A method and system for predicting the snow melting time of a photovoltaic panel based on machine vision

Abstract

The present invention relates to a method and system for predicting snowmelt time of photovoltaic panels based on machine vision, the method comprising: using a drone equipped with a high-definition camera, regularly collecting images of photovoltaic panels without any coverage in the photovoltaic panel area in winter according to a preset flight path, and collecting images of photovoltaic panels covered with snow in the photovoltaic panel area in winter snowy weather; performing data enhancement preprocessing on the images of photovoltaic panels covered with snow, using them as training image data sets, building a YOLO model to detect whether there is snow on the photovoltaic panels, and if so, calculating the snow coverage rate and snow thickness; according to the actual snow thickness calculated, combined with the snow density measured on the spot by a snow analyzer, obtaining the snow water equivalent, and then combining the snow coverage rate to obtain the actual snowmelt amount of the snow area, and obtaining the snowmelt time based on the actual snowmelt amount and degree-day factor model. The present invention can timely and accurately predict the snowmelt time, ensure the normal operation of photovoltaic power stations and improve power generation efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy, and particularly relates to a method and system for predicting the snow melting time of a photovoltaic panel based on machine vision. Background Art

[0002] In the prior art, there are related technologies for snow cover detection and simulation modeling of the snow melting process, but there is a lack of discussion on local snow cover detection, snow melting rate, and snow melting time. The specific technical defects are as follows:

[0003] The snow cover state is affected by various factors, such as light changes, different weather conditions, etc. In the face of different environmental conditions, the Naive Bayes classifier may need to be retrained or adjusted to adapt to these changes, resulting in poor environmental adaptability and affected real-time performance. At the same time, the performance of the Naive Bayes classifier depends to a large extent on the quality and quantity of the training data, and the accuracy is insufficient when dealing with complex image data.

[0004] Traditional image processing methods perform well on specific data sets, but have poor generalization ability when dealing with unseen data or environmental changes, which affects the detection accuracy. In the case of complex backgrounds or multiple targets, traditional methods may be difficult to accurately segment and identify the targets, especially when the target is similar to the background or there is occlusion. Traditional image processing methods do not have the ability to learn from data, so it is difficult to adapt to the changing external environment.

[0005] The size of the photovoltaic panel is relatively small, while remote sensing images usually cover a large area, and the image acquisition may not be carried out in real time, resulting in unclear details of the photovoltaic panel, which limits the monitoring ability of the dynamic changes of the snow cover. At the same time, it may be difficult to accurately estimate the thickness of the snow cover in remote sensing images, and different thicknesses will significantly affect the snow melting rate and the power generation efficiency of the photovoltaic panel. Atmospheric conditions (such as fog, clouds, smoke, etc.) will also affect the quality of remote sensing images, thereby affecting the accuracy of snow cover detection. Frequent acquisition and processing of remote sensing images also involve high costs and resource consumption, especially in the case of high-resolution and high-frequency monitoring.

[0006] Most of the prior art constructs snow melting models for remote sensing images, lacking discussions on the snow melting rate and time of local areas. Summary of the Invention

[0007] The object of the present invention is to provide a method and system for predicting the snow melting time of a photovoltaic panel based on machine vision. The detection of snow accumulation on the photovoltaic panel is realized through an image processing unit. Through the calculation method of the snow accumulation thickness on the photovoltaic panel, the detailed detection ability of the photovoltaic panel is improved, and the implementation difficulty and implementation cost are reduced. By calculating the actual snow melting amount of the snow accumulation area and calculating the snow melting amount per unit time through a temperature-dominated degree-day factor model, the snow melting scenario of a local area is simulated, and the time required for the final local area to melt snow is obtained.

[0008] The present invention provides a method for predicting the snow melting time of a photovoltaic panel based on machine vision, including the following steps:

[0009] Step 1, use a drone equipped with a high-definition camera to regularly collect images of the photovoltaic panel without any coverage in the photovoltaic panel area according to a preset flight path in winter, and collect images of the photovoltaic panel covered with snow in the photovoltaic panel area in snowy weather in winter, and correspond the images of the photovoltaic panel without any coverage with the images of the photovoltaic panel covered with snow one by one;

[0010] Step 2, perform data augmentation preprocessing on the images of the photovoltaic panel covered with snow, regard it as a training image data set, build a YOLO model to detect whether there is snow on the photovoltaic panel. If so, calculate the snow coverage rate and the snow accumulation thickness;

[0011] Step 3, according to the calculated actual snow accumulation thickness, combine with the snow density measured on-site by a snow accumulation analyzer to obtain the snow water equivalent, and then combine with the snow coverage rate to obtain the actual snow melting amount of the snow accumulation area, and obtain the snow melting time based on the actual snow melting amount and the degree-day factor model.

[0012] Further, the data augmentation preprocessing of the images of the photovoltaic panel covered with snow in step 2, regarded as a training image data set, and building a YOLO model to detect whether there is snow on the photovoltaic panel includes:

[0013] 1) Data set production

[0014] Collect the images required for training, and use rotation, flipping, and contrast enhancement methods to perform data augmentation as the training image data set;

[0015] Use labelme to label the photovoltaic panel area and the photovoltaic panel snow accumulation area of each image in the training set, determine its position through a bounding box, divide it into snow accumulation classes and photovoltaic panel classes, and at the same time preprocess the images, adjust the image size to a unified size and perform normalization, use a Gaussian filter to perform convolution operations and weight distribution to remove the noise in the image, and perform convolution operations on the convolved image with Sobel operators in the horizontal and vertical directions respectively;

[0016] Calculate the gradient magnitude of the convolution result, and use the non-maximum suppression method to perform local maximum detection on each pixel of the gradient image to remove redundancy;

[0017] 2) Object detection

[0018] Perform image prediction based on the YOLO model. If the drone transmits an image file and both snow class and photovoltaic panel class labels exist in the image prediction result, then transmit the image and the corresponding image of the uncovered photovoltaic panel to the snow parameter calculation module for processing. If only the snow class or the photovoltaic panel class label exists in the image prediction result, then end the operation; if the drone transmits a video file, extract the images from the video stream according to the set frame rate. If both snow class and photovoltaic panel class labels exist in the image prediction result, then transmit the image and the corresponding image of the uncovered photovoltaic panel to the snow parameter calculation module for processing. If only the snow class or the photovoltaic panel class label exists in the image prediction result, then continue to process the next frame of the image.

[0019] Furthermore, the images required for collection and training include photovoltaic panel images covered with snow under different weather conditions, different lighting conditions, and different environments.

[0020] Furthermore, the performing image prediction based on the YOLO model includes:

[0021] Based on the YOLO model, divide the labeled dataset into small batches for training and obtain the weight file; when processing the input image, first perform the same preprocessing operation on the image transmitted by the drone, and then use the YOLO model to divide the image into grids. For the photovoltaic panel target image whose center point falls within the grid, each grid predicts multiple bounding boxes of the target and the confidence of the bounding boxes. The bounding boxes have different shapes and ratios to adapt to the sizes of different targets. The confidence represents the probability that the bounding box contains the target and the matching degree between the predicted box and the actual box. At the same time, predict the conditional probability that the target matches the snow class or the photovoltaic panel class. After the network predicts multiple bounding boxes, remove the overlapping bounding boxes and only retain the best prediction result. Finally, what YOLO outputs are the position, size, confidence of each bounding box, and the probability of the matching class with the existing label.

[0022] Furthermore, the calculation method of the snow coverage rate includes:

[0023] Respectively calculate the number of pixel blocks occupied by the snow class label box and the number of pixel blocks occupied by the photovoltaic panel class label box in the snow-covered photovoltaic panel image, and calculate the ratio of the two, which is the coverage rate of the snow on the photovoltaic panel.

[0024] Furthermore, the calculation method of the snow depth includes:

[0025] Perform edge detection on the two sets of images transmitted, divide edge pixels into strong edges and weak edges through double threshold processing, compare the calculated image gradient amplitude with the threshold, and form a complete edge by connecting adjacent strong edge pixels;

[0026] The edge length and boundary point coordinates of the photovoltaic panel without snow in the image and the boundary point coordinates of the photovoltaic panel covered with snow in the image are recorded, and the proportionality coefficient is calculated using the ratio of the edge length to the actual length of the photovoltaic panel obtained by field measurement.

[0027] By comparing the coordinates of the boundary points before and after snow accumulation on the photovoltaic panel, the thickness of snow at each point on the upper and lower edges of the photovoltaic panel is obtained, and the actual thickness of snow on the photovoltaic panel is obtained based on the proportional coefficient.

[0028] Furthermore, the step 3 comprises:

[0029] The actual snowmelt amount is calculated by the following formula:

[0030] M=SWE*F=H*ρ*F

[0031] Among them, M is the actual snowmelt, H is the snow thickness, ρ is the snow density, and F is the snow cover rate;

[0032] At the same time, a simple linear regression model was constructed. The long-term series temperature and the corresponding snowmelt data were linearly fitted by the least squares method, and a degree-day factor model dominated by temperature was built. The temperature was used as the independent variable and the snowmelt was used as the dependent variable to obtain the following snowmelt model:

[0033]

[0034] Where SM is the amount of snowmelt, T a is the outside temperature, T r is the critical temperature for snow melting;

[0035] The snowmelt volume per unit time is calculated based on the real-time temperature of the photovoltaic array at a certain moment. Combined with the actual snowmelt volume in the snow-covered area, the actual snowmelt time t of the snow-covered area is obtained. The formula is as follows:

[0036]

[0037] The present invention also provides a photovoltaic panel snowmelt time prediction system based on machine vision, comprising a snowmelt time prediction module, and the snowmelt time prediction module executes the photovoltaic panel snowmelt time prediction method based on machine vision.

[0038] The present invention also provides a non-transitory computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the method for predicting the snow melting time of a photovoltaic panel based on machine vision.

[0039] The present invention also provides an electronic device, including:

[0040] a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to implement the method for predicting the snow melting time of a photovoltaic panel based on machine vision.

[0041] By means of the above solution, the method and system for predicting the snow melting time of a photovoltaic panel based on machine vision have the following technical effects:

[0042] 1) In winter or low-temperature regions, snow accumulation will cover the photovoltaic panel, seriously affecting its photoelectric conversion efficiency and service life. By building a local snow melting model, the present invention can accurately predict the snow melting time in a timely manner, ensuring the normal operation of the photovoltaic power station and improving the power generation efficiency.

[0043] 2) Based on machine vision technology, accurate measurement, judgment, and prediction of the snow accumulation situation and snow melting time of the photovoltaic panel are carried out. That is, through a high-resolution image acquisition device, combined with a deep learning algorithm, the snow accumulation situation on the photovoltaic panel is monitored in real time, and the snow melting time is predicted by analyzing the image data. Compared with traditional manual observation or simple temperature sensors, this method can provide more accurate and real-time data, thus improving the accuracy of prediction.

[0044] 3) By combining the machine vision system with an automated snow removal device, intelligent maintenance of the photovoltaic panel can also be realized.

[0045] The above description is only an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and implement it according to the content of the specification, the following describes the preferred embodiments of the present invention in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flowchart of the method for predicting the snow melting time of a photovoltaic panel based on machine vision according to the present invention;

[0047] Figure 2 is a schematic structural diagram of the electronic device according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The following further describes the specific embodiments of the present invention in detail with reference to the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention but not to limit the scope of the present invention.

[0049] As shown Figure 1 in the figure, this embodiment provides a method for predicting the snow melting time of a photovoltaic panel based on machine vision, which is characterized by including the following steps:

[0050] Step 1: Use a drone equipped with a high-definition camera to regularly collect images of the photovoltaic panel area without any coverage in winter according to a preset flight path, collect images of the photovoltaic panel covered with snow in the snow weather in winter, and correspond the images of the photovoltaic panel without any coverage with the images of the photovoltaic panel covered with snow one by one;

[0051] Step 2: Perform data augmentation preprocessing on the images of the photovoltaic panel covered with snow, use them as the image dataset for training, build a YOLO model to detect whether there is snow on the photovoltaic panel. If so, calculate the snow coverage rate and the snow thickness;

[0052] Step 3: According to the calculated actual snow thickness, combine with the snow density measured on-site by a snow analyzer to obtain the snow water equivalent, and then combine with the snow coverage rate to obtain the actual snow melting amount of the snow-covered area. Based on the actual snow melting amount and the degree-day factor model, obtain the snow melting time.

[0053] The following further elaborates on the present invention in detail:

[0054] 1. Data collection

[0055] Equip a high-definition camera on the drone, preset the flight path of the drone, regularly collect images of the photovoltaic panel area in winter, obtain the images of the photovoltaic panel without any coverage as reference images and store them in the computer. Obtain the images of the photovoltaic panel covered with snow in the snow weather, and correspond the front and back images one by one. The images and videos taken by the drone are transmitted to the ground station through wireless communication and processed by the image processing unit.

[0056] 2. Image processing

[0057] The image processing unit is divided into an object detection module and a snow parameter calculation module. The object detection module is divided into two steps: dataset production and object detection. Dataset production requires collecting the images needed for training, which should include images of photovoltaic panels covered with snow under different weather conditions, different lighting conditions, and different environments. Data augmentation is performed using methods such as rotation, flipping, and contrast enhancement to obtain the image dataset for training. Labelme is used to annotate the photovoltaic panel area and the snow-covered area of the photovoltaic panel in each image of the training set, and determine its position (through bounding boxes), specifically divided into two categories: "snow" and "panel", where "snow" refers to the snow class existing on the photovoltaic panel, and "panel" refers to the photovoltaic panel class. At the same time, the images are preprocessed, the image size is adjusted to a unified size and normalized, a Gaussian filter is used for convolution operation and weight assignment to remove the noise in the image. For the image after convolution, convolution operations are performed using Sobel operators in the horizontal and vertical directions respectively, and then the gradient magnitude of the convolution result is calculated. Finally, the non-maximum suppression method is used to detect the local maximum value of each pixel point in the gradient image to remove redundancy.

[0058] Object detection requires building a YOLO model, dividing the labeled dataset into small batches for training, and obtaining a weight file. When processing the input image, first perform the same preprocessing operations on the image transmitted by the drone. Then, use the YOLO model to divide the image into grids. For the photovoltaic panel target image whose center point falls within the grid, each grid predicts multiple bounding boxes of the target and the confidence levels of these bounding boxes. These bounding boxes have different shapes and proportions to adapt to the sizes of different targets. The confidence level represents the probability that the bounding box contains the target and the matching degree between the predicted box and the actual box. At the same time, the conditional probability of the target matching "snow" or "panel" is predicted. After the network predicts multiple bounding boxes, the overlapping bounding boxes are removed, and only the best prediction result is retained. Finally, what YOLO outputs is the position, size, confidence level of each bounding box, and the probability of matching the existing label category.

[0059] If the drone transmits an image file and both "snow" and "panel" labels exist in the image prediction result, then the image and the corresponding image of the uncovered photovoltaic panel are transmitted to the snow parameter calculation module for processing. If only the "snow" or "panel" label exists in the image prediction result, then the operation ends; if the drone transmits a video file, then extract the images in the video stream according to the set frame rate. If both "snow" and "panel" labels exist in the image prediction result, then the image and the corresponding image of the uncovered photovoltaic panel are transmitted to the snow parameter calculation module for processing. If only the "snow" or "panel" label exists in the image prediction result, then continue to process the next frame of the image.

[0060] The snow parameter calculation module is divided into snow coverage calculation and snow thickness calculation. The snow coverage calculation requires calculating the number of pixel blocks occupied by the "snow" label box and the number of pixel blocks occupied by the "panel" label box in the snow-covered photovoltaic panel image, and then calculating the ratio of the two to obtain the snow coverage on the photovoltaic panel.

[0061] Snow thickness calculation requires edge detection of the two sets of images transmitted. The edge pixels are divided into strong edges and weak edges through double threshold processing. The calculated image gradient amplitude is compared with the threshold. The adjacent strong edge pixels are connected to form a complete edge. First, the edge length and boundary point coordinates of the snow-free photovoltaic panel in the image and the boundary point coordinates of the snow-covered photovoltaic panel image are recorded. The ratio of the edge length to the actual length of the photovoltaic panel obtained by field measurement is used to calculate the proportionality coefficient; then, by comparing the boundary point coordinates before and after the snow on the photovoltaic panel, the snow thickness at each point on the upper and lower edges of the photovoltaic panel (the upper and lower edges here are relative to the collected two-dimensional image) can be obtained. Finally, the actual thickness of the snow on the photovoltaic panel is obtained according to the proportionality coefficient. The snow image and the calculated snow coverage rate and actual snow thickness are transmitted to the time prediction unit for processing.

[0062] 3. Snowmelt time prediction

[0063] Based on the calculated actual snow thickness and the snow density measured by the snow analyzer, the snow water equivalent (SWE) can be obtained, which represents the depth or volume of water formed by the melting of snow per unit area. Combined with the snow cover rate, the actual amount of snow melted in the snowy area can be obtained. The calculation formula can be expressed as:

[0064] M=SWE*F=H*ρ*F

[0065] Where M is the actual snowmelt, H is the snow thickness, ρ is the snow density, and F is the snow coverage. At the same time, a simple linear regression model was constructed, and the least squares method was used to linearly fit the long-term temperature series and the corresponding snowmelt data based on the "Research on Simple Snowmelt Model Based on Temperature Changes" to build a degree-day factor model dominated by temperature, in which the temperature is the independent variable and the snowmelt is the dependent variable to obtain the following snowmelt model:

[0066]

[0067] Where SM is the amount of snowmelt, T a is the outside temperature, T r is the critical temperature for snowmelt. According to the formula, the amount of snowmelt per unit time can be calculated based on the real-time temperature of the photovoltaic array at a certain moment. Combined with the actual amount of snowmelt in the snow-covered area, the actual snowmelt time in the snow-covered area can be obtained. The formula is as follows:

[0068]

[0069] This embodiment also provides a photovoltaic panel snow melting time prediction system based on machine vision, including a snow melting time prediction module, and the snow melting time prediction module executes the photovoltaic panel snow melting time prediction method based on machine vision.

[0070] This embodiment also provides a non-transitory computer-readable storage medium, and the non-transitory computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the photovoltaic panel snow melting time prediction method based on machine vision is implemented.

[0071] Refer Figure 2 As shown, this embodiment also provides an electronic device, including:

[0072] A memory 201 and a processor 202, the memory 201 and the processor 202 are communicatively connected to each other, the memory 201 stores computer instructions, and the processor 202 executes the computer instructions to execute the photovoltaic panel snow melting time prediction method based on machine vision.

[0073] The photovoltaic panel snow melting time prediction method and system based on machine vision have the following technical effects:

[0074] 1) In winter or low-temperature regions, snow cover will affect photovoltaic panels, seriously affecting their photoelectric conversion efficiency and service life. By building a local snow melting model, the present invention can accurately predict the snow melting time in a timely manner, ensuring the normal operation of the photovoltaic power station and improving the power generation efficiency.

[0075] 2) Based on machine vision technology, accurate measurement and judgment prediction of the snow accumulation situation and snow melting time of photovoltaic panels are carried out, that is, through a high-resolution image acquisition device, combined with a deep learning algorithm, the snow accumulation situation on the photovoltaic panels is monitored in real time, and the snow melting time is predicted by analyzing the image data. Compared with traditional manual observation or simple temperature sensors, this method can provide more accurate and real-time data, thus improving the accuracy of prediction.

[0076] 3) Combining the machine vision system with an automated snow removal device can also achieve intelligent maintenance of photovoltaic panels.

[0077] The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting the snow melting time of a photovoltaic panel based on machine vision, characterized in that, The steps include: Step 1: using a drone equipped with a high-definition camera to regularly collect images of uncovered photovoltaic panels in the photovoltaic panel area in winter according to a preset flight path, and collecting images of photovoltaic panels covered with snow in the photovoltaic panel area in snowy weather in winter, and making a one-to-one correspondence between the images of the uncovered photovoltaic panels and the images of the photovoltaic panels covered with snow; Step 2: Perform data enhancement preprocessing on the image of the photovoltaic panel covered with snow, use it as a training image data set, build a YOLO model to detect whether there is snow on the photovoltaic panel, and if so, calculate the snow coverage rate and actual snow thickness; the construction of the YOLO model to detect whether there is snow on the photovoltaic panel includes: 1) Dataset creation Collect the images needed for training, and use rotation, flipping, and contrast enhancement to perform data augmentation as training image datasets; Labelme is used to annotate the photovoltaic panel area and the snow-covered area of ​​each image in the training set. The bounding box is used to determine their location and classify them into snow and photovoltaic panel categories. The images are preprocessed, the image size is adjusted to a uniform size and normalized, and the Gaussian filter is used to perform convolution operations and weight allocation to remove noise in the image. The convolution images are convolved with Sobel operators in the horizontal and vertical directions respectively. Calculate the gradient amplitude of the convolution result, and use the non-maximum suppression method to detect the local maximum value of each pixel in the gradient image to remove redundancy; 2) Object Detection Image prediction is performed based on the YOLO model. If the drone transmits an image file and the image prediction result contains both snow and photovoltaic panel labels, the image and the corresponding uncovered photovoltaic panel image are transmitted to the snow parameter calculation module for processing. If the image prediction result contains only snow or photovoltaic panel labels, the operation is terminated. If the drone transmits a video file, the image in the video stream is extracted according to the set frame rate. If the image prediction result contains both snow and photovoltaic panel labels, the image and the corresponding uncovered photovoltaic panel image are transmitted to the snow parameter calculation module for processing. If the image prediction result contains only snow or photovoltaic panel labels, the processing of the next frame of image continues. Step 3: Based on the actual snow thickness calculated and the snow density measured by the snow analyzer, the snow water equivalent is obtained, and then the actual snowmelt amount in the snow area is obtained in combination with the snow cover rate. The snowmelt time is obtained based on the actual snowmelt amount and the degree-day factor model, including: The actual snowmelt amount is calculated by the following formula: ； Among them, is the actual snowmelt volume, is the actual snow depth, is the snow density, is the snow cover rate, is the snow water equivalent; At the same time, a simple linear regression model was constructed. The long-term series temperature and the corresponding snowmelt data were linearly fitted by the least squares method, and a degree-day factor model dominated by temperature was built. The temperature was used as the independent variable and the snowmelt was used as the dependent variable to obtain the following snowmelt model: ； wherein is the snowmelt volume, is the outside air temperature, is the critical temperature for snowmelt; Calculate the snow melting amount per unit time based on the real-time temperature of the photovoltaic array at a certain moment, and combine it with the actual snow melting amount of the snow-covered area to obtain the actual snow melting time of the snow-covered area , and the formula is as follows: ； The calculation method of the snow cover rate includes: Calculate the number of pixel blocks occupied by the snow-covered photovoltaic panel label frame and the number of pixel blocks occupied by the photovoltaic panel label frame in the snow-covered photovoltaic panel image respectively, and calculate the ratio of the two to obtain the coverage rate of snow on the photovoltaic panel; The calculation method of the actual snow thickness includes: Perform edge detection operations on the two sets of transmitted images. By double-threshold processing, edge pixels are divided into strong edges and weak edges. Compare the calculated image gradient magnitude with the threshold, and form a complete edge by connecting adjacent strong edge pixels. Record the edge length of the snow-free photovoltaic panel and the coordinates of the boundary points in the image, as well as the coordinates of the boundary points of the snow-covered photovoltaic panel image. Calculate the proportionality coefficient using the ratio of the edge length to the actual length of the photovoltaic panel obtained by on-site measurement. By comparing the boundary point coordinates of the photovoltaic panel before and after snow accumulation, obtain the snow depth at each point on the upper and lower edges of the photovoltaic panel, and calculate the actual snow depth on the photovoltaic panel according to the proportionality coefficient.

2. The method for predicting the snow melting time of a photovoltaic panel based on machine vision according to claim 1, wherein The images required for collection and training include photovoltaic panel images covered with snow under different weather conditions, different lighting conditions, and different environments.

3. The method for predicting the snow melting time of a photovoltaic panel based on machine vision according to claim 2, wherein The image prediction based on the YOLO model includes: Based on the YOLO model, divide the labeled dataset into small batches for training and obtain a weight file. When processing the input image, first perform the same preprocessing operations on the image transmitted by the drone, and then use the YOLO model to divide the image into grids. For the photovoltaic panel target image whose center point falls within the grid, each grid predicts multiple bounding boxes of the target and the confidence of the bounding boxes. The bounding boxes have different shapes and proportions to adapt to the sizes of different targets. The confidence represents the probability that the bounding box contains the target and the matching degree between the predicted box and the actual box. At the same time, predict the conditional probability of the target matching the snow class or the photovoltaic panel class. After the network predicts multiple bounding boxes, remove the overlapping bounding boxes and only retain the best prediction result. Finally, what YOLO outputs are the position, size, confidence of each bounding box, and the probability of matching the category with the existing label.

4. A photovoltaic panel snow melting time prediction system based on machine vision, characterized in that, It includes a snow melting time prediction module, and the snow melting time prediction module executes the method for predicting the snow melting time of a photovoltaic panel based on machine vision according to any one of claims 1-3.

5. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method for predicting the snow melting time of a photovoltaic panel based on machine vision according to any one of claims 1-3 is implemented.

6. An electronic device, characterized in that, It includes: A memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the method for predicting the snow melting time of a photovoltaic panel based on machine vision according to any one of claims 1-3.

Images