Infrared photovoltaic module defect identification method, device and equipment and storage medium

Abstract

The embodiment of the invention discloses an infrared photovoltaic module defect identification method and device, equipment and a storage medium. According to the technical scheme provided by the embodiment of the invention, the to-be-recognized infrared image obtained by shooting the infrared photovoltaic module is obtained, defect recognition is carried out on the to-be-recognized infrared image through the defect recognition model, the defect recognition result of the infrared photovoltaic module is obtained, the defect recognition model is obtained by building the target detection network, and the defect recognition result of the infrared photovoltaic module is obtained. According to the defect identification model, a deformable convolution kernel is configured at a preset convolution position of a backbone network, the deformable convolution kernel configures a sampling position of an input feature map as a dynamic position through learning offset, and the deformable convolution kernel is more suitable for a target form of an infrared image compared with conventional convolution. Therefore, the defect identification model has the capability of accurately identifying infrared target deformation and spatial irregular features, and the efficiency and accuracy of photovoltaic module defect identification are effectively improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to a method, apparatus, device, and storage medium for identifying defects in infrared photovoltaic modules. Background Art

[0002] With the large-scale deployment of photovoltaic energy, PV modules are prone to various defects such as hot spots, broken grids, cracks, and burns during long-term use. These defects can reduce PV module power generation efficiency, shorten module lifespan, and even pose safety hazards such as fires. Rapid and accurate identification of PV module defects is essential to ensure the stable operation of PV systems.

[0003] Currently, PV module defect identification is typically performed through manual inspections, where workers rely on experience to identify defects in PV modules. However, this reliance on manual experience makes it difficult to accurately identify PV module defects, resulting in low efficiency and accuracy. Summary of the Invention

[0004] The embodiments of the present application provide an infrared photovoltaic module defect recognition method, device, equipment and storage medium to solve the technical problems of low efficiency and accuracy in photovoltaic module defect recognition in related technologies, and can effectively improve the efficiency and accuracy of photovoltaic module defect recognition.

[0005] In a first aspect, an embodiment of the present application provides a method for identifying defects in an infrared photovoltaic module, comprising: Acquire an infrared image to be identified, where the infrared image to be identified is obtained by photographing an infrared photovoltaic module; Inputting the infrared image to be identified into a trained defect recognition model, performing defect recognition on the infrared image to be identified by the defect recognition model, and obtaining a defect recognition result of the infrared photovoltaic module; Among them, the defect recognition model is built based on the target detection network. The defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset.

[0006] Furthermore, the deformable convolution kernel processes the input feature map through the following formula:

[0007] in, The output feature map is at position The characteristics of the place, For the input feature map The characteristics of the place, is the number of sampling points of the deformable convolution kernel, is the weight parameter of the deformable convolution kernel, To preset the offset coordinates, is the learned offset, is the learned modulation scalar.

[0008] Furthermore, the defect recognition model is configured with a similarity attention module in the backbone network, and the similarity attention module is used to determine the retained neurons based on the energy functions of multiple neurons in the input feature map.

[0009] Furthermore, the energy function is determined by the following formula:

[0010] in, For the i The energy function of a neuron, For the i The eigenvalue of a neuron, To remove i The mean of the characteristic values ​​of neurons other than To remove i The variance of the eigenvalues ​​of neurons other than is the balance parameter.

[0011] Furthermore, the defect recognition model is configured with a global attention mechanism module in front of the detection head, and the global attention mechanism module is used to enhance the context perception capability of multi-scale information in the input features.

[0012] Furthermore, the defect recognition model is trained based on a bounding box loss function, which is determined by the following formula:

[0013] in, is the intersection area of ​​the predicted box and the real box, is the union area of ​​the predicted box and the true box, is the distance between the center point of the predicted box and the center point of the real box, is the normal distribution penalty weight.

[0014] Furthermore, the defect recognition model is configured with four detection heads, and the downsampling multiples of the four detection heads are 32×, 16×, 8× and 4× respectively.

[0015] In a second aspect, an embodiment of the present application provides an infrared photovoltaic module defect recognition device, including an image acquisition module and a defect recognition module, wherein: The image acquisition module is used to acquire the infrared image to be identified, and the infrared image to be identified is obtained by photographing the infrared photovoltaic module; The defect recognition module is used to input the infrared image to be recognized into the trained defect recognition model, perform defect recognition on the infrared image to be recognized through the defect recognition model, and obtain a defect recognition result of the infrared photovoltaic module; Among them, the defect recognition model is built based on the target detection network. The defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset.

[0016] In a third aspect, an embodiment of the present application provides an infrared photovoltaic module defect identification device, comprising: a memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the infrared photovoltaic component defect identification method as described in the first aspect.

[0017] In a fourth aspect, an embodiment of the present application provides a storage medium storing computer-executable instructions, which, when executed by a computer processor, are used to execute the infrared photovoltaic module defect identification method as described in the first aspect.

[0018] The embodiment of the present application obtains an infrared image to be identified by photographing an infrared photovoltaic component, and performs defect identification on the infrared image to be identified through a defect recognition model to obtain a defect recognition result of the infrared photovoltaic component, wherein the defect recognition model is constructed based on a target detection network, and the defect recognition model is configured with a deformable convolution kernel at a preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning an offset. Compared with conventional convolution, the deformable convolution kernel is more adaptable to the target morphology of the infrared image, so that the defect recognition model has the ability to accurately identify the deformation and spatial irregular features of the infrared target, enhances the defect recognition model's perception of defects with blurred boundaries and irregular shapes, and effectively improves the efficiency and accuracy of photovoltaic component defect recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of a method for identifying defects in infrared photovoltaic modules provided by an embodiment of the present application; Figure 2 This is a schematic diagram of the network structure of a defect recognition model provided by an embodiment of the present application; Figure 3 This is a schematic structural diagram of an infrared photovoltaic module defect identification device provided in an embodiment of the present application; Figure 4This is a structural diagram of an infrared photovoltaic module defect identification device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] To further clarify the objectives, technical solutions, and advantages of this application, specific embodiments of the present application are described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are intended only to illustrate this application and are not intended to limit it. It should also be noted that, for ease of description, the drawings only illustrate portions relevant to this application, not all of them. Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts depict the various operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The processes described above may terminate upon completion, but may also include additional steps not shown in the accompanying drawings. These processes may correspond to methods, functions, procedures, subroutines, subprograms, and the like.

[0021] With the large-scale deployment of photovoltaic energy, PV modules are prone to various defects such as hot spots, broken grids, cracks, and burns over long-term use. These defects can lead to reduced module power generation efficiency, shortened module lifespan, and even fire hazards. Therefore, quickly and accurately identifying PV module defects is crucial for the stable operation of PV systems. Existing PV module defect detection solutions typically rely on manual inspections. However, manual inspections are not only inefficient but also difficult to detect hidden defects in PV modules (such as initial hot spots). Consequently, infrared thermal imaging-based object detection is becoming increasingly popular. Among common object detection algorithms, object recognition networks (such as YOLOv8 (You Only Look Once Version 8)) offer advantages such as end-to-end modeling and high inference speed, making them widely used for image detection tasks. However, these networks use conventional convolutional architectures for feature extraction, making it difficult to accurately perceive hot spots and cracks with blurred boundaries and irregular shapes, leading to missed or false detections. Furthermore, these networks also lack the ability to express features. The overall contrast of infrared images is low, and small defects are often drowned in background noise. The target recognition network has limited attention capacity and is prone to ignoring key areas. The target recognition network's target saliency modeling is insufficient. The target recognition network's default three-scale detection head performs poorly when faced with tiny defects such as fine hot spots and broken grids. Especially in images taken at a long distance, small targets are easily ignored, and the target recognition network has weak detection capabilities for small targets. The loss function used by the target recognition network (such as the scalable intersection over union loss function, SIoU (Scalable Intersection over Union Loss) loss function) may experience regression jitter in infrared images and is insensitive to irregular hot spot boundaries. The regression loss function of the target recognition network is unstable, affecting the training effect. It can be seen that the current efficiency and accuracy of photovoltaic module defect recognition based on infrared thermal imaging are low. Based on this, an infrared photovoltaic module defect recognition method according to an embodiment of the present application is provided to solve the technical problem of low efficiency and accuracy of existing photovoltaic module defect recognition.

[0022] Figure 1 A flowchart of an infrared photovoltaic module defect identification method provided in an embodiment of the present application is given. The infrared photovoltaic module defect identification method provided in an embodiment of the present application can be executed by an infrared photovoltaic module defect identification device, which can be implemented by hardware and / or software and integrated into an infrared photovoltaic module defect identification device.

[0023] The following describes the infrared photovoltaic module defect recognition method using an infrared photovoltaic module defect recognition device as an example. Figure 1 , the infrared photovoltaic module defect recognition method includes: S110: Acquire an infrared image to be identified, where the infrared image to be identified is obtained by photographing an infrared photovoltaic module.

[0024] Exemplarily, an infrared image to be identified corresponding to an infrared photovoltaic module that needs to be defect-identified is obtained. The infrared image to be identified is obtained by photographing the infrared photovoltaic module, for example, by photographing the infrared photovoltaic module that needs to be defect-identified using an infrared thermal imaging device to obtain an infrared image to be identified of the infrared photovoltaic module. The infrared image to be identified can reflect the surface temperature distribution of the infrared photovoltaic module. In the defect identification of infrared photovoltaic modules, defects such as hot spots, broken grids, and burns in the infrared photovoltaic modules will appear as abnormally high-temperature areas in the infrared image with obvious temperature differences. Infrared photovoltaic module defects can be understood as damage that occurs to the infrared photovoltaic module during manufacturing or use, including hot spots, broken grids, burns, cracks, etc. These defects may affect the power generation efficiency of the infrared photovoltaic module and even cause safety problems. It is necessary to promptly identify possible defects in the infrared photovoltaic module to ensure the normal operation of the infrared photovoltaic module.

[0025] S120: Inputting the infrared image to be identified into the trained defect recognition model, performing defect recognition on the infrared image to be identified by the defect recognition model, and obtaining a defect recognition result of the infrared photovoltaic module.

[0026] For example, after acquiring an infrared image to be identified, the image can be input into a trained defect recognition model. Upon receiving the infrared image, the defect recognition model will perform defect recognition on the image to be identified, generating defect recognition results for the infrared photovoltaic module. The defect recognition results for the infrared photovoltaic module can be used to indicate the location, range (for example, by using a positioning box to reflect the location and range of the defect), and defect type of the infrared photovoltaic module in the infrared image to be identified. Based on the defect recognition results, personnel can determine the defects and their locations in the infrared photovoltaic module and perform repairs on the module.

[0027] The defect recognition model provided in this application is built based on a target detection network, which includes a backbone network (backbone), a neck network (neck), and a head network (head) connected in sequence. The backbone network is responsible for extracting features of the input data, the neck network is responsible for further processing and integrating the features extracted by the backbone network, and the head network is responsible for performing specific tasks, such as classification, target detection, semantic segmentation, etc. The defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset.

[0028] Optionally, the target detection network provided in this application may be a YOLO (e.g., YOLOv8) target detection network. Based on the target detection network, this application replaces multiple default convolution kernels at preset convolution positions in the target detection network with deformable convolution kernels. The default convolution kernel may be a standard convolution kernel C2f, the preset convolution position may be a position where the standard convolution kernel C2f is configured in the original target detection network, and the deformable convolution kernel may be a deformable convolution kernel DCNv2 (Deformable Convolution v2). The deformable convolution kernel DCNv2 can be used as a module for improving the spatial modeling capabilities of convolutional neural networks. It can adaptively adjust the convolution kernel sampling position based on the features of the input image, and can accurately process images with obvious target deformation or blurred boundaries, such as infrared images with hot spots and cracks.

[0029] In one possible embodiment, the deformable convolution kernel provided in this application processes the input feature map through the following formula:

[0030] in, The output feature map is at position The characteristics of the place, For the input feature map The characteristics of the place, is the number of sampling points of the deformable convolution kernel, is the weight parameter of the deformable convolution kernel, To preset the offset coordinates, is the offset learned during the defect recognition model training process, is the modulation scalar learned during the defect recognition model training process.

[0031] This application uses standard convolution kernels for feature extraction in the C2f module. The fixed convolution receptive field has limitations in processing defects such as hot spots and cracks with irregular shapes or blurred boundaries in infrared images. The standard convolution kernels at key convolution positions (preset convolution positions) in the backbone network of the target detection network are replaced with deformable convolution kernels. The deformable convolution kernels are trained by learning the offset. , the sampling position of the convolution kernel can be changed from a fixed grid to a learnable dynamic position, so that the defect recognition model can focus on the key area, enhance the spatial modeling capability of the features, accurately process defects such as hot spots and cracks with irregular shapes or blurred boundaries in infrared images, and better perceive the contours of deformed targets in infrared images, thereby improving the recognition accuracy of the defect recognition model in photovoltaic hot spot and crack detection.

[0032] In one possible embodiment, the defect recognition model provided by the present application is configured with a similarity attention module (such as SimAM (Simple Attention Module)) in the backbone network. The similarity attention module can be used to determine the retained neurons based on the energy function of multiple neurons in the input feature map. The similarity attention module is based on a lightweight attention mechanism and judges the significance of neurons by calculating the energy distribution of neurons in three-dimensional space. It does not require adding model parameters and can enhance the model's attention to key areas in the image. Since the similarity attention module has no parameters, the computational cost of the similarity attention module is low. The similarity attention module can be directly embedded in the connection between the backbone network and the neck network to enhance the local significance judgment ability. Optionally, the similarity attention module can be set at the end of the backbone network, for example, between the output of the spatial pyramid pooling fast module (SPPF module) of the backbone network and the input of the neck network.

[0033] Optionally, the energy function provided in this application is determined by the following formula:

[0034] in, For the i The energy function of a neuron, For the i The eigenvalue of a neuron, To remove i The mean of the characteristic values ​​of neurons other than To remove i The variance of the eigenvalues ​​of neurons other than The similarity attention module uses a channel attention mechanism based on the neuron energy function. For each neuron in the input feature map, if it has more obvious independence or significance in the current feature map, its energy is higher and its retention is more significant. Neurons whose energy function in the input feature map reaches the preset energy threshold are retained, improving the defect recognition model's ability to identify the saliency of local key areas.

[0035] In one possible embodiment, the defect recognition model provided by the present application is configured with a global attention mechanism module (such as GAM (Global Attention Module)) before the detection head. The global attention mechanism module is based on an attention mechanism that combines spatial and channel dimensions, and can establish a connection between global context information at different scales, enhance feature representation capabilities, and is particularly suitable for target differentiation in complex scenes. The global attention mechanism module can be used to enhance the context perception capability of multi-scale information in the input features. Optionally, given the input feature map of the global attention mechanism module, , the mixed attention map can be generated based on the following formula:

[0036] in, is the intermediate state feature, is the output feature, is the channel attention, For spatial attention, Represents element-wise multiplication operation, is the number of channels, width, and height of the input feature map R. The global attention mechanism module combines spatial and channel attention mechanisms to amplify interactive features in the global dimension while reducing information diffusion. It has stronger global perception capabilities and effectively suppresses background interference, effectively improving the defect recognition model's ability to identify the saliency of local key areas.

[0037] It needs to be explained that due to the characteristics of low contrast and weak local information of infrared images, the defect area often does not have obvious significance in the feature map corresponding to the infrared image to be identified. This application introduces a similarity attention module and a global attention mechanism module into the defect recognition model. The similarity attention module improves the saliency recognition ability of local key areas, and the global attention mechanism module enhances the context perception ability of multi-scale information, thereby enhancing the defect recognition model's saliency perception of areas such as tiny hot spots and broken grids, and improving the defect recognition accuracy of the defect recognition model in complex backgrounds.

[0038] In one possible embodiment, training samples (infrared images) corresponding to different defect types and defect locations can be collected, and the defect types and defect locations in the training samples can be labeled. A training sample set is constructed based on the labeled training samples, and the defect recognition model is trained using the training sample set based on a preset loss function. The defect recognition model provided in this application can be trained based on a bounding box loss function (for example, a bounding box loss function PIoU (Probabilistic Intersection over Union)). The bounding box loss function can introduce an uncertainty modeling mechanism based on the traditional intersection over union (IoU, (Intersection over Union)), which can more accurately evaluate the difference between the model prediction box and the true box, helping to train a more stable and robust defect recognition model. Optionally, the bounding box loss function can be determined by the following formula:

[0039] in, is the bounding box loss function, is the intersection area of ​​the predicted box (the positioning box predicted by the defect recognition model) and the true box (the defect positioning box marked in the training sample), is the union area of ​​the predicted box and the true box, is the distance between the center point of the predicted box and the center point of the real box, is the normal distribution penalty weight.

[0040] It needs to be explained that the conventional target detection network uses the scalable intersection-over-union loss function as the positioning regression loss function by default, but the scalable intersection-over-union loss function is not stable enough when processing defective targets with blurred boundaries or occlusion in infrared images. This application uses a probabilistic bounding box loss function to train the defect recognition model. The bounding box loss function can model the matching uncertainty between boxes, optimize the bounding box fitting effect, effectively reduce the regression jitter that may occur in infrared images, and is insensitive to irregular hot spot boundaries. The regression loss function of the target recognition network is unstable, which affects the training effect. The training of the defect recognition model is more stable, which improves the defect positioning accuracy and is more suitable for the precise positioning of irregular and unclear-boundary targets in infrared images.

[0041] In one embodiment, the defect recognition model provided by this application is configured with 4 detection heads, and the downsampling multiples of the 4 detection heads are 32×, 16×, 8×, and 4×, respectively. It should be explained that the conventional target detection network uses 3 detection heads by default, and the corresponding input downsampling multiples are 32×, 16×, and 8×, respectively. In response to the situation where the size of small defect targets in infrared images accounts for a small proportion of the infrared image area, this application adds a detection head with a downsampling multiple of 4× to the head network, thereby improving the defect recognition model's ability to detect small defects (such as fine cracks, slight ablation, etc.), making up for the shortcomings of the original target detection network (such as YOLOv8) in small target detection, and can improve the recall rate of small targets in infrared images, effectively improving the defect recognition accuracy of the defect recognition model.

[0042] like Figure 2 As shown in the network structure diagram of a defect recognition model provided, based on a conventional target detection network (such as YOLOv8), this application addresses the problems of conventional target detection networks in infrared images, such as blurred target boundaries, inaccurate recognition of small targets, and insufficient saliency perception. Adjustments are made to the backbone network, attention mechanism, loss function, and detection head structure of the conventional target detection network, making the defect recognition model more adaptable to the low contrast and non-rigid defect characteristics of infrared photovoltaic images.

[0043] Specifically, if Figure 2As shown, this application replaces the default standard convolution kernel C2f of the target detection network (which may include the standard convolution kernel C2f of the backbone network, or the standard convolution kernel C2f of the backbone network and the neck network) with the deformable convolution kernel DCNv2, wherein the connection module is used to perform the connection operation (concat operation), the upsampling module is used to perform the upsampling operation (Upsample operation), and the convolution layer is the conv layer. The defect recognition model is configured with a similarity attention module SimAM in the backbone network. Specifically, the similarity attention module SimAM is connected between the output end of the spatial pyramid pooling fast module (SPPF module) of the backbone network and the input end of the neck network (the upsampling module in the figure). The defect recognition model is configured with 4 detection heads (Detect), ( Figure 2 The downsampling multiples of the four detection heads (from top to bottom) are 32×, 16×, 8×, and 4×, respectively. The defect recognition model is also equipped with a global attention mechanism module (GAM) in front of each of the four detection heads.

[0044] This application addresses the shortcomings of conventional target detection networks in terms of recognition accuracy, small target detection capabilities, boundary fitting, and model robustness. By introducing deformable convolution kernels in the feature extraction stage, this method enhances the model's ability to perceive defects with blurred boundaries and irregular shapes (such as hot spots and broken grids). Compared to conventional convolution, this method is more adaptable to target morphology in infrared images. This application incorporates a similarity attention module (SimAM) and a global attention mechanism module (GAM) in the feature fusion stage. The similarity attention module (SimAM) can effectively improve the ability to identify the saliency of local key areas, while the global attention mechanism module (GAM) can effectively enhance the contextual perception of multi-scale information and improve the model's recognition accuracy in complex backgrounds. This application also adds a set of high-resolution detection heads with a downsampling factor of 4× to the technology of conventional target detection networks. This can effectively improve the model's ability to detect small defects (such as fine cracks and slight ablation), thus addressing the shortcomings of conventional target detection networks in small target detection.

[0045] In the above, by acquiring the infrared image to be identified by photographing the infrared photovoltaic component, the defect recognition result of the infrared photovoltaic component is obtained by performing defect recognition on the infrared image to be identified through the defect recognition model, wherein the defect recognition model is built based on the target detection network, and the defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset. Compared with conventional convolution, the deformable convolution kernel is more adaptable to the target morphology of the infrared image, so that the defect recognition model has the ability to accurately identify the deformation and spatial irregular features of the infrared target, enhances the defect recognition model's perception of defects with blurred boundaries and irregular shapes, and effectively improves the efficiency and accuracy of photovoltaic component defect recognition.

[0046] Figure 3 A schematic diagram of the structure of an infrared photovoltaic module defect recognition device provided in an embodiment of the present application is given. Figure 3 The infrared photovoltaic module defect recognition device includes an image acquisition module 31 and a defect recognition module 32.

[0047] Among them, the image acquisition module 31 is used to acquire the infrared image to be identified, and the infrared image to be identified is obtained by photographing the infrared photovoltaic component; the defect recognition module 32 is used to input the infrared image to be identified into the trained defect recognition model, and perform defect recognition on the infrared image to be identified through the defect recognition model to obtain the defect recognition result of the infrared photovoltaic component; wherein, the defect recognition model is constructed based on the target detection network, and the defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network, and the deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset.

[0048] In the above, by acquiring the infrared image to be identified by photographing the infrared photovoltaic component, the defect recognition result of the infrared photovoltaic component is obtained by performing defect recognition on the infrared image to be identified through the defect recognition model, wherein the defect recognition model is built based on the target detection network, and the defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset. Compared with conventional convolution, the deformable convolution kernel is more adaptable to the target morphology of the infrared image, so that the defect recognition model has the ability to accurately identify the deformation and spatial irregular features of the infrared target, enhances the defect recognition model's perception of defects with blurred boundaries and irregular shapes, and effectively improves the efficiency and accuracy of photovoltaic component defect recognition.

[0049] In one possible embodiment, the deformable convolution kernel processes the input feature map through the following formula:

[0050] in, The output feature map is at position The characteristics of the place, For the input feature map The characteristics of the place, is the number of sampling points of the deformable convolution kernel, is the weight parameter of the deformable convolution kernel, To preset the offset coordinates, is the learned offset, is the learned modulation scalar.

[0051] In one possible embodiment, the defect recognition model is configured with a similarity attention module in the backbone network. The similarity attention module is used to determine the retained neurons based on the energy functions of multiple neurons in the input feature map.

[0052] In one possible embodiment, the energy function is determined by the following formula:

[0053] in, For the i The energy function of a neuron, For the i The eigenvalue of a neuron, To remove i The mean of the characteristic values ​​of neurons other than To remove i The variance of the eigenvalues ​​of neurons other than is the balance parameter.

[0054] In one possible embodiment, the defect recognition model is configured with a global attention mechanism module before the detection head. The global attention mechanism module is used to enhance the context perception capability of multi-scale information in the input features.

[0055] In one possible embodiment, the defect recognition model is trained based on a bounding box loss function, which is determined by the following formula:

[0056] in, is the intersection area of ​​the predicted box and the real box, is the union area of ​​the predicted box and the true box, is the distance between the center point of the predicted box and the center point of the real box, is the normal distribution penalty weight.

[0057] In a possible embodiment, the defect recognition model is configured with four detection heads, and the downsampling multiples of the four detection heads are 32×, 16×, 8×, and 4×, respectively.

[0058] It is worth noting that in the embodiment of the above-mentioned infrared photovoltaic component defect identification device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the embodiments of this application.

[0059] The embodiment of the present application also provides an infrared photovoltaic module defect recognition device, which can be integrated with the infrared photovoltaic module defect recognition apparatus provided in the embodiment of the present application. Figure 4 This is a schematic diagram of the structure of an infrared photovoltaic module defect recognition device provided by an embodiment of the present application. Figure 4 The infrared photovoltaic module defect recognition device includes: an input device 43, an output device 44, a memory 42, and one or more processors 41; the memory 42 is used to store one or more programs; when the one or more programs are executed by the one or more processors 41, the one or more processors 41 implement the infrared photovoltaic module defect recognition method provided in the above embodiment. The input device 43, the output device 44, the memory 42, and the processor 41 can be connected by a bus or other means. Figure 4 The bus connection is taken as an example.

[0060] Memory 42, as a computing device-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the infrared photovoltaic module defect identification method provided in any embodiment of the present application (e.g., the image acquisition module 31 and defect identification module 32 in the infrared photovoltaic module defect identification device). Memory 42 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data generated based on device usage. Furthermore, memory 42 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, memory 42 may further include memory located remotely from processor 41, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0061] The input device 43 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 44 may include a display device such as a display screen.

[0062] The processor 41 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 42, that is, realizes the above-mentioned infrared photovoltaic module defect identification method.

[0063] The infrared photovoltaic module defect identification device, equipment and computer provided above can be used to execute the infrared photovoltaic module defect identification method provided in any of the above embodiments, and have corresponding functions and beneficial effects.

[0064] An embodiment of the present application also provides a storage medium storing computer-executable instructions, which, when executed by a computer processor, are used to execute the infrared photovoltaic module defect recognition method provided in the above embodiment, wherein the infrared photovoltaic module defect recognition method includes: obtaining an infrared image to be recognized, wherein the infrared image to be recognized is obtained by photographing the infrared photovoltaic module; inputting the infrared image to be recognized into a trained defect recognition model, performing defect recognition on the infrared image to be recognized through the defect recognition model, and obtaining a defect recognition result of the infrared photovoltaic module; wherein the defect recognition model is constructed based on a target detection network, and the defect recognition model is configured with a deformable convolution kernel at a preset convolution position of the backbone network, and the deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning an offset.

[0065] Storage medium - any of various types of memory devices or storage devices. The term "storage medium" is intended to include: installation media, such as CD-ROMs, floppy disks, or tape drives; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (such as hard disks or optical storage); registers or other similar types of memory elements, etc. Storage media may also include other types of memory or combinations thereof. In addition, the storage medium may be located in the first computer system in which the program is executed, or it may be located in a different second computer system that is connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term "storage medium" may include two or more storage media that can reside in different locations (e.g., in different computer systems connected via a network). The storage medium can store program instructions (e.g., embodied as a computer program) that can be executed by one or more processors.

[0066] Of course, the computer executable instructions of the storage medium for storing computer executable instructions provided in the embodiment of the present application are not limited to the infrared photovoltaic module defect identification method provided above, and can also execute related operations in the infrared photovoltaic module defect identification method provided in any embodiment of the present application.

[0067] The infrared photovoltaic module defect identification device, equipment and storage medium provided in the above embodiments can execute the infrared photovoltaic module defect identification method provided in any embodiment of the present application. For technical details not described in detail in the above embodiments, please refer to the infrared photovoltaic module defect identification method provided in any embodiment of the present application.

[0068] The above are only preferred embodiments of the present application and the technical principles employed. The present application is not limited to the specific embodiments provided herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments. Without departing from the concept of the present application, it may also include more other equivalent embodiments, and the scope of the present application is determined by the scope of the claims.

Claims

1. A method for identifying defects in infrared photovoltaic modules, characterized in that: include: Acquire an infrared image to be identified, where the infrared image to be identified is obtained by photographing an infrared photovoltaic module; Inputting the infrared image to be identified into a trained defect recognition model, performing defect recognition on the infrared image to be identified by the defect recognition model, and obtaining a defect recognition result of the infrared photovoltaic module; Among them, the defect recognition model is built based on the target detection network. The defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset.

2. The infrared photovoltaic module defect identification method according to claim 1, characterized in that: The deformable convolution kernel processes the input feature map through the following formula: in, The output feature map is at position The characteristics of the place, For the input feature map The characteristics of the place, is the number of sampling points of the deformable convolution kernel, is the weight parameter of the deformable convolution kernel, To preset the offset coordinates, is the learned offset, is the learned modulation scalar.

3. The infrared photovoltaic module defect identification method according to claim 1, characterized in that: The defect recognition model is configured with a similarity attention module in the backbone network, and the similarity attention module is used to determine the retained neurons based on the energy functions of multiple neurons in the input feature map.

4. The infrared photovoltaic module defect identification method according to claim 3, characterized in that: The energy function is determined by the following formula: in, For the i The energy function of a neuron, For the i The eigenvalue of a neuron, To remove i The mean of the characteristic values ​​of neurons other than To remove i The variance of the eigenvalues ​​of neurons other than is the balance parameter.

5. The infrared photovoltaic module defect identification method according to claim 1, characterized in that: The defect recognition model is configured with a global attention mechanism module in front of the detection head, and the global attention mechanism module is used to enhance the context perception capability of multi-scale information in the input features.

6. The infrared photovoltaic module defect identification method according to claim 1, characterized in that: The defect recognition model is trained based on a bounding box loss function, which is determined by the following formula: in, is the intersection area of ​​the predicted box and the real box, is the union area of ​​the predicted box and the true box, is the distance between the center point of the predicted box and the center point of the real box, is the normal distribution penalty weight.

7. The infrared photovoltaic module defect identification method according to claim 1, characterized in that: The defect recognition model is configured with four detection heads, and the downsampling multiples of the four detection heads are 32×, 16×, 8× and 4× respectively.

8. An infrared photovoltaic module defect recognition device, characterized in that: It includes an image acquisition module and a defect recognition module, wherein: The image acquisition module is used to acquire the infrared image to be identified, and the infrared image to be identified is obtained by photographing the infrared photovoltaic module; The defect recognition module is used to input the infrared image to be recognized into the trained defect recognition model, perform defect recognition on the infrared image to be recognized through the defect recognition model, and obtain a defect recognition result of the infrared photovoltaic module; Among them, the defect recognition model is built based on the target detection network. The defect recognition model is configured with a deformable convolution kernel at the preset convolution position of the backbone network. The deformable convolution kernel configures the sampling position of the input feature map to a dynamic position by learning the offset.

9. An infrared photovoltaic module defect identification device, characterized in that: include: memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the infrared photovoltaic component defect identification method according to any one of claims 1 to 7.

10. A storage medium storing computer executable instructions, characterized in that: The computer executable instructions, when executed by a computer processor, are used to execute the infrared photovoltaic assembly defect identification method according to any one of claims 1 to 7.

Images