Photovoltaic module EL defect detection system based on machine vision and deep learning

Abstract

The present invention discloses a photovoltaic module EL defect detection system based on machine vision and deep learning, which belongs to the field of photovoltaic module technology, and includes an image acquisition module, an image processing module, a model training module, a test optimization module, a defect detection module and a data management module. The present invention solves the problem that the existing photovoltaic module EL defect detection cannot well locate and classify photovoltaic module EL defects, cannot effectively identify various types of photovoltaic module EL defects, resulting in low detection accuracy, and cannot quickly process a large number of photovoltaic module EL defect images, resulting in long detection time and low defect detection efficiency. The present invention can well locate and classify photovoltaic module EL defects, can effectively identify various types of photovoltaic module EL defects, resulting in high detection accuracy, and can quickly process a large number of photovoltaic module EL defect images, resulting in short detection time, which can improve defect detection efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic modules, and in particular to a photovoltaic module EL defect detection system based on machine vision and deep learning. Background Art

[0002] With the rapid development of the solar energy industry, the production and sales of photovoltaic modules have increased year by year. However, in the production and application of photovoltaic modules, EL (charge-coupled device) defects have become increasingly prominent; EL defects will affect the photoelectric conversion efficiency and life of photovoltaic modules, cause the performance of photovoltaic modules to decline, and even affect the safety and reliability of the entire system. Therefore, timely detection and elimination are required.

[0003] The Chinese patent with publication number CN215987376U discloses a photovoltaic module hot spot drone automatic detection system based on dual-light technology, including an aircraft and a controller, the aircraft is equipped with a machine vision module; the controller integrates an intelligent diagnosis module, a hot spot positioning module and a control module; the aircraft can be used to identify and locate the hot spots of photovoltaic modules in photovoltaic power stations, complete the automatic inspection of photovoltaic power station modules and the automatic positioning of hot spots, and provide a reliable and efficient inspection method for photovoltaic power station operation and maintenance personnel; however, the patent has the following defects:

[0004] The existing photovoltaic module EL defect detection cannot locate and classify photovoltaic module EL defects well, cannot effectively identify various types of photovoltaic module EL defects, resulting in low detection accuracy, and cannot quickly process a large number of photovoltaic module EL defect images, resulting in long detection time and low defect detection efficiency. Summary of the invention

[0005] The purpose of the present invention is to provide a photovoltaic module EL defect detection system based on machine vision and deep learning, which can better locate and classify photovoltaic module EL defects, effectively identify various types of photovoltaic module EL defects, and achieve high detection accuracy. It can also quickly process a large number of photovoltaic module EL defect images, shorten the detection time, improve the defect detection efficiency, and solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Photovoltaic module EL defect detection system based on machine vision and deep learning, including:

[0008] Image acquisition module, used to acquire real-time images of photovoltaic modules EL based on machine vision;

[0009] An image processing module is used to process the real-time EL image of the photovoltaic module and determine the EL characteristic image of the photovoltaic module;

[0010] Model training module, used to build a photovoltaic module EL defect detection model based on deep learning;

[0011] A test optimization module is used to test and optimize the photovoltaic module EL defect detection model to determine the best photovoltaic module EL defect detection model;

[0012] A defect detection module is used to identify and detect EL defects of photovoltaic modules and determine the EL defect detection results of photovoltaic modules;

[0013] The data management module is used to display the EL characteristic images and EL defects of photovoltaic modules, store the information of defective photovoltaic modules and non-defective photovoltaic modules after classification, and store the information of all photovoltaic modules that have been tested.

[0014] Preferably, the image acquisition module includes:

[0015] A light source lighting unit, used to provide a suitable light source to illuminate the photovoltaic assembly to be photographed;

[0016] The image acquisition unit is used to photograph the photovoltaic modules using a high-resolution industrial camera to obtain clear and accurate real-time EL images of the photovoltaic modules based on machine vision.

[0017] Preferably, the image processing module includes:

[0018] An image filtering unit, used for filtering the real-time image of the photovoltaic module EL;

[0019] Obtain real-time EL images of photovoltaic modules based on machine vision;

[0020] Based on the median filter, the real-time EL image of the photovoltaic module is filtered;

[0021] The real-time EL image of the photovoltaic module is divided into multiple blocks, and the median value of the pixels in each block is taken;

[0022] If a pixel value falls near the mean, it is considered to be the true value in the image, otherwise, the pixel is considered as noise and ignored;

[0023] An image enhancement unit, used for enhancing the real-time image of the photovoltaic module EL;

[0024] Obtain the real-time EL image of the photovoltaic module after filtering;

[0025] Enhance the filtered photovoltaic module EL real-time image;

[0026] Based on the method of adaptive histogram equalization, the difference between adjacent pixels in the real-time EL image of photovoltaic modules is improved, and the readability and contrast of the real-time EL image of photovoltaic modules are increased;

[0027] A feature extraction unit, used for extracting features from the real-time EL image of the photovoltaic module;

[0028] Obtain enhanced EL real-time images of photovoltaic modules;

[0029] Extract features from the enhanced EL real-time image of photovoltaic modules;

[0030] Determine the EL characteristic image of photovoltaic modules based on machine vision.

[0031] Preferably, the image enhancement unit comprises:

[0032] An image block extraction module, used to extract a plurality of image blocks corresponding to the real-time image of the photovoltaic module EL;

[0033] The grayscale parameter value acquisition module is used to obtain the grayscale parameter value corresponding to each image block according to the grayscale value of the pixel block contained in each image block; wherein the grayscale parameter value is obtained by the following formula:

[0034]

[0035] Where J represents the grayscale parameter value; n represents the number of pixel blocks contained in the image block; H i Represents the grayscale value corresponding to the i-th pixel block; H p Indicates the grayscale average value corresponding to the image block; H z Represents the grayscale median value corresponding to all pixel blocks contained in the image block;

[0036] An adjacent grayscale parameter value acquisition module is used to extract the grayscale parameter values ​​of adjacent image blocks corresponding to each image block;

[0037] A target grayscale value acquisition module, used to acquire a target grayscale value of a pixel block contained in each image block according to the grayscale parameter value of an adjacent image block corresponding to each image block;

[0038] The grayscale value adjustment module is used to adjust the grayscale value of each pixel block according to the target grayscale value corresponding to the pixel block contained in each image block, and obtain the image block where the grayscale value of all pixel blocks has been adjusted; wherein the image block where the grayscale value of all pixel blocks has been adjusted is the image block that has completed the image enhancement processing.

[0039] Preferably, the target gray value acquisition module includes:

[0040] A grayscale parameter value information extraction module is used to extract the grayscale parameter values ​​of adjacent image blocks corresponding to each image block;

[0041] The grayscale value adjustment coefficient acquisition module is used to obtain the grayscale value adjustment coefficient corresponding to each image block according to the grayscale parameter value of the adjacent image block corresponding to each image block; wherein the grayscale value adjustment coefficient is obtained by the following formula:

[0042]

[0043] Where, f represents the gray value adjustment coefficient corresponding to each image block; m represents the number of image blocks adjacent to each image block; J i represents the grayscale parameter value corresponding to the i-th adjacent image block; J d Indicates the grayscale parameter value of the current image block; J p Represents the average value of the grayscale parameter values ​​corresponding to m adjacent pixel blocks;

[0044] A gray value parameter information extraction module is used to extract the gray value of the pixel block contained in each image block;

[0045] The target grayscale value acquisition execution module is used to obtain the target grayscale value corresponding to each pixel block according to the grayscale value adjustment coefficient corresponding to each image block and the grayscale value of the pixel block contained in each image block; wherein the target grayscale value is obtained by the following formula:

[0046]

[0047] Among them, H m represents the target grayscale value corresponding to the pixel block; H0 represents the grayscale value of the pixel block; f represents the grayscale value adjustment coefficient corresponding to each image block; H z Represents the grayscale median value corresponding to all pixel blocks contained in the image block.

[0048] Preferably, the model training module includes:

[0049] An image collection unit, used to collect historical images of EL defects of photovoltaic modules;

[0050] According to the needs of photovoltaic module EL defect detection, a large number of photovoltaic module EL defect historical images based on dark spots, current unevenness, broken grid, black spots, velvet, SE offset, debris, linear cracks, virtual prints, branch-shaped cracks, black pieces, invalid pieces, clear traces, and mixed light and dark pieces are collected;

[0051] Based on the historical images of EL defects of photovoltaic modules as model training samples;

[0052] A sample division unit, used for dividing the photovoltaic module EL defect history image;

[0053] Obtain historical images of EL defects of photovoltaic modules;

[0054] Divide the historical images of EL defects of photovoltaic modules;

[0055] Determine the training set and test set;

[0056] A model building unit, used to build a photovoltaic module EL defect detection model;

[0057] According to the requirements of photovoltaic module EL defect detection, select a model architecture suitable for photovoltaic module EL defect detection;

[0058] Based on the training set, the selected model architecture suitable for EL defect detection of photovoltaic modules is trained;

[0059] A photovoltaic module EL defect detection model based on deep learning was determined.

[0060] Preferably, the test optimization module includes:

[0061] Performance test unit, used to perform performance test on photovoltaic module EL defect detection model;

[0062] Obtain a photovoltaic module EL defect detection model based on deep learning;

[0063] Based on the test set, the performance of the photovoltaic module EL defect detection model is tested;

[0064] Determine the performance test results based on the EL defect detection model of photovoltaic modules;

[0065] An optimization and adjustment unit, used to optimize and adjust the photovoltaic module EL defect detection model;

[0066] Obtain performance test results based on PV module EL defect detection model;

[0067] Conduct in-depth research and analysis on the performance test results based on the photovoltaic module EL defect detection model;

[0068] Determine the optimization adjustment scheme based on the EL defect detection model of photovoltaic modules;

[0069] Based on the optimization and adjustment scheme, the photovoltaic module EL defect detection model is optimized and adjusted;

[0070] Determine the best EL defect detection model for photovoltaic modules.

[0071] Preferably, the defect detection module includes:

[0072] An image extraction unit, used for extracting EL characteristic images of photovoltaic modules based on machine vision;

[0073] According to the requirements of photovoltaic module EL defect detection, the EL feature image of photovoltaic modules based on machine vision is extracted;

[0074] Defect detection unit, used to identify and detect EL defects of photovoltaic modules;

[0075] Obtain the best EL defect detection model for photovoltaic modules;

[0076] Based on the best photovoltaic module EL defect detection model, the photovoltaic module EL defect recognition and detection is performed on the photovoltaic module EL feature image;

[0077] Determine the EL defect detection results of photovoltaic modules based on deep learning;

[0078] Based on the EL defect detection results of photovoltaic modules, the EL defect areas of photovoltaic modules and their corresponding types are marked with identification boxes;

[0079] A threshold adjustment unit is used to adjust the threshold interface and adjust the detection accuracy;

[0080] According to actual industrial needs, different detection thresholds and sensitivities are set for different types of EL defects of photovoltaic modules, and the threshold interface is adjusted to adjust the detection accuracy.

[0081] Preferably, the data management module includes:

[0082] A picture display unit, used to display EL characteristic images of photovoltaic modules;

[0083] Among them, when detecting EL defects of photovoltaic modules, the EL characteristic image of the photovoltaic modules after detection is displayed, and it is automatically updated in real time according to the detection status of EL defects of photovoltaic modules;

[0084] When the EL defect detection of photovoltaic modules is stopped, the EL characteristic image of the photovoltaic modules that meets the user's needs is displayed according to the user's needs;

[0085] A defect display unit, used to display relevant information of EL defects of photovoltaic modules;

[0086] When detecting EL defects of photovoltaic modules, the relevant information of the detected EL defects of photovoltaic modules will be automatically updated and displayed in real time;

[0087] A user control unit for setting PV module EL defect detection parameters and detection selection;

[0088] Among them, photovoltaic module EL defect detection includes all-type defect detection and single-type defect detection;

[0089] The user performs single-type defect detection based on the type of defect he wants to detect;

[0090] Among them, the EL defect detection parameter of photovoltaic modules is 0.25-1.00. The larger the parameter value, the stricter the defect screening.

[0091] Preferably, the data management module further includes:

[0092] The data display unit is used for EL defect detection of photovoltaic modules, to screen out photovoltaic modules with defects, and to update and display their image file names and all defect information in real time for viewing;

[0093] A data classification unit is used to classify information of defective photovoltaic modules and non-defective photovoltaic modules, so that users can query according to their needs, wherein the classification is performed according to date;

[0094] The data storage unit is used to store the information of the classified defective photovoltaic modules and non-defective photovoltaic modules, and to store the information of all the photovoltaic modules that have been tested.

[0095] Compared with the prior art, the present invention has the following beneficial effects:

[0096] 1. The present invention collects real-time EL images of photovoltaic modules based on machine vision, processes the real-time EL images of photovoltaic modules, determines the EL feature images of photovoltaic modules, constructs a photovoltaic module EL defect detection model based on deep learning according to the requirements of EL defect detection of photovoltaic modules, tests and optimizes the EL defect detection model of photovoltaic modules, determines the best EL defect detection model of photovoltaic modules, identifies and detects EL defects of photovoltaic modules based on the best EL defect detection model of photovoltaic modules, determines the EL defect detection results of photovoltaic modules, and marks the EL defect areas of photovoltaic modules and their corresponding types with identification boxes.

[0097] 2. The present invention can display the EL characteristic images and EL defects of photovoltaic modules, and store the information of defective photovoltaic modules and non-defective photovoltaic modules after classification, store all the information of photovoltaic modules that have been tested, and can better locate and classify the EL defects of photovoltaic modules, and can effectively identify various types of EL defects of photovoltaic modules, so that the detection accuracy is high, and a large number of EL defect images of photovoltaic modules can be processed quickly, so that the detection time is short, and the defect detection efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 It is a framework structure diagram of the photovoltaic module EL defect detection system of the present invention;

[0099] Figure 2 This is an interactive control console interface diagram of the photovoltaic module EL defect detection system of the present invention;

[0100] Figure 3This is a diagram of the defect selection interface of the photovoltaic module EL defect detection system of the present invention. DETAILED DESCRIPTION

[0101] 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.

[0102] In order to solve the problems that the existing photovoltaic module EL defect detection cannot locate and classify photovoltaic module EL defects well, cannot effectively identify various types of photovoltaic module EL defects, resulting in low detection accuracy, and cannot quickly process a large number of photovoltaic module EL defect images, resulting in long detection time and low defect detection efficiency, please refer to Figure 1-Figure 3 , this embodiment provides the following technical solutions:

[0103] The photovoltaic module EL defect detection system based on machine vision and deep learning includes an image acquisition module, an image processing module, a model training module, a test optimization module, a defect detection module and a data management module.

[0104] It should be noted that through the interactive communication between the image acquisition module, the image processing module, the model training module, the test optimization module, the defect detection module and the data management module, the EL defects of the photovoltaic modules can be better located and classified, and various types of EL defects of photovoltaic modules can be effectively identified, so that the detection accuracy is high, and a large number of EL defect images of photovoltaic modules can be processed quickly, so that the detection time is short, which can improve the defect detection efficiency.

[0105] Among them, the image acquisition module is used to collect real-time images of photovoltaic modules EL based on machine vision;

[0106] In this embodiment, the image acquisition module includes:

[0107] A light source lighting unit, used to provide a suitable light source to illuminate the photovoltaic assembly to be photographed;

[0108] The image acquisition unit is used to photograph the photovoltaic modules using a high-resolution industrial camera to obtain clear and accurate real-time EL images of the photovoltaic modules based on machine vision.

[0109] The image processing module is used to process the real-time EL image of the photovoltaic module and determine the EL characteristic image of the photovoltaic module;

[0110] In this embodiment, the image processing module includes:

[0111] An image filtering unit, used for filtering the real-time image of the photovoltaic module EL;

[0112] Obtain real-time EL images of photovoltaic modules based on machine vision;

[0113] Based on the median filter, the real-time EL image of the photovoltaic module is filtered;

[0114] The real-time EL image of the photovoltaic module is divided into multiple blocks, and the median value of the pixels in each block is taken;

[0115] If a pixel value falls near the mean, it is considered to be the true value in the image, otherwise, the pixel is considered as noise and ignored;

[0116] An image enhancement unit, used for enhancing the real-time image of the photovoltaic module EL;

[0117] Obtain the real-time EL image of the photovoltaic module after filtering;

[0118] Enhance the filtered photovoltaic module EL real-time image;

[0119] Based on the method of adaptive histogram equalization, the difference between adjacent pixels in the real-time EL image of photovoltaic modules is improved, and the readability and contrast of the real-time EL image of photovoltaic modules are increased;

[0120] A feature extraction unit, used for extracting features from the real-time EL image of the photovoltaic module;

[0121] Obtain enhanced EL real-time images of photovoltaic modules;

[0122] Extract features from the enhanced EL real-time image of photovoltaic modules;

[0123] Determine the EL characteristic image of photovoltaic modules based on machine vision.

[0124] Among them, the model training module is used to build a photovoltaic module EL defect detection model based on deep learning;

[0125] Specifically, the image enhancement unit includes:

[0126] An image block extraction module, used to extract a plurality of image blocks corresponding to the real-time image of the photovoltaic module EL;

[0127] The grayscale parameter value acquisition module is used to obtain the grayscale parameter value corresponding to each image block according to the grayscale value of the pixel block contained in each image block; wherein the grayscale parameter value is obtained by the following formula:

[0128]

[0129] Where J represents the grayscale parameter value; n represents the number of pixel blocks contained in the image block; H i Represents the grayscale value corresponding to the i-th pixel block; H p Indicates the grayscale average value corresponding to the image block; H z Represents the grayscale median value corresponding to all pixel blocks contained in the image block;

[0130] An adjacent grayscale parameter value acquisition module is used to extract the grayscale parameter values ​​of adjacent image blocks corresponding to each image block;

[0131] A target grayscale value acquisition module, used to acquire a target grayscale value of a pixel block contained in each image block according to the grayscale parameter value of an adjacent image block corresponding to each image block;

[0132] The grayscale value adjustment module is used to adjust the grayscale value of each pixel block according to the target grayscale value corresponding to the pixel block contained in each image block, and obtain the image block where the grayscale value of all pixel blocks has been adjusted; wherein the image block where the grayscale value of all pixel blocks has been adjusted is the image block that has completed the image enhancement processing.

[0133] The technical effect of the above technical solution is: by calculating the gray parameter value (J) of each image block, which comprehensively considers the gray value (Hi), gray mean value (Hp) and gray median value (Hz) of the pixels in the image block, this method helps to suppress random noise while enhancing image details. The use of gray mean and median can smooth the image and reduce the impact of extreme gray values, thereby improving the signal-to-noise ratio of the image.

[0134] The grayscale value adjustment in the technical solution is based on the grayscale parameter values ​​of each image block and its adjacent image blocks. This local processing method makes the image enhancement process more flexible and adaptable. Image characteristics (such as brightness, contrast, etc.) in different areas may be different, and local adjustment can better preserve and enhance these characteristics.

[0135] Since the grayscale value adjustment is based on the grayscale parameter values ​​of the image block and its adjacent image blocks, this method helps to maintain the edge information of the image while adjusting the grayscale value. The edge is one of the most important features in the image, which is particularly critical for defect detection of photovoltaic module EL (electroluminescence) images.

[0136] By adjusting the grayscale value of each pixel block, the grayscale distribution of the entire image can be made more uniform and the contrast more reasonable, thereby improving the overall quality of the image. This is crucial for subsequent image analysis, defect detection and other tasks.

[0137] The entire image enhancement process is completed automatically through multiple modules without human intervention, which greatly improves processing efficiency and accuracy. At the same time, since the processing is based on image blocks, it can be processed in parallel to a certain extent, further improving the processing speed.

[0138] In summary, this technical solution achieves local adaptability enhancement of photovoltaic module EL images by comprehensively considering the grayscale characteristics of image blocks and their relationship with adjacent image blocks, effectively improves image quality, and provides strong support for subsequent image analysis and defect detection.

[0139] Specifically, the target gray value acquisition module includes:

[0140] A grayscale parameter value information extraction module is used to extract the grayscale parameter values ​​of adjacent image blocks corresponding to each image block;

[0141] The grayscale value adjustment coefficient acquisition module is used to obtain the grayscale value adjustment coefficient corresponding to each image block according to the grayscale parameter value of the adjacent image block corresponding to each image block; wherein the grayscale value adjustment coefficient is obtained by the following formula:

[0142]

[0143] Where, f represents the gray value adjustment coefficient corresponding to each image block; m represents the number of image blocks adjacent to each image block; J i represents the grayscale parameter value corresponding to the i-th adjacent image block; J d Indicates the grayscale parameter value of the current image block; J p Represents the average value of the grayscale parameter values ​​corresponding to m adjacent pixel blocks;

[0144] A gray value parameter information extraction module is used to extract the gray value of the pixel block contained in each image block;

[0145] The target grayscale value acquisition execution module is used to obtain the target grayscale value corresponding to each pixel block according to the grayscale value adjustment coefficient corresponding to each image block and the grayscale value of the pixel block contained in each image block; wherein the target grayscale value is obtained by the following formula:

[0146]

[0147] Among them, H m represents the target grayscale value corresponding to the pixel block; H0 represents the grayscale value of the pixel block; f represents the grayscale value adjustment coefficient corresponding to each image block; H z Represents the grayscale median value corresponding to all pixel blocks contained in the image block.

[0148] The technical effect of the above technical solution is: by calculating the gray value adjustment coefficient (f), the coefficient comprehensively considers the gray parameter values ​​of the current image block (Jd) and its adjacent image blocks (Ji) and the average value (Jp) of the gray parameter values ​​of these adjacent image blocks. This method can adaptively adjust the gray value of the current image block based on local neighborhood information, making the image enhancement process more refined and in line with the actual situation.

[0149] The introduction of the grayscale value adjustment coefficient allows the grayscale adjustment of the pixel blocks within each image block to different degrees. By combining the original grayscale value (H0) of each pixel block and the grayscale adjustment coefficient (f), as well as the grayscale median (Hz) within the image block, noise can be effectively suppressed while maintaining image details. This processing method helps to enhance image quality without losing important image features. At the same time, since the grayscale value adjustment coefficient is calculated based on the grayscale parameter values ​​of adjacent image blocks, this method can enhance the edge and texture information in the image. Edges and textures are important visual features in images and are crucial for tasks such as defect detection of photovoltaic module EL images.

[0150] By adjusting the grayscale value of each image block, the technical solution can optimize the local contrast of the image. The improvement of local contrast helps to better display the details and layers of the image, and improve the visual effect and readability of the image. At the same time, the grayscale value adjustment coefficient and the calculation method of the target grayscale value in the technical solution have certain flexibility and robustness. They can adapt to EL images of photovoltaic modules with different lighting conditions, different image quality and different defect types, thereby providing a stable image enhancement effect.

[0151] The entire target grayscale value acquisition process is completed automatically through multiple modules without manual intervention. At the same time, since the processing is based on image blocks and pixel blocks, it can be processed in parallel to a certain extent, further improving the processing speed and efficiency.

[0152] In summary, this technical solution realizes adaptive and refined grayscale adjustment of the EL image of photovoltaic modules by introducing a grayscale value adjustment coefficient and a target grayscale value calculation method based on the coefficient, effectively improving the image quality and providing more reliable and accurate image data for subsequent image analysis and defect detection.

[0153] In this embodiment, the model training module includes:

[0154] An image collection unit, used to collect historical images of EL defects of photovoltaic modules;

[0155] According to the requirements of EL defect detection of photovoltaic modules, historical images of EL defects of photovoltaic modules based on dark spots, uneven current, broken grid, black spots, velvet, SE offset, debris, linear cracks, virtual prints, branch-shaped cracks, black pieces, invalid pieces, clear traces, and mixed light and dark pieces are collected;

[0156] Based on the historical images of EL defects of photovoltaic modules as model training samples;

[0157] A sample division unit, used for dividing the photovoltaic module EL defect history image;

[0158] Obtain historical images of EL defects of photovoltaic modules;

[0159] Divide the historical images of EL defects of photovoltaic modules;

[0160] Determine the training set and test set;

[0161] A model building unit, used to build a photovoltaic module EL defect detection model;

[0162] According to the requirements of photovoltaic module EL defect detection, select a model architecture suitable for photovoltaic module EL defect detection;

[0163] Based on the training set, the selected model architecture suitable for EL defect detection of photovoltaic modules is trained;

[0164] A photovoltaic module EL defect detection model based on deep learning was determined.

[0165] Among them, the test optimization module is used to test and optimize the photovoltaic module EL defect detection model to determine the best photovoltaic module EL defect detection model;

[0166] In this embodiment, the test optimization module includes:

[0167] Performance test unit, used to perform performance test on photovoltaic module EL defect detection model;

[0168] Obtain a photovoltaic module EL defect detection model based on deep learning;

[0169] Based on the test set, the performance of the photovoltaic module EL defect detection model is tested;

[0170] Determine the performance test results based on the EL defect detection model of photovoltaic modules;

[0171] An optimization and adjustment unit, used to optimize and adjust the photovoltaic module EL defect detection model;

[0172] Obtain performance test results based on PV module EL defect detection model;

[0173] Conduct in-depth research and analysis on the performance test results based on the photovoltaic module EL defect detection model;

[0174] Determine the optimization adjustment scheme based on the EL defect detection model of photovoltaic modules;

[0175] Based on the optimization and adjustment scheme, the photovoltaic module EL defect detection model is optimized and adjusted;

[0176] Determine the best EL defect detection model for photovoltaic modules.

[0177] Among them, the defect detection module is used to identify and detect EL defects of photovoltaic modules and determine the EL defect detection results of photovoltaic modules;

[0178] In this embodiment, the defect detection module includes:

[0179] An image extraction unit, used for extracting EL characteristic images of photovoltaic modules based on machine vision;

[0180] According to the requirements of photovoltaic module EL defect detection, the EL feature image of photovoltaic modules based on machine vision is extracted;

[0181] Defect detection unit, used to identify and detect EL defects of photovoltaic modules;

[0182] Obtain the best EL defect detection model for photovoltaic modules;

[0183] Based on the best photovoltaic module EL defect detection model, the photovoltaic module EL defect recognition and detection is performed on the photovoltaic module EL feature image;

[0184] Determine the EL defect detection results of photovoltaic modules based on deep learning;

[0185] Based on the EL defect detection results of photovoltaic modules, the EL defect areas of photovoltaic modules and their corresponding types are marked with identification boxes;

[0186] A threshold adjustment unit is used to adjust the threshold interface and adjust the detection accuracy;

[0187] According to actual industrial needs, different detection thresholds and sensitivities are set for different types of EL defects of photovoltaic modules, and the threshold interface is adjusted to adjust the detection accuracy.

[0188] The data management module is used to display the EL characteristic images and EL defects of photovoltaic modules, store the classified information of defective photovoltaic modules and non-defective photovoltaic modules, and store the information of all tested photovoltaic modules.

[0189] In this embodiment, the data management module includes:

[0190] A picture display unit, used to display EL characteristic images of photovoltaic modules;

[0191] Among them, when detecting EL defects of photovoltaic modules, the EL characteristic image of the photovoltaic modules after detection is displayed, and it is automatically updated in real time according to the detection status of EL defects of photovoltaic modules;

[0192] When the EL defect detection of photovoltaic modules is stopped, the EL characteristic image of the photovoltaic modules that meets the user's needs is displayed according to the user's needs;

[0193] It should be noted that the picture display unit has two operating modes. One is that when the detection function is running, the pictures of the photovoltaic modules that have been inspected will be displayed at the same time and automatically updated in real time. The other is that when the detection is stopped, the user can use the "View" button on the console to select the picture he wants to view and display it on the system interface.

[0194] A defect display unit, used to display relevant information of EL defects of photovoltaic modules;

[0195] When detecting EL defects of photovoltaic modules, the relevant information of the detected EL defects of photovoltaic modules will be automatically updated and displayed in real time;

[0196] It should be noted that when the detection function is running, the relevant information of the detected photovoltaic module EL defects will be updated in real time and displayed in the text box.

[0197] A user control unit for setting PV module EL defect detection parameters and detection selection;

[0198] Among them, photovoltaic module EL defect detection includes all-type defect detection and single-type defect detection;

[0199] The user performs single-type defect detection based on the type of defect he wants to detect;

[0200] Among them, the EL defect detection parameter of photovoltaic modules is 0.25-1.00. The larger the parameter value, the stricter the defect screening.

[0201] It should be noted that single-type defect detection allows users to select the type of defect they want to detect through a drop-down selection box.

[0202] In this embodiment, the data management module further includes:

[0203] The data display unit is used for EL defect detection of photovoltaic modules, to screen out photovoltaic modules with defects, and to update and display their image file names and all defect information in real time for viewing;

[0204] It should be noted that during the detection process, defective photovoltaic modules are screened out, and their image file names and all defect information are updated in real time to the database display interface for viewing.

[0205] A data classification unit is used to classify information of defective photovoltaic modules and non-defective photovoltaic modules, so that users can query according to their needs, wherein the classification is performed according to date;

[0206] The data storage unit is used to establish a SQlite database to store the classified information of defective photovoltaic modules and non-defective photovoltaic modules, and store the information of all tested photovoltaic modules.

[0207] Therefore, the use of machine vision and deep learning can realize the positioning and classification of EL defects of photovoltaic modules. By reading the EL imaging pictures of photovoltaic modules, 14 types of EL defects can be detected, including dark spots, uneven current, broken grid, black spots, velvet, SE offset, fragments, linear cracks, virtual prints, branch-like cracks, black films, invalid films, clear traces, and mixed light and dark films. The defective areas and their corresponding types are marked with identification boxes. At the same time, an adjustable threshold interface is set up, and users can adjust the detection accuracy according to actual industrial needs.

[0208] It should be noted that when the photovoltaic module EL defect detection system is used to detect the EL defects of the photovoltaic module, the EL defect detection situation of the photovoltaic module is shown in Table 1:

[0209] Table 1: EL defect detection of photovoltaic modules

[0210] Defect Type Accuracy Recall Dark spots 0.927 0.861 Uneven current 0.933 0.867 Broken fence 0.881 0.829 Dark spots 0.936 0.728 Velvet 0.925 0.993 SE offset 0.914 0.885 Fragments 0.898 0.867 Linear cracks 0.889 0.841 Virtual Print 0.879 0.867 Dendrite crack 0.887 0.953 Black Film 0.974 0.895 Invalid film 0.943 0.972 Clear traces 0.967 0.879 Mixed light and dark film 0.895 0.865

[0211] Therefore, the photovoltaic module EL defect detection system has a high detection accuracy for these 14 types of defects and can meet the needs of industrial production.

[0212] In terms of detection rate, the system takes about 2.4 seconds to complete the detection of an image in a CPU environment, while in a GPU environment, the system's detection rate reaches 0.35 seconds per image, which can meet the industry's requirements for real-time detection. In terms of detection algorithms, the system optimizes the post-processing end, which greatly improves the detection system's detection effect in dense scenes and the quality of detection frame positioning.

[0213] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0214] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Photovoltaic module EL defect detection system based on machine vision and deep learning, characterized by: include: Image acquisition module, used to acquire real-time images of photovoltaic modules EL based on machine vision; An image processing module is used to process the real-time EL image of the photovoltaic module and determine the EL characteristic image of the photovoltaic module; Model training module, used to build a photovoltaic module EL defect detection model based on deep learning; A test optimization module is used to test and optimize the photovoltaic module EL defect detection model to determine the best photovoltaic module EL defect detection model; A defect detection module is used to identify and detect EL defects of photovoltaic modules and determine the EL defect detection results of photovoltaic modules; A data management module is used to display the EL characteristic images and EL defects of photovoltaic modules, store the information of defective photovoltaic modules and non-defective photovoltaic modules after classification, and store the information of all tested photovoltaic modules; The image processing module comprises: An image filtering unit, used for filtering the real-time image of the photovoltaic module EL; Obtain real-time EL images of photovoltaic modules based on machine vision; Based on the median filter, the real-time EL image of the photovoltaic module is filtered; The real-time EL image of the photovoltaic module is divided into multiple blocks, and the median value of the pixels in each block is taken; If a pixel value falls near the mean, it is considered to be the true value in the image, otherwise, the pixel is considered as noise and ignored; An image enhancement unit, used for enhancing the real-time image of the photovoltaic module EL; Obtain the real-time EL image of the photovoltaic module after filtering; Enhance the filtered photovoltaic module EL real-time image; Based on the method of adaptive histogram equalization, the difference between adjacent pixels in the real-time EL image of photovoltaic modules is improved, and the readability and contrast of the real-time EL image of photovoltaic modules are increased; A feature extraction unit, used for extracting features from the real-time EL image of the photovoltaic module; Obtain enhanced EL real-time images of photovoltaic modules; Extract features from the enhanced EL real-time image of photovoltaic modules; Determine the EL characteristic image of photovoltaic modules based on machine vision; Image enhancement unit, including: An image block extraction module, used to extract a plurality of image blocks corresponding to the real-time image of the photovoltaic module EL; The grayscale parameter value acquisition module is used to obtain the grayscale parameter value corresponding to each image block according to the grayscale value of the pixel block contained in each image block; wherein the grayscale parameter value is obtained by the following formula: Where J represents the grayscale parameter value; n represents the number of pixel blocks contained in the image block; H i Represents the grayscale value corresponding to the i-th pixel block; H p Indicates the grayscale average value corresponding to the image block; H z Represents the grayscale median value corresponding to all pixel blocks contained in the image block; An adjacent grayscale parameter value acquisition module is used to extract the grayscale parameter values ​​of adjacent image blocks corresponding to each image block; A target grayscale value acquisition module, used to acquire a target grayscale value of a pixel block contained in each image block according to the grayscale parameter value of an adjacent image block corresponding to each image block; A grayscale value adjustment module is used to adjust the grayscale value of each pixel block according to the target grayscale value corresponding to the pixel blocks contained in each image block, and obtain an image block in which all pixel blocks have completed the grayscale value adjustment; wherein the image block in which all pixel blocks have completed the grayscale value adjustment is the image block that has completed the image enhancement processing; The target gray value acquisition module includes: A grayscale parameter value information extraction module is used to extract the grayscale parameter values ​​of adjacent image blocks corresponding to each image block; The grayscale value adjustment coefficient acquisition module is used to obtain the grayscale value adjustment coefficient corresponding to each image block according to the grayscale parameter value of the adjacent image block corresponding to each image block; wherein the grayscale value adjustment coefficient is obtained by the following formula: Where, f represents the gray value adjustment coefficient corresponding to each image block; m represents the number of image blocks adjacent to each image block; J i represents the grayscale parameter value corresponding to the i-th adjacent image block; J d Indicates the grayscale parameter value of the current image block; J p Represents the average value of the grayscale parameter values ​​corresponding to m adjacent pixel blocks; A gray value parameter information extraction module is used to extract the gray value of the pixel block contained in each image block; The target grayscale value acquisition execution module is used to obtain the target grayscale value corresponding to each pixel block according to the grayscale value adjustment coefficient corresponding to each image block and the grayscale value of the pixel block contained in each image block; wherein the target grayscale value is obtained by the following formula: Among them, H m represents the target grayscale value corresponding to the pixel block; H0 represents the grayscale value of the pixel block; f represents the grayscale value adjustment coefficient corresponding to each image block; H z Represents the grayscale median value corresponding to all pixel blocks contained in the image block.

2. The photovoltaic module EL defect detection system based on machine vision and deep learning according to claim 1 is characterized in that: The image acquisition module comprises: A light source lighting unit, used to provide a suitable light source to illuminate the photovoltaic assembly to be photographed; The image acquisition unit is used to photograph the photovoltaic modules using a high-resolution industrial camera to obtain clear and accurate real-time EL images of the photovoltaic modules based on machine vision.

3. The photovoltaic module EL defect detection system based on machine vision and deep learning according to claim 2, characterized in that: The model training module includes: An image collection unit, used to collect historical images of EL defects of photovoltaic modules; According to the requirements of EL defect detection of photovoltaic modules, historical images of EL defects of photovoltaic modules based on dark spots, uneven current, broken grid, black spots, velvet, SE offset, debris, linear cracks, virtual prints, branch-shaped cracks, black pieces, invalid pieces, clear traces, and mixed light and dark pieces are collected; Based on the historical images of EL defects of photovoltaic modules as model training samples; A sample division unit, used for dividing the photovoltaic module EL defect history image; Obtain historical images of EL defects of photovoltaic modules; Divide the historical images of EL defects of photovoltaic modules; Determine the training set and test set; A model building unit, used to build a photovoltaic module EL defect detection model; According to the requirements of photovoltaic module EL defect detection, select a model architecture suitable for photovoltaic module EL defect detection; Based on the training set, the selected model architecture suitable for EL defect detection of photovoltaic modules is trained; A photovoltaic module EL defect detection model based on deep learning was determined.

4. The photovoltaic module EL defect detection system based on machine vision and deep learning according to claim 3 is characterized in that: The test optimization module includes: Performance test unit, used to perform performance test on photovoltaic module EL defect detection model; Obtain a photovoltaic module EL defect detection model based on deep learning; Based on the test set, the performance of the photovoltaic module EL defect detection model is tested; Determine the performance test results based on the EL defect detection model of photovoltaic modules; An optimization and adjustment unit, used to optimize and adjust the photovoltaic module EL defect detection model; Obtain performance test results based on PV module EL defect detection model; Conduct in-depth research and analysis on the performance test results based on the photovoltaic module EL defect detection model; Determine the optimization adjustment scheme based on the EL defect detection model of photovoltaic modules; Based on the optimization and adjustment scheme, the photovoltaic module EL defect detection model is optimized and adjusted; Determine the best PV module EL defect detection model.

5. The photovoltaic module EL defect detection system based on machine vision and deep learning according to claim 4 is characterized in that: The defect detection module comprises: An image extraction unit, used for extracting EL characteristic images of photovoltaic modules based on machine vision; According to the requirements of photovoltaic module EL defect detection, the EL feature image of photovoltaic modules based on machine vision is extracted; Defect detection unit, used to identify and detect EL defects of photovoltaic modules; Obtain the best EL defect detection model for photovoltaic modules; Based on the best photovoltaic module EL defect detection model, the photovoltaic module EL defect recognition and detection is performed on the photovoltaic module EL feature image; Determine the EL defect detection results of photovoltaic modules based on deep learning; Based on the EL defect detection results of photovoltaic modules, the EL defect areas of photovoltaic modules and their corresponding types are marked with identification boxes; A threshold adjustment unit is used to adjust the threshold interface and adjust the detection accuracy; According to actual industrial needs, different detection thresholds and sensitivities are set for different types of EL defects of photovoltaic modules, and the threshold interface is adjusted to adjust the detection accuracy.

6. The photovoltaic module EL defect detection system based on machine vision and deep learning according to claim 5 is characterized in that: The data management module comprises: A picture display unit, used to display EL characteristic images of photovoltaic modules; Among them, when detecting EL defects of photovoltaic modules, the EL characteristic image of the photovoltaic modules after detection is displayed, and it is automatically updated in real time according to the detection status of EL defects of photovoltaic modules; When the EL defect detection of photovoltaic modules is stopped, the EL characteristic image of the photovoltaic modules that meets the user's needs is displayed according to the user's needs; A defect display unit, used to display relevant information of EL defects of photovoltaic modules; When detecting EL defects of photovoltaic modules, the relevant information of the detected EL defects of photovoltaic modules will be automatically updated and displayed in real time; A user control unit for setting PV module EL defect detection parameters and detection selection; Among them, photovoltaic module EL defect detection includes all-type defect detection and single-type defect detection; The user performs single-type defect detection based on the type of defect he wants to detect; Among them, the EL defect detection parameter of photovoltaic modules is 0.25-1.

00. The larger the parameter value, the stricter the defect screening.

7. The photovoltaic module EL defect detection system based on machine vision and deep learning according to claim 6, characterized in that: The data management module also includes: The data display unit is used for EL defect detection of photovoltaic modules, to screen out photovoltaic modules with defects, and to update and display their image file names and all defect information in real time for viewing; A data classification unit is used to classify information of defective photovoltaic modules and non-defective photovoltaic modules, so that users can query according to their needs, wherein the classification is performed according to date; The data storage unit is used to store the information of the classified defective photovoltaic modules and non-defective photovoltaic modules, and to store the information of all the photovoltaic modules that have been tested.

Images