Photovoltaic panel AI defect detection method and equipment

Abstract

The present application is applicable to the field of photovoltaic cell detection technology, and in particular to an AI defect detection method and device for photovoltaic panels, the method comprising: obtaining an EL image of a target photovoltaic panel and starting the entire defect detection process; using a pre-trained photovoltaic panel detection model to process the EL image and identify various types of defects in the target photovoltaic panel; obtaining a defect result post-processing rule, based on a first defect analysis result, re-inspecting the target photovoltaic panel through the defect result post-processing rule, obtaining a second defect analysis result of the target photovoltaic panel, and further verifying and correcting the detection result; based on the historical defect analysis results and the second defect analysis results, comprehensively evaluating the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset time period, not only evaluating the quality of a single photovoltaic panel, but also monitoring and optimizing the quality control of the entire production line, thereby improving production efficiency and product qualification rate.

Description

Technical Field

[0001] The present application belongs to the field of photovoltaic cell detection technology, and in particular, relates to a photovoltaic cell panel AI defect detection method and device. Background Art

[0002] Photovoltaic panels are the core components of solar power generation systems. During production, transportation and use, they are prone to various defects such as cracks, broken grids, black spots, dirt, etc. due to manufacturing process defects, mechanical stress or environmental factors (such as humidity, temperature changes, etc.). These defects will not only reduce the power generation efficiency of photovoltaic panels and shorten their service life, but may also cause hot spot effects and even cause safety hazards such as fire.

[0003] In the existing technology, defect detection of photovoltaic panels relies on traditional AOI (automatic optical inspection) systems and manual visual inspection, which requires precise setting of detection standards and parameters. It takes a long time to adapt to new products, and is prone to false positives and missed positives under complex or boundary conditions. The lack of effective detection tools leads to low detection and classification accuracy and weak generalization. Summary of the invention

[0004] The embodiments of the present application provide a photovoltaic panel AI defect detection method and device, which can solve the problem of low detection and classification accuracy and weak generalization due to the lack of effective detection tools in the defect detection process of photovoltaic panels.

[0005] In a first aspect, an embodiment of the present application provides a photovoltaic panel AI defect detection method, comprising:

[0006] Acquire the EL image of the target photovoltaic panel;

[0007] Using a photovoltaic panel detection model to process the EL image of the target photovoltaic panel to obtain a first defect analysis result of the target photovoltaic panel; wherein the photovoltaic panel detection model is a machine learning model pre-trained using the EL image of a sample photovoltaic panel;

[0008] Obtaining defect result post-processing rules and historical defect analysis results; wherein the defect result post-processing rules are used to re-inspect the first defect analysis results; the historical defect analysis results are used to reflect the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset period of time before the target photovoltaic panel is detected;

[0009] Based on the first defect analysis result, re-inspecting the target photovoltaic panel using the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel;

[0010] Based on the historical defect analysis results and the second defect analysis results of the target photovoltaic panel, the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset time period is determined, including: based on the historical defect analysis results and the second defect analysis results, the diffusion path and rate of defects in the production line are simulated to obtain glass defect propagation data within a future preset time period, and based on the glass defect propagation data, the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within the preset time period is determined.

[0011] The above technical solutions in the embodiments of the present application have at least the following technical effects:

[0012] The photovoltaic panel AI defect detection method provided in the embodiment of the present application starts the entire defect detection process by acquiring the EL image of the target photovoltaic panel. The EL image is processed using a pre-trained photovoltaic panel detection model to identify various types of defects in the target photovoltaic panel. The defect result post-processing rules are obtained, and based on the first defect analysis result, the target photovoltaic panel is re-inspected through the defect result post-processing rules to obtain the second defect analysis result of the target photovoltaic panel, and the detection result is further verified and corrected. Based on the historical defect analysis results and the second defect analysis results, the overall quality of all photovoltaic panels of the corresponding production line of the target photovoltaic panel within a preset time period is comprehensively evaluated, not only the quality of a single photovoltaic panel is evaluated, but also the quality control of the entire production line is monitored and optimized, thereby improving production efficiency and product qualification rate, and comprehensively and systematically managing and improving the production quality of photovoltaic panels.

[0013] In a second aspect, an embodiment of the present application provides a photovoltaic panel AI defect detection system, comprising:

[0014] An acquisition unit, used for acquiring an EL image of a target photovoltaic panel;

[0015] a processing unit, configured to process the EL image of the target photovoltaic panel using a photovoltaic panel detection model to obtain a first defect analysis result of the target photovoltaic panel; wherein the photovoltaic panel detection model is a machine learning model pre-trained using the EL image of a sample photovoltaic panel;

[0016] A rule unit, used to obtain a defect result post-processing rule and a historical defect analysis result; wherein the defect result post-processing rule is used to re-check the first defect analysis result; the historical defect analysis result is used to reflect the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset period before the target photovoltaic panel is detected;

[0017] a re-inspection unit, configured to re-inspect the target photovoltaic panel based on the first defect analysis result and by using the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel;

[0018] A result unit is used to determine the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset time period based on the historical defect analysis results and the second defect analysis results of the target photovoltaic panel, including: simulating the diffusion path and rate of defects in the production line based on the historical defect analysis results and the second defect analysis results, obtaining glass defect propagation data within a future preset time period, and determining the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within the preset time period based on the glass defect propagation data.

[0019] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method as described in any one of the above aspects when executing the computer program.

[0020] In a fourth aspect, an embodiment of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device executes any one of the methods described in the above aspects.

[0021] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant descriptions in the above aspects and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 It is a flow chart of a photovoltaic panel AI defect detection method provided by an embodiment of the present application;

[0024] Figure 2 It is a schematic diagram of the operation of the photovoltaic panel AI defect detection method provided in one embodiment of the present application;

[0025] Figure 3 It is a structural schematic diagram of a photovoltaic panel AI defect detection system provided in one embodiment of the present application;

[0026] Figure 4 It is a schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0027] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0028] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0029] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] As used in the specification of this application and the appended claims, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrase "if it is determined" or "if the described condition or event is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce the described condition or event is detected" or "in response to detecting the described condition or event" depending on the context.

[0031] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0032] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0033] In the existing technology, defect detection of photovoltaic panels relies on traditional AOI (automatic optical inspection) systems and manual visual inspection, which requires precise setting of detection standards and parameters. It takes a long time to adapt to new products, and is prone to false positives and missed positives under complex or boundary conditions. The lack of effective detection tools leads to low detection and classification accuracy and weak generalization.

[0034] To solve the above problems, an embodiment of the present application provides an AI defect detection method and device for photovoltaic panels. In this method, the entire defect detection process is started by acquiring the EL image of the target photovoltaic panel. The EL image is processed using a pre-trained photovoltaic panel detection model to identify various types of defects in the target photovoltaic panel. The defect result post-processing rules are obtained, and based on the first defect analysis result, the target photovoltaic panel is re-inspected through the defect result post-processing rules to obtain the second defect analysis result of the target photovoltaic panel, and the detection result is further verified and corrected. Based on the historical defect analysis results and the second defect analysis results, the overall quality of all photovoltaic panels of the corresponding production line of the target photovoltaic panel within a preset time period is comprehensively evaluated, not only the quality of a single photovoltaic panel is evaluated, but also the quality control of the entire production line is monitored and optimized, thereby improving production efficiency and product qualification rate, and comprehensively and systematically managing and improving the production quality of photovoltaic panels.

[0035] The photovoltaic panel AI defect detection method provided in the embodiment of the present application can be applied to electronic devices. In this case, the electronic device is the executor of the photovoltaic panel AI defect detection method provided in the embodiment of the present application. The embodiment of the present application does not impose any restrictions on the specific type of electronic device.

[0036] For example, the electronic device may be an ultra-mobile personal computer (UMPC), a netbook, a desktop computer, a computer, a laptop computer, a communication device, a computing device, a satellite wireless device, etc.

[0037] In order to better understand the photovoltaic panel AI defect detection method provided in the embodiment of the present application, the specific implementation process of the photovoltaic panel AI defect detection method provided in the embodiment of the present application is exemplarily introduced below.

[0038] Figure 1 A schematic flow chart of a photovoltaic panel AI defect detection method provided in an embodiment of the present application is shown, and the photovoltaic panel AI defect detection method includes:

[0039] S100, acquiring an EL image of a target photovoltaic panel.

[0040] It can be understood that the EL image (Electroluminescence Image) is a visual image of the inside of a photovoltaic panel captured by electroluminescence technology. The EL image of the production line panel can be automatically captured from the production equipment (machine), and the image result detection can be performed, and the detection results and data statistics can be displayed in real time. The EL image of the photovoltaic panel can be manually uploaded and stored through the file upload portal, and the image format can be verified. The original image can be converted into a format suitable for target detection processing, such as an RGB image or a grayscale image. After obtaining the EL image of the target photovoltaic panel, image preprocessing operations can be performed on it, including denoising, grayscale, and normalization, so that the image quality meets the input requirements of the target detection algorithm. Image processing libraries (such as OpenCV) can be used to implement noise removal, grayscale processing, and image normalization operations, thereby improving the accuracy and efficiency of subsequent analysis.

[0041] S200, using a photovoltaic panel detection model to process the EL image of the target photovoltaic panel to obtain a first defect analysis result of the target photovoltaic panel; wherein the photovoltaic panel detection model is a machine learning model pre-trained using the EL image of a sample photovoltaic panel.

[0042] It can be understood that the photovoltaic panel detection model is a convolutional neural network built based on a deep learning framework (such as ResNet, YOLO or U-Net), which extracts the mapping relationship of defect features (such as hidden cracks, broken grids, and black cores) from EL images through training learning. The input of the photovoltaic panel detection model is the EL image of the target photovoltaic panel, and the output is the defect label. The photovoltaic panel detection model extracts image features through multi-layer convolution and pooling operations, locates the defect area using a fully connected layer or anchor frame mechanism, and determines the defect type through classification branches. After obtaining the defect label, the target photovoltaic panel can be judged by the defect label. According to a series of parameters such as the defect category, size, location, area, grayscale value, etc., it is automatically judged whether the sample meets the quality standards, and the initial defect analysis result of the target photovoltaic panel, that is, the first defect analysis result, is obtained.

[0043] In a possible implementation, S200, using a photovoltaic panel detection model to process an EL image of a target photovoltaic panel to obtain a first defect analysis result of the target photovoltaic panel includes:

[0044] S210, using the photovoltaic panel detection model to process the EL image of the target photovoltaic panel to obtain a defect label of the EL image of the target photovoltaic panel.

[0045] It can be understood that the EL image of the target photovoltaic panel can be input into the photovoltaic panel detection model, and the photovoltaic panel detection model will generate a defect label corresponding to the target photovoltaic panel. The photovoltaic panel detection model can not only locate the target, but also classify each detected target, so as to achieve accurate detection of panel defects. The photovoltaic panel detection model can identify each defective area in the input EL image, and provide the type, location and size information of the defective area. For example, defect types can include dirt, broken grid, cross mark, hidden crack, black edge and black corner, black spot, concentric circle, scratch, fingerprint, etc.

[0046] S220, making a judgment based on the defect label of the EL image to obtain a first defect analysis result of the target photovoltaic panel; wherein the first defect analysis result includes any one of qualified and unqualified.

[0047] It can be understood that the determination of the first defect analysis result is based on the preset quality standard rules, and the specific rules set thresholds according to the defect type, size, location and quantity. For example: Hidden crack: If the length is ≥3mm or the number is ≥2, it is determined to be "unqualified", and the first defect analysis result is "unqualified"; Broken grid: If the length of continuous broken grid is ≥2mm or the total area of ​​the broken grid area is ≥5mm², it is determined to be "unqualified", and the first defect analysis result is "unqualified"; Black core: If the area of ​​the black core area is ≥1mm² and is located within 5mm of the center of the battery cell, it is determined to be "unqualified", and the first defect analysis result is "unqualified". For example, the defect label of an EL image contains:

[0048] Hidden crack: length 4.2mm (coordinate X: 130-134mm, Y: 10-14mm);

[0049] Broken grid: continuous length 1.8mm (coordinate X: 80-81.8mm, Y: 50-50mm).

[0050] According to the preset quality standard rules, if the length of the hidden crack exceeds the threshold of 3mm, it will be directly judged as "unqualified". If all defects do not trigger the threshold, it will be marked as "qualified". The process of judging the target photovoltaic panel by defect label can be configured as an editable rule library, which supports production line managers to dynamically adjust the threshold according to process improvements.

[0051] Optionally, before using the photovoltaic panel detection model to process the EL image of the target photovoltaic panel to obtain a first defect analysis result of the target photovoltaic panel, the method further includes:

[0052] S600 , acquiring a sample EL image using a sample photovoltaic panel.

[0053] It can be understood that the sample EL images are the basic data for building the photovoltaic panel inspection model, which can include sample EL images of typical defect types (such as hidden cracks, broken grids, black spots, dirt, etc.) of photovoltaic panels under different production lines and different process conditions, as well as normal sample EL images. Each photovoltaic panel can be inspected for appearance in a standardized environment: the sample photovoltaic panel is placed in a dark room, a constant current (such as 1.2 times the nominal current) is applied to stimulate electroluminescence, and a high-resolution industrial camera (such as 4096×4096 pixels, 16-bit grayscale depth) is used to capture the image, with an exposure time of 30-60 seconds to balance image quality and acquisition efficiency. For example, the sample library contains 180 million EL sample images, and the detection items include 17 defect types such as broken grid, hidden cracks, black spots, black dots, belt marks, scratches, over-engraving, concentric circles, roller marks, dirt, fog and black, offset, boat marks, frame marks, missing corners, black edges, and black lines. The resolution is unified at 5μm / pixel to capture micron-level defects. It also includes samples with different lighting conditions (such as overexposure, underexposure) and noise interference (such as sensor noise and dust interference) to simulate the complex scenes in the real production line environment.

[0054] S700 , determining a defect label corresponding to a sample EL image of a sample photovoltaic panel.

[0055] It can be understood that the defect label is a supervisory signal for model training, which can accurately mark the defect type, location coordinates, size and severity. The defect label generation can adopt a hybrid process of "automatic pre-labeling + manual precision labeling": the defect area is initially identified through traditional image processing algorithms (such as Canny edge detection, Otsu threshold segmentation) or pre-trained models to generate rough labels; then the precise labels are obtained by quality inspection experts using labeling tools (such as LabelImg, CVAT) to correct the pre-labeling results to ensure label accuracy. For example, the boundary of the hidden crack needs to be accurate to ±0.1mm, the broken grid area needs to mark the starting and ending points of the continuous grid line break, and the black spot needs to circle the precise outline of the low grayscale area. The defect label format can adopt the COCO standard and be stored as a JSON file, including category_id (defect type code), bbox (bounding box coordinates, such as [120,10,5,0.2] represents a hidden crack area with X=120mm, Y=10mm, length 5mm, width 0.2mm) and area (pixel area). For complex defects (such as the coexistence of hidden cracks and broken grids), they need to be labeled separately and associated with the same photovoltaic panel ID to support multi-label training.

[0056] In a possible implementation, S700, determining a defect label corresponding to a sample EL image of a sample photovoltaic panel, includes:

[0057] S710, performing visual feature recognition on a sample EL image of a sample photovoltaic panel to obtain a recognition analysis result of the sample EL image.

[0058] It can be understood that visual feature recognition is to extract potential defect areas and their attributes in sample EL images through image processing technology, providing a quantitative basis for subsequent label generation. Uneven illumination and sensor noise can be eliminated by performing Gaussian filtering denoising, CLAHE contrast enhancement and grayscale normalization preprocessing on the original EL image; adaptive threshold segmentation (such as local binarization) or semantic segmentation model (such as U-Net) is used to separate suspected defect areas, for example, areas with grayscale values ​​lower than 2 times the standard deviation of the background mean are marked as candidate defects; morphological features (area, perimeter, aspect ratio), texture features (LBP, Haralick texture) and grayscale statistics (mean, variance, gradient distribution) of candidate areas can be extracted, for example, hidden crack areas usually present an elongated shape (aspect ratio>5) and the grayscale gradient is continuous along the grain boundary direction, while broken grid areas appear as fracture bands along the grid line direction.

[0059] S720 , determining a defect label corresponding to the sample EL image of the sample photovoltaic panel based on the recognition and analysis result of the sample EL image.

[0060] It can be understood that the visual feature recognition results can be mapped into standardized defect labels in combination with the predefined rule base. For example, the feature combination of elongated areas (aspect ratio>5) and continuous grayscale gradient is judged as hidden cracks, the broken area along the grid line direction is marked as broken grid, and the circular low grayscale area is classified as black spots. The position annotation takes the lower left corner of the photovoltaic panel as the origin, records the center coordinates (X, Y) of the defect area and the size of the bounding box (width W, height H), for example, the hidden crack is marked as bbox:[120,10,5,0.2]. The severity classification is based on the size threshold, such as the hidden crack length <2mm is mild, 2-5mm is moderate, and >5mm is severe. For complex defects (such as the coexistence of hidden cracks and broken grids), the spatial relative position can be independently labeled and recorded to support the model to learn the associated features of multiple defects.

[0061] S800, taking the sample EL image of the sample photovoltaic panel as the expected input and the corresponding defect label as the expected output, the initial machine learning model is trained to obtain a trained photovoltaic panel detection model.

[0062] It can be understood that the photovoltaic panel detection model training can adopt a supervised learning framework, and can choose the target detection architecture (such as YOLOv8 or Faster R-CNN) or the segmentation model (such as Mask R-CNN). The backbone network can use ResNet-50 or EfficientNet-B7 to balance accuracy and speed; apply data enhancement to the training set to improve the robustness of the model; define a multi-task loss function, and jointly optimize the classification loss (Focal Loss solves category imbalance), positioning loss (Smooth L1Loss optimizes bounding box regression) and segmentation loss (Dice Loss improves mask accuracy).

[0063] For example, 18,000 sample EL images covering 17 types of defects such as hidden cracks, broken grids, black spots, and concentric circles are used to train the initial machine model. The resolution of each image is 2048×2048 pixels (16-bit grayscale). The data distribution shows a long-tail characteristic (e.g., hidden cracks account for 30% and concentric circles account for only 2%). In order to improve the generalization ability of the model, a dynamic data enhancement strategy can be adopted: including random horizontal flipping (probability 50%), rotation (±15°), Gaussian noise (σ=0.03) and brightness jitter (±15%), and the introduction of synthetic defects generated by GAN (such as simulated black edges and black corners) to supplement minority category samples. YOLOv8 is selected as the basic model to adapt to the photovoltaic detection scene. The sample EL image is adjusted to a resolution of 640×640 (keeping the aspect ratio filling), the backbone network uses CSPDarknet53, and the neck introduces BiFPN to enhance multi-scale feature fusion. The training parameters are set as follows: batch size 8 (NVIDIAA100 GPU), initial learning rate 3e-5 (cosine annealing schedule), and training rounds of 300 rounds (about 67,500 iterations). The loss function uses the improved Focal Loss (α=0.25, γ=2.0), and dynamically adjusts the loss weight according to the sample ratio of 17 types of defects (such as the weight coefficient of concentric circles is set to 3.0, and the hidden crack is set to 0.8). During the training process, the validation set (3,600 images) is evaluated every 10 rounds. After training, the photovoltaic panel detection model can achieve real-time inference of ≤150ms for a single EL image, output defect type, location and size, with an accuracy rate of ≥90% (verified by the 10,000-level test set) and a misjudgment rate of ≤8%, as shown in the following table:

[0064]

[0065] S300, obtaining defect result post-processing rules and historical defect analysis results; wherein, the defect result post-processing rules are used to re-inspect the first defect analysis result; the historical defect analysis results are used to reflect the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset period before the target photovoltaic panel is detected.

[0066] It can be understood that the defect result post-processing rules are a set of dynamically adjustable decision logics used to conduct a second re-inspection and verification of the first defect analysis results (such as "unqualified") output by the photovoltaic panel detection model to reduce the misjudgment rate and improve the detection reliability. The defect result post-processing rules can be constructed based on expert experience through a real-time interactive interface or based on historical data, and can include the following core logic:

[0067] Type-size threshold linkage: Dynamic size thresholds are set for different defect types. For example, re-inspection is triggered when the length of a hidden crack is ≥3mm or the area of ​​a broken grid is ≥2mm², while the threshold for black spots is relaxed to ≥5mm² because they have less impact on performance.

[0068] Position sensitivity correction: stricter judgment is adopted for defects in key areas (such as 5mm range in the center of the battery cell and ±2mm near the electrode). For example, a hidden crack in the center area needs to be re-inspected even if it is only 1mm in length, while a hidden crack of the same size at the edge may be released directly.

[0069] Multiple defect superposition rule: If the same photovoltaic panel has both hidden cracks (length ≥ 2mm) and broken grids (length ≥ 1mm), even if a single defect does not exceed the threshold, re-inspection is still mandatory.

[0070] The historical defect analysis results are statistical data sets extracted from the database of the target photovoltaic panel production line, covering the defect data of all photovoltaic panels that have been inspected within a preset period of time (such as the past 72 hours). The integration of defect result post-processing rules and historical defect analysis results can embed single-board inspection results into the global quality portrait of the production line, realizing closed-loop control from "single-point judgment" to "system optimization".

[0071] S400, based on the first defect analysis result, re-inspect the target photovoltaic panel using a defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel.

[0072] It can be understood that the target photovoltaic panel with an unqualified first defect analysis result can be re-inspected by identifying the first defect analysis result. The dynamic defect result post-processing rules and multi-dimensional re-inspection process are used for secondary verification to reduce the risk of misjudgment and improve the reliability of detection. The defect label (including defect type, location coordinates and size) of the target photovoltaic panel EL image with an unqualified first defect analysis result can be extracted, and the final re-inspection result, i.e., the second defect analysis result, can be generated by combining the defect result post-processing rules and physical coordinate system mapping analysis.

[0073] In a possible implementation, S400, based on the first defect analysis result, re-inspecting the target photovoltaic panel by using a defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel includes:

[0074] S410, when the first defect analysis result of the target photovoltaic panel is unqualified, re-inspect the target photovoltaic panel according to the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel.

[0075] It can be understood that for photovoltaic panels with a first defect analysis result of "unqualified", a second verification is performed through dynamic defect result post-processing rules and multi-dimensional re-inspection processes to reduce the risk of misjudgment and improve detection reliability. The defect label (including defect type, location coordinates and size) of the target photovoltaic panel EL image with a first defect analysis result of unqualified can be extracted, and the final re-inspection result, i.e., the second defect analysis result, is generated by combining the defect result post-processing rules and physical coordinate system mapping analysis.

[0076] Optionally, in S410, when the first defect analysis result of the target photovoltaic panel is unqualified, re-inspecting the target photovoltaic panel according to the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel includes:

[0077] S411, obtaining a defect label of the EL image of the target photovoltaic panel; wherein the defect label includes the defect type, defect position, and defect size of each defect area in the EL image.

[0078] It can be understood that the defect label is the structured data output by the photovoltaic panel detection model after detecting the EL image of the target photovoltaic panel, including the defect type (such as hidden cracks, broken grids, black spots), defect location (pixel coordinates with the upper left corner of the image as the origin or millimeter coordinates in the physical coordinate system) and defect size (such as hidden crack length, broken grid area) of each defect area. For example, the defect label of a photovoltaic panel can include defect type, defect location, defect size, etc.:

[0079] Hidden crack: type code C1, position coordinates (X: 120-125mm, Y: 10-10mm), length 5.2mm;

[0080] Broken grid: type code C3, position coordinates (X:80-82mm, Y:50-50mm), length 2.1mm.

[0081] Tag data is usually stored in JSON or XML format to support subsequent rule engine parsing and processing.

[0082] S412, judging the defect type, defect position, and defect size of each defect area in the EL image through the defect result post-processing rule, and obtaining the first defect re-inspection result of the target photovoltaic panel.

[0083] It can be understood that the defect result post-processing rules are constructed in real time through an interactive interface based on expert experience or a rule database based on historical data, which may include the following core logic:

[0084] Type-size threshold linkage: Dynamic size thresholds are set for different defect types. For example, re-inspection is triggered when the length of a hidden crack is ≥3mm or the area of ​​a broken grid is ≥2mm², while the threshold for black spots is relaxed to ≥5mm² because they have less impact on performance.

[0085] Position sensitivity correction: stricter judgment is adopted for defects in key areas (such as 5mm range in the center of the battery cell and ±2mm near the electrode). For example, even if the length of the hidden crack in the center area is only 1mm, it will not pass the re-inspection and will be judged as unqualified, while the hidden crack of the same size at the edge may be released directly.

[0086] Multiple defect superposition rule: If the same photovoltaic panel has both hidden cracks (length ≥ 2mm) and broken grids (length ≥ 1mm), even if a single defect does not exceed the threshold, it will still be forced to fail the re-inspection and be judged as unqualified.

[0087] The first defect re-inspection result is the result of re-inspecting the target photovoltaic panel using the defect result post-processing rule after the photovoltaic panel detection model determines that the target photovoltaic panel is unqualified. The defect result post-processing rule can be applied to the re-inspection process of the target photovoltaic panel through the rule engine to obtain the first defect re-inspection result.

[0088] The defect result post-processing rules include size determination rules, sensitivity analysis rules and defect type priority; illustratively, in S412, the defect type, defect position and defect size of each defect area in the EL image are determined by the defect result post-processing rules to obtain the first defect re-inspection result of the target photovoltaic panel, including:

[0089] S4121, performing quantitative analysis on the defect size according to the defect type of each defect area in the EL image through the defect result post-processing rule, and determining whether the defect area meets the size determination rule.

[0090] It can be understood that the size determination rules in the defect result post-processing rules can be dynamically matched according to the defect type. For example, the determination threshold of hidden cracks may be set to a length ≥ 3mm, while the determination threshold of broken grids is a continuous length ≥ 2mm or a total area ≥ 4mm². During quantitative analysis, the actual size can be compared with the threshold in the size determination rule. For example, the length of a hidden crack is 2.9mm (slightly lower than the threshold of 3mm), and the re-inspection is in compliance with the size determination rule. For defects with less impact on performance, such as black spots, the threshold may be relaxed to an area ≥ 5mm², and dynamically adjusted in combination with position sensitivity (such as a larger acceptable area in the edge area).

[0091] S4122, when the defect area meets the size determination rule, coordinate transformation is performed on the defect position of the defect area, the defect position is mapped to the physical coordinate system of the target photovoltaic panel, and the sensitivity analysis rule analysis result of the defect position is determined.

[0092] It can be understood that coordinate transformation is to map the defect position from the image pixel coordinate system to the photovoltaic panel physical coordinate system (such as X: 120mm, Y: 50mm) to adapt the production line process parameters (such as welding point location, cell partitioning). The sensitivity analysis rule is based on the defect result post-processing rule to evaluate the impact of the defect position on the performance of the photovoltaic panel. For example, the ±5mm range of the cell center is defined as a high-sensitivity area. Even if the hidden cracks in this area are small in size, they may cause significant performance degradation. Even if they meet the size judgment rules in the defect result post-processing rules, they are judged as not meeting the sensitivity analysis rules; while the edge buffer zone (< or = 10mm from the edge) is a low-sensitivity area, and defects of the same size may be marked as "meeting the sensitivity analysis rules". For example, a hidden crack is located in the central area (coordinates X: 75mm, Y: 75mm), even if the length is only 1.5mm (below the threshold of 3mm), it is still judged as "not meeting the sensitivity analysis rules".

[0093] S4123, obtaining the first defect re-inspection result of the target photovoltaic panel according to the sensitivity analysis rule analysis results of all defect areas and the defect type priority.

[0094] It can be understood that the defect type priority is a pre-defined rule in the defect result post-processing rules, which is used to identify the severity level of the impact of different defects on the performance of photovoltaic panels, such as hidden cracks (priority 1) > broken grid (priority 2) > black spots (priority 3). The defect type with the highest priority will dominate the judgment logic - that is, the target photovoltaic panel is qualitatively divided into defects by priority level. For example, if the target photovoltaic panel has hidden cracks (priority 1), broken grid (priority 2) and black spots (priority 3) and other defects, the first defect re-inspection result of the target photovoltaic panel is judged to be [unqualified, hidden cracks].

[0095] S413, performing grayscale detection on the EL image of the target photovoltaic panel to determine a second defect re-inspection result of the target photovoltaic panel.

[0096] It can be understood that the EL image of the target photovoltaic panel can be finely analyzed through grayscale detection technology to verify or correct the first defect re-inspection result, reduce the misjudgment rate and improve the detection accuracy. Grayscale detection focuses on the grayscale value distribution characteristics of image pixels, combines anomaly detection algorithms with spatial distribution analysis, identifies grayscale abnormal areas caused by real defects (such as hidden cracks and broken grids), and eliminates pseudo-abnormal areas caused by uneven lighting or noise.

[0097] Exemplarily, S413, performing grayscale detection on the EL image of the target photovoltaic panel to determine a second defect re-inspection result of the target photovoltaic panel includes:

[0098] S4131, performing grayscale detection on the EL image of the target photovoltaic panel to obtain grayscale distribution characteristics of the EL image.

[0099] It can be understood that the grayscale detection of the EL image of the target photovoltaic panel can be performed to generate a grayscale histogram and count the pixel distribution of each grayscale level. For example, the grayscale value of the normal area is concentrated in 120-180 (8-bit grayscale range), while the grayscale value of the cracked area may drop to 80-100 due to the reduced electroluminescence intensity. The high-frequency noise can be removed by using Gaussian filtering (kernel size 5×5, σ=1.5) during preprocessing, and the low-frequency grayscale gradient characteristics can be retained.

[0100] S4132, based on the grayscale distribution characteristics of the EL image, the grayscale abnormal area in the EL image is identified by using grayscale value statistics and anomaly detection algorithm, and the anomaly detection threshold is dynamically adjusted according to the peak and valley values ​​of the grayscale distribution to obtain the grayscale abnormal area.

[0101] It can be understood that the analysis can be based on the peak and valley values ​​of the grayscale histogram, and the abnormal detection threshold can be dynamically set through the preset peak and valley value rules. For example, if the main peak of the grayscale distribution is at 150 (normal area) and the secondary peak is at 90 (defective area), the abnormal area is preliminarily divided by the valley value between the two peaks (such as 120). The grayscale mean and standard deviation of the local area can be calculated in combination with statistical methods (such as the 3σ principle): if the grayscale mean of a grayscale abnormal area is lower than the global mean by 2 times the standard deviation (such as the mean is 100 and the standard deviation is 20, then the threshold = 100-2×20=60), it is marked as a candidate abnormality. Dynamically setting the abnormality detection threshold can adapt to different lighting conditions, such as increasing the threshold by 10% in overexposed images.

[0102] S4133, perform spatial distribution analysis on the identified grayscale abnormal area, and eliminate the pseudo-abnormal area according to the shape, texture and grayscale change trend of the defect area corresponding to the grayscale abnormal area, and obtain the area size and grayscale difference degree of each remaining grayscale abnormal area; wherein the pseudo-abnormal area is the abnormal area caused by uneven lighting and image noise.

[0103] It can be understood that the candidate grayscale abnormal area can be analyzed for spatial distribution, and whether it is a pseudo abnormal area can be determined by the preset spatial position rules. Pseudo abnormal areas are abnormal areas caused by uneven illumination and image noise. For example, the grayscale abnormal area is concentrated in a specific position of the photovoltaic panel (such as the edge or near the welding point). If the grayscale abnormal area is isolated and distributed in a non-process sensitive area (such as a buffer zone ≥15mm from the edge), it may be marked as a pseudo abnormality. Irregular spots can be filtered based on shape features (such as aspect ratio, area perimeter ratio): real hidden cracks are usually long and thin strips (aspect ratio>5, area perimeter ratio<0.2), while noise artifacts are mostly nearly circular (aspect ratio≈1, area perimeter ratio>0.5). For example, a grayscale abnormal area with an aspect ratio of 1.3 and an area perimeter ratio of 0.6 is determined to be noise and removed. The nature of the grayscale abnormal area can be further verified by texture analysis. For example, by calculating the local binary pattern (LBP) or Haralick texture features (such as contrast, energy), the real defect area shows low texture complexity (LBP contrast < 0.5) due to structural regularity (such as hidden cracks extending along grain boundaries), while the pseudo-abnormal area shows high texture complexity (LBP contrast > 0.8) due to random noise. For example, the LBP contrast of a grayscale abnormal area is 0.9, which is determined to be an artifact caused by uneven illumination. The grayscale gradient direction consistency can be analyzed by grayscale change trend. The grayscale gradient of the real defect (such as broken grid) along the grid line is continuous and consistent in direction (gradient direction standard deviation < 10°), while the grayscale change of the pseudo-abnormal area is irregular (gradient direction standard deviation > 30°). After the pseudo-abnormal area is removed, the area of ​​the remaining grayscale abnormal area (such as 5mm²) and the degree of grayscale difference can be calculated. The grayscale difference is the absolute difference between the grayscale mean of the abnormal area and the global grayscale mean (such as the regional grayscale mean is 60 lower than the global mean). For example, a grayscale abnormal area with an area of ​​4.8mm² and a grayscale difference of 55 is retained as a real defect; while a grayscale abnormal area with an area of ​​1.2mm² and a grayscale difference of 20 is removed because its shape and texture do not match.

[0104] S4134, performing a quantitative score on each grayscale abnormal region according to the area size and grayscale difference degree of the grayscale abnormal region.

[0105] It can be understood that the physical characteristics (area, grayscale difference) of the grayscale abnormal area are converted into comparable numerical indicators through quantitative scoring, providing an objective basis for subsequent judgment. The weighted scoring formula can be used: score = w1×area+w2×grayscale difference; among which, the weight w1=0.6 (area has a greater impact on the performance of photovoltaic panels), w2=0.4 (grayscale difference reflects the severity of defects). Grayscale difference is defined as the absolute difference between the grayscale mean of the abnormal area and the global grayscale mean (such as the global mean is 150, the mean of the abnormal area is 90, and the difference is 60). For example, a certain hidden crack area has an area of ​​5mm² and a grayscale difference of 60, and its score is 0.6×5+0.4×60=3+24=27; while a certain black spot area has an area of ​​2mm² and a grayscale difference of 30, and its score is 0.6×2+0.4×30=1.2+12=13.2. The scoring threshold can be adjusted dynamically based on historical data: if the missed detection rate of a production line has increased recently, the threshold may be reduced from 20 to 18 to increase sensitivity.

[0106] S4135, determining the second defect re-inspection result of the target photovoltaic panel based on the grayscale distribution characteristics of the EL image and the quantitative scores of all grayscale abnormal areas.

[0107] It can be understood that the global grayscale distribution characteristics and the quantitative scores of grayscale abnormal areas can be analyzed through preset rules to generate the final second defect re-inspection results. The main peak position of the grayscale distribution of the EL image and the overall shift trend can be evaluated. For example, the grayscale main peak of a normal photovoltaic panel is usually located at 150 (8-bit grayscale range). If the main peak shifts significantly to the left to 130, it indicates that the electroluminescence intensity is reduced as a whole, which may imply batch process defects (such as hidden crack clusters or material contamination). At this time, even if the score of a single abnormal area does not exceed the threshold (such as a score of 18), manual review can be triggered to check for potential risks. For local abnormal areas, the quantitative score combines the area and grayscale difference (the formula is: score = 0.6×area+0.4×grayscale difference). If there is at least one area with a score of >25 (such as a hidden crack area of ​​5mm², a grayscale difference of 60, and a score of 27), it is directly judged as "unqualified"; if the highest score is between 15-25 (such as a broken grid score of 22), it is judged in combination with the shift amplitude of the main peak of the grayscale distribution (such as a left shift of 10%). For example, the main peak of a photovoltaic panel shifted 12% to the left and there were two abnormal areas (scores 23 and 17). Due to the shift of the main peak, it was confirmed that there were micro cracks, and the result of the second defect re-inspection was determined to be "unqualified".

[0108] S414, obtaining a second defect analysis result of the target photovoltaic panel based on the first defect re-inspection result and the second defect re-inspection result of the target photovoltaic panel.

[0109] It can be understood that the final judgment conclusion can be generated by logically fusing the first defect re-inspection result and the second defect re-inspection result, that is, the second defect analysis result of the target photovoltaic panel is obtained. If the first defect re-inspection result is "unqualified" (triggered by a high-priority defect, such as a hidden crack length exceeding the threshold), the second re-inspection result is directly overwritten and the final judgment is "unqualified". For example, a photovoltaic panel triggers the first re-inspection "unqualified" due to a hidden crack (priority 1, length 3.5mm). Even if the grayscale score of the second re-inspection is only 18 (lower than the threshold 20), the "unqualified" conclusion is still maintained, so that high-risk defects are strictly intercepted.

[0110] S500, based on the historical defect analysis results and the second defect analysis results of the target photovoltaic panel, determine the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset time period, including: based on the historical defect analysis results and the second defect analysis results, simulate the diffusion path and rate of defects in the production line to obtain glass defect propagation data within a future preset time period, and determine the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within the preset time period based on the glass defect propagation data.

[0111] It can be understood that the global quality portrait of the production line can be constructed and future process risks can be predicted through the historical defect analysis results and the current detection results (second defect analysis results). The historical defect analysis results (such as the defect type, location and size of all photovoltaic panels on the production line in the past 72 hours) and the second defect analysis results of the target photovoltaic panel (the final judgment confirmed by re-inspection) can be integrated to form a defect data set. The defect data set can include the unique identification of each photovoltaic panel, the detection timestamp, the defect type code, the physical coordinates (such as X: 120mm, Y: 50mm), the size (such as the length of the hidden crack 3.2mm) and the judgment result (qualified / unqualified). For example, among the 1,000 photovoltaic panels detected by a production line within 24 hours, the hidden cracks accounted for 35% (350 cases) and the broken grids accounted for 20% (200 cases). The second defect analysis result of the target photovoltaic panel showed hidden cracks (length 4.1mm, coordinates X: 80mm, Y: 30mm), which were added to the defect data set. The spatiotemporal feature modeling of the defect data set can be performed. The surface of the photovoltaic panel is divided into grid units, and the number and type of defects in each grid unit are counted to generate a spatial probability density map. At the same time, the preset time period (such as 24 hours) can be divided into a time interval per hour, and the trend of the number of defects can be counted to generate a time trend curve. The spatial probability density map and the time trend curve can be fused to obtain the spatiotemporal characteristic data of the target photovoltaic panel corresponding to the production line in the preset time period. Based on the spatiotemporal characteristic data, a defect association network can be constructed, and the defect area of ​​each photovoltaic panel is regarded as a node. The association relationship between nodes (such as the co-occurrence probability of hidden cracks and broken grids) is defined as an edge to form a network topology map. By simulating the defect diffusion path, the number and spatial distribution of new defects that may be added to the production line within the preset time period in the future are predicted, and defect propagation data is generated. Based on the defect data set and the defect propagation data in the future preset time period, the overall quality report of the production line is output, that is, the overall quality of all photovoltaic panels of the target photovoltaic panel corresponding to the production line in the preset time period.

[0112] In a possible implementation, S500, based on the historical defect analysis result and the second defect analysis result of the target photovoltaic panel, determines the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset period of time, including:

[0113] S510, constructing a defect data set including defect data of all photovoltaic panels of a production line corresponding to a target photovoltaic panel within a preset time period according to the historical defect analysis result and the second defect analysis result.

[0114] It can be understood that the historical defect analysis results and the second defect analysis results can be integrated through data extraction and aggregation algorithms to construct a defect data set containing the defect data of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset period of time, providing a data basis for production line quality analysis and prediction.

[0115] Exemplarily, data fields can be defined in advance, for example, each record can contain fields: photovoltaic panel ID, detection time, defect type, location coordinates (X / Y), size, grayscale score, judgment result, data source (historical defect analysis results / second defect analysis results), and the historical defect analysis results and the second defect analysis results are unified to the same time zone (such as UTC+8) according to the current detection time, and sorted by detection time to obtain a defect data set.

[0116] S520, modeling the defect distribution law of all photovoltaic panels within a preset time period based on the defect data set to obtain spatiotemporal characteristic data of the defect distribution; wherein the spatiotemporal characteristic data is used to reflect the high probability areas and time clustering of the defect distribution of all photovoltaic panels within the preset time period.

[0117] It can be understood that the distribution pattern and evolution law of defects in the production line can be revealed through the joint analysis of space and time dimensions, providing a quantitative basis for process optimization. The surface of the photovoltaic panel can be divided into grid units, and the frequency and spatial density of different defect types in each grid can be counted to generate a spatial probability density map. The preset time period can be divided into multiple time intervals by hour, and the number and type distribution changes of defects in the interval can be counted to generate a time trend curve. By matrixing the time trend curve and the spatial probability density map, the spatiotemporal characteristic data of the defect distribution can be obtained. The spatiotemporal characteristic data is used to reflect the high probability areas and time aggregation of the defect distribution of all photovoltaic panels in the preset time period.

[0118] Optionally, S520, modeling the defect distribution law of all photovoltaic panels within a preset period of time according to the defect data set to obtain spatiotemporal characteristic data of defect distribution, including:

[0119] S521, dividing the surface of the photovoltaic panel into a number of grid cells of equal size according to the defect data set, and counting the number of defects and the proportion of their types in each grid cell respectively.

[0120] It can be understood that the surface of the photovoltaic panel can be divided into grid units of equal size (such as 10mm×10mm), and each unit corresponds to a two-dimensional coordinate interval. For each grid unit, the number of defects of all photovoltaic panels in the unit within a preset period of time and the number ratio of each defect type (type ratio) can be counted.

[0121] S522, based on the number of defects and the proportion of their types in each grid unit, a spatial probability density map of defect distribution is generated; wherein the probability density map is used to reflect the spatial concentration area of ​​defect distribution and its distribution boundary.

[0122] It can be understood that based on the number and type of defects in each grid cell, a spatial probability density map can be generated using kernel density estimation (KDE) or histogram statistical methods. The probability density value represents the relative frequency of a certain type of defect in a specific grid cell. The spatial probability density map reflects the spatial concentration area and distribution boundary of the defect distribution of all photovoltaic panels within a preset period of time.

[0123] S523, dividing the preset time period into a plurality of continuous time intervals according to the defect data set, and counting the total number of defects and the spatial distribution change trend in each time interval.

[0124] It can be understood that the preset time period (such as 24 hours) can be divided into continuous time windows (such as one window per hour), and the total number of each defect type in each window and the spatial distribution changes are counted. For example, 40 broken grids were detected in window W1 (9:00-10:00), of which 70% were concentrated in the coordinate area X:80-120mm, while the number of broken grids in window W2 (10:00-11:00) dropped to 25, and the spatial distribution spread to X:120-150mm.

[0125] S524, based on the total number of defects and the spatial distribution change trend in each time interval, segmented modeling is performed on the frequency of occurrence of defects to obtain a time trend curve of the defect changing over time.

[0126] It can be understood that the frequency of defects can be modeled in segments using the moving average method or the autoregressive integrated moving average model (ARIMA). Periodic signals (such as a small peak of broken grids every 4 hours) can be extracted through Fourier transform or wavelet analysis to distinguish process fluctuations from environmental interference. By using spatial distribution changes (such as coordinate offsets and density fluctuations in high-incidence areas) as covariates, a multivariate time series model (such as a VAR model) is constructed. For example, the number of broken grids is significantly correlated with the density change in area X: 80-120mm (correlation coefficient r=0.85), and the model expression is: N t =α+β1D t−1 +β2ΔX t +ϵ t Among them, N t is the number of gate breaks in period t, D t−1 is the density value of the previous window, ΔX t It is the coordinate offset of the high-incidence area and outputs the time trend curve.

[0127] S525, based on the spatial probability density map and the time trend curve, obtain the spatiotemporal characteristic data of the defect distribution.

[0128] It can be understood that the spatial probability density map and the time trend curve can be mapped into a three-dimensional matrix with the dimension of space-time unit (space grid X, space grid Y, time window T), and the matrix element value is the joint probability density in the corresponding space-time unit. The three-dimensional matrix can be reduced in dimension to extract the main space-time pattern, and the matrix can be decomposed into a combination of spatial factors, time factors and defect type factors. The Pearson correlation coefficient or mutual information can be used to quantify the correlation strength between spatial density and time trend, and obtain the space-time characteristic data of defect distribution. For example, the density of hidden cracks in area X: 80-120mm is significantly correlated with the number of defects in time window W1 (9:00-10:00) (r=0.82, p<0.01).

[0129] S530, considering the defective area of ​​each photovoltaic panel as a node, and the association relationship between defects as an edge, constructing a defect association network, and determining the association network topology of the defect distribution; wherein the association network topology is used to reflect the key defect nodes and their impact range of the defect distribution of all photovoltaic panels within a preset time period.

[0130] It can be understood that discrete defect data can be converted into a network structure through a graph model to reveal the potential association and propagation path between defects. Each independent defect area of ​​each photovoltaic panel is regarded as a node, and the attributes include defect type, location coordinates, size and detection timestamp. For example, edges can be defined based on two types of relationships, such as the co-occurrence probability relationship of different defects in the same photovoltaic panel (such as the co-occurrence probability of hidden cracks and broken grids is 25%); the spatial proximity relationship of defect areas of different photovoltaic panels. The network can be represented by using an adjacency matrix or an edge list, and the edge weight can be quantified by statistical methods (such as Jaccard similarity, inverse of spatial distance) to obtain the association network topology of the defect distribution. The association network topology reflects the key defect nodes and their impact range of the defect distribution of all photovoltaic panels within a preset period of time.

[0131] S540, based on the spatiotemporal characteristic data of defect distribution and the associated network topology, simulate the diffusion path and rate of defects in the production line to obtain defect propagation data within a preset time period in the future.

[0132] It can be understood that the defect propagation data is a predictive data set generated by dynamic modeling based on the spatiotemporal characteristics and network topology of the production line, which is used to quantify the diffusion path, rate and impact range of photovoltaic panel defects in the production line corresponding to the target photovoltaic panel in a certain period of time in the future. The associated network topology can be used as the diffusion path skeleton to define the node status (such as "activated" indicates the occurrence of defects) and propagation rules. The partial differential equation can be constructed to describe the defect diffusion rate through spatiotemporal characteristic data (such as regional density and trend slope), for example, calculated by the following formula: ; Where D is the defect density, α is the diffusion coefficient, β is the natural growth rate, and K is the environmental capacity. The defect distribution in the future period can be simulated by numerical methods (such as the finite difference method), and the output includes the prediction of the number of defects in each future time window (such as 120 new cases of hidden cracks after 6 hours), spatial hot spots (such as the density of the X:100-150mm area increased to 0.9) and the change in type proportion (such as the broken grid rate increased from 20% to 25%), that is, the defect propagation data in the future preset period.

[0133] S550, based on the defect data set and the defect propagation data within a future preset time period, determine the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within the preset time period.

[0134] It can be understood that based on the defect data set and the defect propagation data within the preset time period in the future, the inspection report of all samples of the target photovoltaic panel corresponding production line within the preset time period can be generated and exported through the template report, and the overall quality of all photovoltaic panels of the target photovoltaic panel corresponding production line within the preset time period can be obtained. The report content can include the inspection results of each sample (qualified / unqualified), the type and quantity of defects detected, and other information; the defect propagation data, spatiotemporal feature data, and associated network topology of the target photovoltaic panel corresponding production line within the preset time period in the future, etc., to improve the accuracy and generalization of detection classification, promote the quality of production lines from passive control to active optimization, and achieve a dual improvement in manufacturing efficiency and product reliability.

[0135] Corresponding to the photovoltaic panel AI defect detection method of the above embodiment, the embodiment of the present application also provides a photovoltaic panel AI defect detection system, and each unit of the system can implement each step of the photovoltaic panel AI defect detection method. Figure 3 A structural block diagram of a photovoltaic panel AI defect detection system provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0136] Reference Figure 3 , the photovoltaic panel AI defect detection system includes:

[0137] An acquisition unit, used for acquiring an EL image of a target photovoltaic panel;

[0138] a processing unit, configured to process the EL image of the target photovoltaic panel using a photovoltaic panel detection model to obtain a first defect analysis result of the target photovoltaic panel; wherein the photovoltaic panel detection model is a machine learning model pre-trained using the EL image of a sample photovoltaic panel;

[0139] A rule unit, used to obtain a defect result post-processing rule and a historical defect analysis result; wherein the defect result post-processing rule is used to re-check the first defect analysis result; the historical defect analysis result is used to reflect the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset period before the target photovoltaic panel is detected;

[0140] a re-inspection unit, configured to re-inspect the target photovoltaic panel based on the first defect analysis result and by using the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel;

[0141] A result unit is used to determine the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset time period based on the historical defect analysis results and the second defect analysis results of the target photovoltaic panel, including: simulating the diffusion path and rate of defects in the production line based on the historical defect analysis results and the second defect analysis results, obtaining glass defect propagation data within a future preset time period, and determining the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within the preset time period based on the glass defect propagation data.

[0142] It should be noted that the information interaction, execution process, etc. between the above-mentioned systems / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0143] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit module can exist physically separately, or two or more unit modules can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0144] The embodiment of the present application also provides an electronic device, Figure 4 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 4 As shown, the electronic device 6 of this embodiment includes: at least one processor 60 ( Figure 4 Only one is shown), at least one memory 61 ( Figure 4 Only one is shown in the figure) and a computer program 62 stored in the at least one memory 61 and executable on the at least one processor 60. When the processor 60 executes the computer program 62, the electronic device 6 implements the steps in any of the above-mentioned photovoltaic panel AI defect detection method embodiments, or implements the functions of each unit in the above-mentioned system embodiments.

[0145] Exemplarily, the computer program 62 may be divided into one or more units, which are stored in the memory 61 and executed by the processor 60 to complete the present application. The one or more units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 62 in the electronic device 6.

[0146] The electronic device 6 may be a computing device or terminal device such as a desktop computer, a notebook, a PDA, or a cloud server. The electronic device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will appreciate that Figure 4 It is only an example of the electronic device 6 and does not constitute a limitation on the electronic device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, buses, etc.

[0147] The processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0148] In some embodiments, the memory 61 may be an internal storage unit of the electronic device 6, such as a hard disk or memory of the electronic device 6. In other embodiments, the memory 61 may also be an external storage device of the electronic device 6, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 6. Further, the memory 61 may also include both an internal storage unit and an external storage device of the electronic device 6. The memory 61 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory 61 may also be used to temporarily store data that has been output or is to be output.

[0149] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0150] An embodiment of the present application provides a computer program product. When the computer program product is executed on an electronic device, the electronic device implements the steps in any of the above method embodiments.

[0151] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electric carrier signal, a telecommunication signal, and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0152] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0153] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0154] In the embodiments provided in the present application, it should be understood that the disclosed photovoltaic panel AI defect detection method and device can be implemented in other ways. For example, the photovoltaic panel AI defect detection method and device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0155] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0156] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A photovoltaic panel AI defect detection method, characterized in that: include: Acquire the EL image of the target photovoltaic panel; Using a photovoltaic panel detection model to process the EL image of the target photovoltaic panel to obtain a first defect analysis result of the target photovoltaic panel; wherein the photovoltaic panel detection model is a machine learning model pre-trained using the EL image of a sample photovoltaic panel; Obtaining defect result post-processing rules and historical defect analysis results; wherein the defect result post-processing rules are used to re-inspect the first defect analysis results; the historical defect analysis results are used to reflect the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset period of time before the target photovoltaic panel is detected; Based on the first defect analysis result, re-inspecting the target photovoltaic panel using the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel; Based on the historical defect analysis results and the second defect analysis results of the target photovoltaic panel, the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within a preset time period is determined, including: based on the historical defect analysis results and the second defect analysis results, the diffusion path and rate of defects in the production line are simulated to obtain glass defect propagation data within a future preset time period, and based on the glass defect propagation data, the overall quality of all photovoltaic panels of the production line corresponding to the target photovoltaic panel within the preset time period is determined.

2. The photovoltaic panel AI defect detection method according to claim 1, characterized in that: Before using the photovoltaic panel detection model to process the EL image of the target photovoltaic panel to obtain a first defect analysis result of the target photovoltaic panel, the method further includes: Acquire sample EL images using a sample photovoltaic panel; determining a defect label corresponding to a sample EL image of the sample photovoltaic panel; The sample EL image of the sample photovoltaic panel is used as the expected input, and the corresponding defect label is used as the expected output, and the initial machine learning model is trained to obtain the trained photovoltaic panel detection model.

3. The photovoltaic panel AI defect detection method according to claim 2, characterized in that: The step of determining a defect label corresponding to the sample EL image of the sample photovoltaic panel comprises: Performing visual feature recognition on the sample EL image of the sample photovoltaic panel to obtain recognition and analysis results of the sample EL image; Based on the recognition and analysis result of the sample EL image, a defect label corresponding to the sample EL image of the sample photovoltaic panel is determined.

4. The photovoltaic panel AI defect detection method according to claim 1, characterized in that: The EL image of the target photovoltaic panel is processed using a photovoltaic panel detection model to obtain a first defect analysis result of the target photovoltaic panel, including: Processing the EL image of the target photovoltaic panel using a photovoltaic panel detection model to obtain a defect label of the EL image of the target photovoltaic panel; A first defect analysis result of the target photovoltaic panel is obtained by making a judgment based on the defect label of the EL image; wherein the first defect analysis result includes any one of qualified and unqualified; The method of re-inspecting the target photovoltaic panel based on the first defect analysis result and using the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel includes: When the first defect analysis result of the target photovoltaic panel is unqualified, the target photovoltaic panel is re-inspected according to the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel.

5. The photovoltaic panel AI defect detection method according to claim 4, characterized in that: When the first defect analysis result of the target photovoltaic panel is unqualified, re-inspecting the target photovoltaic panel according to the defect result post-processing rule to obtain a second defect analysis result of the target photovoltaic panel includes: Obtaining a defect label of the EL image of the target photovoltaic panel; wherein the defect label includes a defect type, a defect position, and a defect size of each defect area in the EL image; Determine the defect type, defect position, and defect size of each defect area in the EL image by using the defect result post-processing rule to obtain a first defect re-inspection result of the target photovoltaic panel; Performing grayscale detection on the EL image of the target photovoltaic panel to determine a second defect re-inspection result of the target photovoltaic panel; Based on the first defect re-inspection result and the second defect re-inspection result of the target photovoltaic panel, a second defect analysis result of the target photovoltaic panel is obtained.

6. The photovoltaic panel AI defect detection method according to claim 5, characterized in that: The defect result post-processing rules include size determination rules, sensitivity analysis rules and defect type priority; The defect type, defect position, and defect size of each defect area in the EL image are judged by the defect result post-processing rule to obtain the first defect re-inspection result of the target photovoltaic panel, including: Quantitatively analyzing the defect size according to the defect type of each defect area in the EL image by using the size determination rule to determine whether the defect area meets the size determination rule; In the case where the defect area meets the defect result post-processing rule, coordinate transformation is performed on the defect position of the defect area, the defect position is mapped to the physical coordinate system of the target photovoltaic panel, and the sensitivity analysis rule analysis result of the defect position is determined; According to the sensitivity analysis rule analysis results of all the defect areas and the defect type priorities, a first defect re-inspection result of the target photovoltaic panel is obtained.

7. The photovoltaic panel AI defect detection method according to claim 5, characterized in that: The performing grayscale detection on the EL image of the target photovoltaic panel to determine a second defect re-inspection result of the target photovoltaic panel includes: Performing grayscale detection on the EL image of the target photovoltaic panel to obtain grayscale distribution characteristics of the EL image; Based on the grayscale distribution characteristics of the EL image, grayscale abnormal areas in the EL image are identified using grayscale value statistics and anomaly detection algorithms, and anomaly detection thresholds are dynamically adjusted according to peak and valley values ​​of grayscale distribution to obtain grayscale abnormal areas; Perform spatial distribution analysis on the identified grayscale abnormal regions, and remove pseudo-abnormal regions according to the shape, texture and grayscale change trend of the defect region corresponding to the grayscale abnormal region, to obtain the area size and grayscale difference degree of each remaining grayscale abnormal region; wherein the pseudo-abnormal region is an abnormal region caused by uneven illumination and image noise; According to the area size and the grayscale difference degree of the grayscale abnormal area, each grayscale abnormal area is quantitatively scored; A second defect re-inspection result of the target photovoltaic panel is determined based on the grayscale distribution characteristics of the EL image and the quantitative scores of all the grayscale abnormal areas.

8. The photovoltaic panel AI defect detection method according to claim 1, characterized in that: Determining the overall quality of all photovoltaic panels of a production line corresponding to the target photovoltaic panel within a preset period of time according to the historical defect analysis result and the second defect analysis result of the target photovoltaic panel includes: Constructing a defect data set including defect data of all photovoltaic panels of a production line corresponding to the target photovoltaic panel within a preset time period according to the historical defect analysis result and the second defect analysis result; Modeling the defect distribution law of all photovoltaic panels within a preset period of time according to the defect data set to obtain spatiotemporal characteristic data of defect distribution; wherein the spatiotemporal characteristic data is used to reflect the high probability area and time aggregation of defect distribution of all photovoltaic panels within the preset period of time; The defective area of ​​each photovoltaic panel is regarded as a node, and the association relationship between defects is regarded as an edge, a defect association network is constructed, and the association network topology of the defect distribution is determined; wherein the association network topology is used to reflect the key defect nodes and their influence range of the defect distribution of all photovoltaic panels within a preset period of time; Based on the spatiotemporal characteristic data of the defect distribution and the associated network topology, the diffusion path and rate of the defect in the production line are simulated to obtain the defect propagation data within a preset time period in the future; Based on the defect data set and the defect propagation data within a future preset time period, the overall quality of all photovoltaic panels of a production line corresponding to the target photovoltaic panel within the preset time period is determined.

9. The photovoltaic panel AI defect detection method according to claim 8, characterized in that: The defect distribution law of all photovoltaic panels within a preset period of time is modeled according to the defect data set to obtain the spatiotemporal characteristic data of the defect distribution, including: Dividing the surface of the photovoltaic panel into a number of grid units of equal size according to the defect data set, and counting the number and type proportion of defects in each of the grid units respectively; Based on the number and type ratio of defects in each grid unit, a spatial probability density map of defect distribution is generated; wherein the spatial probability density map is used to reflect the spatial concentration area of ​​defect distribution and its distribution boundary; Divide the preset time period into multiple continuous time intervals according to the defect data set, and count the total number of defects and spatial distribution change trend in each time interval; According to the total number of defects and the spatial distribution change trend in each time interval, the occurrence frequency of defects is modeled in sections to obtain a time trend curve of the defect changing with time; Based on the spatial probability density map and the time trend curve, the spatiotemporal characteristic data of the defect distribution are obtained.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

Images