning path for the stained areas is planned first, and then the auxiliary cleaning path for the non-stained areas is planned. Cleaning parameters include cleaning pressure, cleaning speed, cleaning medium dosage, and number of cleaning cycles. These parameters are determined based on the type and severity of the stain, and adjusted using a correction factor K. The specific adjustments are as follows: ， In the formula, P represents the final cleaning pressure. S represents the basic cleaning pressure, and S represents the stain level coefficient.

8. The method for stain identification and precise cleaning of a photovoltaic panel intelligent cleaning vehicle according to claim 1, characterized in that: S6, the cleaning actuator includes a liftable cleaning brush, a high-pressure spray head, and a negative pressure adsorption device. The height of the liftable cleaning brush is adjusted in real time according to the flatness of the photovoltaic panel surface. The height adjustment is achieved as follows: ， In the formula, H represents the final height of the cleaning brush. For the basic height of the cleaning brush, The flatness deviation value of the photovoltaic panel surface; the spray angle and spray pressure of the high-pressure spray head are synchronously matched with the cleaning path and cleaning pressure; the negative pressure adsorption device works behind the cleaning brush to adsorb the sewage and residual stains after cleaning in real time. During the cleaning process, the control module collects the working parameters of the cleaning actuator in real time, compares them with the planned cleaning parameters, and makes timely adjustments when deviations occur.

9. The method for stain identification and precise cleaning of a photovoltaic panel intelligent cleaning vehicle according to claim 1, characterized in that: In step S7, the secondary detection uses a multispectral acquisition module to acquire image data of the cleaned area. Through image preprocessing and a stain recognition model, it detects whether there are still unremoved stains in the cleaned area.

10. A method for stain identification and precise cleaning of photovoltaic panel intelligent cleaning carts according to claim 9, characterized in that: In S7, the standard for judging whether the cleaning effect meets the standard is: there are no heavy stains in the cleaning area, the area ratio of light stains and moderate stains is lower than the preset threshold, and the light transmittance of the cleaning area is restored to a preset ratio of the normal photovoltaic panel light transmittance. If the cleaning effect is not up to standard, extract the characteristic parameters of the unremoved stains, return to S5 to replan the cleaning path and cleaning parameters, adjust the key parameters of cleaning pressure and number of cleaning cycles, and perform the cleaning operation again; If the cleaning effect meets the standard, record the relevant data of this cleaning, including stain type, cleaning parameters, cleaning time, etc., and store them in the storage module of the vehicle.

Images