Photovoltaic module hot spot detection method, system, equipment, medium and product

Abstract

The invention discloses a photovoltaic module hot spot detection method, system and equipment, a medium and a product, and relates to the field of photovoltaic equipment detection.The method comprises the steps that an infrared image of a sample photovoltaic module is collected and subjected to data enhancement, and an expanded data set is obtained; a YOLOv5-ECA model is constructed on the basis of a YOLOv5s network; the YOLOv5-ECA model comprises a trunk feature extraction network, a reinforced feature extraction network and a detection head which are connected in sequence, and an efficient channel attention mechanism is introduced into the trunk feature extraction network; training the YOLOv5-ECA model by adopting the extended data set to obtain a photovoltaic module hot spot detection model; acquiring an infrared image of a target photovoltaic module and inputting the infrared image into the photovoltaic module hot spot detection model to obtain a detection result; the detection result comprises existence of hot spots and hot spot areas. According to the method, the dependence on high-quality data samples can be reduced, and meanwhile, the detection precision and the reasoning speed are improved.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic equipment detection, and in particular to a method, system, equipment, medium and product for detecting hot spots in photovoltaic modules. Background Art

[0002] When photovoltaic modules encounter obstructions, cracks, bubbles, delamination, dirt, or internal connection failures, they can generate localized heating, known as the hot spot effect. This is a major cause of photovoltaic module failure. Hot spots not only consume the energy generated by the photovoltaic modules, reducing overall output power, but can also cause permanent damage to the modules and even cause fires. Therefore, technical means are needed to promptly detect hot spot failures in photovoltaic modules and implement remedial measures to ensure the safe operation of photovoltaic power plants.

[0003] Because hot spots manifest as high-temperature, bright areas in infrared images, hot spot detection can be achieved by analyzing infrared images of photovoltaic modules. These techniques can be broadly categorized into two types. One relies on traditional image processing methods. Yang Yanan proposed a hot spot detection algorithm based on a support vector machine (SVM). This algorithm extracts temperature features from sub-image blocks and constructs a classification model, which involves complex matrix calculations. Tsanakas et al. first perform edge detection on infrared array images using the Canny operator and then analyze the temperature histogram of the extracted area. This method requires manual labeling of the array and background areas after edge detection, resulting in a low degree of automation. The other type of approach is based on deep learning. Guo Menghao et al. used the Faster-CNN algorithm combined with transfer learning to train the model. Compared to single-stage detection algorithms, this model achieves higher recognition accuracy, but its larger number of parameters results in slower inference speed. Wang Daolei's team proposed a detection algorithm based on a modified YOLOv4-tiny. This algorithm uses a convolutional block attention module (CBAM) to focus on channel features, but it fails to effectively capture attention interactions between channels. Although deep learning-based methods achieve high detection accuracy, they are heavily dependent on high-quality data samples. Summary of the Invention

[0004] The purpose of this application is to provide a photovoltaic module hot spot detection method, system, equipment, medium and product, which can reduce the dependence on high-quality data samples while improving detection accuracy and reasoning speed.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for detecting hot spots in photovoltaic modules, comprising:

[0007] Collect infrared images of sample photovoltaic modules and perform data enhancement to obtain an expanded dataset;

[0008] A YOLOv5-ECA model is constructed based on the YOLOv5s network; the YOLOv5-ECA model includes a backbone feature extraction network, an enhanced feature extraction network, and a detection head connected in sequence, and an efficient channel attention mechanism is introduced into the backbone feature extraction network; the backbone feature extraction network is used to extract features from the input image to obtain a first feature map; the enhanced feature extraction network is used to extract features from the first feature map to obtain a second feature map; and the detection head is used to classify and locate the second feature map;

[0009] The YOLOv5-ECA model was trained using the expanded dataset to obtain a photovoltaic module hot spot detection model;

[0010] An infrared image of the target photovoltaic module is collected and input into a photovoltaic module hot spot detection model to obtain a detection result; the detection result includes the presence or absence of hot spots and the hot spot area.

[0011] Optionally, infrared images of sample photovoltaic modules are collected and data augmentation is performed to obtain an extended dataset, including:

[0012] Collect infrared images of several sample photovoltaic modules as basic data set 1;

[0013] Perform grid mask processing on all images in basic dataset 1 to obtain basic dataset 2;

[0014] Perform elastic deformation processing on all images in basic dataset 2 to obtain basic dataset 3;

[0015] The basic dataset 1, basic dataset 2 and basic dataset 3 are merged to obtain the expanded dataset.

[0016] Optionally, infrared images of sample photovoltaic modules are collected and data augmentation is performed to obtain an extended dataset, including:

[0017] Collect infrared images of several sample photovoltaic modules as basic data set 1;

[0018] Perform grid mask processing on all images in basic dataset 1 to obtain basic dataset 2;

[0019] Perform elastic deformation processing on all images in basic dataset 2 to obtain basic dataset 3;

[0020] Perform elastic deformation processing on all images in basic dataset 1 to obtain basic dataset 4;

[0021] The basic dataset 1, basic dataset 2, basic dataset 3 and basic dataset 4 are merged to obtain the expanded dataset.

[0022] Optionally, the backbone feature extraction network includes a convolution plus normalization structure, a first inverse residual module, a first efficient channel attention module, a second inverse residual module, a second efficient channel attention module, a third inverse residual module, a third efficient channel attention module, a fourth inverse residual module and a fourth efficient channel attention module connected in sequence; the first inverse residual module includes a one-layer inverse residual structure; the second inverse residual module includes a two-layer inverse residual structure; the third inverse residual module includes a five-layer inverse residual structure; the fourth inverse residual module includes a three-layer inverse residual structure.

[0023] Optionally, the enhanced feature extraction network includes a first convolutional layer, a first upsampling layer, a first fusion layer, a first C3 module, a second convolutional layer, a second upsampling layer, a second fusion layer, a second C3 module, a third convolutional layer, a third fusion layer, a third C3 module, a fourth convolutional layer, a fourth fusion layer and a fourth C3 module connected in sequence; the first fusion layer is also connected to the third inverse residual module; the second fusion layer is also connected to the second inverse residual module; the third fusion layer is also connected to the second convolutional layer; and the fourth fusion layer is also connected to the fourth inverse residual module.

[0024] Optionally, the YOLOv5-ECA model is trained using an expanded dataset to obtain a photovoltaic module hot spot detection model, including:

[0025] Label the hot spot areas of all images in the expanded data set to obtain a labeling file;

[0026] The stochastic gradient descent optimizer was used to train the YOLOv5-ECA model based on the expanded dataset and annotation files to obtain a photovoltaic module hot spot detection model. The initial learning rate was 0.0001, the batch size was 4, the weight decay was 0.0005, and a total of 400 rounds of training were performed, with the first 100 rounds using the frozen training method.

[0027] In a second aspect, the present application provides a photovoltaic module hot spot detection system, comprising:

[0028] An infrared image acquisition unit, used for acquiring infrared images of photovoltaic modules;

[0029] a storage unit for storing a computer program;

[0030] The processing unit is used to run a computer program to implement the photovoltaic module hot spot detection method.

[0031] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the photovoltaic module hot spot detection method.

[0032] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the photovoltaic module hot spot detection method when executed by a processor.

[0033] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which implements the photovoltaic module hot spot detection method when executed by a processor.

[0034] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0035] The present application provides a method, system, device, medium and product for detecting hot spots in photovoltaic modules. By performing data enhancement on the infrared images of collected sample photovoltaic modules, a diverse and abundant expanded data set can be obtained, reducing the dependence on high-quality data samples, effectively compensating for the problem of sparse and poor quality infrared images of hot spots in photovoltaic modules, and thus effectively improving the detection accuracy. By improving the YOLOv5s network, introducing an efficient channel attention mechanism, and constructing a YOLOv5-ECA model, the interaction between channels can be captured in a non-dimensionality reduction manner, thereby improving the detection accuracy. Based on the improved backbone network of YOLOv5s, the inference speed of the detection model can also be greatly improved while minimizing the loss of model accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0037] Figure 1 Flowchart of the photovoltaic module hot spot detection method provided in this application;

[0038] Figure 2 The YOLOv5-ECA model structure diagram provided for this application;

[0039] Figure 3 The Conv3BN network structure diagram provided for this application;

[0040] Figure 4 Inverted Residual network structure diagram provided for this application;

[0041] Figure 5 ECANet network structure diagram provided for this application. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0043] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0044] In an exemplary embodiment, the present application provides a photovoltaic module hot spot detection method, which is performed using a photovoltaic module hot spot detection system based on a computer. In the embodiment of the present application, Figure 1 As shown, the method includes the following steps 1 to 4.

[0045] Step 1: Collect infrared images of sample photovoltaic modules and perform data enhancement to obtain an expanded dataset.

[0046] The infrared image data of the photovoltaic module is enhanced using data enhancement technology to obtain enhanced and expanded infrared images of the photovoltaic module, thereby obtaining an expanded dataset. The data enhancement technology is a hybrid enhancement technology that includes grid masking and elastic deformation. The specific implementation process is as follows.

[0047] The infrared images of photovoltaic modules captured by the infrared image acquisition unit in the photovoltaic module hot spot detection system are used as basic dataset 1. All images in basic dataset 1 are first grid-masked to obtain basic dataset 2. Then, all images in basic dataset 2 are elastically deformed to obtain basic dataset 3. Basic datasets 1, 2, and 3 are merged to obtain the expanded dataset.

[0048] Define the grid mask M={r,d,δ x ,δ y}, where the parameter r determines the retention ratio of the input image, the parameter d represents the length of the cell, and δ x and δ y They represent the x-direction distance and y-direction distance from the first unit to the edge of the image respectively.

[0049] The retention ratio k of a given mask M r The calculation is as follows:

[0050]

[0051] k r =1-(1-r) 2 =2r-r 2 (2)

[0052] δ x (δ y )=random(0,d-1) (3)

[0053] Here, h and w represent the height and width of the image respectively.

[0054] Alternatively, grid masking can be performed on only base dataset 1 to obtain base dataset 2. Elastic deformation can be performed on only base dataset 1 to obtain base dataset 4. The data in base dataset 4 can be diversified by adjusting the activation factor α. Base dataset 1 is then combined with base datasets 2, 3, and 4 obtained through data augmentation to form the augmented dataset.

[0055] Step 2: Build the YOLOv5-ECA model based on the YOLOv5s network.

[0056] Based on YOLOv5s, a YOLOv5-ECA model is constructed for hot spot detection. The overall framework of the model is as follows: Figure 2 As shown, the system includes a backbone feature extraction network (Backbone), a reinforcement feature extraction network (Neck), and a detection head (Detect) connected in sequence, and an efficient channel attention mechanism is introduced into the backbone feature extraction network. The backbone feature extraction network is used to extract features from the input image to obtain a first feature map; the reinforcement feature extraction network is used to extract features from the first feature map to obtain a second feature map; and the detection head is used to classify and locate the second feature map.

[0057] The backbone feature extraction network (Backbone) is the input to the hotspot detection model, extracting feature maps from infrared images of photovoltaic modules. This application uses MobileNetv3 as the backbone feature extraction network based on YOLOv5s and improves MobileNetv3 by adding a lightweight ECA attention mechanism. The improved backbone feature extraction network consists of a Conv3BN, an Inverted Residual module, and an Efficient Channel Attention (ECA) module connected in sequence.

[0058] Specifically, the backbone feature extraction network includes a convolution plus normalization structure (Conv3BN), a first inverse residual module, a first efficient channel attention module, a second inverse residual module, a second efficient channel attention module, a third inverse residual module, a third efficient channel attention module, a fourth inverse residual module and a fourth efficient channel attention module connected in sequence; the first inverse residual module includes a layer of inverse residual structure, namely Inverted Residual; the second inverse residual module includes a two-layer inverse residual structure, namely 2*Inverted Residual; the third inverse residual module includes a five-layer inverse residual structure, namely 5*Inverted Residual; the fourth inverse residual module includes a three-layer inverse residual structure, namely 3*InvertedResidual. The following is a detailed introduction to each structure in the backbone feature extraction network.

[0059] Conv3BN is a standard convolution plus normalization structure. The network structure is as follows Figure 3 As shown in the figure, it includes the input layer (Input), the convolution layer (3*3Conv2d) with a convolution kernel size of 3*3, the normalization operation + activation function layer (BN+Hardwish) and the output layer (Output).

[0060] The Inverted Residual module is the basic module in MobileNetv3. The network structure is as follows: Figure 4 As shown, it includes the input layer (Input), the convolution layer with a convolution kernel size of 1*1 (1*1Conv2d), the normalization operation + activation function layer (BN+Hardwish), the depth-separable convolution layer with a convolution kernel size of 3*3 (3*3DW-Conv2d), the normalization layer (BN), the channel attention layer (SEblock), the activation function layer (Hardwish), the convolution layer with a convolution kernel size of 1*1 (1*1Conv2d), the normalization layer (BN) and the output layer (Output).

[0061] The Efficient Channel Attention module (ECA) is an improvement of the channel attention in SENet. The network structure is as follows: Figure 5 As shown, it includes an input layer (Input), a global average pooling layer (GAP), a convolution layer (k*k Conv2d) whose convolution kernel size is adaptively determined to be k*k (k=5 in this application), an activation function layer (sigmoid), and an output layer (Output). Where C represents the number of feature map channels, W represents the feature map width, H represents the feature map height, and X represents the input of the convolution block. represents the output of the convolutional block, and σ represents the activation function.

[0062] The enhanced feature extraction network (Neck) uses the Feature Pyramid Network (FPN) + Path Aggregation Network (PAN) structure to combine upsampling and downsampling to form a feature pyramid structure to further extract feature map information. The input backbone data is (640*640*3), and the data obtained after the Conv3BN structure is (320*320*64) and transmitted to the Inverted Residual structure; the data obtained after the Inverted Residual structure is (160*160*128) and transmitted to the ECA and 2*Inverted Residual structures; the data obtained after the ECA and 2*Inverted Residual structures is (80*80*256) and transmitted to the Concat structure in Neck. For other detailed data output, see Figure 2 As shown in the figure, Conv represents the convolution layer, Upsample represents the upsampling layer, Concat represents the fusion layer, and C3 represents the C3 module.

[0063] Specifically, the enhanced feature extraction network includes a first convolutional layer, a first upsampling layer, a first fusion layer, a first C3 module, a second convolutional layer, a second upsampling layer, a second fusion layer, a second C3 module, a third convolutional layer, a third fusion layer, a third C3 module, a fourth convolutional layer, a fourth fusion layer and a fourth C3 module connected in sequence; the first fusion layer is also connected to the third inverse residual module; the second fusion layer is also connected to the second inverse residual module; the third fusion layer is also connected to the second convolutional layer; the fourth fusion layer is also connected to the fourth inverse residual module.

[0064] The detection head (Detect) performs the final classification and positioning of the enhanced feature map, that is, classifies the presence or absence of hot spots and locates the hot spot area in the image with hot spots.

[0065] Step 3: Use the expanded dataset to train the YOLOv5-ECA model to obtain a photovoltaic module hot spot detection model.

[0066] The images in the expanded dataset were randomly screened in an 8:2 ratio. 80% of the selected images were defined as the training dataset, and the remaining 20% were defined as the test dataset. Hotspot areas in the expanded dataset images were manually annotated, with the hotspot areas outlined, and a corresponding annotation file was generated for each image.

[0067] After multiple rounds of model training and parameter adjustment optimization, the initialization parameters of the model training are set as follows: the Stochastic Gradient Descent (SGD) optimizer is used, the initial learning rate is 0.0001, the batch size is 4, the weight decay is set to 0.0005, and a total of 400 rounds of training are performed. The first 100 rounds are frozen training to speed up the training.

[0068] Step 4: Collect an infrared image of the target photovoltaic module and input it into the photovoltaic module hot spot detection model to obtain a detection result. The detection result includes the presence and area of the hot spot.

[0069] The infrared image acquisition unit of the photovoltaic module hot spot detection system is used to acquire an infrared image of the photovoltaic module as a photovoltaic module test image, and then the test image is input into a trained photovoltaic module hot spot detection model to obtain and output a detection result.

[0070] In an exemplary embodiment, the present application provides a photovoltaic module hot spot detection system. The system includes core modules such as an infrared image acquisition unit, a storage unit, and a processing unit. The infrared image acquisition unit is an infrared camera used to capture infrared images of photovoltaic modules. The storage unit is used to store a computer program that implements the photovoltaic module hot spot detection method described herein, as well as other relevant computer programs and data for the system. The processing unit is used to run the aforementioned computer program, including the photovoltaic module hot spot detection method, to enable the photovoltaic module hot spot detection system to execute the photovoltaic module hot spot detection method described herein.

[0071] The system takes the above-mentioned units as core components and is not limited to hardware composition and resource scale, such as "cloud computing system (for example, camera + server)", "edge computing system (for example, portable inspection equipment)", and "cloud and edge combined computing system (for example, camera + drone / robot)".

[0072] This application addresses the problem of insufficient hot spot feature extraction capabilities by introducing an ECA module to capture the interaction between channels in a non-dimensionality reduction manner, thereby improving detection accuracy. This application is based on an improved backbone network based on YOLOv5s, which significantly improves the inference speed of the detection model while minimizing the loss of model accuracy. This application uses data enhancement technology to help obtain a diverse and abundant training data set of photovoltaic module infrared images, effectively compensating for the scarcity and poor quality of photovoltaic module hot spot infrared images, and effectively improving detection accuracy. Compared with the existing technology, this application has the following advantages:

[0073] In the Chinese patent "Photovoltaic Panel Hot Spot Identification Method, Storage Medium, and Electronic Device" (CN202310866947.X), a semantic segmentation model is implemented using the U2Net network and RSU-L structure. A Yolov5 detection model is then constructed by incorporating the SENet network. This model can detect faults such as cell failures, reflections, hot spots, diode failures, and occlusions with a certain degree of detection accuracy. Compared to CN202310866947.X, this application implements a hot spot detection model using an instance segmentation algorithm based on target detection. It also upgrades the SENet attention network to an ECA network, achieving higher detection accuracy while reducing computational overhead.

[0074] In the Chinese patent "A method for detecting hot spots in infrared images of photovoltaic panels based on an improved BETR model" (CN202211290886.9), an improved BETR model was constructed and the BETR model was trained in two stages using the concept of transfer learning. This resulted in an optimized BETR model, which improved the recognition accuracy of small target hot spots. Compared to CN202211290886.9, this application improves the backbone feature extraction network by introducing MobileNetv3. MobileNetv3, based on a lightweight architecture, can significantly increase the inference speed, that is, the real-time detection rate, without a significant loss in detection accuracy, while facilitating end-to-end deployment.

[0075] In the Chinese patent “A method and system for detecting hot spots in photovoltaic panel infrared images based on YOLOv5” (CN202111647466.7), an improved YOLOv5 model was constructed. Its backbone network backbone adopts the Focus structure and CSP structure, and a convolutional attention module CBAM is added to improve the accuracy of hot spot detection of photovoltaic panels, which is suitable for distinguishing long strip hot spot defects from small hot spots. Compared with CN202111647466.7, this application integrates the inverse residual module and the efficient channel attention mechanism by improving the backbone network, which greatly reduces the number of model parameters with little reduction in accuracy. Moreover, the ECA attention mechanism used in this application captures the interaction between channels without reducing the dimension like CBAM, which is more conducive to the distribution of channel weights and is beneficial to improving recognition accuracy. The ECA attention captures the interaction between channels without reducing the dimension, which is very important for the distribution of channel weights.

[0076] In summary, although the above-mentioned related schemes claim to be able to realize the function of detecting hot spots of photovoltaic modules to a certain extent, the technical scheme adopted in this application is different from the above-mentioned related schemes, and is superior to the above-mentioned related schemes in at least one indicator such as computational complexity, detection accuracy, inference speed, and system complexity.

[0077] In an exemplary embodiment, the present application further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0078] In an exemplary embodiment, the present application further provides a computer-readable storage medium storing a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor.

[0079] In an exemplary embodiment, the present application further provides a computer program product, including a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor.

[0080] In this application, all actions to obtain signals, information, or data are performed in compliance with the relevant data protection laws and policies of the country in which they are located and with the authorization of the corresponding device owner. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, storage, and display, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data must comply with relevant laws and regulations.

[0081] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0082] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0083] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for detecting hot spots in photovoltaic modules, characterized in that: include: Collect infrared images of sample photovoltaic modules and perform data enhancement to obtain an expanded dataset; A YOLOv5-ECA model is constructed based on the YOLOv5s network; the YOLOv5-ECA model includes a backbone feature extraction network, an enhanced feature extraction network, and a detection head connected in sequence, and an efficient channel attention mechanism is introduced into the backbone feature extraction network; the backbone feature extraction network is used to extract features from the input image to obtain a first feature map; the enhanced feature extraction network is used to extract features from the first feature map to obtain a second feature map; and the detection head is used to classify and locate the second feature map; The YOLOv5-ECA model was trained using the expanded dataset to obtain a photovoltaic module hot spot detection model; An infrared image of the target photovoltaic module is collected and input into a photovoltaic module hot spot detection model to obtain a detection result; the detection result includes the presence or absence of hot spots and the hot spot area.

2. The photovoltaic module hot spot detection method according to claim 1, characterized in that: Collect infrared images of sample photovoltaic modules and perform data augmentation to obtain an expanded dataset, including: Collect infrared images of several sample photovoltaic modules as basic data set 1; Perform grid mask processing on all images in basic dataset 1 to obtain basic dataset 2; Perform elastic deformation processing on all images in basic dataset 2 to obtain basic dataset 3; The basic dataset 1, basic dataset 2 and basic dataset 3 are merged to obtain the expanded dataset.

3. The photovoltaic module hot spot detection method according to claim 1, characterized in that: Collect infrared images of sample photovoltaic modules and perform data augmentation to obtain an expanded dataset, including: Collect infrared images of several sample photovoltaic modules as basic data set 1; Perform grid mask processing on all images in basic dataset 1 to obtain basic dataset 2; Perform elastic deformation processing on all images in basic dataset 2 to obtain basic dataset 3; Perform elastic deformation processing on all images in basic dataset 1 to obtain basic dataset 4; The basic dataset 1, basic dataset 2, basic dataset 3 and basic dataset 4 are merged to obtain the expanded dataset.

4. The photovoltaic module hot spot detection method according to claim 1, characterized in that: The backbone feature extraction network includes a convolution plus normalization structure, a first inverse residual module, a first efficient channel attention module, a second inverse residual module, a second efficient channel attention module, a third inverse residual module, a third efficient channel attention module, a fourth inverse residual module and a fourth efficient channel attention module connected in sequence; the first inverse residual module includes a one-layer inverse residual structure; the second inverse residual module includes a two-layer inverse residual structure; the third inverse residual module includes a five-layer inverse residual structure; the fourth inverse residual module includes a three-layer inverse residual structure.

5. The photovoltaic module hot spot detection method according to claim 4, characterized in that: The enhanced feature extraction network includes a first convolutional layer, a first upsampling layer, a first fusion layer, a first C3 module, a second convolutional layer, a second upsampling layer, a second fusion layer, a second C3 module, a third convolutional layer, a third fusion layer, a third C3 module, a fourth convolutional layer, a fourth fusion layer and a fourth C3 module connected in sequence; the first fusion layer is also connected to the third inverse residual module; the second fusion layer is also connected to the second inverse residual module; the third fusion layer is also connected to the second convolutional layer; and the fourth fusion layer is also connected to the fourth inverse residual module.

6. The photovoltaic module hot spot detection method according to claim 1, characterized in that: The YOLOv5-ECA model is trained using the expanded dataset to obtain a photovoltaic module hot spot detection model, including: Label the hot spot areas of all images in the expanded data set to obtain a labeling file; The stochastic gradient descent optimizer was used to train the YOLOv5-ECA model based on the expanded dataset and annotation files to obtain a photovoltaic module hot spot detection model. The initial learning rate was 0.0001, the batch size was 4, the weight decay was 0.0005, and a total of 400 rounds of training were performed, with the first 100 rounds using the frozen training method.

7. A photovoltaic module hot spot detection system, characterized in that: include: An infrared image acquisition unit, used for acquiring infrared images of photovoltaic modules; a storage unit for storing a computer program; A processing unit, configured to run a computer program to implement the photovoltaic module hot spot detection method according to any one of claims 1 to 6.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the photovoltaic component hot spot detection method according to any one of claims 1 to 6.

9. 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.

10. A computer program product comprising a computer program, 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