A method for identifying hot spots of solar photovoltaic modules based on infrared images

Abstract

The present invention relates to the field of image recognition, and specifically to a method for identifying hot spots of solar photovoltaic components based on infrared images. A grayscale image of the surface of a solar panel to be identified is collected, and a grayscale histogram is established for curve fitting to determine whether the grayscale image of the surface of the solar panel to be identified has noise; pixels within a set grayscale value range are obtained as target pixels; a window is established with each target pixel as the center to determine whether the target pixel is a noise pixel; the grayscale weight value of each non-central pixel in the window is obtained; the denoised grayscale value of the target pixel in the corresponding window is obtained; the optimal threshold of the segmentation algorithm is obtained to segment the grayscale image to obtain the target area and the background area; and the target area is identified for hot spots. The present invention can accurately identify the hot spot area in the image by screening out solar panel images with noise, and then denoising the noise pixels to obtain an accurate segmentation threshold.

Description

Technical Field

[0001] The invention relates to the field of image recognition, and in particular to a method for identifying hot spots of solar photovoltaic components based on infrared images. Background Art

[0002] With the widespread use of drones in photovoltaic inspections, the efficiency of photovoltaic panel fault detection has been greatly improved. The high-definition cameras on drones continuously transmit photos of tens of thousands of photovoltaic panels in real time. It is difficult to achieve accurate and timely processing by manual labor, and computers are needed to detect defects in real time. When a photovoltaic panel fails, due to the mismatch between current and voltage, a large amount of electricity will be consumed in the fault area and its vicinity to generate heat, and the brightness on the infrared image is higher than the normal part. This phenomenon is called "hot spot". Based on the infrared thermal imager equipped with drones to collect infrared image information of photovoltaic power station solar panels, when hot spot detection is performed on the infrared image later through image recognition technology, the complex background and serious thermal noise interference in the solar photovoltaic module image will have a serious impact on the recognition result of the hot spot. Therefore, it is necessary to denoise the image before detecting the hot spot in the photovoltaic module image.

[0003] Among the means of spot recognition for photovoltaic components in the prior art, one is based on the electrical characteristics of solar cells, that is, a detection method that monitors the output voltage, output current, output power, etc. of the power generation component; the other is a detection method based on image analysis of the power generation component, which mainly analyzes the temperature distribution thermal imaging of the photovoltaic panel under different working conditions, and uses machine learning or image processing methods to realize hot spot recognition. In the detection method based on image recognition, Gaussian filtering is usually used to remove noise in the image. However, Gaussian filtering has a high usage rate of computing resources, memory, and power consumption. When a large number of photovoltaic panels need to be detected and denoised, it cannot meet the actual detection needs. Summary of the invention

[0004] In order to solve the problem of high utilization rate of computing resources, memory and power consumption in the prior art, the present invention provides a method for identifying hot spots of solar photovoltaic components based on infrared images, including: collecting grayscale images of the surface of the solar panel to be identified, and establishing a grayscale histogram for curve fitting to determine whether there is noise in the grayscale image of the surface of the solar panel to be identified; obtaining pixels within a set grayscale value range as target pixels; establishing a window with each target pixel as the center to determine whether the target pixel is a noise pixel; obtaining the grayscale weight value of each non-central pixel in the window; obtaining the denoised grayscale value of the target pixel in the corresponding window; obtaining the optimal threshold of the segmentation algorithm to segment the grayscale image to obtain the target area and the background area; and identifying hot spots in the target area.

[0005] The present invention converts the infrared image of the solar panel into a grayscale image, and establishes a grayscale histogram for straight line fitting, thereby performing a preliminary screening of the image noise degree, which can reduce the amount of calculation of the subsequent denoising algorithm; further determines the denoising target in the image, that is, the noise pixel point, and uses the idea of ​​median filtering to adopt different size windows to denoise the noise pixel points, which has the advantage of adaptively denoising different noise pixel points in the image, and can achieve accurate denoising of each noise pixel point; finally, the optimal threshold of the segmentation algorithm is selected according to the grayscale value of the denoised pixel point, so as to obtain an accurate target area, which can ensure the accurate identification of the hot spot area in the solar panel, and can solve the problem that the background in the solar photovoltaic module image is complex and the thermal noise interference is serious, which affects the identification result of the hot spot.

[0006] The present invention adopts the following technical solution: a method for identifying hot spots of solar photovoltaic components based on infrared images.

[0007] Collect infrared images of the surface of the solar panel to be identified, convert the infrared images of the surface of the solar panel to be identified into grayscale images, and establish a grayscale histogram.

[0008] A curve fitting is performed on the grayscale histogram, and whether there is noise in the grayscale image of the solar panel surface to be identified is determined according to the number of peaks on the fitting curve.

[0009] When noise exists in the grayscale image of the solar panel surface to be identified, all pixel points within a set grayscale value range in the grayscale image of the solar panel surface to be identified are obtained as target pixel points.

[0010] A window is established with each target pixel as the center, and whether the target pixel is a noise pixel is determined based on the grayscale value of the target pixel and the maximum and minimum grayscale values ​​of the pixels in the window.

[0011] If the target pixel is a noise pixel, obtain the median grayscale value of the pixels in the window where the noise pixel is located, and obtain the grayscale weight value of each non-center pixel in the window where the noise pixel is located according to the grayscale value of each non-center pixel in the window and the median grayscale value.

[0012] The denoised grayscale value of the target pixel in the corresponding window is obtained according to the grayscale values ​​of all non-central pixels in the window and the corresponding grayscale weight values.

[0013] If the target pixel is not a noise pixel, the grayscale value of the target pixel is used as the denoising grayscale value; and the denoising grayscale values ​​of all target pixels within the set grayscale value range are obtained.

[0014] The optimal threshold of the segmentation algorithm is obtained according to the denoised grayscale values ​​of all target pixels in the grayscale image of the solar panel surface to be identified, and the grayscale image of the solar panel surface to be identified is segmented according to the optimal threshold to obtain the target area in the grayscale image of the solar panel surface to be identified.

[0015] Hot spot identification is performed based on the grayscale mean value of the target area in the grayscale image of the solar panel surface to be identified.

[0016] Furthermore, a method for identifying hot spots of solar photovoltaic components based on infrared images is provided, and a method for determining whether there is noise in the grayscale image of the surface of the solar panel to be detected according to the number of peaks on the fitting curve is as follows:

[0017] When the number of peaks on the fitting curve is greater than 3, noise exists in the grayscale image of the surface of the solar panel to be inspected.

[0018] Furthermore, a method for identifying hot spots of solar photovoltaic modules based on infrared images is provided to determine whether a target pixel is a noise pixel:

[0019] Get the maximum grayscale value and the minimum grayscale value in the window where the target pixel is located;

[0020] Get the difference between the maximum grayscale value in the window and the grayscale value of the target pixel; get the difference between the grayscale value of the target pixel and the minimum grayscale value in the window;

[0021] When the difference is less than 0, the target pixel is a noise pixel.

[0022] Furthermore, a method for identifying hot spots of solar photovoltaic modules based on infrared images, after determining whether a target pixel is a noise pixel, further includes:

[0023] When the target pixel is a noise pixel, the median grayscale value in the window where the target pixel is located is used as the first grayscale value of the target pixel;

[0024] Respectively obtain the difference between the maximum grayscale value in the window and the first grayscale value of the target pixel, and the difference between the first grayscale value of the target pixel and the minimum grayscale value in the window;

[0025] If there is a difference less than 0, change the window size where the target pixel is located, and use the median grayscale value in the window after the size change as the second grayscale value of the target pixel; iterate in sequence until there is no difference less than 0 or the window size reaches the set size.

[0026] Furthermore, a method for identifying hot spots of solar photovoltaic modules based on infrared images, a method for obtaining the gray weight value of each non-central pixel in the window is:

[0027] Get the median grayscale value of the window where the target pixel is located after iteration, and get the difference between the grayscale value of each non-central pixel in the window after iteration and the median grayscale value;

[0028] The grayscale weight value of each non-central pixel is obtained according to the difference between the grayscale value of each non-central pixel and the median grayscale value. The expression is:

[0029]

[0030] in, Represents the grayscale weight value of the jth non-central pixel, represents the gray value of the jth non-central pixel, It represents the median grayscale value of the window where the target pixel is located after iteration, and N represents the number of non-central pixels in the window.

[0031] Furthermore, a method for identifying hot spots of solar photovoltaic modules based on infrared images, and a method for obtaining the denoised grayscale value of the target pixel point in the corresponding window is as follows:

[0032] Get the gray value of each non-center pixel in the window;

[0033] According to the sum of the products of the grayscale value of each non-central pixel in the window and the corresponding grayscale weight value, the denoised grayscale value of the target pixel in the corresponding window is obtained.

[0034] Furthermore, a method for identifying hot spots of solar photovoltaic components based on infrared images is provided, wherein the method for identifying hot spots of target areas in grayscale images of the surface of solar panels to be identified with noise is as follows:

[0035] Obtain multiple grayscale images of the surface of the solar panel with hot spot areas, calculate the difference between the grayscale mean of the hot spot area and the grayscale mean of the background area in each image, and use the average of the grayscale mean difference between the hot spot area and the background area in the multiple images as the judgment threshold;

[0036] When the difference between the grayscale mean of the target area in the grayscale image of the solar panel surface to be identified with noise and the grayscale mean of the background area is greater than the judgment threshold, the target area in the grayscale image of the solar panel surface to be identified with noise is a hot spot area.

[0037] The beneficial effects of the present invention are as follows: the present invention converts the infrared image of the solar panel into a grayscale image, and establishes a grayscale histogram for straight line fitting, thereby performing preliminary screening of the image noise degree, which can reduce the amount of calculation of the subsequent denoising algorithm; further determines the denoising target in the image, that is, the noise pixel point, and uses different size windows to denoise the noise pixel points based on the idea of ​​median filtering, which has the advantage of adaptive denoising of different noise pixel points in the image, and can achieve accurate denoising of each noise pixel point; finally, the optimal threshold of the segmentation algorithm is selected according to the grayscale value of the denoised pixel point, so as to obtain an accurate target area, which can ensure the accurate identification of the hot spot area in the solar panel, and can solve the problem that the background in the solar photovoltaic module image is complex and the thermal noise interference is serious, which affects the identification result of the hot spot, and improves the accuracy of the image recognition of the hot spot area in the solar panel. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 The present invention is a schematic structural diagram of a method for identifying hot spots of solar photovoltaic components based on infrared images according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1 As shown, a schematic diagram of the structure of a method for identifying hot spots of solar photovoltaic components based on infrared images according to an embodiment of the present invention is provided, including:

[0042] 101. Collect a grayscale image of the surface of the solar panel to be identified, and determine whether there is noise in the grayscale image of the surface of the solar panel to be identified.

[0043] The present invention firstly collects infrared images of all solar panels in a photovoltaic power station by using an infrared thermal imager equipped with an unmanned aerial vehicle, and then uses Hough line detection to obtain the angles of the main grid lines and the secondary grid lines on the solar panels. According to the detected angles, a rotation transformation method is used to obtain a corrected image, in which the image is the secondary grid lines in the vertical direction and the main grid lines in the horizontal direction. Then, the SIFT algorithm is used to splice the corrected images to obtain a top-down infrared image of the complete solar panel surface in the photovoltaic power station.

[0044] The present invention requires a clear infrared image of the solar panel surface, so as to identify the hot spot area according to the infrared radiation energy distribution in the image. Therefore, a drone equipped with an infrared thermal imager is required to take aerial photos and collect infrared images of solar panels on the ground to identify the characteristic information of the solar panel surface in the image. The present invention adopts DNN semantic segmentation to identify the target in the segmented image. The relevant content of the DNN network is as follows:

[0045] The dataset used is an infrared image dataset of solar panels on the ground collected from a bird's-eye view.

[0046] The pixels that need to be segmented are divided into two categories, that is, the labeling process of the training set is: single-channel semantic label, the corresponding position pixel belongs to the background class is labeled as 0, and the corresponding position pixel belongs to the surface of the solar panel is labeled as 1.

[0047] The task of the network is classification, so the loss function used is the cross entropy loss function.

[0048] So far, the infrared image of solar panels on the ground has been processed through DNN to obtain the infrared image of the surface of a single solar panel in the image.

[0049] Since the thermal infrared image collector uses an infrared detector and an optical imaging objective lens to receive the infrared radiation energy distribution pattern of the target to be measured and reflects it on the photosensitive element of the infrared detector, thereby obtaining an infrared thermal image. This thermal image corresponds to the heat distribution field on the surface of the object. Therefore, the infrared image of the solar panel surface can be divided into two parts according to the pixel grayscale value of the image, namely the grid line area that does not absorb heat and the solar panel area that absorbs heat, and then the solar panel area is segmented to identify the hot spot area that absorbs too much heat.

[0050] The present invention grayscales the infrared image of the solar panel surface, and counts its grayscale histogram, then uses the least square method to perform smooth curve fitting on the grayscale histogram, and counts the horizontal coordinate values ​​of the peak points on the fitting curve to obtain the set , where n is the number of peaks on the fitting curve. The present invention can divide the collected images into three categories for hot spot detection according to the peak data.

[0051] Due to the grayscale difference between the grid line area and the solar panel area, it can be seen that if , indicating that there is no large temperature difference on the solar panel, and it is judged that this solar panel is normal.

[0052] like , it means that the quality of the collected image is good and the sunlight is evenly distributed, that is, the radiation energy is evenly distributed. At this time, the peak of the pixel grayscale in the grid area is removed on the grayscale histogram, and the pixel grayscale value in the solar panel area is in a double peak state on the histogram, that is, the horizontal coordinate value of the first trough on the grayscale histogram curve is taken. , take the value of the horizontal axis on the histogram greater than The standard Otsu algorithm can be used to accurately segment the part. According to the segmentation characteristics of the Otsu algorithm in the bimodal state of the histogram, the optimal segmentation threshold is between the two peaks, so The image is segmented into the target area larger than the optimal segmentation threshold and the background area in the image.

[0053] like , which means that the infrared image is affected by the uneven distribution of radiation energy caused by noise interference and uneven illumination. Directly performing the Otsu algorithm will cause misclassification, that is, there may be noise in the image. Therefore, it is necessary to first perform denoising and enhancement on the image, and then perform threshold segmentation.

[0054] 102. For a grayscale image of a solar panel surface to be identified with noise, all pixels within a set grayscale value range are obtained as target pixels.

[0055] It is known that the hot spot effect is a local temperature on the solar panel that is much higher than the surrounding temperature. Therefore, it can be considered that the infrared image at this time is composed of high-radiation bright areas and low-radiation dark areas. First, a rough segmentation is performed to extract the brightest and darkest parts of the image, that is, the pixels that are most likely to be the target and the background are first segmented, and the remaining parts may contain both the target and the background. The transfer of heat gradually increases or decreases, so these pixels are processed accordingly in combination with their neighborhood information to increase the accuracy of the Otsu algorithm threshold segmentation.

[0056] From left to right, the horizontal coordinate values ​​of the second trough and the last trough on the grayscale histogram curve are respectively and According to the surface characteristics of the solar panel, the first peak on the grayscale image is the grid line area, the second peak is the darkest area on the panel, and the last peak is the brightest area. Therefore, it is determined that the pixel grayscale value is less than or equal to The pixel point is the background area, and the pixel gray value is greater than or equal to The pixel point is the target pixel point.

[0057] 103. Establish a window with each target pixel as the center to determine whether the target pixel is a noise pixel.

[0058] The method to determine whether the target pixel is a noise pixel is:

[0059] Get the maximum grayscale value and the minimum grayscale value in the window where the target pixel is located;

[0060] Get the difference between the maximum grayscale value in the window and the grayscale value of the target pixel; get the difference between the grayscale value of the target pixel and the minimum grayscale value in the window;

[0061] When the difference is less than 0, the target pixel is a noise pixel.

[0062] For the grayscale image of the solar panel surface, the pixel grayscale value is greater than and less than The target pixel is combined with its neighborhood information for corresponding denoising and enhancement processing, and a The window is traversed, and the coordinates of the target pixel point are set to , sort the grayscale values ​​of the pixels in the window from small to large, and take the maximum, minimum, median and grayscale values ​​of the target pixel, respectively. , and as well as .

[0063] Then, it is determined whether the target pixel, that is, the central pixel of the window, is a noise pixel. The grayscale value of the known pixel is close to that of the surrounding pixels. Therefore, the present invention calculates the minimum difference between the grayscale value of the central pixel of the window and the maximum value thereof for determination:

[0064]

[0065]

[0066] when and When both are greater than 0, the target pixel in the current window It is not a noise point, the data in the window remains unchanged, and the window size and the grayscale value of the pixel in the window are recorded. Otherwise is a noise point.

[0067] 104. If the target pixel is a noise pixel, obtain the grayscale weight value of each non-target pixel in the window.

[0068] After determining whether the target pixel is a noise pixel, the following steps are also included:

[0069] When the target pixel is a noise pixel, the median grayscale value in the window where the target pixel is located is used as the first grayscale value of the target pixel;

[0070] Respectively obtain the difference between the maximum grayscale value in the window and the first grayscale value of the target pixel, and the difference between the first grayscale value of the target pixel and the minimum grayscale value in the window;

[0071] If there is a difference less than 0, change the window size where the target pixel is located, and use the median grayscale value in the window after the size change as the second grayscale value of the target pixel; iterate in sequence until there is no difference less than 0 or the window size reaches the set threshold.

[0072] When the target pixel When the target pixel is a noise point, the present invention uses the idea of ​​median filtering to convert the grayscale median value in the window where the target pixel is located into Assign center point , so we need to determine the median Whether it is a noise point, calculate the minimum difference between the grayscale value of the central pixel of the window and its maximum value:

[0073]

[0074]

[0075] when and When both are greater than 0, It is not a noise point, the median Assign center point , record the window size and the pixel grayscale value in the window at this time, otherwise is a noise point, and the window size is increased to , the window size is an odd number, return the judgment center point Whether it is noise.

[0076] When the window size is small, some details in the image can be well protected, but the noise filtering effect is not very good. On the contrary, when the window size is increased, the noise filtering effect is relatively good, but the image will become blurred and the image details will be lost. Therefore, the window sizes are set as follows: , and , thus performing noise filtering by adaptively changing the window size.

[0077] It should be noted that if the loop is continued until the window size reaches the maximum When , that is, the threshold value reached by selecting the window size in the present invention is , window center point and median When it is still judged as noise, the data in the window remains unchanged, and the window size and the pixel grayscale value in the window are recorded. At this time, the adaptive selection of the window size is completed.

[0078] The method to obtain the grayscale weight value of each non-central pixel in the window is:

[0079] Get the median grayscale value of the window where the target pixel is located after iteration, and get the difference between the grayscale value of each non-central pixel in the window after iteration and the median grayscale value;

[0080] The grayscale weight value of each non-central pixel is obtained according to the difference between the grayscale value of each non-central pixel and the median grayscale value. The expression is:

[0081]

[0082] in, Represents the grayscale weight value of the jth non-central pixel. The greater the grayscale difference of the pixels in the window, the smaller the weight. represents the gray value of the jth non-central pixel, It represents the median grayscale value of the window where the target pixel is located after iteration, and N-1 represents the number of non-central pixels in the window. Indicates the grayscale difference between this point and the target pixel. Indicates that the grayscale difference is normalized. To prevent the denominator from being 0.

[0083] The greater the grayscale difference between the non-center point and the center point, the greater the probability that the non-center point is a noise point, that is, the smaller the weight. Therefore, when the target pixel is in the same type of area, the weights of each pixel are similar. When the target pixel is at the edge, the weights of the same type of pixels in the window are large, and the weights of different types of pixels are small. The present invention increases the grayscale difference between the high-temperature area and the background area, that is, the weight is set according to the grayscale value of each non-center pixel in the window and the grayscale value difference of the target pixel, thereby protecting the edge of the image, making the histogram smooth and approaching a 3-peak state, and thus ensuring the accuracy of subsequent image segmentation.

[0084] 105. Get the denoised grayscale value of the target pixel in the corresponding window.

[0085] The method for obtaining the denoised grayscale value of the target pixel in the corresponding window is:

[0086] Get the gray value of each non-center pixel in the window;

[0087] According to the sum of the products of the grayscale value of each non-central pixel in the window and the corresponding grayscale weight value, the denoised grayscale value of the target pixel in the corresponding window is obtained.

[0088] The coordinates after denoising and enhancement are obtained as The gray value of the traversed pixel for:

[0089]

[0090] in Represents the gray value of the jth pixel in the window at this time, is the corresponding weight, N-1 is the number of pixels in the window at this time, The denoising grayscale value of the target pixel is obtained by weighted summing the grayscale value of each pixel in the window multiplied by its weight, and then taking the obtained value as the denoising grayscale value of the target pixel, thereby ensuring the denoising accuracy of the target pixel.

[0091] Similarly, for all grayscale values ​​greater than and less than The target pixel is enhanced and denoised. Since the final enhancement is a weighted summation using the normalized weights, the denoised grayscale value of the updated target pixel is Still in range Then the Otsu algorithm is used to segment the region.

[0092] If the target pixel is not a noise pixel, the grayscale value of the target pixel is used as the denoising grayscale value; and the denoising grayscale values ​​of all target pixels within the set grayscale value range are obtained.

[0093] 106. Get the optimal threshold of the segmentation algorithm for segmentation to obtain the target area and background area.

[0094] At this time, the formula for selecting the optimal threshold T is:

[0095]

[0096] in is the standard Otsu algorithm formula, and is the horizontal coordinate value of the second trough and the last trough on the grayscale histogram curve, , which limits the grayscale range of the threshold traversal, reduces the area of ​​image detection in the algorithm, and denoises and enhances the segmented target pixels, thereby improving the speed and accuracy of the algorithm.

[0097] The image is segmented according to the obtained optimal threshold T, where the pixel grayscale value is greater than T is the target area, otherwise it is the background area.

[0098] 107. Hot spot identification is performed on the target area in the grayscale image of the solar panel surface to be identified with noise.

[0099] Generally speaking, when a solar panel is operating normally, if the temperature of a certain local area is 6.5°C higher than the ambient temperature, the local area of ​​the solar panel assembly can be considered a hot spot area.

[0100] Based on this logic, the present invention selects 10 solar panel surface images containing hot spots, calculates the grayscale mean difference between the hot spot area and the background area in each image, and then takes the average of these 10 differences as , let it be the judgment parameter, when the difference between the grayscale mean of the target area and the grayscale mean of the background area in the solar panel surface image is greater than When, that is:

[0101]

[0102] in Indicates the area where the grayscale value of the pixel in the solar panel surface image is greater than the optimal segmentation threshold T, that is, the grayscale mean value in the target area. It indicates the area where the gray value of the pixel in the image is not greater than the threshold T, that is, the gray mean value in the background area. It can be judged that when the temperature difference between the target area and the background area is above 6.5℃, the target area on the solar panel is a hot spot area.

[0103] The method for identifying hot spots in the target area of ​​the grayscale image of the solar panel surface to be identified with noise is as follows:

[0104] Obtain multiple grayscale images of the surface of the solar panel with hot spot areas, calculate the difference between the grayscale mean of the hot spot area and the grayscale mean of the background area in each image, and use the average of the grayscale mean difference between the hot spot area and the background area in the multiple images as the judgment threshold;

[0105] When the difference between the grayscale mean of the target area in the grayscale image of the solar panel surface to be identified with noise and the grayscale mean of the background area is greater than the judgment threshold, the target area in the grayscale image of the solar panel surface to be identified with noise is a hot spot area.

[0106] The present invention converts the infrared image of the solar panel into a grayscale image and establishes a grayscale histogram for straight line fitting, thereby performing a preliminary screening of the image noise degree, which can reduce the amount of calculation of the subsequent denoising algorithm; further determining the denoising target in the image, that is, the noise pixel point, and using the idea of ​​median filtering to denoise the noise pixel points using windows of different sizes, which has the advantage of adaptively denoising different noise pixel points in the image, and can achieve accurate denoising of each noise pixel point; finally, the optimal threshold of the segmentation algorithm is selected according to the grayscale value of the denoised pixel point, so as to obtain an accurate target area, which can ensure the accurate identification of the hot spot area in the solar panel, and can solve the problem that the background in the solar photovoltaic module image is complex and the thermal noise interference is serious, which affects the identification result of the hot spot, and improves the accuracy of the image recognition of the hot spot area.

[0107] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying hot spots of solar photovoltaic modules based on infrared images, characterized in that: include: Collect infrared images of the surface of the solar panel to be identified, convert the infrared images of the surface of the solar panel to be identified into grayscale images, and establish a grayscale histogram; Perform curve fitting on the grayscale histogram, and determine whether there is noise in the grayscale image of the solar panel surface to be identified according to the number of peaks on the fitting curve; When there is noise in the grayscale image of the solar panel surface to be identified, all pixels within a set grayscale value range in the grayscale image of the solar panel surface to be identified are obtained as target pixels; A window is established with each target pixel as the center, and whether the target pixel is a noise pixel is determined based on the grayscale value of the target pixel and the maximum and minimum grayscale values ​​of the pixels in the window; If the target pixel is a noise pixel, obtain the median grayscale value of the pixels in the window where the noise pixel is located, and obtain the grayscale weight value of each non-center pixel in the window where the noise pixel is located according to the grayscale value of each non-center pixel in the window and the median grayscale value; Obtain the denoised grayscale value of the target pixel in the corresponding window according to the grayscale values ​​of all non-central pixels in the window and the corresponding grayscale weight values; If the target pixel is not a noise pixel, the grayscale value of the target pixel is used as the denoising grayscale value; the denoising grayscale values ​​of all target pixels within the set grayscale value range are obtained; The optimal threshold of the segmentation algorithm is obtained according to the denoised grayscale values ​​of all target pixels in the grayscale image of the solar panel surface to be identified, and the grayscale image of the solar panel surface to be identified is segmented according to the optimal threshold to obtain the target area in the grayscale image of the solar panel surface to be identified; Hot spot identification is performed based on the grayscale mean value of the target area in the grayscale image of the solar panel surface to be identified; The method to determine whether the target pixel is a noise pixel is: Get the maximum grayscale value and the minimum grayscale value in the window where the target pixel is located; Get the difference between the maximum grayscale value in the window and the grayscale value of the target pixel; get the difference between the grayscale value of the target pixel and the minimum grayscale value in the window; When the difference is less than 0, the target pixel is a noise pixel; After determining whether the target pixel is a noise pixel, the following steps are also included: When the target pixel is a noise pixel, the median grayscale value in the window where the target pixel is located is used as the first grayscale value of the target pixel; Respectively obtain the difference between the maximum grayscale value in the window and the first grayscale value of the target pixel, and the difference between the first grayscale value of the target pixel and the minimum grayscale value in the window; If there is a difference less than 0, change the window size where the target pixel is located, and use the median grayscale value in the window after the size change as the second grayscale value of the target pixel; iterate in sequence until there is no difference less than 0 or the window size reaches the set size; The method to obtain the grayscale weight value of each non-central pixel in the window is: Get the median grayscale value of the window where the target pixel is located after iteration, and get the difference between the grayscale value of each non-central pixel in the window after iteration and the median grayscale value; The grayscale weight value of each non-central pixel is obtained according to the difference between the grayscale value of each non-central pixel and the median grayscale value. The expression is: in, Represents the grayscale weight value of the jth non-central pixel, represents the gray value of the jth non-central pixel, It represents the median grayscale value of the window where the target pixel is located after iteration, and N represents the number of non-central pixels in the window; The method for obtaining the denoised grayscale value of the target pixel in the corresponding window is: Get the gray value of each non-center pixel in the window; According to the sum of the products of the grayscale value of each non-central pixel in the window and the corresponding grayscale weight value, the denoised grayscale value of the target pixel in the corresponding window is obtained; The method for identifying hot spots in the target area of ​​the grayscale image of the solar panel surface to be identified with noise is as follows: Obtain multiple grayscale images of the surface of the solar panel with hot spot areas, calculate the difference between the grayscale mean of the hot spot area and the grayscale mean of the background area in each image, and use the average of the grayscale mean difference between the hot spot area and the background area in the multiple images as the judgment threshold; When the difference between the grayscale mean of the target area in the grayscale image of the solar panel surface to be identified with noise and the grayscale mean of the background area is greater than the judgment threshold, the target area in the grayscale image of the solar panel surface to be identified with noise is a hot spot area.

2. A method for identifying hot spots of solar photovoltaic components based on infrared images according to claim 1, characterized in that: The method for determining whether there is noise in the grayscale image of the solar panel surface to be detected based on the number of peaks on the fitting curve is: When the number of peaks on the fitting curve is greater than 3, noise exists in the grayscale image of the surface of the solar panel to be inspected.

Images