Photovoltaic module hot spot detection method, device and storage medium

Abstract

The present invention discloses a method, device and storage medium for detecting hot spots of photovoltaic components. The detection method includes: obtaining an infrared image of a photovoltaic panel within a preset area; inputting the infrared image of the photovoltaic panel into a trained photovoltaic panel segmentation model, and outputting multiple groups of photovoltaic panel data, each group of photovoltaic panel data including a mask area of ​​each photovoltaic panel obtained by segmentation and a corresponding segmentation quality score of the photovoltaic panel; detecting the mask area of ​​the photovoltaic panel with a segmentation quality score higher than a preset score, and obtaining an abnormal photovoltaic panel containing abnormal hot spots and a hot spot area of ​​the abnormal photovoltaic panel. The present application combines deep learning image instance segmentation with small target detection, and adopts a deep learning instance segmentation method with segmentation quality evaluation capability, which can accurately segment the photovoltaic panel area, and is suitable for abnormal hot spots of different scales and shapes in typical photovoltaic panel infrared images captured by drone vision. The algorithm is fast and efficient, and has strong practical value.

Description

Technical Field

[0001] The present invention belongs to the field of hot spot detection of photovoltaic solar photovoltaic modules, and in particular relates to a method, equipment and storage medium for hot spot detection of photovoltaic modules. Background Art

[0002] Using photovoltaic panels to generate electricity through solar energy is an important mode of new energy generation. However, once a certain unit component of the photovoltaic panel is damaged and cannot work properly, it will directly affect the thermal conversion efficiency of the wafer component in that area, and then show abnormal heating phenomenon in infrared thermal imaging, where the temperature is significantly higher than that of the surrounding area. In daily photovoltaic inspection operations, using infrared thermal imaging to detect and locate abnormal hot spot areas of photovoltaic panels is an effective means to determine whether the unit components of photovoltaic modules are operating normally.

[0003] In the prior art, using drones to perform infrared thermal imaging of photovoltaic panels, and then using computer vision calculations to quickly and automatically detect and locate abnormal hot spots, is an important solution that can significantly improve the efficiency of detecting abnormal hot spots on photovoltaic panels. However, the infrared image resolution of drone aerial long-range thermal imaging is limited, and the pixel size of abnormal hot spots is small, and there is no fixed morphological feature. The imaging is easily interfered by complex background heat source clutter. These factors have brought difficulties to the hot spot detection based on drone infrared imaging. At the same time, the existing hot spot detection method for photovoltaic power stations based on thermal infrared images also has problems such as low detection accuracy, the backbone network is insensitive to the extraction of low-resolution target features such as infrared hot spots, the trained model is not universal enough for complex hot spots, and it is impossible to determine which photovoltaic panel wafer component the abnormal hot spot is located in. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the defects in the prior art such as low detection accuracy, insensitivity of the backbone network to low-resolution target feature extraction of infrared hot spots, insufficient universality of the trained model for complex hot spots, and inability to determine which photovoltaic panel wafer component the abnormal hot spot is located, and to provide a photovoltaic component hot spot detection method, device and storage medium.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] A method for detecting hot spots in photovoltaic modules, the method comprising:

[0007] Acquire infrared images of photovoltaic panels within a preset area;

[0008] Inputting the photovoltaic panel infrared image into a trained photovoltaic panel segmentation model, and outputting multiple groups of photovoltaic panel data, each group of photovoltaic panel data including a mask area of ​​each photovoltaic panel obtained by segmentation and a corresponding photovoltaic panel segmentation quality score;

[0009] The mask area of ​​the photovoltaic panel whose segmentation quality score is higher than the preset score is detected to obtain the abnormal photovoltaic panel containing the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel.

[0010] Preferably, the photovoltaic panel segmentation model is obtained by training in the detection method through the following steps, specifically including:

[0011] Constructing a photovoltaic panel training sample set, wherein the photovoltaic panel training sample set includes a plurality of photovoltaic panel infrared training images, and each photovoltaic panel infrared training image includes an actual mask area of ​​each training photovoltaic panel;

[0012] Using the photovoltaic panel training sample set as training data, training to obtain an initial photovoltaic panel segmentation model, and obtaining a predicted mask area and a corresponding segmentation confidence of each training photovoltaic panel;

[0013] Calculating a segmentation quality score for each training photovoltaic panel based on the predicted mask area, the actual mask area and the segmentation confidence;

[0014] The segmentation quality score is added to the training data to continue training, thereby obtaining the photovoltaic panel segmentation model.

[0015] Preferably, the step of calculating the segmentation quality score of each training photovoltaic panel based on the predicted mask area, the actual mask area and the segmentation confidence comprises:

[0016] Calculate the IOU value of each training photovoltaic panel based on the predicted mask area and the actual mask area;

[0017] The segmentation quality score is obtained based on the product of the IOU value and the segmentation confidence.

[0018] Preferably, the step of detecting the mask area of ​​the photovoltaic panel having a segmentation quality score higher than a preset score to obtain the abnormal photovoltaic panel containing abnormal hot spots and the hot spot area of ​​the abnormal photovoltaic panel specifically comprises:

[0019] Select the photovoltaic panels with segmentation quality scores higher than the preset score as the target photovoltaic panels;

[0020] Based on the infrared image small target detection algorithm, the photovoltaic panel image corresponding to the target photovoltaic panel is detected to obtain the abnormal photovoltaic panel containing the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel.

[0021] Preferably, the step of detecting the photovoltaic panel image corresponding to the target photovoltaic panel based on the infrared image small target detection algorithm to obtain the abnormal photovoltaic panel including the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel specifically includes:

[0022] Performing grayscale processing on the photovoltaic panel image corresponding to the target photovoltaic panel to obtain a single-channel grayscale image;

[0023] The single-channel grayscale image is subjected to binary segmentation, and a pixel region where the image pixels are greater than a determination threshold is determined as a hot spot region.

[0024] Preferably, before the step of performing binary segmentation on the single-channel grayscale image, the detection method further comprises:

[0025] Preprocessing the single-channel grayscale image, wherein the preprocessing includes at least one of a convolution filter process, an exponential enhancement operation, and a noise filtering operation;

[0026] Among them, the convolution filtering processing includes performing high-pass convolution filtering on each pixel point in the single-channel grayscale image, the exponential enhancement operation includes performing exponential enhancement on each pixel point in the single-channel grayscale image, and the noise filtering operation includes performing Gaussian low-pass filtering on each pixel point in the single-channel grayscale image.

[0027] Preferably, in the detection method, the determination threshold is determined by the following formula, specifically including:

[0028] t i =mean(M i )+k·std(M i )

[0029] Among them, t i is the decision threshold corresponding to the i-th target photovoltaic panel, mean(M i ) represents the mask area M of the i-th target photovoltaic panel in the single-channel grayscale image i The mean grayscale value of all pixels in the i ) represents the mask area M of the i-th target photovoltaic panel in the single-channel grayscale image i The standard deviation of all pixel grayscales in , k is the standard deviation weight coefficient.

[0030] Preferably, before the step of detecting the mask area of ​​the photovoltaic panel with a segmentation quality score higher than a preset score, the detection method further comprises:

[0031] The regions in the photovoltaic panel infrared image whose segmentation quality scores are not higher than a preset score are filtered out.

[0032] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned photovoltaic component hot spot detection method when executing the computer program.

[0033] A computer-readable storage medium stores a computer program, which implements the above-mentioned photovoltaic component hot spot detection method when executed by a processor.

[0034] The positive progressive effects of the present invention are: the present application combines deep learning image instance segmentation with small target detection, which is suitable for abnormal hot spots of different scales and shapes in typical photovoltaic panel infrared images taken by drone vision; the method adopts a deep learning instance segmentation method with segmentation quality evaluation capability, which can accurately segment the photovoltaic panel area and filter out the mis-segmented areas that do not conform to the shape of the photovoltaic panel. The algorithm is fast and efficient, and has strong practical value in the field of drone photovoltaic inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of a photovoltaic module hot spot detection method according to Embodiment 1 of the present invention.

[0036] Figure 2 It is a schematic diagram of a grayscale infrared image of a photovoltaic panel captured in the photovoltaic module hot spot detection method according to Example 1 of the present invention.

[0037] Figure 3 It is a schematic diagram of the infrared image segmentation of photovoltaic panels and the corresponding segmentation quality scoring results in the photovoltaic module hot spot detection method according to Example 1 of the present invention.

[0038] Figure 4 This is a flow chart of step 13 in the photovoltaic module hot spot detection method according to embodiment 1 of the present invention.

[0039] Figure 5 This is a flow chart of step 132 in the photovoltaic module hot spot detection method according to embodiment 1 of the present invention.

[0040] Figure 6 This is a flow chart of training the photovoltaic panel segmentation model in the photovoltaic module hot spot detection method of Example 1 of the present invention.

[0041] Figure 7 This is a schematic diagram of the structure of an electronic device according to Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0042] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0043] Example 1

[0044] A method for detecting hot spots in photovoltaic modules, such as Figure 1 As shown, the detection method comprises:

[0045] Step 11, obtaining an infrared image of the photovoltaic panel within a preset area;

[0046] Among them, see Figure 2 , the drone can fly parallel to the arrangement direction of each row of the photovoltaic array, and the optical axis viewing angle of the airborne infrared thermal imager is looking down vertically to the ground, thereby collecting infrared image data of the photovoltaic panels.

[0047] Step 12: Input the photovoltaic panel infrared image into the trained photovoltaic panel segmentation model, and output multiple groups of photovoltaic panel data, each group of photovoltaic panel data includes the mask area of ​​each photovoltaic panel obtained by segmentation and the corresponding photovoltaic panel segmentation quality score;

[0048] Among them, the Mask Scoring R-CNN instance segmentation method with accurate segmentation quality evaluation capability can be used to train and optimize the network model, and obtain the network weight parameters of the MaskScoring R-CNN instance segmentation model that can be used to accurately detect and segment each photovoltaic panel area. The model and the corresponding weight parameters are applied to the detection and segmentation of each photovoltaic panel in the image to be detected, and the mask area corresponding to each photovoltaic panel in the image to be detected and the corresponding segmentation quality score of each area are output. The training of the photovoltaic panel segmentation model will be further explained in the following text.

[0049] It should be noted that, during the segmentation process, the circumscribed rectangular frame corresponding to each photovoltaic panel can be automatically generated according to the circumscribed rectangle of each photovoltaic panel mask area (it can also be set during model training, and the model can directly output the corresponding external rectangular frame position). In the subsequent hot spot detection process, the regional image of each photovoltaic panel can be obtained based on the circumscribed rectangular frame and input into the subsequent hot spot detection model for detection.

[0050] Step 13: Filter out areas in the photovoltaic panel infrared image whose segmentation quality scores are not higher than a preset score.

[0051] For example, setting the segmentation quality score threshold t mask , remove all s in the image mask <t mask The detection and segmentation results of s are retained. mask ≥t mask The area is used as a candidate photovoltaic panel area Get the photovoltaic panel information set retained in the image Among them, M i represents the pixel-level segmentation mask area of ​​the i-th photovoltaic panel, B i Indicates M i The corresponding circumscribed rectangular bounding box;

[0052] Step 14: Detect the mask area of ​​the photovoltaic panel whose segmentation quality score is higher than the preset score to obtain the abnormal photovoltaic panel containing the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel.

[0053] For details, see Figure 3 , shows the schematic diagram of the photovoltaic panel infrared image segmentation and the corresponding segmentation quality scoring results, respectively, for all candidate photovoltaic panel regions in the image After processing, the abnormal hot spot detection results of all photovoltaic panels in the image are finally obtained. The image block contained in is taken as input, combined with its corresponding segmentation mask M i , further adopts infrared small target detection methods such as fast visual saliency filtering to quickly extract abnormal hot spots in the photovoltaic panel area and output the abnormal hot spot target area.

[0054] In the above scheme, a deep learning instance segmentation method with segmentation quality evaluation capability is adopted, which can accurately segment the photovoltaic panel area and filter out the mis-segmented areas that do not conform to the shape of the photovoltaic panel, and then detect abnormal hot spots.

[0055] In this embodiment, Figure 4 As shown, step 13 specifically includes:

[0056] Step 131, selecting a photovoltaic panel with a segmentation quality score higher than a preset score as a target photovoltaic panel;

[0057] Step 132: Detect the photovoltaic panel image corresponding to the target photovoltaic panel based on the infrared image small target detection algorithm to obtain the abnormal photovoltaic panel including the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel.

[0058] In the above scheme, considering that the hot spot area in the photovoltaic panel area is generally of unconventional size, a small target detection algorithm is used to detect the hot spot area to improve the detection sensitivity and detection accuracy. Specifically, Figure 5 As shown, step 132 specifically includes:

[0059] Step 1321, grayscale processing is performed on the photovoltaic panel image corresponding to the target photovoltaic panel to obtain a single-channel grayscale image;

[0060] Specifically, for example, image B i Convert to a single channel grayscale image I i ;

[0061] Step 1322: preprocess the single-channel grayscale image, where the preprocessing includes at least one of a convolution filter process, an exponential enhancement operation, and a noise filtering operation;

[0062] Among them, the convolution filtering processing includes performing high-pass convolution filtering on each pixel in the single-channel grayscale image, the exponential enhancement operation includes performing exponential enhancement on each pixel in the single-channel grayscale image, and the noise filtering operation includes performing Gaussian low-pass filtering on each pixel in the single-channel grayscale image.

[0063] Specifically, a 5×5 visual saliency convolution kernel is used to map the mask M i Each pixel point in the region is subjected to high-pass convolution filtering, and the weight parameter matrix of the convolution kernel is: [-4, -1, 0, -1, -4; -1, 2, 3, 2, -1; 0, 3, 4, 3, 0; -1, 2, 3, 2, -1; -4, -1, 0, -1, -4]; after convolution, the photovoltaic panel saliency map S is obtained. i ; Then, the exponential enhancement method is used to enhance the saliency map S i Each pixel value is enhanced, that is Usually α is set to 1.0~2.0; the enhanced image E of the photovoltaic panel saliency map is obtained i ; Then, a Gaussian filter is used, and the Gaussian kernel parameter σ is set to 1.5, and the enhanced image E i The mask M i Gaussian low-pass filtering is performed on each pixel within the area to filter out noise and highlight the visual effective area of ​​the hot spot area;

[0064] Step 1323: perform binary segmentation on the single-channel grayscale image, and determine that the pixel area where the image pixels are greater than the judgment threshold is the hot spot area.

[0065] Specifically, an adaptive threshold segmentation method is used to enhance the image E i The mask M i The pixels in the area are divided into two values, which are greater than the threshold t i The pixel area is the abnormal hot spot target area.

[0066] The detection method can specifically determine the judgment threshold through the following formula, including:

[0067] t i =mean(M i )+k·std(M i )

[0068] Among them, t i is the decision threshold corresponding to the i-th target photovoltaic panel, mean(M i ) represents the mask area M of the i-th target photovoltaic panel in the single-channel grayscale image i The mean grayscale value of all pixels in the i ) represents the mask area M of the i-th target photovoltaic panel in the single-channel grayscale image iThe standard deviation of all pixel grayscales in the image, k is the standard deviation weight coefficient, and k is usually 3.0 to 5.0.

[0069] In this embodiment, the photovoltaic panel segmentation model is trained by the following steps in the detection method: Figure 6 As shown, specifically including:

[0070] Step 21, constructing a photovoltaic panel training sample set; the photovoltaic panel training sample set includes a plurality of photovoltaic panel infrared training images, and each photovoltaic panel infrared training image includes an actual mask area of ​​each training photovoltaic panel;

[0071] Specifically, the drone is flown parallel to the arrangement direction of each row of the photovoltaic array, and the optical axis viewing angle of the onboard infrared thermal imager is looking down vertically to the ground, so as to collect infrared image data of the photovoltaic panels, and the collected images are manually annotated, and the annotated content includes the pixel-level position of each photovoltaic panel area (and may also include the position of the circumscribed rectangular frame), and then a training sample set containing multiple photovoltaic panel infrared training images is constructed; wherein the coordinates of the four vertices of each photovoltaic panel are specifically annotated, and the quadrilateral closed area formed by the connecting lines is the mask area occupied by the photovoltaic panel, and the circumscribed rectangular frame position corresponding to each photovoltaic panel is automatically generated according to the circumscribed rectangle of the mask area of ​​each photovoltaic panel.

[0072] Step 22: Using the photovoltaic panel training sample set as training data, train to obtain an initial photovoltaic panel segmentation model, and obtain a predicted mask area and a corresponding segmentation confidence of each training photovoltaic panel;

[0073] Step 23, calculating the segmentation quality score of each training photovoltaic panel based on the predicted mask area, the actual mask area and the segmentation confidence;

[0074] Wherein, step 23 specifically includes:

[0075] The IOU value of each training photovoltaic panel is calculated based on the predicted mask area and the actual mask area;

[0076] The segmentation quality score is obtained based on the product of the IOU value and the segmentation confidence.

[0077] Step 24: Add the segmentation quality score to the training data and continue training to obtain a photovoltaic panel segmentation model.

[0078] Specifically, the Mask Scoring R-CNN instance segmentation model is based on the Mask R-CNN instance segmentation model network structure, and further adds a MaskIoU Head network branch module to regress and calculate the intersection over union (IOU) value s between the predicted mask area and the actual mask area predicted by the Mask R-CNN model. iou, and obtain the segmentation confidence s by combining it with the Mask R-CNN object detection branch cls Multiply them together to get the target segmentation quality score, i.e., s mask =s cls ×s iou .

[0079] Among them, the Mask R-CNN backbone network adopts the classic ResNet-50 structure and combines it with the FPN (Feature Pyramid Networks) model architecture. The network input image scale is 800×1333 pixels, the basic size of the detection anchor box is 8, and three aspect ratios of 0.5, 1.0, and 2.0 are set respectively. Five feature layers with commonly used step sizes of 4, 8, 16, 32, and 645 are selected for fine detection and classification. The input channel feature dimensions of FPN are 256, 512, 1024, and 2048, and the output channel feature dimension is 256.

[0080] More specifically, the specific structure of the MaskIOU Head network branch module in the above-mentioned Mask Scoring R-CNN instance segmentation model is as follows: the multi-dimensional feature map output by the RoIAlign layer in the Mask R-CNN network is concatenated with the predicted output single-dimensional target segmentation mask Mask map as the input of the MaskIoU Head network. The MaskIOU Head consists of 4 convolutional layers and 3 fully connected layers. The MaskIOU Head contains several convolutional layers and fully connected layers with a convolution kernel of 3×3. The convolution kernel scale in the convolution layer is 3×3 and the output feature channel dimension is 256. The first two output feature channel dimensions in the fully connected layer are 1024, and the last output feature channel dimension is 1. The output of the last fully connected layer is the photovoltaic panel category segmentation score s. mask Regression prediction of .

[0081] In the test and evaluation experiment, the training set contained 2,994 sample images of 128,874 photovoltaic panels and 34,462 hot spot targets for iterative training. Then the trained model was tested on the test sample set containing 749 sample images of 33,242 photovoltaic panels and 8,962 hot spot targets. The final result was that the photovoltaic panel segmentation accuracy was higher than 97%, and the abnormal hot spot detection accuracy was higher than 70%.

[0082] In this embodiment, deep learning image instance segmentation is combined with small target detection, which is suitable for abnormal hot spots of different scales and shapes in typical photovoltaic panel infrared images taken by drone vision. This method uses a deep learning instance segmentation method with segmentation quality evaluation capability, which can accurately segment the photovoltaic panel area and filter out the mis-segmented areas that do not conform to the shape of the photovoltaic panel. The algorithm is fast and efficient, and has strong practical value in the field of drone photovoltaic inspection.

[0083] Example 2

[0084] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the photovoltaic module hot spot detection method described in Example 1 when executing the computer program.

[0085] Figure 7 A schematic diagram of the structure of an electronic device provided in this embodiment. Figure 7 A block diagram of an exemplary electronic device 90 suitable for use in implementing embodiments of the present invention is shown. Figure 7 The electronic device 90 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0086] like Figure 7 As shown, the electronic device 90 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0087] The bus 93 includes a data bus, an address bus, and a control bus.

[0088] The memory 92 may include a volatile memory, such as a random access memory (RAM) 921 and / or a cache memory 922 , and may further include a read-only memory (ROM) 923 .

[0089] The memory 92 may also include a program tool 925 having a set (at least one) of program modules 924, such program modules 924 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the implementation of a network environment.

[0090] The processor 91 executes various functional applications and data processing by running computer programs stored in the memory 92 .

[0091] The electronic device 90 may also communicate with one or more external devices 94 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 95. Furthermore, the electronic device 90 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 96. The network adapter 96 communicates with other modules of the electronic device 90 via a bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0092] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0093] Example 3

[0094] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the photovoltaic module hot spot detection method described in Example 1.

[0095] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0096] In a possible implementation manner, the present invention may also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the photovoltaic module hot spot detection method described in Example 1.

[0097] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on a user device, partially on a user device, as an independent software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0098] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A method for detecting hot spots in photovoltaic modules, characterized in that: The detection method comprises: Acquire infrared images of photovoltaic panels within a preset area; Inputting the photovoltaic panel infrared image into a trained photovoltaic panel segmentation model, and outputting multiple groups of photovoltaic panel data, each group of photovoltaic panel data including a mask area of ​​each photovoltaic panel obtained by segmentation and a corresponding photovoltaic panel segmentation quality score; Detecting the mask area of ​​the photovoltaic panel whose segmentation quality score is higher than a preset score to obtain an abnormal photovoltaic panel containing an abnormal hot spot and a hot spot area of ​​the abnormal photovoltaic panel; The photovoltaic panel segmentation model is obtained by training through the following steps in the detection method, specifically including: Constructing a photovoltaic panel training sample set, wherein the photovoltaic panel training sample set includes a plurality of photovoltaic panel infrared training images, and each photovoltaic panel infrared training image includes an actual mask area of ​​each training photovoltaic panel; Using the photovoltaic panel training sample set as training data, training to obtain an initial photovoltaic panel segmentation model, and obtaining a predicted mask area and a corresponding segmentation confidence of each training photovoltaic panel; Calculating a segmentation quality score for each training photovoltaic panel based on the predicted mask area, the actual mask area and the segmentation confidence; Adding the segmentation quality score to the training data and continuing the training to obtain the photovoltaic panel segmentation model; The step of calculating the segmentation quality score of each training photovoltaic panel based on the predicted mask area, the actual mask area and the segmentation confidence specifically includes: Calculate the IOU value of each training photovoltaic panel based on the predicted mask area and the actual mask area; The segmentation quality score is obtained based on the product of the IOU value and the segmentation confidence.

2. The photovoltaic module hot spot detection method according to claim 1, characterized in that: The step of detecting the mask area of ​​the photovoltaic panel whose segmentation quality score is higher than the preset score to obtain the abnormal photovoltaic panel containing the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel specifically includes: Select the photovoltaic panels with segmentation quality scores higher than the preset score as the target photovoltaic panels; Based on the infrared image small target detection algorithm, the photovoltaic panel image corresponding to the target photovoltaic panel is detected to obtain the abnormal photovoltaic panel containing the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel.

3. The photovoltaic module hot spot detection method according to claim 2, characterized in that: The step of detecting the photovoltaic panel image corresponding to the target photovoltaic panel based on the infrared image small target detection algorithm to obtain the abnormal photovoltaic panel including the abnormal hot spot and the hot spot area of ​​the abnormal photovoltaic panel specifically includes: Performing grayscale processing on the photovoltaic panel image corresponding to the target photovoltaic panel to obtain a single-channel grayscale image; The single-channel grayscale image is subjected to binary segmentation, and a pixel region where the image pixels are greater than a determination threshold is determined as a hot spot region.

4. The photovoltaic module hot spot detection method according to claim 3, characterized in that: Before the step of performing binary segmentation on the single-channel grayscale image, the detection method further comprises: Preprocessing the single-channel grayscale image, wherein the preprocessing includes at least one of a convolution filter process, an exponential enhancement operation, and a noise filtering operation; Among them, the convolution filtering processing includes performing high-pass convolution filtering on each pixel point in the single-channel grayscale image, the exponential enhancement operation includes performing exponential enhancement on each pixel point in the single-channel grayscale image, and the noise filtering operation includes performing Gaussian low-pass filtering on each pixel point in the single-channel grayscale image.

5. The photovoltaic module hot spot detection method according to claim 3, characterized in that: In the detection method, the determination threshold is determined by the following formula, specifically including: t i =mean(M i )+k·std(M i ) Among them, t i is the decision threshold corresponding to the i-th target photovoltaic panel, mean(M i ) represents the mask area M of the i-th target photovoltaic panel in the single-channel grayscale image i The mean grayscale value of all pixels in the i ) represents the mask area M of the i-th target photovoltaic panel in the single-channel grayscale image i The standard deviation of all pixel grayscales in , k is the standard deviation weight coefficient.

6. The photovoltaic module hot spot detection method according to claim 1, characterized in that: Before the step of detecting the mask area of ​​the photovoltaic panel whose segmentation quality score is higher than a preset score, the detection method further includes: The regions in the photovoltaic panel infrared image whose segmentation quality scores are not higher than a preset score are filtered out.

7. 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 photovoltaic module hot spot detection method according to any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the photovoltaic module hot spot detection method according to any one of claims 1 to 6 is implemented.

Images