Method and device for detecting surface dust accumulation amount of photovoltaic panel

Abstract

The invention provides a photovoltaic panel surface dust accumulation amount detection method and device. The method comprises the following steps: acquiring a photovoltaic panel dust accumulation image of a target photovoltaic panel; the photovoltaic panel dust deposition image is input into a target convolutional neural model, target image features are obtained, the target convolutional neural model comprises a preset number of convolutional layers, and the number of convolution kernels included in the adjacent convolutional layers is progressively increased in an equal-product sequence; obtaining a target scattering coefficient and a target ambient light parameter according to the target image feature and a target convolutional neural model; and obtaining the dust accumulation amount of the target photovoltaic panel according to the photovoltaic panel dust accumulation image, the target scattering coefficient and the target ambient light parameter by using an atmospheric scattering model. According to the method, the image features extracted by the CNN model are combined with the physical features corresponding to the atmospheric scattering model, the interference of illumination on the dust accumulation amount evaluation result when the dust accumulation amount is directly evaluated through the CNN model is reduced, the dust accumulation amount detection precision is improved, and thus the cleaning operation of the photovoltaic panel can be guided more accurately.

Description

Technical Field

[0001] This application relates to the field of photovoltaic panel cleaning and inspection technology, and in particular to a method and device for detecting dust accumulation on the surface of photovoltaic panels. Background Technology

[0002] Solar energy, as a clean energy source, has enormous development potential. Photovoltaic power generation technology utilizes the photovoltaic effect to directly convert solar energy into electrical energy, and its core device is the photovoltaic panel. Photovoltaic panels are typically installed in well-lit outdoor environments to maximize sunlight absorption. However, in practical applications, especially in windy and sandy areas such as deserts, photovoltaic panels are affected by dust and sand particles. These particles deposit on the surface of the photovoltaic panels, significantly reducing their light transmittance and photoelectric conversion efficiency, and potentially causing safety hazards such as localized overheating. Therefore, it is necessary to clean the photovoltaic panels according to the amount of dust accumulation. Thus, it is essential to monitor the amount of dust accumulated on the photovoltaic panels.

[0003] Dust accumulation can significantly alter the color and texture of photovoltaic (PV) panels. The degree of dust accumulation on PV panels can be determined by observing their images. Therefore, existing technologies utilize image recognition algorithms such as convolutional neural networks (CNNs) to assess dust levels. Essentially, detecting dust accumulation on PV panels using CNNs establishes a mapping between image features and dust accumulation. By extracting image features from the PV panel image using the CNN, the amount of dust on the panel surface can be directly obtained.

[0004] However, photovoltaic panel images are easily affected by external factors such as lighting, resulting in different images of the same photovoltaic panel under different lighting conditions. This leads to variations in the dust accumulation readings obtained using convolutional neural networks. Therefore, the method of detecting dust accumulation on photovoltaic panels using convolutional neural networks is easily affected by lighting conditions, resulting in low accuracy in dust accumulation calculations. Summary of the Invention

[0005] This application provides a method and apparatus for detecting dust accumulation on the surface of a photovoltaic panel, which solves the problem mentioned in the background art that the method of detecting dust accumulation on a photovoltaic panel using a convolutional neural network is easily affected by light, resulting in low accuracy in dust accumulation calculation.

[0006] In a first aspect, this application provides a method for detecting dust accumulation on the surface of a photovoltaic panel, comprising: Obtain images of dust accumulation on the target photovoltaic panel; The dust accumulation image of the photovoltaic panel is input into the target convolutional neural model to obtain the target image features. The target convolutional neural model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence. Based on the target image features and the target convolutional neural model, the target scattering coefficient and target ambient light parameters are obtained; Using an atmospheric scattering model, the amount of dust accumulated on the target photovoltaic panel is obtained based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters.

[0007] Optionally, the target convolutional neural model includes three convolutional layers. The step of inputting the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain target image features includes: The first convolutional layer is used to extract the first layer of features from the dust accumulation image of the photovoltaic panel to obtain a first feature map. The first layer of features includes: edge information and color changes. The first feature map is subjected to a second convolutional layer to extract second-layer features, and the second-layer features include: particle outline and texture information. The second feature map is subjected to a third convolutional layer to extract the third layer of features, thereby obtaining the target image features, which include: dust distribution pattern and dust density gradient.

[0008] Optionally, the step of using an atmospheric scattering model to obtain the amount of dust accumulated on the target photovoltaic panel based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters includes: Based on the dust accumulation image of the photovoltaic panel, obtain the corresponding dust-free image of the photovoltaic panel; Obtain the target clean pixel value and the target dust-accumulated pixel value corresponding to the dust-free image of the photovoltaic panel and the dust-accumulated image of the photovoltaic panel, respectively. The dust accumulation thickness of the target photovoltaic panel is obtained based on the target mapping relationship and the target dust-free pixel value, the target dust-accumulated pixel value, the target scattering coefficient, and the target ambient light parameter. The target mapping relationship is obtained based on the atmospheric scattering model, which is the mapping relationship between the dust accumulation thickness and the dust-free pixel value, the dust-accumulated pixel value, the scattering coefficient, and the ambient light parameter. The amount of dust accumulated on the target photovoltaic panel is obtained based on the dust accumulation thickness and the surface area of ​​the target photovoltaic panel.

[0009] Optionally, the target mapping relationship is as follows:

[0010] Wherein, d(x, y) represents the dust thickness at coordinate (x, y) on the target photovoltaic panel, P(x, y) represents the dust-accumulated pixel value, P0(x, y) represents the dust-free pixel value, A represents the ambient light parameter, and β represents the scattering coefficient.

[0011] Optionally, before inputting the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain the target image features, the method further includes: Obtain N sets of photovoltaic panel image samples. Each set of photovoltaic panel image samples includes: historical photovoltaic panel dust accumulation image, historical true scattering coefficient, and historical true ambient light parameters. The initial convolutional neural model is trained based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural model. The initial convolutional neural model includes the preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence.

[0012] Optionally, the step of training the initial convolutional neural model based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural model includes: Based on the photovoltaic panel dust accumulation image samples corresponding to the N sets of photovoltaic panel image samples, the target image features of the samples are obtained; Based on the sample target image features and the initial convolutional neural model, the predicted scattering coefficient and predicted ambient light parameters are obtained; The initial convolutional neural model is trained based on the historical true scattering coefficient, the predicted scattering coefficient, the historical true ambient light parameters, the predicted ambient light parameters, and the loss function to obtain the target convolutional neural model.

[0013] Secondly, this application provides a device for detecting dust accumulation on the surface of a photovoltaic panel, comprising: The acquisition module is used to acquire images of dust accumulation on the target photovoltaic panel. The model processing module is used to input the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain target image features, wherein the target convolutional neural model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence; and is used to obtain the target scattering coefficient and target ambient light parameters based on the target image features and the target convolutional neural model. The detection module is used to obtain the amount of dust accumulation on the target photovoltaic panel based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters using an atmospheric scattering model.

[0014] Optionally, the target convolutional neural model includes three convolutional layers. The model processing module inputs the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain target image features, specifically for: The first convolutional layer is used to extract the first layer of features from the dust accumulation image of the photovoltaic panel to obtain a first feature map. The first layer of features includes: edge information and color changes. The first feature map is subjected to a second convolutional layer to extract second-layer features, and the second-layer features include: particle outline and texture information. The second feature map is subjected to a third convolutional layer to extract the third layer of features, thereby obtaining the target image features, which include: dust distribution pattern and dust density gradient.

[0015] Optionally, the detection module utilizes an atmospheric scattering model to obtain the amount of dust accumulated on the target photovoltaic panel based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters. Specifically, it is used for: Based on the dust accumulation image of the photovoltaic panel, obtain the corresponding dust-free image of the photovoltaic panel; Obtain the target clean pixel value and the target dust-accumulated pixel value corresponding to the dust-free image of the photovoltaic panel and the dust-accumulated image of the photovoltaic panel, respectively. The dust accumulation thickness of the target photovoltaic panel is obtained based on the target mapping relationship and the target dust-free pixel value, the target dust-accumulated pixel value, the target scattering coefficient, and the target ambient light parameter. The target mapping relationship is obtained based on the atmospheric scattering model, which is the mapping relationship between the dust accumulation thickness and the dust-free pixel value, the dust-accumulated pixel value, the scattering coefficient, and the ambient light parameter. The amount of dust accumulated on the target photovoltaic panel is obtained based on the dust accumulation thickness and the surface area of ​​the target photovoltaic panel.

[0016] Optionally, the target mapping relationship is as follows:

[0017] Wherein, d(x, y) represents the dust thickness at coordinate (x, y) on the target photovoltaic panel, P(x, y) represents the dust-accumulated pixel value, P0(x, y) represents the dust-free pixel value, A represents the ambient light parameter, and β represents the scattering coefficient.

[0018] Optionally, before the model processing module inputs the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain the target image features, the training module is used for: Obtain N sets of photovoltaic panel image samples. Each set of photovoltaic panel image samples includes: historical photovoltaic panel dust accumulation image, historical true scattering coefficient, and historical true ambient light parameters. The initial convolutional neural model is trained based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural model. The initial convolutional neural model includes the preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence.

[0019] Optionally, the training module trains the initial convolutional neural model based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural model, the parent model being used for: Based on the photovoltaic panel dust accumulation image samples corresponding to the N sets of photovoltaic panel image samples, the target image features of the samples are obtained; Based on the sample target image features and the initial convolutional neural model, the predicted scattering coefficient and predicted ambient light parameters are obtained; The initial convolutional neural model is trained based on the historical true scattering coefficient, the predicted scattering coefficient, the historical true ambient light parameters, the predicted ambient light parameters, and the loss function to obtain the target convolutional neural model.

[0020] Thirdly, this application provides an electronic device, including: a processor and a memory; The memory stores the instructions that the computer executes; The processor executes computer execution instructions stored in memory, causing the processor to perform the method as described in any of the first aspects.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium including a program or instructions that, when run on a computer, execute the method described in any of the first aspects above.

[0022] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the first aspects.

[0023] The method and apparatus for detecting dust accumulation on the surface of photovoltaic panels provided in this application acquire a dust accumulation image of a target photovoltaic panel; input the dust accumulation image into a target convolutional neural network model to obtain target image features, wherein the target convolutional neural network model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence; based on the target image features and the target convolutional neural network model, the target scattering coefficient and target ambient light parameters are obtained; using an atmospheric scattering model, the dust accumulation amount of the target photovoltaic panel is obtained based on the photovoltaic panel dust accumulation image, target scattering coefficient, and target ambient light parameters. This method combines the image features extracted by the CNN model with the corresponding physical features of the atmospheric scattering model, reducing the interference of illumination on the dust accumulation assessment results when directly evaluating dust accumulation amount through the CNN model, improving the accuracy of dust accumulation detection, and thus providing more precise guidance for photovoltaic panel cleaning operations. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart of a method for detecting dust accumulation on the surface of a photovoltaic panel provided in an embodiment of this application; Figure 2 A flowchart illustrating the training process of a target convolutional neural model provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a photovoltaic panel surface dust detection device provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application are described clearly and completely below. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of this application.

[0027] Photovoltaic panels are typically installed in sunny outdoor environments to maximize sunlight absorption. However, in practical applications, especially in areas with high winds and sandstorms such as deserts, photovoltaic panels are affected by dust and sand particles. Dust accumulation significantly impacts the output power of photovoltaic panels, and the surface temperature of photovoltaic panels is higher when dusty than when clean. When there is excessive dust accumulation in a local area, hot spots can easily form, leading to component failure.

[0028] Currently, the main methods for detecting and evaluating dust accumulation on photovoltaic panels include: 1. Visual inspection: This method relies on on-site maintenance personnel at photovoltaic power plants to assess the degree of dust accumulation based on their personal observations and experience. This method is overly dependent on manual labor and the individual skill level of the maintenance personnel, resulting in poor accuracy and real-time performance. It is therefore insufficient to meet the development needs of intelligent operation and maintenance and component health management for distributed photovoltaic power generation systems. 2. Weighing technology: Dust accumulated on photovoltaic panels is collected by manual sampling and weighed using laboratory equipment such as weighing devices. This method is complicated and time-consuming, not suitable for real-time on-site testing, and is prone to dust loss during multiple operations, thus affecting the accuracy of the results. 3. Optical attenuation measurement: This method assesses dust accumulation by installing expensive optical sensors to detect the degree of light attenuation. However, it can only monitor dust accumulation within a limited range and cannot cover the dust accumulation status of the entire photovoltaic power station. Furthermore, the sensor maintenance cost is high. 4. Performance Comparison Method: Relying on power generation efficiency data makes it difficult to distinguish the impact of dust accumulation on efficiency from other factors (such as hardware aging and hotspot effects). This method calculates efficiency loss by monitoring the output power of the photovoltaic panel and comparing it with the power under theoretically clean conditions.

[0029] This method is based on the assumption of efficiency loss ( The main cause is dust accumulation, which is used to estimate the amount of dust. However, the output power of photovoltaic panels is not only related to dust accumulation, but also to radiation intensity, panel material degradation, hot spot effect, inverter loss or line abnormality, etc. Moreover, it relies only on electrical performance parameters and lacks visual information or environmental parameters to support it.

[0030] Therefore, in recent years, with the development of artificial intelligence, methods for monitoring dust accumulation on photovoltaic panels using artificial intelligence have been studied. Since dust accumulation can cause significant changes in the color and texture of photovoltaic panels, the degree of dust accumulation can be determined by observing images of the dust accumulation on the panels. Therefore, image recognition algorithms such as convolutional neural networks can be used to determine the degree of dust accumulation on photovoltaic panels.

[0031] However, photovoltaic panel images are easily affected by external factors such as lighting, resulting in different images of the same photovoltaic panel under different lighting conditions. This leads to variations in the dust accumulation readings obtained using convolutional neural networks. Therefore, the method of detecting dust accumulation on photovoltaic panels using convolutional neural networks is easily affected by lighting conditions, resulting in low accuracy in dust accumulation calculations.

[0032] Therefore, to address the technical problems existing in the prior art, this application proposes a method and device for detecting dust accumulation on the surface of photovoltaic panels. It utilizes a Convolutional Neural Network (CNN) model to extract features from photovoltaic panel images. Furthermore, the CNN model is designed such that the number of convolutional kernels corresponding to each convolutional layer increases in an equal-product sequence according to the abstraction level and semantic richness of the features to be extracted by the corresponding convolutional layer. This ensures matching with the features to be extracted by the corresponding convolutional layer, improving the accuracy and semantic richness of the target image features, and thus improving the accuracy of the output results. Moreover, by combining an atmospheric scattering model, the difference between dust-free and dust-accumulated images of photovoltaic panels is analyzed, thereby achieving accurate quantification of dust accumulation.

[0033] The technical solution of this application will be described below with reference to specific embodiments: Scattering coefficient β β represents the probability of light being scattered by the medium per unit path length. The larger β is, the higher the proportion of light being scattered as it propagates through the medium, resulting in a stronger attenuation of transmitted light intensity. In photovoltaic panel dust accumulation, the dust acts as a scattering medium; the greater the dust thickness, the larger β is.

[0034] Ambient light A The background brightness produced by unobstructed global illumination on the imaging plane is physically derived from diffuse light (indirect light) from the sky and surrounding environment. Specifically, for photovoltaic panels, when light passes through dust accumulation on the panel, some light is scattered to form ambient light. In this application, the value of ambient light is referred to as the ambient light parameter, and is also used... A This represents ambient light parameters.

[0035] Figure 1 This is a flowchart illustrating a method for detecting dust accumulation on the surface of a photovoltaic panel according to an embodiment of this application. The execution entity of the method shown in this embodiment can be a server corresponding to the photovoltaic panel control system, such as... Figure 1 As shown, the method in this embodiment includes: S101. Obtain an image of dust accumulation on the target photovoltaic panel.

[0036] In this step, the target photovoltaic panel is any photovoltaic panel in the photovoltaic power station. To acquire the dust accumulation image of the photovoltaic panel, multiple calibration points can be set on the surface of the target photovoltaic panel to correct geometric distortions in the dust accumulation image. Furthermore, a high-resolution camera is typically used to capture the image of the photovoltaic panel, ensuring that subtle surface features can be distinguished.

[0037] To avoid interference from sunlight, controlled lighting conditions should be used to minimize the impact of environmental variables. Furthermore, since photovoltaic panels are typically installed outdoors, shade panels should be used or sunny weather should be chosen when taking photos.

[0038] After obtaining the image of dust accumulation on the photovoltaic panel, image preprocessing can be performed, including: Geometric correction: Detect the position of calibration points in the dust accumulation image of the photovoltaic panel, calculate the perspective distortion parameters of the photovoltaic panel, and use the perspective transformation method to correct the dust accumulation image of the photovoltaic panel into a rectangle to ensure that the proportion of the dust accumulation image of the photovoltaic panel is consistent with the actual physical size. In addition, bilinear interpolation can be applied to fill pixels in the corrected image to reduce image loss caused by distortion. Denoising: Median filtering is used to remove random noise in the photovoltaic panel dust image, such as high bright spots caused by reflections on the photovoltaic panel surface; Gaussian filtering is used to smooth the image, preserving edge features while reducing texture noise between dust particles; based on the color and grayscale characteristics of the photovoltaic panel, color segmentation technology is applied to remove background interference.

[0039] It should be noted that other methods can be used to preprocess images of dust accumulation on photovoltaic panels, and this embodiment does not limit this.

[0040] S102. Input the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain the target image features. The target convolutional neural model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-integer sequence.

[0041] In this step, the image of dust accumulation on the photovoltaic panel is input into the target CNN model. The target CNN model extracts key information related to the dust accumulation lines, such as particle size, color variations, and dust distribution patterns. The training process of the target CNN model is described below.

[0042] Since photovoltaic panels are located outdoors, the images captured are affected by factors such as sunlight and cloud cover, resulting in image noise. This noise causes the edges, colors, and textures in the captured images to differ from the actual images of the photovoltaic panels. Therefore, it is necessary to process features of different frequency bands in segmented layers. Furthermore, the dust accumulated on the photovoltaic panels contains particles of varying sizes. Thus, traditional CNN models using single-scale convolutional layers struggle to capture all of these features comprehensively.

[0043] Therefore, in this application, the target CNN model has 3 convolutional layers. This layered design of convolutional layers can separate noise (such as reflective highlights) from real dust features. Furthermore, the number of convolutional kernels in each convolutional layer increases in an equal product sequence, making the information contained in the features extracted by each convolutional layer more accurate.

[0044] Optionally, the specific implementation of S102 includes: S1021. The first feature map is obtained by performing the first layer feature extraction on the dust accumulation image of the photovoltaic panel through the first convolutional layer. The first layer features include: edge information and color change.

[0045] Specifically, for the input layer of the target CNN, the input size... H * W * C The RGB image, i.e., the dust accumulation image of the photovoltaic panel, ensures that the entire photovoltaic panel area is captured. Here, C represents the number of channels, typically three channels: red, green, and blue.

[0046] The first convolutional layer contains 32 convolutional kernels. Since a larger kernel size allows for a wider focusing area, the first convolutional layer primarily captures the global structure and extracts basic features, i.e., the first-layer features. Therefore, each kernel in the first convolutional layer is 7×7 in size, used to capture the global structure and extract basic features. These basic features include edge information and color variations.

[0047] Edge information includes the boundaries between dusty areas and other areas (e.g., dust-free areas) in an image. In the image, this manifests as abrupt changes in pixel brightness, obtainable through variations in pixel values. Edge information can be used to locate the starting point of dusty areas (such as dust accumulation at the edge of a photovoltaic panel), providing a spatial reference for subsequent analysis of dust particles. For example, dust accumulation forms jagged edges on the frame of a photovoltaic panel, distinct from natural scratches on glass.

[0048] Color variation refers to the difference in spectral reflectance caused by dust accumulation. Specifically, it can manifest as at least one of two things: brightness reduction and color shift. For example, the average brightness of a dust-free area is approximately 180±20, while in a dust-accumulated area it drops to 80±15, resulting in a brightness reduction of 100±20. Another example is that sand dust is yellowish-brown, corresponding to RGB values ​​of 180, 150, 90, while coal ash is dark gray, corresponding to RGB values ​​of 70, 70, 70; the color shift would then be the array {110, 80, 20}. Color variation can be used to distinguish the type of dust accumulation (sand dust / coal ash) and to make a preliminary estimate of the dust accumulation thickness.

[0049] The feature extraction of the convolutional layer is represented by formula (1): (1) in, X This represents the input image of this convolutional layer. W This represents the weights corresponding to the convolution kernel. m , n These represent the rows and columns of the convolution kernel, respectively. b f It is the bias term corresponding to the convolution kernel. Z i，j，f This represents the feature map output by the convolutional layer. i and j It is the location of the feature map. f Indicates the first f There are 1 feature map, where K represents the size of the convolution kernel.

[0050] For the first convolutional layer, 32 convolutional kernels are used to extract features from the dust accumulation image of the photovoltaic panel, and the first feature map is obtained according to formula (1).

[0051] After obtaining the first feature map in the convolutional layer, an activation function (such as ReLU) is used to introduce a nonlinear transformation, which increases the nonlinear features of the network and preserves the significant feature responses in the first feature map.

[0052] The expression for ReLU is shown in formula (2): (2) Then, the first feature map after linear transformation is reduced in dimensionality using a pooling layer, which reduces computational cost while retaining important features in the first feature map. The max pooling operation is represented by formula (3): (3) In this application, the feature map obtained after the pooling layer retains important features from the first feature map, and the feature map obtained after the pooling layer is also referred to as the first feature map. Therefore, for the next convolutional layer, its input feature map is the feature map output by the pooling layer.

[0053] S1022. The first feature map is extracted by the second convolutional layer to obtain the second feature map. The second feature map includes: particle outline and texture information.

[0054] Specifically, compared to the first convolutional layer, the second convolutional layer extracts deeper, more abstract, and more detailed features. Therefore, the second convolutional layer contains 64 convolutional kernels, each with a size of 5×5, to extract intermediate features, namely the second-layer features. These intermediate features include: particle contours and texture information.

[0055] The particle profile includes the geometric features of individual dust particles. For example, when the dust is floating dust or pollen, its geometric features are circular outlines, and when the dust is gravel or industrial dust, the corresponding geometric features are irregular polygons. Therefore, the particle size distribution can be quantified (by calculating the outline area) and the physical wear risk of the dust can be assessed through the particle profile.

[0056] Texture information is the spatial arrangement pattern of dust accumulation, including: the uniformity of the spatial distribution of dust particles and the distribution trend information in different spatial directions, which is used to assess the distribution of dust accumulation and determine whether the dust accumulation is uniform or clumped together.

[0057] The second convolutional layer performs convolution, non-linear activation, and pooling on the first feature map to obtain the second feature map, which is then input into the third convolutional layer.

[0058] The second convolutional layer performs convolution, nonlinear activation and pooling operations on the first feature map, which can be referred to by formulas (1)-(3) respectively, and will not be repeated here.

[0059] S1023. The second feature map is extracted by the third convolutional layer to obtain the target image features, which include: dust distribution pattern and dust density gradient.

[0060] Specifically, the third convolutional layer is used to extract more refined features. In the case of dust accumulation on photovoltaic panels, the third layer of features extracted by the third convolutional layer includes: dust distribution patterns and dust density gradients. Therefore, the third convolutional layer contains 128 convolutional kernels, each with a size of 3×3. The small kernels focus on local areas, thus enabling the extraction of more abstract and semantically rich features.

[0061] Among them, the dust distribution pattern refers to the macroscopic coverage pattern of dust on the surface of the photovoltaic panel, such as the relationship between the dust thickness at the edge of the dust accumulation area and the dust thickness at the center.

[0062] The dust density gradient is the rate of change of dust thickness per unit area, which can be used to adjust the measurement deviation of ambient light A.

[0063] The convolutional operation, nonlinear activation and pooling operation of the third convolutional layer on the second feature map can be referred to formulas (1)-(3) respectively, which will not be elaborated here.

[0064] The first convolutional layer filters out image noise in the photovoltaic panel dust image to obtain the first feature map. Then, the second convolutional layer locates the dust accumulation area. Finally, the third convolutional layer quantifies the particle density of the dust to obtain the target image features, thus improving the accuracy of the target image features.

[0065] S103. Based on the target image features and the target convolutional neural model, obtain the target scattering coefficient and target ambient light parameters.

[0066] In this step, after obtaining the target image features, a 128-dimensional feature vector is output through the fully connected layer of the target CNN model. The target scattering coefficient and target ambient light parameters are obtained from the target CNN model.

[0067] The target scattering coefficient is obtained according to formula (4): β=σ ( W β * Z + b β (4) in, This represents the Sigmoid activation function. Z This represents the 128-dimensional feature vector output by the fully connected layer. W β This represents the weight matrix corresponding to the scattering coefficients. b β This represents the bias term corresponding to the scattering coefficient.

[0068] The target ambient light parameters are obtained according to formula (5): A=σ (W A * Z + b A (5) in, W A This represents the weight matrix corresponding to the ambient light parameters. b A This represents the bias term corresponding to the ambient light parameters.

[0069] Optionally, camera parameters affect image brightness, thus affecting the accuracy of the obtained ambient light parameter A. Therefore, in order to improve the accuracy of ambient light parameter A, camera parameters are considered when the target CNN model calculates ambient light parameter A, that is, the target ambient light parameter is calculated using formula (6): A=Pmax * σ (W A * Z + b A (6) in, P max This indicates camera parameters.

[0070] S104. Using an atmospheric scattering model, the amount of dust accumulated on the target photovoltaic panel is obtained based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters.

[0071] In this step, in the existing technology, atmospheric scattering models are mainly used to calculate the concentration of haze in the air (such as the distribution of particulate matter in the atmosphere). The principle is that when there are suspended particulate matter in the atmosphere, light is scattered and absorbed by the suspended particulate matter when it propagates in the atmosphere. Therefore, by measuring or simulating the attenuation and scattering of light, the concentration and size distribution of suspended particulate matter can be deduced.

[0072] In photovoltaic panels, when there is dust accumulation on the surface of the photovoltaic panel, factors such as the particle size, uneven distribution, and humidity of the dust will affect the light scattering behavior. Therefore, the atmospheric scattering model can be applied to the detection of dust accumulation on photovoltaic panels, and the thickness of dust accumulation at various points on the surface of the photovoltaic panel can be obtained by using the atmospheric scattering model.

[0073] The specific process includes: S1041. Based on the dust accumulation image of the photovoltaic panel, obtain the corresponding dust-free image of the photovoltaic panel.

[0074] Specifically, a dust-free image of a photovoltaic panel is an image taken when there is no dust accumulation on the photovoltaic panel, such as when it is put into use after installation or after cleaning.

[0075] The shooting conditions for the dusty photovoltaic panel images and the corresponding dust-free photovoltaic panel images need to be as consistent as possible. For example, the shooting angle, the lighting during shooting, and the camera used should be as consistent as possible to ensure that the difference between the dusty photovoltaic panel images and the dust-free photovoltaic panel images is due to dust accumulation and to avoid other factors causing the difference between the dusty photovoltaic panel images and the dust-free photovoltaic panel images.

[0076] In this process, after capturing a dust-free image of a photovoltaic panel, a correlation between the photovoltaic panel and the image can be established. Similarly, when capturing a dust-covered image of a photovoltaic panel, a correlation between the dust-covered image and the photovoltaic panel can be established. Therefore, after obtaining a dust-covered image of a photovoltaic panel, the corresponding photovoltaic panel can be identified through the dust-covered image, thereby determining the corresponding dust-free image of the photovoltaic panel.

[0077] S1042. Obtain the target dust-free pixel value and the target dust-accumulated pixel value corresponding to the photovoltaic panel dust-free image and the photovoltaic panel dust-accumulated image, respectively.

[0078] Specifically, the pixel value is the brightness value captured by the camera sensor, typically ranging from 0-255 or 0-1023. It represents the intensity of light reflection after it hits the surface of the photovoltaic panel. When there is dust on the surface of the photovoltaic panel, the reflectivity is low due to the dust blocking the light, resulting in a lower pixel value. Therefore, the pixel value when there is dust on the surface of the photovoltaic panel is different from the pixel value when there is no dust. Thus, the dust accumulation status on the surface of the photovoltaic panel can be reflected by the target dust-free pixel value and the target dust-covered pixel value.

[0079] The methods for obtaining the target dust-free pixel value and the target dust-accumulated pixel value can refer to existing technologies, and this embodiment is not limited thereto. For example, for the target dust-free pixel value, a dust-free image of the photovoltaic panel can be input into Photoshop software to obtain the target dust-free pixel value. Similarly, the target dust-accumulated pixel value can be obtained.

[0080] S1043. Based on the target mapping relationship and the target dust-free pixel value, the target dust-accumulated pixel value, the target scattering coefficient, and the target ambient light parameters, obtain the dust accumulation thickness of the target photovoltaic panel.

[0081] The target mapping relationship is obtained based on the atmospheric scattering model, which shows the mapping relationship between dust accumulation thickness and dust-free pixel value, dust accumulation pixel value, scattering coefficient, and ambient light parameters.

[0082] Specifically, the atmospheric scattering model is given by formula (7): P ( x,y )= P 0( x,y ) τ ( x,y )+ A (1- τ ( x,y ))(7) in, P ( x,y ) indicates the location of dust accumulation on the photovoltaic panel in the image. x,y The dust pixel value at () P 0( x,y ) indicates the location in the dust-free image of the photovoltaic panel ( x,y The dust-free pixel value at () τ ( x,y ) represents the projection function. τ ( x,y )= e -βd(x,y) , d ( x,y ) indicates that the photovoltaic panel ( x,y The thickness of dust accumulation at that location.

[0083] Based on formula (7) and the transmission function, the target mapping relationship is obtained as formula (8): (8) After obtaining the target dust-free pixel value, the target dust-accumulated pixel value, the target scattering coefficient, and the target ambient light parameters, substitute them into formula (8) to obtain the dust accumulation thickness of the target photovoltaic panel.

[0084] S1044. Obtain the amount of dust accumulated on the target photovoltaic panel based on the dust accumulation thickness and surface area of ​​the target photovoltaic panel.

[0085] Specifically, the thickness of dust accumulation varies at different locations on the photovoltaic panel; therefore, the dust accumulation thickness of the target photovoltaic panel is... d ( x, y Integrating over the surface area S of the target photovoltaic panel, the total dust accumulation is obtained, where the total dust accumulation can be obtained according to formula (9): (9) In this embodiment, an image of dust accumulation on a target photovoltaic panel is acquired. This image is then input into a target convolutional neural network (CNN) model to obtain target image features. The CNN model includes a preset number of convolutional layers, with adjacent layers containing an increasing number of convolutional kernels in an equal-product sequence. Based on the target image features and the CNN model, the target scattering coefficient and ambient light parameters are obtained. Finally, using an atmospheric scattering model, the amount of dust accumulated on the target photovoltaic panel is determined based on the image, scattering coefficient, and ambient light parameters. This approach combines the image features extracted by the CNN model with the corresponding physical features from the atmospheric scattering model. This reduces the interference of illumination on the dust accumulation assessment results when directly evaluating dust accumulation using the CNN model, improving the accuracy of dust accumulation detection and providing more precise guidance for photovoltaic panel cleaning operations.

[0086] Figure 2 This is a flowchart illustrating the training process of a target convolutional neural model according to an embodiment of this application. The execution entity of the method shown in this embodiment can be a server corresponding to a photovoltaic panel control system, or other servers. For example... Figure 2 As shown, the method illustrated in this embodiment includes: S201. Obtain N sets of photovoltaic panel image samples. Each set of photovoltaic panel image samples includes: historical photovoltaic panel dust accumulation images, historical true scattering coefficients, and historical true ambient light parameters.

[0087] In this embodiment, when training the initial convolutional neural model, in order to improve the applicability and robustness of the trained target convolutional neural model, historical images of photovoltaic panel dust accumulation under different regions, times, and lighting conditions are collected. Furthermore, the historical photovoltaic panel dust accumulation images are preprocessed according to the image preprocessing method for predicting the amount of photovoltaic panel dust accumulation using the target convolutional neural model, thereby further improving the accuracy of the target convolutional neural model in predicting the amount of photovoltaic panel dust accumulation.

[0088] The historical true scattering coefficient can be calculated using the Mie scattering formula, which will not be elaborated here.

[0089] Historical ambient light parameters can be obtained through the cumulative calculation of Rayleigh scattering and Mie scattering, which will not be elaborated here.

[0090] S202. Train the initial convolutional neural model based on N sets of photovoltaic panel image samples to obtain the target convolutional neural model. The initial convolutional neural model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-incremental sequence.

[0091] In this embodiment, N sets of photovoltaic panel image samples are input into the initial convolutional neural model. The structure of the initial convolutional neural model is consistent with that of the target convolutional neural model, and will not be described again here. The initial convolutional neural model assigns weights W to each convolutional kernel in the convolutional layer and bias terms to the convolutional kernels based on the N sets of photovoltaic panel image samples. b f and the weight matrix corresponding to the scattering coefficients in formula (4) W β The bias term corresponding to the scattering coefficient b β And the weight matrices corresponding to the ambient light parameters in formulas (5) and (6). W A Indicates the bias term corresponding to the ambient light parameters. b A The model parameters are then used for training.

[0092] Specifically, after N sets of photovoltaic panel image samples are input into the initial convolutional neural model, the initial convolutional neural model outputs the predicted scattering coefficient and predicted ambient light parameters based on historical photovoltaic panel dust accumulation images. Thus, the model parameters are trained based on the predicted scattering coefficient and predicted ambient light parameters, as well as the historical true scattering coefficient and historical true ambient light parameters.

[0093] The loss function can be the mean squared error, as shown in formula (10): (10) in, Indicates the predicted scattering coefficient. y 1i Represents the historical true scattering coefficient, This indicates the predicted ambient light parameters. y 2i This represents the historical ambient light parameters.

[0094] In this embodiment, N sets of photovoltaic panel image samples are acquired, and an initial convolutional neural network model is trained based on these N sets of photovoltaic panel image samples to obtain a target convolutional neural network model. Each set of photovoltaic panel image samples includes: historical photovoltaic panel dust accumulation images, historical true scattering coefficients, and historical true ambient light parameters. The initial convolutional neural network model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence to train the initial convolutional neural network model. This establishes a mapping relationship between photovoltaic panel dust accumulation images and scattering coefficients and ambient light parameters in the target convolutional neural network model, which is equivalent to associating vision with physics through the convolutional neural network model. This allows the amount of dust accumulation on the photovoltaic panel to be calculated using an atmospheric scattering model, thereby improving the accuracy of photovoltaic panel dust accumulation detection.

[0095] Figure 3 This is a schematic diagram of a photovoltaic panel surface dust detection device provided in one embodiment of this application. Figure 3 As shown, the photovoltaic panel surface dust detection device includes: an acquisition module 301, a model processing module 302, and a detection module 303. Optionally, the photovoltaic panel surface dust detection device may further include: a training module 304.

[0096] The acquisition module 301 is used to acquire images of dust accumulation on the target photovoltaic panel. The model processing module 302 is used to input the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain target image features, wherein the target convolutional neural model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence; and is used to obtain the target scattering coefficient and target ambient light parameters based on the target image features and the target convolutional neural model. The detection module 303 is used to obtain the amount of dust accumulation on the target photovoltaic panel based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters using an atmospheric scattering model.

[0097] Optionally, the target convolutional neural model includes three convolutional layers. The model processing module 302 inputs the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain target image features, specifically for: The first convolutional layer is used to extract the first layer of features from the dust accumulation image of the photovoltaic panel to obtain a first feature map. The first layer of features includes: edge information and color changes. The first feature map is subjected to a second convolutional layer to extract second-layer features, and the second-layer features include: particle outline and texture information. The second feature map is subjected to a third convolutional layer to extract the third layer of features, thereby obtaining the target image features, which include: dust distribution pattern and dust density gradient.

[0098] Optionally, the detection module 303 utilizes an atmospheric scattering model to obtain the amount of dust accumulated on the target photovoltaic panel based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters. Specifically, it is used for: Based on the dust accumulation image of the photovoltaic panel, obtain the corresponding dust-free image of the photovoltaic panel; Obtain the target clean pixel value and the target dust-accumulated pixel value corresponding to the dust-free image of the photovoltaic panel and the dust-accumulated image of the photovoltaic panel, respectively. The dust accumulation thickness of the target photovoltaic panel is obtained based on the target mapping relationship and the target dust-free pixel value, the target dust-accumulated pixel value, the target scattering coefficient, and the target ambient light parameter. The target mapping relationship is obtained based on the atmospheric scattering model, which is the mapping relationship between the dust accumulation thickness and the dust-free pixel value, the dust-accumulated pixel value, the scattering coefficient, and the ambient light parameter. The amount of dust accumulated on the target photovoltaic panel is obtained based on the dust accumulation thickness and the surface area of ​​the target photovoltaic panel.

[0099] Optionally, the target mapping relationship is as follows:

[0100] Wherein, d(x, y) represents the dust thickness at coordinate (x, y) on the target photovoltaic panel, P(x, y) represents the dust-accumulated pixel value, P0(x, y) represents the dust-free pixel value, A represents the ambient light parameter, and β represents the scattering coefficient.

[0101] Optionally, before the model processing module 302 inputs the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain the target image features, the training module 304 is used for: Obtain N sets of photovoltaic panel image samples. Each set of photovoltaic panel image samples includes: historical photovoltaic panel dust accumulation image, historical true scattering coefficient, and historical true ambient light parameters. The initial convolutional neural model is trained based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural model. The initial convolutional neural model includes the preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence.

[0102] Optionally, the training module 304 trains the initial convolutional neural model based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural model, the parent model being used for: Based on the photovoltaic panel dust accumulation image samples corresponding to the N sets of photovoltaic panel image samples, the target image features of the samples are obtained; Based on the sample target image features and the initial convolutional neural model, the predicted scattering coefficient and predicted ambient light parameters are obtained; The initial convolutional neural model is trained based on the historical true scattering coefficient, the predicted scattering coefficient, the historical true ambient light parameters, the predicted ambient light parameters, and the loss function to obtain the target convolutional neural model.

[0103] The dust accumulation detection device on the surface of the photovoltaic panel provided in this application embodiment can be referred to the above method embodiment for its specific implementation process. The implementation principle and technical effect are similar, and will not be repeated here.

[0104] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may be a server, such as... Figure 4 As shown, the electronic device includes a processor 401 and a memory 402.

[0105] The memory 402 stores computer-executed instructions.

[0106] The processor 401 executes the computer execution instructions stored in the memory 402, causing the processor 401 to perform the method described in any of the above embodiments.

[0107] The electronic device provided in this application embodiment can be referred to the above method embodiment for its specific implementation process. The implementation principle and technical effect are similar, and will not be repeated here.

[0108] In the above Figure 4In the illustrated embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0109] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage.

[0110] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0111] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method shown in the above-described method embodiments.

[0112] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0113] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0114] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting dust accumulation on the surface of a photovoltaic panel, characterized in that, include: Obtain images of dust accumulation on the target photovoltaic panel; The dust accumulation image of the photovoltaic panel is input into the target convolutional neural model to obtain the target image features. The target convolutional neural model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence. Based on the target image features and the target convolutional neural model, the target scattering coefficient and target ambient light parameters are obtained; Using an atmospheric scattering model, the amount of dust accumulated on the target photovoltaic panel is obtained based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters.

2. The method according to claim 1, characterized in that, The target convolutional neural model includes three convolutional layers. The process of inputting the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain target image features includes: The first convolutional layer is used to extract the first layer of features from the dust accumulation image of the photovoltaic panel to obtain a first feature map. The first layer of features includes: edge information and color changes. The first feature map is subjected to a second convolutional layer to extract second-layer features, and the second-layer features include: particle outline and texture information. The second feature map is subjected to a third convolutional layer to extract the third layer of features, thereby obtaining the target image features, which include: dust distribution pattern and dust density gradient.

3. The method according to claim 1, characterized in that, The step of obtaining the dust accumulation amount of the target photovoltaic panel using an atmospheric scattering model based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters includes: Based on the dust accumulation image of the photovoltaic panel, obtain the corresponding dust-free image of the photovoltaic panel; Obtain the target clean pixel value and the target dust-accumulated pixel value corresponding to the dust-free image of the photovoltaic panel and the dust-accumulated image of the photovoltaic panel, respectively. The dust accumulation thickness of the target photovoltaic panel is obtained based on the target mapping relationship and the target dust-free pixel value, the target dust-accumulated pixel value, the target scattering coefficient, and the target ambient light parameter. The target mapping relationship is obtained based on the atmospheric scattering model, which is the mapping relationship between the dust accumulation thickness and the dust-free pixel value, the dust-accumulated pixel value, the scattering coefficient, and the ambient light parameter. The amount of dust accumulated on the target photovoltaic panel is obtained based on the dust accumulation thickness and the surface area of ​​the target photovoltaic panel.

4. The method according to claim 3, characterized in that, The target mapping relationship is as follows: ,in, d ( x , y ) represents the coordinates on the target photovoltaic panel ( x , y The thickness of dust accumulation at the location, P ( x , y ) represents the dust accumulation pixel value, P 0( x , y ) represents the dust-free pixel value, A This represents the ambient light parameters. β This represents the scattering coefficient.

5. The method according to any one of claims 1-4, characterized in that, Before inputting the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain the target image features, the method further includes: Obtain N sets of photovoltaic panel image samples. Each set of photovoltaic panel image samples includes: historical photovoltaic panel dust accumulation image, historical true scattering coefficient, and historical true ambient light parameters. The initial convolutional neural model is trained based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural model. The initial convolutional neural model includes the preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence.

6. The method according to claim 5, characterized in that, The step of training the initial convolutional neural network model based on the N sets of photovoltaic panel image samples to obtain the target convolutional neural network model includes: Based on the photovoltaic panel dust accumulation image samples corresponding to the N sets of photovoltaic panel image samples, the target image features of the samples are obtained; Based on the sample target image features and the initial convolutional neural model, the predicted scattering coefficient and predicted ambient light parameters are obtained; The initial convolutional neural model is trained based on the historical true scattering coefficient, the predicted scattering coefficient, the historical true ambient light parameters, the predicted ambient light parameters, and the loss function to obtain the target convolutional neural model.

7. A device for detecting dust accumulation on the surface of a photovoltaic panel, characterized in that, include: The acquisition module is used to acquire images of dust accumulation on the target photovoltaic panel. The model processing module is used to input the dust accumulation image of the photovoltaic panel into the target convolutional neural model to obtain target image features, wherein the target convolutional neural model includes a preset number of convolutional layers, and the number of convolutional kernels in adjacent convolutional layers increases in an equal-product sequence; and is used to obtain the target scattering coefficient and target ambient light parameters based on the target image features and the target convolutional neural model. The detection module is used to obtain the amount of dust accumulation on the target photovoltaic panel based on the dust accumulation image of the photovoltaic panel, the target scattering coefficient, and the target ambient light parameters using an atmospheric scattering model.

8. An electronic device, characterized in that, include: Processor and memory; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images