Photovoltaic module subfissure detection method

Abstract

The invention discloses a photovoltaic module subfissure detection method, relates to the technical field of photovoltaic module detection, and mainly aims to provide a photovoltaic module subfissure detection method capable of improving the characteristic extraction capability and identification efficiency of photovoltaic module defects. The invention provides a method for detecting hidden cracks of a photovoltaic module, and the method comprises the steps: replacing a Conv structure block in an existing RT-DETR model with a block-DualConv structure block to extract the shape and structure information of a target image, replacing an AIFI structure block in the existing RT-DETR model with a HiLo structure block, replacing a RepC3 structure block in the existing RT-DETR model with a KANC3 structure block, and carrying out the detection of the hidden cracks of the photovoltaic module. Compared with the prior art, the method provided by the invention has the advantages that the feature extraction capability of the defects of the photovoltaic module is improved, hidden cracks in the photovoltaic panel can be identified and positioned more accurately, the detection accuracy and robustness are improved, the feature maps extracted by the backbone are fused more thoroughly, and the detection accuracy and robustness are improved. The feature extraction capability and the noise robustness enable the KAN convolution to be excellent in improving the photovoltaic subfissure detection precision and stability.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic component detection, and in particular to a method for detecting hidden cracks in photovoltaic components. Background Art

[0002] Hidden cracks in photovoltaic modules refer to tiny cracks that appear in the cells. These cracks are difficult to detect with the naked eye, but they will affect the performance and life of the modules. Hidden cracks will cause the power of the cells to decay, and even expand under mechanical loads, leading to open circuit damage or hot spot effects. The main reasons for the formation of hidden cracks in photovoltaic modules include the following aspects: In the production process of photovoltaic modules, improper processes such as heat treatment and antimony film treatment will lead to excessive internal stress, thus generating hidden cracks; changes in ambient temperature, especially under high temperature conditions, will cause changes in the elastic stress of photovoltaic modules, thereby causing hidden cracks; during transportation and installation, external forces or vibrations may cause hidden cracks; defects in raw materials are also an important factor leading to hidden cracks.

[0003] There are many ways to detect hidden cracks in photovoltaic modules. For example, by measuring the open circuit voltage, short circuit current, maximum power point voltage and current and other electrical performance parameters of photovoltaic modules, it can be determined whether the module has performance abnormalities. Hidden cracks will increase the internal resistance of the module, thus affecting its electrical performance. Alternatively, infrared thermal imaging detection: use infrared thermal imaging technology to scan photovoltaic modules and observe the temperature distribution of the modules. Hidden cracks will cause the internal heat conduction of the module to be blocked, resulting in abnormal temperature distribution on the infrared thermal image. By comparing and analyzing the infrared thermal images of normal modules and suspected hidden crack modules, it can be determined whether the module has hidden cracks.

[0004] The existing hidden crack detection uses the EfficientNet image classification network for feature extraction. However, this method has the following problems:

[0005] 1. When training with very large image sizes, the training speed of EfficientNet becomes very slow. This is because larger image sizes require smaller batch sizes for training, which will cause the training speed to drop significantly and the accuracy will also be affected;

[0006] 2. Using Depthwise convolutions in shallow layers of the network will slow down the processing speed. Although Depthwise convolutions can reduce the number of parameters, it will affect the overall network training speed when used in early layers;

[0007] 3. EfficientNet scales up the model at each stage, which is not optimal. Scaling up the model at each stage may lead to uneven resource allocation and affect the performance of the model. Summary of the invention

[0008] In view of this, an embodiment of the present invention provides a method for detecting hidden cracks in photovoltaic modules, the main purpose of which is to provide a method for detecting hidden cracks in photovoltaic modules that can improve the feature extraction capability and recognition efficiency of photovoltaic module defects.

[0009] In order to achieve the above object, the present invention mainly provides the following technical solutions:

[0010] An embodiment of the present invention provides a method for detecting hidden cracks in a photovoltaic module, comprising the following steps:

[0011] Acquire the battery cell assembly image through an image acquisition device;

[0012] The existing RT-DETR model is improved to obtain an improved RT-DETR model, wherein the improvement includes:

[0013] The Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image, the AIFI structure block in the existing RT-DETR model is replaced with the HiLo structure block, and the RepC3 structure block in the existing RT-DETR model is replaced with the KANC3 structure block;

[0014] Inputting the image of the battery cell assembly to be inspected into the improved RT-DETR model for inspection;

[0015] Output the image feature map of the battery cell assembly.

[0016] Furthermore, the Conv structure block is replaced by the block-DualConv structure block;

[0017] Input the feature map to the block-DualConv structure block, input X, first pass a CBR, its output is then subjected to a DualConv convolution and then a residual structure is used to add X and the output of DualConv together before performing a RELU.

[0018] Furthermore, the input feature map is fed into the DualConv convolution, first processed by a group convolution, then processed by a point-by-point convolution, and finally the output results of the two convolutions are added together to generate the final output feature map.

[0019] Furthermore, the input tensor is processed by the HiLo building block;

[0020] The processing is done through a feed-forward network with normalization and residual connections between the two stages.

[0021] Furthermore, the head in the attention layer is divided into the first layer and the second layer;

[0022] The first layer encodes high frequencies through self-attention within each local window;

[0023] The second layer performs attention to model the global relationship between the average pooled low-frequency keys of each window and each query position in the input feature map.

[0024] Furthermore, the input feature map is preliminarily processed through two 1x1 convolutional layers to generate a first feature map and a second feature map respectively;

[0025] The first feature map is input into a sequence of N KAN modules, and repeated KAN convolution operations are performed to extract deeper features.

[0026] Add the output of repeated convolution operations to the second feature map to achieve feature fusion;

[0027] The fused feature map is finally processed through a 1x1 convolutional layer to obtain the output feature map.

[0028] Furthermore, the input data is passed through multiple convolution kernels for feature extraction;

[0029] The extracted features are further processed using polynomial transformation;

[0030] The processed features are weighted and aggregated to output enhanced feature representations.

[0031] Compared with the prior art, the present invention has the following technical effects:

[0032] First, the Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image, thereby improving the feature extraction capability of photovoltaic module defects;

[0033] Second, the AIFI structure block in the existing RT-DETR model is replaced with the HiLo structure block, making the feature content extracted by the backbone more comprehensive, thereby being able to more accurately identify and locate hidden cracks in photovoltaic panels and improve the accuracy and robustness of detection;

[0034] Third, the RepC3 structure block in the existing RT-DETR model is replaced with the KANC3 structure block, which makes it more thoroughly integrate the feature maps extracted by the backbone. This feature extraction capability and noise robustness make the KAN convolution perform well in improving the accuracy and stability of photovoltaic hidden crack detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1A flow chart of a method for detecting hidden cracks in a photovoltaic module provided by an embodiment of the present invention;

[0036] Figure 2 Flow chart of another method for detecting hidden cracks in photovoltaic modules provided by an embodiment of the present invention

[0037] Figure 3 A structural diagram of an improved RT-DETR model provided by an embodiment of the present invention;

[0038] Figure 4 A structural diagram of a block-DualConv structural block provided by an embodiment of the present invention;

[0039] Figure 5 A structural diagram of a HiLo structural block provided in an embodiment of the present invention;

[0040] Figure 6 A structural diagram of a KANC3 structural block provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0042] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0043] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0045] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0046] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0047] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0048] like Figure 1 and Figure 3 As shown, an embodiment of the present invention provides a method for detecting hidden cracks in a photovoltaic module, the method comprising the following steps:

[0049] Acquire the battery cell assembly image through an image acquisition device;

[0050] The existing RT-DETR model is improved to obtain an improved RT-DETR model, wherein the improvement includes:

[0051] The Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image, the AIFI structure block in the existing RT-DETR model is replaced with the HiLo structure block, and the RepC3 structure block in the existing RT-DETR model is replaced with the KANC3 structure block;

[0052] Inputting the image of the battery cell assembly to be inspected into the improved RT-DETR model for inspection;

[0053] Output the image feature map of the battery cell assembly.

[0054] Compared with the prior art, the present invention has the following technical effects:

[0055] First, the Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image, thereby improving the feature extraction capability of photovoltaic module defects;

[0056] Second, the AIFI structure block in the existing RT-DETR model is replaced with the HiLo structure block, making the feature content extracted by the backbone more comprehensive, thereby being able to more accurately identify and locate hidden cracks in photovoltaic panels and improve the accuracy and robustness of detection;

[0057] Third, the RepC3 structure block in the existing RT-DETR model is replaced with the KANC3 structure block, which makes it more thoroughly integrate the feature maps extracted by the backbone. This feature extraction capability and noise robustness make the KAN convolution perform well in improving the accuracy and stability of photovoltaic hidden crack detection.

[0058] Example 1

[0059] like Figure 1 and Figure 3 As shown, an embodiment of the present invention provides a method for detecting hidden cracks in a photovoltaic module, the method comprising the following steps:

[0060] 101. Acquire a battery cell assembly image through an image acquisition device.

[0061] The image acquisition device mainly uses infrared cameras and visible light cameras to photograph the battery cells, and then uploads the captured images to the detection module.

[0062] 102. The existing RT-DETR model is improved to obtain an improved RT-DETR model, wherein the improvement includes:

[0063] The Conv structure block in the existing RT-DETR model is replaced with a block-DualConv structure block to extract the shape and structure information of the target image, the AIFI structure block in the existing RT-DETR model is replaced with a HiLo structure block, and the RepC3 structure block in the existing RT-DETR model is replaced with a KANC3 structure block.

[0064] The Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structural information of the target image, so as to achieve the goal of using whether there is a hidden crack as the main classification and different types of products as sub-classifications. Then, the AIFI structure block in the existing RT-DETR model is replaced with the HiLo structure block, so that the high frequency in the image focuses on capturing fine-grained high-precision features, which is important for identifying tiny dimensional feature cracks, and the low frequency focus is used to capture global boundary information, which helps to understand the position and shape of the crack in the overall structure and extract structural information in a larger range. Then, the RepC3 structure block in the existing RT-DETR model is replaced with the KANC3 structure block, which can capture more complex local and global features in the feature extraction process and effectively enhance the recognition ability of subtle cracks. In addition, the design of the KAN convolution makes it more flexible in processing features of various scales and directions, which is particularly important for defects such as cracks on the surface of photovoltaic panels that are complex and difficult to predict.

[0065] 103. Input the image of the battery cell assembly to be inspected into the improved RT-DETR model for inspection.

[0066] The improved RT-DETR model can detect photovoltaic modules with complex background and small object features, and can more accurately identify and locate hidden cracks in photovoltaic panels.

[0067] 104. Output an image feature map of the battery cell assembly.

[0068] The improved RT-DETR model outputs the recognized image feature map in the form of a picture.

[0069] Compared with the prior art, the present invention has the following technical effects:

[0070] First, the Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image, thereby improving the feature extraction capability of photovoltaic module defects;

[0071] Second, the AIFI structure block in the existing RT-DETR model is replaced with the HiLo structure block, making the feature content extracted by the backbone more comprehensive, thereby being able to more accurately identify and locate hidden cracks in photovoltaic panels and improve the accuracy and robustness of detection;

[0072] Third, the RepC3 structure block in the existing RT-DETR model is replaced with the KANC3 structure block, which makes it more thoroughly integrate the feature maps extracted by the backbone. This feature extraction capability and noise robustness make the KAN convolution perform well in improving the accuracy and stability of photovoltaic hidden crack detection.

[0073] Example 2

[0074] like Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, an embodiment of the present invention provides a method for detecting hidden cracks in a photovoltaic module, the method comprising the following steps:

[0075] 201. Acquire a battery cell assembly image through an image acquisition device.

[0076] 202. The Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image.

[0077] like Figure 3 and Figure 4 As shown in the figure, the Conv structure block is replaced by the block-DualConv structure block; the input feature map is sent to the block-DualConv structure block, X is input, and it first passes through a CBR. Its output is then subjected to a DualConv convolution and a residual structure is used to add X and the output of DualConv together before a RELU is performed. The input feature map is sent to the DualConv convolution and first processed by a group convolution. The group convolution divides the input channels into multiple groups, and performs several volume operations on each group. Then, it is processed by a point-by-point convolution. The point-by-point convolution uses a 1x1 convolution kernel to change the number of channels. Finally, the output results of the two convolutions are added together to generate the final output feature map. This process not only retains the computational efficiency of the group convolution, but also realizes cross-channel information fusion through point-by-point convolution.

[0078] 203. Replace the AIFI structure block in the existing RT-DETR model with the HiLo structure block.

[0079] like Figure 3 and Figure 5As shown, the input tensor is processed by the HiLo structure block, dividing the head in the attention layer into the first and second layers; the first layer encodes the high frequency through self-attention within each local window; the second layer performs attention to model the global relationship between the average pooled low-frequency keys of each window and each query position in the input feature map. The HiLo structure block enables the high frequency in the image to capture local fine details, the low frequency to focus on the global structure, and the multi-head self-attention layer ignores the characteristics of different frequencies; it is processed through a feedforward network, and normalization and residual connections are performed between the two stages to maintain the flow of information and the stability of the model.

[0080] 204. Replace the RepC3 building block in the existing RT-DETR model with the KANC3 building block.

[0081] like Figure 3 and Figure 6 As shown in the figure, the workflow of the KANC3 structural block is as follows: the input feature map is preliminarily processed through two 1x1 convolutional layers to generate the first feature map and the second feature map respectively; the first feature map is input into a sequence containing N KAN modules, and repeated KAN convolution operations are performed to extract deeper features; the output of the repeated convolution operation is added to the second feature map to achieve feature fusion; the fused feature map is finally processed through a 1x1 convolutional layer to obtain an output feature map, wherein the KAN convolution includes: extracting features from the input data through multiple convolution kernels; further processing the extracted features using polynomial transformations; weighting the processed features and outputting enhanced feature representations. This process improves the expressiveness and stability of the model, enabling it to more effectively capture complex and diverse features.

[0082] 205. The measured cell assembly image is input into the improved RT-DETR model for detection.

[0083] 206. Image feature diagram of the pool sheet assembly.

[0084] Compared with the prior art, the present invention has the following technical effects:

[0085] First, the Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image, thereby improving the feature extraction capability of photovoltaic module defects;

[0086] Second, the AIFI block in the existing RT-DETR model is replaced with the HiLo block. The attention module of the HiLo block enables the high frequencies in the image to focus on capturing fine-grained high-precision features, which is important for identifying tiny dimensional feature cracks. The low-frequency attention is used to capture global boundary information, which helps to understand the position and shape of the cracks in the overall structure and extract structural information in a larger range. In this way, by combining high-frequency and low-frequency features, the HiLo block can more accurately identify and locate hidden cracks in photovoltaic panels, thereby improving the accuracy and robustness of detection.

[0087] Third, the RepC3 structural block in the existing RT-DETR model is replaced with the KANC3 structural block. The KAN convolution can capture more complex local and global features in the feature extraction process, effectively enhancing the recognition ability of subtle cracks. In addition, the design of the KAN convolution makes it more flexible in processing features of various scales and directions, which is especially important for defects such as cracks on the surface of photovoltaic panels that are complex and difficult to predict. This feature extraction capability and noise robustness make the KAN convolution perform well in improving the accuracy and stability of photovoltaic hidden crack detection.

[0088] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for detecting hidden cracks in photovoltaic modules, characterized in that: The following steps are involved: Acquire the battery cell assembly image through an image acquisition device; The existing RT-DETR model is improved to obtain an improved RT-DETR model, wherein the improvement includes: The Conv structure block in the existing RT-DETR model is replaced with the block-DualConv structure block to extract the shape and structure information of the target image, the AIFI structure block in the existing RT-DETR model is replaced with the HiLo structure block, and the RepC3 structure block in the existing RT-DETR model is replaced with the KANC3 structure block; Inputting the image of the battery cell assembly to be inspected into the improved RT-DETR model for inspection; Output the image feature map of the battery cell assembly.

2. The detection method according to claim 1, characterized in that: The method of replacing the Conv structure block in the existing RT-DETR model with the block-DualConv structure block to extract the shape and structure information of the target image includes: Replace the Conv structure block with the block-DualConv structure block; Input the feature map to the block-DualConv structure block, input X, first pass a CBR, its output is then subjected to a DualConv convolution and then a residual structure is used to add X and the output of DualConv together before performing a RELU.

3. The detection method according to claim 2, characterized in that: The DualConv convolution includes: The input feature map is fed into the DualConv convolution, first processed by a group convolution, then processed by a point-by-point convolution, and finally the output results of the two convolutions are added together to generate the final output feature map.

4. The detection method according to claim 3, characterized in that: The replacing of the AIFI structural block in the existing RT-DETR model with the HiLo structural block comprises: The input tensor is processed by the HiLo structure block; The processing is done through a feed-forward network with normalization and residual connections between the two stages.

5. The detection method according to claim 4, characterized in that: The HiLo building block includes: Split the head in the attention layer into the first and second layers; The first layer encodes high frequencies through self-attention within each local window; The second layer performs attention to model the global relationship between the average pooled low-frequency keys of each window and each query position in the input feature map.

6. The detection method according to claim 4, characterized in that: The replacing of the RepC3 structural block in the existing RT-DETR model with the KANC3 structural block comprises: The input feature map is preliminarily processed through two 1x1 convolutional layers to generate the first feature map and the second feature map respectively; The first feature map is input into a sequence of N KAN modules, and repeated KAN convolution operations are performed to extract deeper features. Add the output of repeated convolution operations to the second feature map to achieve feature fusion; The fused feature map is finally processed through a 1x1 convolutional layer to obtain the output feature map.

7. The detection method according to claim 6, characterized in that: The KAN convolution includes: The input data is passed through multiple convolution kernels for feature extraction; The extracted features are further processed using polynomial transformation; The processed features are weighted and aggregated to output enhanced feature representations.

Images