Photovoltaic panel cleaning detection method and system based on artificial intelligence and storage medium

Abstract

The invention discloses a photovoltaic panel cleaning detection method and system based on artificial intelligence and a storage medium. The method comprises the following steps: acquiring photovoltaic panel images according to a preset sampling interval, and splicing the photovoltaic panel images by using an image splicing algorithm based on an enhanced KAZE algorithm; constructing a photovoltaic panel cleaning detection model based on a BP neural network; and inputting the spliced photovoltaic panel image into a photovoltaic panel cleaning detection model based on a BP neural network to obtain a photovoltaic panel cleaning detection result. According to the invention, artificial intelligence and an image processing technology are combined, so that the surface pollutant characteristics of the photovoltaic panel can be accurately captured, local detection omission is avoided, and compared with manual inspection, subjective errors are greatly reduced, and the detection efficiency is remarkably improved. And whether cleaning is needed or not can be judged according to the actual pollution degree, invalid operation and resource waste of regular cleaning are avoided, unnecessary equipment loss and labor cost are reduced, and the photovoltaic power station is assisted to achieve on-demand operation and maintenance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, more particularly to a photovoltaic panel cleaning detection method and system based on artificial intelligence and a storage medium. BACKGROUND

[0002] Currently, in the long-term operation of photovoltaic panels, dust, bird droppings, leaves and other pollutants are easily attached to the surface, resulting in a decrease in light transmittance, which in turn reduces the power generation efficiency of the photovoltaic system. When pollution is severe, the efficiency can be reduced by 20-30%. Current photovoltaic panel cleaning mainly uses two methods: timed cleaning and manual inspection followed by cleaning. These methods have obvious defects: The timed cleaning mode does not take into account the actual pollution level, which may result in ineffective cleaning when the pollution is light or not timely cleaning when the pollution is severe. Manual inspection relies on the experience of personnel and is highly subjective and inefficient, and it is difficult to meet the real-time monitoring needs of large-scale photovoltaic power stations. With the rapid development of artificial intelligence and image processing technology, the ability to acquire, process and analyze large-scale image data can be significantly improved. Advanced image processing algorithms can extract features and recognize patterns from complex images, providing a technical basis for photovoltaic panel cleaning prediction.

[0003] Therefore, how to detect whether the photovoltaic panel needs to be cleaned based on artificial intelligence and image processing technology, and thereby improve the overall operation and maintenance efficiency of the photovoltaic system is a problem that needs to be solved by those skilled in the art. SUMMARY

[0004] Therefore, the present application provides a photovoltaic panel cleaning detection method and system based on artificial intelligence and a storage medium to solve the problems in the background art.

[0005] To achieve the above purpose, the present application adopts the following technical solutions: A photovoltaic panel cleaning detection method based on artificial intelligence, comprising: acquiring a photovoltaic panel image at a preset sampling interval, the photovoltaic panel image including an infrared thermal imaging image and a visible light image; splicing the photovoltaic panel image using an image splicing algorithm based on the enhanced KAZE algorithm; constructing a photovoltaic panel cleaning detection model based on a BP neural network; inputting the spliced photovoltaic panel image into the photovoltaic panel cleaning detection model based on the BP neural network to obtain a photovoltaic panel cleaning detection result.

[0006] Optionally, the acquiring of the photovoltaic panel image at the preset sampling interval comprises: The infrared thermal imaging camera and the visible light camera are mounted on a photovoltaic module inspection aircraft or a mobile photovoltaic module detection device, and infrared thermal imaging images and visible light images are synchronously collected.

[0007] Optionally, the collected photovoltaic panel image is preprocessed, including image graying, image filtering, image enhancement and image edge detection, and the specific steps are as follows: After the graying processing of the collected color image, an image P gray is obtained. P gray = 0.30R + 0.59G + 0.11B. Wherein, R represents the pixel value of the red component in the color image, G represents the pixel value of the green component in the color image, and B represents the pixel value of the blue component in the color image. The image is filtered by using a discrete Gaussian filter function, the weighted average of the image is performed, the weighted average of the pixel neighborhood is used to replace the gray value of the center of the Gaussian template, and the discrete Gaussian filter function is: ; Wherein, represents the coordinates of a point in the neighborhood, δ represents the standard deviation. The image is enhanced by changing the pixel gray value of the image, and the pixel value of the processed image is g(x, y), which is represented as: g(x,y)=[ f ( x , y )] 2 ; Wherein, f (x,y) represents the pixel value of the image at (x, y) after the image graying and image filtering processing, and the image gray range is [0, 255]. If the calculated result g(x, y) exceeds 255, it is set to 255. The Canny detection operator is selected for image edge detection: first, the convolution operation is performed on the Gaussian mask and the image after the image graying and image filtering processing, the information of a single pixel is unchanged, then the gradient amplitude and direction are calculated by using the first-order partial difference, and finally the double threshold method is used to detect and connect the image edge, and the gradient amplitude and direction are respectively represented as: ; Wherein, , Partial derivatives of the image gray scale in x, y directions, respectively.

[0008] Optionally, the image splicing algorithm based on the enhanced KAZE algorithm is used to splice the photovoltaic panel images, and specifically includes: constructing a nonlinear scale space; feature point detection and positioning; feature vector description; and feature vector matching. Optionally, the photovoltaic panel cleaning detection model based on the BP neural network sets the number of input layer nodes n, the number of hidden layer nodes l and the number of output layer nodes m, sets the weight from the input layer to the hidden layer as ω ij , the weight from the hidden layer to the output layer as ω jk , the threshold value from the input layer to the hidden layer as a j , the threshold value from the hidden layer to the output layer as b k , the learning rate as η , and the excitation function as g(x).

[0009] Optionally, the excitation function g(x) adopts a Sigmoid function, and is expressed as: ; wherein x is an input matrix.

[0010] Optionally, the training step of the photovoltaic panel cleaning detection model based on the BP neural network specifically includes: a: inputting the spliced photovoltaic panel images into the BP neural network as training samples; b: judging whether the training samples are loaded completely, if yes, executing the next step, and if not, executing step a; c: setting the output of the hidden layer as H j , and calculating the output of the hidden layer neuron: ; wherein n is the number of input layer nodes, is the weight from the input layer to the hidden layer, is an input matrix, is the threshold value from the input layer to the hidden layer; d: setting the output of the output layer as O k , and calculating the output of the output layer neuron: ; wherein l is the number of hidden layer nodes, is the weight from the hidden layer to the output layer, is the threshold value from the hidden layer to the output layer; e: calculating the error: ; wherein, is an error, m is the number of output layer nodes, is an expected output, is the output of the output layer; f: weight update: ; g: threshold value update: ; h: judge whether the difference between the two adjacent errors is less than a set value, if less than the set value, the BP neural network training is ended, if not less than the set value, steps c-g are executed cyclically.

[0011] Optionally, the BP neural network-based photovoltaic panel cleaning detection model is evaluated, and the evaluation index is accuracy, and the calculation formula is as follows: ; In the formula, Acc is the accuracy, N is the total number of samples, and n is the total number of detection result classifications.

[0012] A photovoltaic panel cleaning detection system based on artificial intelligence comprises: An image acquisition module acquires photovoltaic panel images at a preset sampling interval, and the photovoltaic panel images comprise infrared thermal imaging images and visible light images; An image stitching module stitches the photovoltaic panel images by using an image stitching algorithm based on the enhanced KAZE algorithm; A model construction module constructs a BP neural network-based photovoltaic panel cleaning detection model; A result output module inputs the stitched photovoltaic panel images into the BP neural network-based photovoltaic panel cleaning detection model to obtain photovoltaic panel cleaning detection results. A storage medium stores instructions, which, when executed by one or more processors of a machine, cause the processors to perform a photovoltaic panel cleaning detection method based on artificial intelligence.

[0013] According to the above technical solution, compared with the prior art, the present application provides a photovoltaic panel cleaning detection method, system and storage medium based on artificial intelligence, which effectively solves the pain points of traditional photovoltaic panel cleaning detection by combining artificial intelligence and image processing technology, and has multi-dimensional practical value.

[0014] 1. Improve detection accuracy and efficiency: Simultaneously acquire infrared thermal imaging and visible light images, and combine them with preprocessing steps such as grayscale conversion and Gaussian filtering to accurately capture the characteristics of contaminants on the surface of photovoltaic panels; Image stitching based on the enhanced KAZE algorithm can completely present the large-scale photovoltaic panel image, avoiding local detection omissions. Compared with manual inspection, it significantly reduces subjective errors and improves detection efficiency significantly.

[0015] 2. Optimize cleaning strategies and reduce operation and maintenance costs. Relying on the BP neural network model to achieve intelligent detection, it can determine whether cleaning is needed based on the actual degree of pollution, avoiding ineffective operations and resource waste from scheduled cleaning, reducing unnecessary equipment wear and labor costs, and helping photovoltaic power plants achieve on-demand operation and maintenance.

[0016] 3. Adapting to the needs of large-scale power plants and enhancing practicality, the image acquisition equipment can be mounted on inspection aircraft or mobile devices to adapt to photovoltaic power plants of different scales; the model is continuously optimized through accuracy indicators to ensure detection stability, providing reliable technical support for the efficient operation and maintenance of large-scale photovoltaic systems, and indirectly improving the overall power generation efficiency of photovoltaic power plants. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the method flow provided by the present invention. Detailed Implementation

[0019] 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.

[0020] This invention discloses an artificial intelligence-based method for cleaning and inspecting photovoltaic panels, such as... Figure 1 As shown, it includes: Photovoltaic panel images are acquired at preset sampling intervals, including infrared thermal imaging images and visible light images; A photovoltaic panel image stitching algorithm based on the enhanced KAZE algorithm was used to stitch together images. Construct a photovoltaic panel cleaning detection model based on BP neural network; The stitched photovoltaic panel image is input into a photovoltaic panel cleaning detection model based on a backpropagation neural network to obtain the photovoltaic panel cleaning detection results.

[0021] In one specific embodiment, acquiring photovoltaic panel images according to a preset sampling interval specifically includes: Infrared thermal imaging cameras and visible light cameras are used, both mounted on photovoltaic module inspection aircraft or mobile photovoltaic module inspection devices, to simultaneously acquire infrared thermal imaging images and visible light images.

[0022] In one specific embodiment, the method further includes preprocessing the acquired photovoltaic panel images. This preprocessing can reduce the significant impact of weather and changes in lighting angle on photovoltaic panel identification, improve anti-interference capabilities, highlight photovoltaic panel information, quickly and accurately detect foreign objects on the photovoltaic panel, and improve the image. The preprocessing includes image grayscale conversion, image filtering, image enhancement, and image edge detection. The specific steps are as follows: After converting the acquired color image to grayscale, the image is obtained. P gray , represented as: P gray =0.30R + 0.59G + 0.11B; Where R represents the pixel value of the red component in the color image, G represents the pixel value of the green component in the color image, and B represents the pixel value of the blue component in the color image. To improve image quality and mitigate image quality issues caused by noise interference, Gaussian filtering is used for image denoising. This embodiment employs 9... A Gaussian template scans each pixel in the image. A Gaussian filter replaces the gray value at the center of the Gaussian template with the weighted mean of the pixel's neighborhood. Since the weight of each neighboring pixel decreases with its distance from the center point, this makes the image smoother. Discrete Gaussian filter function. for: ; in, Represents the coordinates of a point within the neighborhood. δ Indicates standard deviation; To improve image sharpness and contrast, the image effect is enhanced by directly changing the grayscale values ​​of image pixels, i.e., contrast stretching. Image enhancement is achieved by changing the grayscale values ​​of image pixels. The processed image pixel value is g(x,y), which is represented as: g(x,y)=[ f ( x , y )] 2 ; in,f (x, y) represents the pixel value at point (x, y) of the image after image grayscale conversion and image filtering. The image grayscale range is [0, 255]. If the calculated result g(x, y) exceeds 255, it is set to 255. The Canny detection operator is selected for image edge detection: First, a Gaussian mask is used to convolve the image after grayscale conversion and filtering, keeping the information of individual pixels unchanged. Then, the magnitude and direction of the gradient are calculated using the first-order partial derivative difference. Next, non-maximum suppression is performed using the gradient magnitude. Finally, a double thresholding method is used to detect and connect image edges. The magnitude and direction of the gradient are expressed as follows: ; in, , These represent the partial derivatives of the image grayscale values ​​in the x and y directions, respectively.

[0023] In one specific embodiment, an image stitching algorithm based on the enhanced KAZE algorithm is used to stitch together photovoltaic panel images, specifically including: constructing a nonlinear scale space; feature point detection and localization; feature vector description; and feature vector matching.

[0024] The specific method for constructing a nonlinear scale space is as follows: constructing a nonlinear scale space through a variable conduction diffusion method.

[0025] The specific method for feature point detection and localization is as follows: In different scale spaces, to determine the position and scale corresponding to each feature point, each point is compared with points in its neighborhood to obtain the local maxima of the normalized matrix. After obtaining the feature point positions, the precise sub-pixel positions are solved using the Taylor expansion.

[0026] The specific method for describing the feature vector is as follows: construct a feature vector for each feature point with a determined position and main direction, and take a rectangular window on the gradient image with each feature point as the center, and divide and weight it.

[0027] The specific method for matching feature vectors is to use the Euclidean distance between two feature vectors for matching.

[0028] In one specific embodiment, the photovoltaic panel cleaning detection model based on a BP neural network sets the number of input layer nodes n and the number of hidden layer nodes. l Given the number of output layers m, and the weights from the input layers to the hidden layers as follows: ω ij The weights from the hidden layer to the output layer are: ω jk The threshold from the input layer to the hidden layer is a j The threshold from the hidden layer to the output layer isb k The learning rate is η The activation function is g(x).

[0029] This example categorizes the foreign objects that need to be cleaned from the photovoltaic panel into three types: dust, bird droppings, and leaves. Therefore, the number of input nodes is set to 20, and the number of output nodes is set to 3. The selection of hidden layer nodes is determined according to the following formula:

[0030] Where n is the number of hidden layer nodes; Input the number of nodes; This will output the number of nodes. a Take a constant between 1 and 10.

[0031] In one specific embodiment, the activation function g(x) adopts the Sigmoid function, expressed as: ; Where x is the input matrix.

[0032] In a specific embodiment, the training steps of the photovoltaic panel cleaning detection model based on BP neural network specifically include: a: Input training samples: Input the stitched photovoltaic panel image into the BP neural network; b: Determine whether the training samples have been fully loaded. If they have been fully loaded, proceed to the next step. If they have not been fully loaded, proceed to step a. c: Let the output of the hidden layer be... H j Calculate the output of the hidden layer neurons: ; Where n is the number of nodes in the input layer. The weights from the input layer to the hidden layer. For the input matrix, The threshold value from the input layer to the hidden layer; d: Let the output of the output layer be... O k Calculate the output of the neurons in the output layer: ; Where l is the number of hidden layer nodes. The weights from the hidden layer to the output layer. The threshold value from the hidden layer to the output layer; e: Calculation error: ; in, The value is the error, and m is the number of nodes in the output layer. For the desired output, For the output of the output layer; f: Weight update: ; g: Threshold update: ; h: Determine whether the difference between two adjacent errors is less than a set value. If it is less than the set value, the BP neural network training ends. If it is not less than the set value, then repeat step cg.

[0033] In one specific embodiment, the evaluation also includes assessing the photovoltaic panel cleaning detection model based on a BP neural network. The evaluation metric is accuracy, calculated using the following formula: ; In the formula, Acc is the accuracy, N is the total number of samples, and n is the total number of classifications of the detection results.

[0034] An artificial intelligence-based photovoltaic panel cleaning and inspection system includes: The image acquisition module acquires photovoltaic panel images according to a preset sampling interval. The photovoltaic panel images include infrared thermal imaging images and visible light images. The image stitching module uses an image stitching algorithm based on the enhanced KAZE algorithm to stitch together images of photovoltaic panels; The model building module constructs a photovoltaic panel cleaning detection model based on a BP neural network. The output module inputs the stitched photovoltaic panel image into a photovoltaic panel cleaning detection model based on a backpropagation neural network to obtain the photovoltaic panel cleaning detection results. A readable storage medium stores instructions that, when executed by one or more processors of a machine, cause the processor to perform an artificial intelligence-based photovoltaic panel cleaning detection method.

[0035] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0036] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting cleaning of a photovoltaic panel based on artificial intelligence, characterized in that, Comprise: According to the preset sampling interval, the photovoltaic panel image is obtained, and the photovoltaic panel image includes infrared thermal imaging image and visible light image; The photovoltaic panel image is spliced by using the image splicing algorithm based on the reinforced KAZE algorithm; A photovoltaic panel cleaning detection model based on BP neural network is constructed; The spliced photovoltaic panel image is input into the photovoltaic panel cleaning detection model based on BP neural network to obtain the photovoltaic panel cleaning detection result.

2. A method for detecting cleaning of a photovoltaic panel based on artificial intelligence according to claim 1, characterized in that, The photovoltaic panel image obtained according to the preset sampling interval specifically comprises: An infrared thermal imaging camera and a visible light camera are used, and the infrared thermal imaging camera and the visible light camera are mounted on a photovoltaic module inspection aircraft or a mobile photovoltaic module detection device to synchronously collect infrared thermal imaging images and visible light images.

3. The method of claim 2, wherein the method further comprises: Further comprising pre-processing the collected photovoltaic panel image, including image graying, image filtering, image enhancement and image edge detection, the specific steps are: After the collected color image is processed to be gray, the image P gray is represented as: P gray = 0.30R + 0.59G + 0.11B; Wherein, R represents the pixel value of the red component in the color image, G represents the pixel value of the green component in the color image, and B represents the pixel value of the blue component in the color image; The image is filtered by using a discrete Gaussian filter function, weighted average is performed on the image, a Gaussian template is used to scan each pixel point in the image, a weighted average value of a pixel neighborhood is used to replace a gray value of a center of the Gaussian template, and the discrete Gaussian filter function is as follows: G(x,y) = exp(-(x2+y2) / 2σ2) ； wherein denotes the coordinates of a point in the neighborhood, Delta denotes the standard deviation; The image pixel gray value is changed for image enhancement, and the processed image pixel value is g(x, y), which is represented as: g(x,y) = [ f ( x , y )] 2 ; wherein, f (x,y) represents the pixel value of the image after image graying and image filtering processing at the point (x,y), the image grayscale range is [0, 255], and if the calculated result g(x,y) exceeds 255, it is set to 255; The Canny detection operator is selected for image edge detection: first, the Gaussian mask and the image processed by image graying and image filtering are convolved, the information of a single pixel remains unchanged, then the gradient amplitude and direction are calculated by first-order partial derivative difference, and finally the gradient amplitude is used to suppress the non-maximum value, and the double threshold method is used to detect and connect the image edge, the gradient amplitude and direction are represented as: ； wherein , respectively represent partial derivatives of the image gray scale in the x, y directions.

4. The method for detecting cleaning of a photovoltaic panel based on artificial intelligence according to claim 1, characterized in that, The photovoltaic panel image is spliced by using the image splicing algorithm based on the reinforced KAZE algorithm, specifically comprising: constructing a nonlinear scale space; feature point detection and positioning; feature vector description; and feature vector matching.

5. The method of claim 1, wherein the method is based on artificial intelligence. The photovoltaic panel cleaning detection model based on BP neural network sets the number of input layer nodes n and the number of hidden layer nodes. l Given the number of output layers m, and the weights from the input layers to the hidden layers as follows: Omega ij The weights from the hidden layer to the output layer are: Omega jk The threshold from the input layer to the hidden layer is a j The threshold from the hidden layer to the output layer is b k The learning rate is Eta The activation function is g(x).

6. The method for detecting cleaning of a photovoltaic panel based on artificial intelligence according to claim 5, characterized in that, The activation function g(x) adopts Sigmoid function, which is represented as: ； Wherein, x is the input matrix.

7. The method of claim 6, wherein the method further comprises: The training steps of the photovoltaic panel cleaning detection model based on BP neural network specifically comprise: a: input training sample: input the spliced photovoltaic panel image into the BP neural network; b: judge whether the training sample is loaded, if loaded, execute the next step, if not loaded, execute step a; c: set the output of the hidden layer to H j , compute the output of the hidden layer neurons: ； wherein n is the number of input layer nodes, is the weight from the input layer to the hidden layer, is the input matrix, is the threshold value from the input layer to the hidden layer; d: set the output of the output layer to O k , compute the output of the output layer neurons: ； wherein, l is the number of nodes in the hidden layer, W is the weight from the hidden layer to the output layer, T is the threshold from the hidden layer to the output layer; e: calculate error: ； wherein, is the error, m is the number of output layer nodes, is the desired output, is the output of the output layer; f: weight update: ； g: threshold update: ； h: judge whether the difference between the adjacent two errors is less than the set value, if less than the set value, the BP neural network training is finished, if not less than the set value, the steps c-g are executed in a loop.

8. The method of claim 1, wherein the method is based on artificial intelligence. Further comprising evaluating the photovoltaic panel cleaning detection model based on BP neural network, and the evaluation index is accuracy, and the calculation formula is as follows: ； In the formula, Acc is the accuracy, N is the total number of samples, and n is the total number of detection result classifications.

9. An artificial intelligence based photovoltaic panel cleaning detection system characterized in that, The application of any one of claims 1-8 comprises: An image acquisition module acquires photovoltaic panel images according to a preset sampling interval, and the photovoltaic panel images include infrared thermal imaging images and visible light images; An image splicing module splices the photovoltaic panel images by using an image splicing algorithm based on a reinforced KAZE algorithm; A model construction module constructs a photovoltaic panel cleaning detection model based on a BP neural network. A result output module inputs the spliced photovoltaic panel image into the photovoltaic panel cleaning detection model based on the BP neural network to obtain a photovoltaic panel cleaning detection result.

10. A readable storage medium storing instructions, characterized in that, The instructions, when executed by one or more processors of a machine, cause the processors to perform a photovoltaic panel cleaning detection method based on artificial intelligence according to any one of claims 1-8.

Images