NES-photovoltaic module big data AI analysis and diagnosis system

Abstract

The invention belongs to the technical field of photovoltaic power station detection, and particularly relates to an NES-photovoltaic module big data AI analysis and diagnosis system which comprises an image processing module, an angular point extraction module, an image classification module, a defect extraction module and a defect positioning module. According to the method, the defect features of the photovoltaic panel can be extracted through identification processing of the inspection image, so that real-time monitoring of the health condition of the photovoltaic panel is realized, specifically, diode faults are judged through gradient-based edge detectors, bright spot defects are identified based on edge corner features, and finally, a Gaussian pyramid image matching algorithm is adopted to realize real-time monitoring of the health condition of the photovoltaic panel. The method is advantaged in that the accuracy and efficiency of defect detection are improved, manual inspection requirements are reduced, operation cost is reduced, powerful data support is provided for maintenance and management of the photovoltaic panel, and improvement of overall performance and reliability of a photovoltaic power station is facilitated.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power station detection, and specifically relates to a NES-photovoltaic component big data AI analysis and diagnosis system. Background Art

[0002] With the development of advanced technologies such as deep learning, cloud computing, and the Internet of Things (IoT), machine vision systems are becoming more and more intelligent, flexible, and powerful. Future machine vision technology will focus more on real-time processing capabilities, higher resolution, and wider adaptability, enabling it to better serve various application needs. As a key technology, machine vision is not only promoting progress in industrial automation, intelligent manufacturing, and other fields, but is also constantly incorporating new scientific and technological elements, becoming one of the key technologies for realizing intelligent production and intelligent life.

[0003] The recognition work of machine vision is generally divided into two steps. The first step is to simulate human vision and use OpenCV to extract the corners and edge features in the image. This part does not require labeling. The second step is to simulate human judgment of image information, mainly through pattern matching and deep learning, to associate the features of the image with the type of fault that needs to be judged. This part requires a large amount of labeling.

[0004] The two steps of machine vision are used in combination in practice, not completely isolated. The method of directly extracting image features has a solid theoretical foundation in image processing. It has high accuracy when the image features are relatively clear, and does not require data pre-training. This has special advantages when there are relatively few fault samples. However, the method of directly extracting image features is not convenient for the generalization of the algorithm. The model recognizes too many types of images, or when it is applied in other scenarios, the model needs to be adjusted. How to improve the generalization of the model and the accuracy of recognition on complex images is the main issue to be considered in the image feature extraction algorithm. Through deep learning, when the specific image features are uncertain, the image is recognized through a complex neural network, which has the advantage of solving general problems. It has good results when the computing power is sufficient and the amount of marking is sufficient; and because the algorithm is based on specific judgments through image information, it can be directly combined with business scenarios. The disadvantage of this method is that in the process of identifying image features, a large amount of accurate manual annotation is required. In many cases, the number of fault samples is too small to form effective annotations. Summary of the invention

[0005] The purpose of the present invention is to provide a NES-photovoltaic module big data AI analysis and diagnosis system, which can realize fault detection on a small number of samples, and then solve the stability of recognition when the size, rotation, and light intensity change through the Gaussian pyramid of the image, and solve the problems of small samples and generalization.

[0006] The technical solution adopted by the present invention is as follows: NES-Photovoltaic module big data AI analysis and diagnosis system, including image processing module, corner point extraction module, image classification module, defect extraction module and defect location module; The image processing module is used to collect inspection images of photovoltaic panels and perform background separation processing on the inspection images to obtain a front target image; The corner point extraction module is used to collect the edge area of ​​the preceding target image, and determine the edge corner points of the preceding target image according to the brightness difference of the edge area; The image classification module is used to describe the neighborhood of the edge corner point to generate multiple image descriptors, and to compare the similarity of the descriptors of different pre-target images to obtain multiple image descriptor sets; The defect extraction module is used to determine the defect characteristics of the photovoltaic panel according to the edge corner points, and the defect characteristics include diode failure and bright spot defect; The defect location module is used to locate the position of the photovoltaic panel with the defect feature, add a coordinate number, and send it to the management end.

[0007] In a preferred embodiment, when the image processing module is executed, the following steps are included: Acquire the inspection image, and perform grayscale processing on the inspection image; Gaussian blur algorithm is used to smooth the inspection image after grayscale processing to reduce image noise; The smoothed inspection image is binarized using an adaptive threshold segmentation method to separate the image background and target area, and the image in the target area is used as the pre-target image.

[0008] In a preferred embodiment, the execution steps of the corner point extraction module include: Acquiring brightness information of the preceding target image; Calculate the brightness difference between adjacent pixels in the previous target image and record it as the condition parameter to be evaluated; Obtaining a screening threshold, and comparing the screening threshold with the condition parameter to be evaluated, retaining pixel points corresponding to the parameter to be evaluated that are greater than the screening threshold, and recording them as edge points to be evaluated; The brightness difference between the edge point to be evaluated and the pixel points in the neighborhood is compared, and when the brightness difference between the edge point to be evaluated is greater than the brightness difference between all points in the neighborhood, the edge point to be evaluated is recorded as an edge corner point.

[0009] In a preferred embodiment, the execution steps of the image classification module include: Acquire a neighborhood image of the edge corner point; Acquire representative features of the neighborhood image, and directional attributes between the representative features and edge corner points; Combining the representative features and the directional attributes between the representative features and the edge corner points to obtain a feature description group; The feature description group is vectorized, and the vectorized feature description group is output as an image descriptor of the edge corner point.

[0010] In a preferred solution, when determining a diode fault, the image gradient of the inspection image is obtained, the neighborhood pixel values ​​are checked along the direction of the gradient, and pixels with non-maximum values ​​are suppressed; Acquire a hysteresis threshold, wherein the hysteresis threshold includes a high hysteresis threshold and a low hysteresis threshold; If the gradient amplitude of the edge corner point is greater than the high hysteresis threshold, the edge corner point is recorded as a strong edge point; If the gradient amplitude of the edge corner point is between the low hysteresis threshold and the high hysteresis threshold, the edge corner point is recorded as a weak edge point; If the gradient amplitude of the edge corner point is less than the hysteresis threshold, the edge corner point is suppressed to zero and does not belong to the edge; Keep all strong edge points and weak edge points connected to the strong edge points, and together they form an edge curve; The length of the edge curve is measured, and when there are two edge curves parallel to each other, and the lengths of the two edge curves exceed 50% of the block in the photovoltaic panel, it is determined that the photovoltaic panel is a diode failure.

[0011] In a preferred embodiment, when determining a bright spot defect, the bright spot area corresponding to the edge corner point and the training samples with bright spots are obtained, and pattern matching is performed based on the edge corner point and the training samples to determine whether the photovoltaic panel has a bright spot defect.

[0012] In a preferred embodiment, when determining whether the photovoltaic panel has a bright spot defect, the feature descriptor of the bright spot area is compared with the feature descriptor of the training sample to obtain a feature matching score; The feature matching score is compared with a preset evaluation threshold, and when the evaluation threshold is smaller than the feature matching score, the photovoltaic panel is determined to be a bright spot defect.

[0013] In a preferred embodiment, when the defect location module is executed, it searches for inspection image positions at all scales, identifies key points in the inspection image that are invariant to scale and rotation, and matches the position of the photovoltaic panel based on the key points.

[0014] The present invention also provides a NES-photovoltaic module big data AI analysis and diagnosis method, which is applied to the above-mentioned NES-photovoltaic module big data AI analysis and diagnosis system, and includes the following steps: Collecting inspection images of photovoltaic panels, and performing background separation processing on the inspection images to obtain a front target image; Acquire the edge area of ​​the preceding target image, and determine the edge corner points of the preceding target image according to the brightness difference of the edge area; Describing the neighborhood of the edge corner point to generate multiple image descriptors, and performing similarity comparison on the descriptors of different pre-target images to obtain multiple image descriptor sets; Determining defect features of the photovoltaic panel based on the edge corner points, wherein the defect features include diode failure and bright spot defects; The position of the photovoltaic panel with the defect feature is located, and a coordinate number is added and sent to the management end.

[0015] And, an electronic device, the electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; Among them, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned NES-photovoltaic component big data AI analysis and diagnosis system.

[0016] The technical effects achieved by the present invention are: The present invention can extract the defect features of photovoltaic panels through recognition and processing of inspection images, thereby realizing real-time monitoring of the health status of photovoltaic panels. Specifically, diode faults are determined by gradient-based edge detection, bright spot defects are identified based on edge corner point features, and finally a Gaussian pyramid image matching algorithm is used to obtain the matching photovoltaic panel positioning, thereby improving the accuracy and efficiency of defect detection, reducing the need for manual inspections, and reducing operating costs. It provides strong data support for the maintenance and management of photovoltaic panels, and helps to improve the overall performance and reliability of photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a system module diagram of the present invention; Figure 2is a schematic diagram of a fixed window in a flat area in the present invention; Figure 3 is a schematic diagram of a fixed window in an edge area in the present invention; Figure 4 is a schematic diagram of a fixed window in a corner area in the present invention; Figure 5 It is a schematic diagram of the gradient change rate of the image in two vertical directions in the window of the present invention; Figure 6 is a pyramid structure of an image formed by downsampling in the present invention; Figure 7 is a schematic diagram of a Gaussian difference image in the present invention; Figure 8 It is a schematic diagram of the difference between the extreme points obtained in the discrete space and the extreme points in the continuous space in the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0019] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" that appears in different places in this specification does not refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0021] Secondly, the present invention is described in detail with reference to the schematic diagram. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0022] See also Figure 1 As shown, the present invention provides a NES-photovoltaic module big data AI analysis and diagnosis system, including an image processing module, a corner point extraction module, an image classification module, a defect extraction module and a defect positioning module; The image processing module is used to collect inspection images of photovoltaic panels and perform background separation processing on the inspection images to obtain a front target image; The corner point extraction module is used to collect the edge area of ​​the preceding target image, and determine the edge corner points of the preceding target image according to the brightness difference of the edge area; The image classification module is used to describe the neighborhood of the edge corner point to generate multiple image descriptors, and to compare the similarity of the descriptors of different pre-target images to obtain multiple image descriptor sets; The defect extraction module is used to determine the defect characteristics of the photovoltaic panel according to the edge corner points, and the defect characteristics include diode failure and bright spot defect; The defect location module is used to locate the position of the photovoltaic panel with the defect feature, add a coordinate number, and send it to the management end.

[0023] In the present invention, with the continuous development of energy storage technology, power generation methods are becoming more and more diversified, and photovoltaic power generation is one of them. However, the area of ​​the photovoltaic panel layout area is relatively large, and the inspection is relatively cumbersome and not accurate enough. In order to improve the overall efficiency and stability of the photovoltaic power generation system, this embodiment proposes a NES-photovoltaic component big data AI analysis and diagnosis system, which has multiple functional modules, including an image processing module, a corner point extraction module, an image classification module, a defect extraction module, and a defect positioning module. These modules work together to jointly realize the detection and diagnosis of photovoltaic panels. First, the image processing module is responsible for collecting the inspection image of the photovoltaic panel. After the inspection image is processed, the background interference can be removed to obtain a clear front target image and highlight the target area. The corner point extraction module will collect the edge area of ​​the front target image. By analyzing the brightness difference of the edge area, the module can accurately determine the edge curve and edge corner point of the front target image. The edge corner point can help the system identify defects on the photovoltaic panel. The image classification module describes the neighborhood of the edge corner point in detail and generates multiple image descriptors. These descriptors can capture the key points in the image. Information helps the system distinguish different photovoltaic panel images. By comparing the similarity of descriptors of different pre-target images, the system can generate multiple sets of image descriptors to facilitate the management of photovoltaic panels. The defect extraction module determines the defect characteristics of photovoltaic panels based on edge corners. The defect characteristics mainly include diode failures and bright spot defects. The defect location module is responsible for locating the specific position of photovoltaic panels with defective characteristics. It can not only accurately identify the defect location, but also add coordinate numbers to the image for easy management and tracking. After the positioning is completed, the relevant information will be sent to the management end for further processing and maintenance by relevant personnel.

[0024] In a preferred embodiment, when the image processing module is executed, the following steps are included: Acquire the inspection image, and perform grayscale processing on the inspection image; Gaussian blur algorithm is used to smooth the inspection image after grayscale processing to reduce image noise; The smoothed inspection image is binarized using an adaptive threshold segmentation method to separate the image background and target area, and the image in the target area is used as the pre-target image.

[0025] In this embodiment, in the process of executing the image processing module, the required inspection images are first obtained, and these inspection images are grayed to simplify the image information. Then, in order to reduce the noise in the inspection images, the Gaussian blur algorithm is applied to smooth the grayed inspection images. In this way, the details and noise in the image are effectively weakened, thereby improving the accuracy of subsequent processing. Then, the adaptive threshold segmentation method is used to binarize the smoothed inspection images, and the threshold is dynamically adjusted according to the local features of the image, so as to more accurately separate the background and target areas of the image. Finally, the image in the target area is extracted as a pre-target image for further analysis and processing.

[0026] In a preferred embodiment, the execution steps of the corner point extraction module include: Acquiring brightness information of the preceding target image; Calculate the brightness difference between adjacent pixels in the previous target image and record it as the condition parameter to be evaluated; Obtaining a screening threshold, and comparing the screening threshold with the condition parameter to be evaluated, retaining pixel points corresponding to the parameter to be evaluated that are greater than the screening threshold, and recording them as edge points to be evaluated; The brightness difference between the edge point to be evaluated and the pixel points in the neighborhood is compared, and when the brightness difference between the edge point to be evaluated is greater than the brightness difference between all points in the neighborhood, the edge point to be evaluated is recorded as an edge corner point.

[0027] In this embodiment, when the corner point extraction module is executed, first, it is necessary to obtain the brightness information of the previous target image, so that the brightness difference of the pixels in the inspection image can be calculated later, so as to further extract the corner point information. The brightness information can be obtained by an image processing algorithm, such as grayscale conversion. Secondly, the brightness difference between adjacent pixels in the previous target image is calculated, and these differences are recorded as condition parameters to be evaluated, and the areas with large brightness changes in the image are found. These areas usually correspond to the positions of the corner points. The brightness difference calculation can be achieved by a difference algorithm, such as using the Sobel operator or P rewitt operator, etc. Next, obtain the screening threshold and compare the screening threshold with the condition parameter to be evaluated. During the comparison process, only the pixel points corresponding to the parameter to be evaluated that are greater than the screening threshold will be retained and recorded as the edge point to be evaluated. The setting of the screening threshold can be adjusted according to the actual situation to adapt to different image features and extraction requirements. Finally, the edge point to be evaluated is further compared. Specifically, the brightness difference between the edge point to be evaluated and the pixel points in its neighborhood is compared. When the brightness difference of the edge point to be evaluated is greater than the brightness difference of all points in the neighborhood, the point can be determined to be an edge corner point. For example, suppose there is a two-dimensional grayscale image I(x,y), whose gradients in the x and y directions are: , ; In the formula, A function representing a grayscale image, represents the horizontal coordinate of the pixel, represents the pixel ordinate, represents the horizontal gradient, Represents the vertical gradient. Use a fixed window to slide in any direction on the image and compare the grayscale change of the pixels in the window before and after sliding. If there is a large grayscale change when sliding in any direction, it can be considered that there are edge corners in the window. Figure 2 ), moving in any direction, without grayscale change, in edge areas (such as Figure 3 ), there is no grayscale change when moving along the edge direction, and there is obvious grayscale change when moving perpendicular to the edge direction. Figure 4 ), there is an obvious grayscale change when moving in any direction.

[0028] Its autocorrelation matrix is: ; Among them, the energy function In the figure, it represents the weighted average of all pixel values ​​of the two images under the premise of horizontal movement u distance and vertical movement v distance. Indicates the horizontal movement distance of the window. Indicates the distance the window moves in the vertical direction; x, y are the pixel coordinates, summed within the window range, is the weight of each pixel in the original image. Multiply each pixel difference by a weight to reflect the contribution to the whole. Here is the Gaussian filter kernel. Represents the value of the pixel after the original coordinates are moved. Represents the value of the pixel at the original coordinate position. It is the grayscale difference between before and after the movement. Because the difference can be positive or negative and needs to be guaranteed not to change suddenly, it is squared. Will Perform a first-order Taylor expansion: ; so ; in: represents the sum of squares of horizontal gradients in the window, represents the sum of squares of vertical gradients in the window, Represents the sum of the products of the horizontal and vertical gradients in the window; remember ,in , , ,so , it can be seen is a quadratic form, is a real symmetric matrix, so we can To standardize: ; yes The value rotated by the vector, yes The value rotated by the vector, and is a matrix In order to evaluate whether each pixel is a corner point, the response function is introduced. , based on the eigenvalue of the autocorrelation matrix, using the eigenvalue is to rotate the matrix to a direction where the two axes are linearly irrelevant. and Respectively represent the gradient change rate of the image in the two vertical directions in the window, and the specific performance is as follows Figure 5 ,when and When both are large, it indicates that the position is a corner point. When one eigenvalue is large and the other is small, it indicates that the position is an edge. When both eigenvalues ​​are small, it indicates that the position is a flat area. Because the calculation of the eigenvalue is more complicated, it can be approximated by the difference between the matrix determinant and the trace. For an n-order square matrix, it has the following properties: The determinant is equal to the sum of the eigenvalues, and the trace is equal to the product of the eigenvalues.

[0029] So define the response function for: ; in: , , is a matrix The determinant value of is a matrix The trace of , which is the sum of the values ​​on the main diagonal; By calculating the response value , we can determine whether the pixel is a corner point: if, >0 and is larger, then the pixel is a corner point. , then this area is a flat area if <0 and is smaller, the pixel is in the edge area.

[0030] In a preferred embodiment, the execution steps of the image classification module include: Acquire a neighborhood image of the edge corner point; Acquire representative features of the neighborhood image, and directional attributes between the representative features and edge corner points; Combining the representative features and the directional attributes between the representative features and the edge corner points to obtain a feature description group; The feature description group is vectorized, and the vectorized feature description group is output as an image descriptor of the edge corner point.

[0031] In this implementation, it is first necessary to obtain a neighborhood image of the edge corner point and capture local information around the edge corner point in order to better understand the environment of the corner point. Specifically, a small area centered on the edge corner point will be extracted from the original image, and the size of the area can be adjusted according to actual needs. Then, representative features need to be extracted from the neighborhood image, which can be visual information such as color, texture, and shape, which can effectively describe the visual content of the neighborhood image. At the same time, it is also necessary to determine the directional attributes between these features and the edge corner point, that is, the relative position and direction of the features around the corner point, so as to further enhance the description ability of the features so that they contain not only visual information but also spatial information. Then, the representative features are combined with the directional attributes between them and the edge corner points. Through this combination, a feature description group can be obtained, and the feature description group not only contains the visual features of the neighborhood image, but also includes the spatial distribution of these features. Such a description group can more comprehensively describe the local features of the edge corner points. Finally, the feature description group is vectorized. The purpose of vectorization is to convert these features into numerical form for computer processing and analysis. After the vectorization is completed, the obtained feature description group is output as an image descriptor of the edge corner points to help identify and classify different objects in the image.

[0032] In a preferred embodiment, when determining a diode failure, the image gradient of the inspection image is obtained, the neighborhood pixel values ​​are checked along the direction of the gradient, and pixels with non-maximum values ​​are suppressed; Acquire a hysteresis threshold, wherein the hysteresis threshold includes a high hysteresis threshold and a low hysteresis threshold; If the gradient amplitude of the edge corner point is greater than the high hysteresis threshold, the edge corner point is recorded as a strong edge point; If the gradient amplitude of the edge corner point is between the low hysteresis threshold and the high hysteresis threshold, the edge corner point is recorded as a weak edge point; If the gradient amplitude of the edge corner point is less than the hysteresis threshold, the edge corner point is suppressed to zero and does not belong to the edge; Keep all strong edge points and weak edge points connected to the strong edge points, and together they form an edge curve; The length of the edge curve is measured, and when there are two edge curves parallel to each other, and the lengths of the two edge curves exceed 50% of the block in the photovoltaic panel, it is determined that the photovoltaic panel is a diode failure.

[0033] When determining that a diode is faulty, the image gradient information of the inspection image needs to be obtained first. By calculating the image gradient, the pixel values ​​in the neighborhood can be checked along the direction of the gradient. In this process, non-maximum pixel points need to be suppressed in order to more accurately identify edge corner points.

[0034] Assume that there is a two-dimensional grayscale image I(x,y) with a pixel intensity value of I. The gradient of the image is expressed as partial derivatives in two directions: Gradient in the x direction: , represents the grayscale change rate in the horizontal direction; Gradient in the y direction: , represents the grayscale change rate in the vertical direction; The magnitude of the gradient, used to indicate the degree of change in the pixel, can be obtained by calculating the Euclidean norm of these two partial derivatives: ; The direction of the gradient θ gives the direction of the edge and is calculated as: ; The filter in the horizontal direction is , used to find the gradient in the horizontal direction, and the filter in the vertical direction is , used to find the vertical gradient. Before horizontal and vertical filtering, Gaussian filtering is performed to remove noise, because the noise in the image must be smoothed. The mathematical expression of Gaussian filtering is: , where (x, y) is the coordinate of the filter center and σ is the standard deviation of the Gaussian function, which is convolved with the image I(x, y): , Represents the convolution of the grayscale value function of the image and the filter function. The convolution operation represents the application of a smoothing filter on the image to obtain a smoothed image, followed by non-maximum suppression. The goal of non-maximum suppression is to accurately locate the edge by suppressing pixels whose gradient amplitude is not the local maximum. For each pixel point, the neighborhood pixel value is checked along the direction of the gradient to suppress pixels that are not the maximum value. The specific operations are as follows: For each pixel, (x, y), find the two adjacent pixels in the gradient direction; If the gradient magnitude G(x,y) of the pixel is not the maximum value, it is suppressed to zero; Next, you need to obtain a hysteresis threshold. The hysteresis threshold consists of two parts: a high hysteresis threshold and a low hysteresis threshold. The high hysteresis threshold and the low hysteresis threshold can be used to more accurately judge the strength of the edge corner points. Specifically, if the gradient amplitude of an edge corner point is greater than the high hysteresis threshold, then the edge corner point is considered to be a strong edge point and is recorded. On the contrary, if the gradient amplitude of the edge corner point is less than the hysteresis threshold, then the edge corner point will be suppressed to zero, which means that it is not part of the edge. For those edge corner points whose gradient amplitude is between the low hysteresis threshold and the high hysteresis threshold, , and record them as weak edge points. Although these weak edge points are not as strong as strong edge points, they still have certain reference value. After processing all edge corner points, all strong edge points will be retained, and the weak edge points connected to these strong edge points will also be retained to form an edge curve. In order to further confirm the fault of the diode, it is necessary to measure the length of the edge curve. If there are two parallel edge curves in the inspection image, and the length of these two edge curves exceeds 50% of the length of a block in the photovoltaic panel, then it can be determined that the photovoltaic panel has a diode fault.

[0035] In a preferred embodiment, when determining the bright spot defect, the bright spot area corresponding to the edge corner point and the training sample with the bright spot are obtained, and pattern matching is performed based on the edge corner point and the training sample to determine whether the photovoltaic panel has a bright spot defect; When determining whether the photovoltaic panel has a bright spot defect, comparing the feature descriptor of the bright spot area with the feature descriptor of the training sample to obtain a feature matching score; The feature matching score is compared with a preset evaluation threshold, and when the evaluation threshold is smaller than the feature matching score, the photovoltaic panel is determined to be a bright spot defect.

[0036] When detecting bright spot defects on photovoltaic panels, it is first necessary to obtain the bright spot area corresponding to the edge corner points, and collect training samples with bright spots. By comparing the edge corner points and the training samples, pattern matching is performed to determine whether there is a bright spot defect on the photovoltaic panel. Specifically, in the process of judging whether there is a bright spot defect on the photovoltaic panel, the feature descriptor of the bright spot area will be compared in detail with the feature descriptors in the training samples, so as to obtain a feature matching score reflecting the similarity between the bright spot area and the training samples. Finally, the obtained feature matching score is compared with a pre-set evaluation threshold. If the evaluation threshold is less than the feature matching score, then it can be determined that the photovoltaic panel has a bright spot defect. Otherwise, there is no bright spot defect.

[0037] In a preferred embodiment, when the defect location module is executed, it searches for inspection image positions at all scales, identifies key points in the inspection image that are invariant to scale and rotation, and matches the position of the photovoltaic panel based on the key points; Search for key points in different scale spaces and calculate the direction of the key points. The key points found are some very prominent points that will not change due to factors such as illumination, affine transformation and noise, such as corner points, edge points, bright spots in dark areas and dark spots in bright areas. This implementation method mainly determines the key points by searching for edge corner points; First, scale space extrema detection: Search for image positions at all scales, and identify potential points of interest that are invariant to scale and rotation through Gaussian differential functions. The basic idea of ​​scale space theory is: introduce a parameter regarded as scale into the image information processing model, obtain scale space representation sequences at multiple scales by continuously changing the scale parameter, extract the main contour of the scale space from these sequences, and use the main contour as a feature vector to achieve edge and corner detection and feature extraction at different resolutions. The scale space method incorporates traditional single-scale image information processing technology into a dynamic analysis framework with constantly changing scales, making it easier to obtain the essential features of the image. The blur degree of images of each scale in the scale space gradually increases, which can simulate the formation process of the target on the retina when the distance from the target changes from near to far. The scale space satisfies visual invariance. The visual interpretation of visual invariance is as follows: When observing an object with the eyes, on the one hand, when the illumination conditions of the background in which the object is located change, the brightness level and contrast of the image perceived by the retina are different. Therefore, the analysis of the image by the scale space operator is not affected by the gray level and contrast changes of the image, that is, it satisfies the gray invariance and contrast invariance. On the other hand, relative to a fixed coordinate system, when the relative position between the observer and the object changes, the position, size, angle and shape of the image perceived by the retina are different. Therefore, the analysis of the image by the scale space operator is independent of the position, size, angle and affine transformation of the image, that is, it satisfies the translation invariance, scale invariance, Euclidean invariance and affine invariance. The scale space L(x, y, σ) of an image is defined as the convolution of a Gaussian function G(x, y, σ) of a varying scale and the original image I(x, y), that is: , ; in, represents the new image scaled by the scale-space factor, Represents the convolution kernel transformed using the scale space factor, m and n represent the dimensions of the Gaussian template, (x, y) represents the position of the image pixel, σ is the scale space factor, the smaller the σ value, the less the image is smoothed, and the smaller the corresponding scale. The small scale corresponds to the detailed features of the image, and the large scale corresponds to the general features of the image; Secondly, the scale space is implemented using a Gaussian pyramid. The construction of the Gaussian pyramid is divided into two parts: First, perform Gaussian blur and downsampling on the image at different scales Downsampling can form an image pyramid (such as Figure 6 ), the original image is continuously downsampled to obtain a series of images of different sizes, which form a tower model from large to small and from top to bottom. The original image is the first layer of the pyramid, and the new image obtained by each downsampling is the previous layer of the pyramid. There is one image on each layer, and each pyramid has n layers in total. The number of layers of the pyramid is determined by the original size of the image and the size of the top image.

[0038] In order to make the scale reflect its continuity, the Gaussian pyramid adds Gaussian filtering on the basis of simple sampling. An image in each layer of the pyramid is Gaussian blurred using different parameters, so that each layer of the pyramid contains multiple Gaussian blurred images. The multiple images in each layer of the pyramid are collectively called a group. There is only one group of images in each layer of the pyramid. The number of groups is equal to the number of pyramid layers. Each group contains multiple layers of images. The adjacent upper and lower layers of images in each group of the Gaussian pyramid are subtracted to obtain a Gaussian difference image (such as Figure 7 ), perform extreme value detection, the selection of key points is based on their stability, the initial exploration of key points is completed by comparing the two adjacent layers of images in the same group, in order to find the extreme points of the function, each pixel point should be compared with all its adjacent points to determine whether it is larger or smaller than its adjacent points in the image domain and scale domain, and determine the extreme points in the discrete space. After that, it is necessary to accurately determine the position and scale of the key points by fitting a three-dimensional quadratic function, and remove low-contrast key points and unstable edge response points to enhance matching stability and improve noise resistance. The extreme points in the discrete space are not true extreme points, such as Figure 8 , which shows the difference between the extreme points obtained in the discrete space of a two-dimensional function and the extreme points in the continuous space. The method of obtaining the extreme points in the continuous space by interpolating known discrete space points is called sub-pixel interpolation. The required extreme points can be obtained by calculating the difference method.

[0039] The NES-photovoltaic module big data AI analysis and diagnosis method is applied to the above-mentioned NES-photovoltaic module big data AI analysis and diagnosis system, and includes the following steps: Collecting inspection images of photovoltaic panels, and performing background separation processing on the inspection images to obtain a front target image; Acquire the edge area of ​​the preceding target image, and determine the edge corner points of the preceding target image according to the brightness difference of the edge area; Describing the neighborhood of the edge corner point to generate multiple image descriptors, and performing similarity comparison on the descriptors of different pre-target images to obtain multiple image descriptor sets; Determining defect features of the photovoltaic panel based on the edge corner points, wherein the defect features include diode failure and bright spot defects; The position of the photovoltaic panel with the defect feature is located, and a coordinate number is added and sent to the management end.

[0040] The execution process of the above-mentioned NES-Photovoltaic Module Big Data AI Analysis and Diagnosis Method is consistent with the NES-Photovoltaic Module Big Data AI Analysis and Diagnosis System, and will not be repeated here.

[0041] An electronic device, comprising: at least one processor; and a memory communicatively coupled to the at least one processor; Among them, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned NES-photovoltaic component big data AI analysis and diagnosis system.

[0042] The processor of the electronic device may be an ARM architecture processor, an x86 architecture processor, or other types of processors, depending on the performance requirements and cost budget of the electronic device. The memory may be a random access memory (RAM), a read-only memory (ROM), a flash memory, or other forms of non-volatile memory, and the computer program may include an operating system, an application, and software modules related to the NES-PV Module Big Data AI Analysis and Diagnosis System.

[0043] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0044] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. NES-Photovoltaic module big data AI analysis and diagnosis system, characterized by: It includes image processing module, corner point extraction module, image classification module, defect extraction module and defect location module; The image processing module is used to collect inspection images of photovoltaic panels and perform background separation processing on the inspection images to obtain a front target image; The corner point extraction module is used to collect the edge area of ​​the preceding target image, and determine the edge corner points of the preceding target image according to the brightness difference of the edge area; The image classification module is used to describe the neighborhood of the edge corner point to generate multiple image descriptors, and to compare the similarity of the descriptors of different pre-target images to obtain multiple image descriptor sets; The defect extraction module is used to determine the defect characteristics of the photovoltaic panel according to the edge corner points, and the defect characteristics include diode failure and bright spot defect; The defect location module is used to locate the position of the photovoltaic panel with the defect feature, add a coordinate number, and send it to the management end.

2. The NES-photovoltaic module big data AI analysis and diagnosis system according to claim 1 is characterized by: When the image processing module is executed, the following steps are included: Acquire the inspection image, and perform grayscale processing on the inspection image; Gaussian blur algorithm is used to smooth the inspection image after grayscale processing to reduce image noise; The smoothed inspection image is binarized using an adaptive threshold segmentation method to separate the image background and target area, and the image in the target area is used as the pre-target image.

3. The NES-photovoltaic module big data AI analysis and diagnosis system according to claim 1 is characterized by: The execution steps of the corner point extraction module include: Acquiring brightness information of the preceding target image; Calculate the brightness difference between adjacent pixels in the previous target image and record it as the condition parameter to be evaluated; Obtaining a screening threshold, and comparing the screening threshold with the condition parameter to be evaluated, retaining pixel points corresponding to the parameter to be evaluated that are greater than the screening threshold, and recording them as edge points to be evaluated; The brightness difference between the edge point to be evaluated and the pixel points in the neighborhood is compared, and when the brightness difference between the edge point to be evaluated is greater than the brightness difference between all points in the neighborhood, the edge point to be evaluated is recorded as an edge corner point.

4. The NES-photovoltaic module big data AI analysis and diagnosis system according to claim 1 is characterized by: The execution steps of the image classification module include: Acquire a neighborhood image of the edge corner point; Acquire representative features of the neighborhood image, and directional attributes between the representative features and edge corner points; Combining the representative features and the directional attributes between the representative features and the edge corner points to obtain a feature description group; The feature description group is vectorized, and the vectorized feature description group is output as an image descriptor of the edge corner point.

5. The NES-photovoltaic module big data AI analysis and diagnosis system according to claim 1 is characterized by: When determining a diode fault, the image gradient of the inspection image is obtained, the neighborhood pixel values ​​are checked along the direction of the gradient, and pixels with non-maximum values ​​are suppressed; Acquire a hysteresis threshold, wherein the hysteresis threshold includes a high hysteresis threshold and a low hysteresis threshold; If the gradient amplitude of the edge corner point is greater than the high hysteresis threshold, the edge corner point is recorded as a strong edge point; If the gradient amplitude of the edge corner point is between the low hysteresis threshold and the high hysteresis threshold, the edge corner point is recorded as a weak edge point; If the gradient amplitude of the edge corner point is less than the hysteresis threshold, the edge corner point is suppressed to zero and does not belong to the edge; Keep all strong edge points and weak edge points connected to the strong edge points, and together they form an edge curve; The length of the edge curve is measured, and when there are two edge curves parallel to each other, and the lengths of the two edge curves exceed 50% of the block in the photovoltaic panel, it is determined that the photovoltaic panel is a diode failure.

6. The NES-photovoltaic module big data AI analysis and diagnosis system according to claim 1 is characterized by: When determining a bright spot defect, the bright spot area corresponding to the edge corner point and the training samples with bright spots are obtained, and pattern matching is performed based on the edge corner point and the training samples to determine whether the photovoltaic panel has a bright spot defect.

7. The NES-photovoltaic module big data AI analysis and diagnosis system according to claim 6 is characterized by: When determining whether the photovoltaic panel has a bright spot defect, comparing the feature descriptor of the bright spot area with the feature descriptor of the training sample to obtain a feature matching score; The feature matching score is compared with a preset evaluation threshold, and when the evaluation threshold is smaller than the feature matching score, the photovoltaic panel is determined to be a bright spot defect.

8. The NES-photovoltaic module big data AI analysis and diagnosis system according to claim 1 is characterized by: When the defect location module is executed, it searches for inspection image positions at all scales, identifies key points in the inspection image that are invariant to scale and rotation, and matches the positions of the photovoltaic panels based on the key points.

9. NES-Photovoltaic module big data AI analysis and diagnosis method, characterized by: The NES-photovoltaic module big data AI analysis and diagnosis system applied to any one of claims 1 to 8 comprises the following steps: Collecting inspection images of photovoltaic panels, and performing background separation processing on the inspection images to obtain a front target image; Acquire the edge area of ​​the preceding target image, and determine the edge corner points of the preceding target image according to the brightness difference of the edge area; Describing the neighborhood of the edge corner point to generate multiple image descriptors, and performing similarity comparison on the descriptors of different pre-target images to obtain multiple image descriptor sets; Determining defect features of the photovoltaic panel based on the edge corner points, wherein the defect features include diode failure and bright spot defects; The position of the photovoltaic panel with the defect feature is located, and a coordinate number is added and sent to the management end.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; Wherein, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the NES-photovoltaic component big data AI analysis and diagnosis system described in any one of claims 1 to 8.

Images