System and method for identifying and detecting hot spots of photovoltaic panel

Abstract

The embodiment of the invention relates to the technical field of photovoltaic panel hot spot recognition and detection, and provides a photovoltaic panel hot spot recognition and detection system and a method thereof.A camera is used for collecting an infrared thermal image of an analyzed photovoltaic panel, and shallow layer features and deep layer features of the infrared thermal image are extracted; according to the method, the shallow-layer feature and the deep-layer feature are fused to obtain an enhanced shallow-layer feature map, and hot spot recognition detection is carried out based on the enhanced shallow-layer feature map, so that automation of hot spot detection and recognition of the photovoltaic panel is effectively realized, and the detection accuracy and robustness are improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of photovoltaic panel hot spot identification and detection, and particularly relates to a photovoltaic panel hot spot identification and detection system and method thereof. Background Art

[0002] A photovoltaic panel, also known as a solar panel or a solar photovoltaic panel, is a device that uses solar energy to generate electricity. It is composed of multiple photovoltaic cells and can convert sunlight into electrical energy. A photovoltaic cell is a semiconductor device that converts photons in sunlight into electrons through the photovoltaic effect, generating a direct current. A photovoltaic panel is usually composed of multiple photovoltaic cell wafers, which are generally made of semiconductor materials such as silicon. The surface of the photovoltaic cell wafers is covered with a reflective layer and an anti-reflection film to improve the absorption efficiency of sunlight. The photovoltaic cell wafers output electrical energy through metal wires and can be used alone or assembled into a photovoltaic array to form a photovoltaic system, i.e., a solar power generation system.

[0003] The working principle of a photovoltaic panel is based on the photovoltaic effect. When sunlight shines on the photovoltaic cell wafers, photons will excite electrons in the semiconductor material, causing them to jump into the conduction band and form an electric current. This electric current can be connected to external devices through a circuit for power supply or storage. The advantages of a photovoltaic panel include being renewable, environmentally friendly, noise-free, and having low maintenance costs, and it can be widely applied in various scenarios, including residential houses, commercial buildings, industrial facilities, and power supply in remote areas.

[0004] A photovoltaic panel hot spot refers to an area with abnormal temperature on the surface of the photovoltaic panel. The formation of hot spots may be caused by various reasons, including component damage, pollution, blockage, or cell wafer failure, etc., which will lead to a decrease in the light absorption ability or current loss in a local area of the photovoltaic panel, thereby causing the local temperature of the photovoltaic panel to rise and form a hot spot.

[0005] The existence of hot spots will have an adverse impact on the performance and lifespan of the photovoltaic panel. The increase in temperature in the hot spot area will cause the operating temperature of the photovoltaic cell to rise, thereby reducing the conversion efficiency of the cell. The larger and more severe the hot spot, the greater the impact on the cell conversion efficiency. Hot spots will also accelerate the aging of the photovoltaic cell and shorten the service life of the photovoltaic panel. The high-temperature environment in the hot spot area will also cause corrosion and damage to the cell material, thereby reducing the reliability and durability of the photovoltaic panel. The existence of the hot spot area will also generate additional maintenance and repair work. For large-scale photovoltaic power stations, the existence of hot spots needs to be detected and repaired in a timely manner to ensure the normal operation of the photovoltaic system.

[0006] To ensure the safe operation of the photovoltaic system and maximize the power generation efficiency, it is very important to detect and identify the hot spots of photovoltaic panels in a timely manner. Timely detection and identification of the hot spots of photovoltaic panels can help the operation and maintenance personnel discover problems early and take corresponding repair and maintenance measures to ensure the safe operation of the photovoltaic system and maximize the power generation efficiency, which is crucial for improving the reliability of the photovoltaic system, extending the service life of the photovoltaic panels, and protecting the return on investment.

[0007] Currently, the following methods and technologies are commonly used in the prior art to identify and detect the hot spots of photovoltaic panels: First, an infrared thermal imager is used to detect the hot spots by measuring the infrared radiation on the surface of the photovoltaic panel. The hot spots usually appear as areas with abnormal temperatures. The infrared thermal imaging technology can provide comprehensive hot spot detection and positioning. Second, a dedicated monitoring system is installed for the photovoltaic system. The monitoring system can monitor the temperature and performance parameters of the photovoltaic panels in real time, collect data and analyze the data, so as to detect the hot spots of the photovoltaic panels, discover abnormal situations and give alarms in a timely manner. By analyzing the operation data of the photovoltaic system, the abnormal hot spots of the photovoltaic panels can be detected. The data analysis can identify indicators such as performance degradation and abnormal temperature, so as to judge whether there is a hot spot problem. Third, the photovoltaic panels are manually visually inspected regularly to observe whether there are obvious hot spots or damage signs on the surface of the photovoltaic panels. This requires professional personnel to conduct the inspection and can assist other detection methods.

[0008] However, among the above methods commonly used in the prior art, installing a dedicated monitoring system for the photovoltaic system will increase the detection cost. Manually visually inspecting the photovoltaic panels not only requires professional personnel to complete, but also results in low detection efficiency. The analysis method based on infrared thermal images is a commonly used method for detecting and identifying the hot spots of photovoltaic panels at present. However, this method itself also has some challenges, making it difficult for traditional infrared thermal image processing methods to accurately extract and classify the hot spot areas. First, the infrared thermal images may be interfered by noise, and the noise will introduce uncertainty, making the boundaries of the hot spots blurred, thus affecting the accurate detection and identification of the hot spots. Second, the infrared thermal images may have non-uniformity, which means that the hot spot intensities and background brightnesses in different regions may vary, making the visual difference between the hot spots and the background small and difficult to accurately segment and extract the hot spot areas. In addition, the contrast of the infrared thermal images is usually low, and the temperature difference between the hot spots and the surrounding areas may not be obvious, resulting in the details of the hot spot areas being difficult to clearly display, which poses challenges to the accurate detection and identification of the hot spots. Summary of the Invention

[0009] This disclosure aims to at least solve one of the problems existing in the prior art, and provides a photovoltaic panel hot spot identification and detection system and method.

[0010] In one aspect of this disclosure, a method for identifying and detecting hot spots of a photovoltaic panel is provided, and the method includes:

[0011] Collect an infrared thermal image of the photovoltaic panel to be analyzed by a camera installed above the photovoltaic panel array;

[0012] Extract the shallow features and deep features of the infrared thermal image respectively to obtain a corresponding shallow feature map of the thermal infrared distribution of the photovoltaic panel and a deep feature map of the thermal infrared distribution of the photovoltaic panel;

[0013] Fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain a semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel;

[0014] Based on the semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel, determine the hot spot identification and detection result of the photovoltaic panel to be analyzed.

[0015] Optionally, extracting the shallow features and deep features of the infrared thermal image respectively to obtain a corresponding shallow feature map of the thermal infrared distribution of the photovoltaic panel and a deep feature map of the thermal infrared distribution of the photovoltaic panel includes:

[0016] Pass the infrared thermal image through a shallow feature extractor of temperature distribution based on a first convolutional neural network model to obtain the shallow feature map of the thermal infrared distribution of the photovoltaic panel;

[0017] Pass the shallow feature map of the thermal infrared distribution of the photovoltaic panel through a deep feature extractor of temperature distribution based on a second convolutional neural network model to obtain the deep feature map of the thermal infrared distribution of the photovoltaic panel.

[0018] Optionally, passing the infrared thermal image through a shallow feature extractor of temperature distribution based on a first convolutional neural network model to obtain the shallow feature map of the thermal infrared distribution of the photovoltaic panel includes:

[0019] Use each layer of the shallow feature extractor of temperature distribution based on the first convolutional neural network model to perform convolution processing, pooling processing, and non-linear activation processing on the infrared thermal image respectively in the forward pass of the layer, and extract the shallow feature map of the thermal infrared distribution of the photovoltaic panel from the shallow layer of the shallow feature extractor of temperature distribution based on the first convolutional neural network model.

[0020] Optionally, fusing the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain a semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel includes:

[0021] Use a residual information enhancement fusion module to fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain the semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel.

[0022] Optionally, a residual information enhanced fusion module is used to fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel, obtaining the shallow feature map of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask, including:

[0023] Upsample and perform convolution processing on the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain a reconstructed deep feature map of the thermal infrared distribution of the photovoltaic panel;

[0024] Calculate the position-wise difference between the reconstructed deep feature map of the thermal infrared distribution of the photovoltaic panel and the shallow feature map of the thermal infrared distribution of the photovoltaic panel to obtain a difference feature map;

[0025] Perform non-linear activation processing on the difference feature map based on the Sigmoid function to obtain a mask feature map;

[0026] Perform element-wise multiplication on the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the mask feature map to obtain a fused feature map;

[0027] Perform an attention-based PMA pooling operation on the fused feature map to obtain the shallow feature map of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic mask.

[0028] Optionally, based on the shallow feature map of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask, determine the hot spot identification and detection result of the photovoltaic panel to be analyzed, including:

[0029] Perform dimensionality reduction processing on the shallow feature map of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask to obtain a shallow feature matrix of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask;

[0030] Perform feature distribution correction on the shallow feature matrix of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask to obtain a corrected shallow feature matrix of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask;

[0031] Use the Softmax classification function to perform class prediction on each pixel in the corrected shallow feature matrix of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask to obtain the hot spot identification and detection result.

[0032] Optionally, perform dimensionality reduction processing on the shallow feature map of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask to obtain a shallow feature matrix of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask, including:

[0033] Perform pooling processing on the shallow feature map of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask along the channel dimension to obtain the shallow feature matrix of the thermal infrared distribution of the photovoltaic panel enhanced by the semantic residual mask.

[0034] Another aspect of the present disclosure provides a photovoltaic panel hot spot identification and detection system, which includes:

[0035] An infrared thermal image acquisition module, configured to acquire an infrared thermal image of the photovoltaic panel to be analyzed through a camera installed above the photovoltaic panel array;

[0036] A shallow feature and deep feature extraction module, configured to extract the shallow feature and deep feature of the infrared thermal image respectively, and obtain a corresponding shallow feature map of the thermal infrared distribution of the photovoltaic panel and a deep feature map of the thermal infrared distribution of the photovoltaic panel;

[0037] A feature map fusion module, configured to fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel, and obtain a semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel;

[0038] A photovoltaic panel hot spot identification and detection result determination module, configured to determine the hot spot identification and detection result of the photovoltaic panel to be analyzed based on the semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel.

[0039] Optionally, the shallow feature and deep feature extraction module includes:

[0040] A shallow feature extraction unit, configured to obtain the shallow feature map of the thermal infrared distribution of the photovoltaic panel by passing the infrared thermal image through a temperature distribution shallow feature extractor based on a first convolutional neural network model;

[0041] A deep feature extraction unit, configured to obtain the deep feature map of the thermal infrared distribution of the photovoltaic panel by passing the shallow feature map of the thermal infrared distribution of the photovoltaic panel through a temperature distribution deep feature extractor based on a second convolutional neural network model.

[0042] Optionally, the shallow feature extraction unit is specifically configured to:

[0043] Use each layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model to perform convolution processing, pooling processing, and non-linear activation processing on the infrared thermal image respectively in the forward pass of the layer, and extract the shallow feature map of the thermal infrared distribution of the photovoltaic panel from the shallow layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model.

[0044] Compared with the prior art, the present disclosure uses a camera to acquire an infrared thermal image of the photovoltaic panel to be analyzed, extracts the shallow feature and deep feature of the infrared thermal image, fuses the shallow feature and deep feature to obtain an enhanced shallow feature map, and performs hot spot identification and detection based on the enhanced shallow feature map, effectively realizing the automation of photovoltaic panel hot spot detection and identification, and improving the detection accuracy and robustness. Description of the Drawings

[0045] One or more embodiments are exemplarily illustrated by the pictures in the corresponding drawings. These exemplary illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements, unless otherwise stated. The drawings in the figures do not constitute a scale limitation.

[0046] Figure 1 It is a flowchart of a method for identifying and detecting hot spots on a photovoltaic panel provided by an embodiment of the present disclosure;

[0047] Figure 2 It is a schematic diagram of data flow of a method for identifying and detecting hot spots on a photovoltaic panel provided by another embodiment of the present disclosure;

[0048] Figure 3 It is a schematic structural diagram of a system for identifying and detecting hot spots on a photovoltaic panel provided by another embodiment of the present disclosure;

[0049] Figure 4 It is a schematic diagram of an application scenario of a method for identifying and detecting hot spots on a photovoltaic panel and a system for identifying and detecting hot spots on a photovoltaic panel provided by another embodiment of the present disclosure. Detailed Embodiments

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the following will elaborate on each embodiment of the present disclosure in conjunction with the drawings. However, those of ordinary skill in the art can understand that in each embodiment of the present disclosure, many technical details are provided to help readers better understand the present disclosure. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions required to be protected by the present disclosure can still be implemented. The division of the following embodiments is for convenience of description and should not constitute any limitation on the specific implementation manners of the present disclosure. The various embodiments can be combined and cross-referenced with each other on the premise of not being contradictory.

[0051] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present disclosure have the same meaning as commonly understood by those skilled in the technical field to which the present disclosure belongs. The terms used in the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present disclosure.

[0052] In the records of the embodiments of the present disclosure, it should be noted that unless otherwise stated and defined, the term "connection" should be understood in a broad sense. For example, it can be an electrical connection, or it can be the communication inside two components. It can be directly connected, or it can be indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meaning of the above terms can be understood according to specific situations.

[0053] It should be noted that the terms "first", "second", and "third" involved in the embodiments of the present disclosure are only used to distinguish similar objects, and do not represent a specific order for the objects. It can be understood that "first", "second", and "third" can be interchanged in a specific order or sequence when permitted. It should be understood that the objects distinguished by "first", "second", and "third" can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described here can be implemented in an order other than those illustrated or described here.

[0054] It should also be pointed out that in the methods and systems of the present disclosure, each step or each module can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of the present disclosure.

[0055] To solve the problems existing in the identification and detection of photovoltaic panel hot spots in the prior art, in recent years, artificial intelligence technology based on deep learning has made remarkable progress in the detection and identification of photovoltaic panel hot spots. The artificial intelligence technology based on deep learning provides an effective solution for the automated detection and identification of photovoltaic panel hot spots. By making full use of large-scale training data and powerful feature extraction capabilities, the deep learning model can overcome the limitations of traditional detection methods and achieve accurate and reliable hot spot analysis. Based on this, the present disclosure provides a photovoltaic panel hot spot identification and detection system and its method. Its technical concept is to use artificial intelligence technology based on deep learning to process and analyze the infrared thermal images of photovoltaic panels, so as to achieve accurate detection and classification of photovoltaic panel hot spots. Specifically, it collects the infrared thermal images of the photovoltaic panel to be analyzed through a camera installed above the photovoltaic panel array, extracts the shallow features and deep features of the infrared thermal images respectively, obtains the shallow feature map of the photovoltaic panel thermal infrared distribution and the deep feature map of the photovoltaic panel thermal infrared distribution, fuses the shallow feature map of the photovoltaic panel thermal infrared distribution and the deep feature map of the photovoltaic panel thermal infrared distribution, obtains the semantic residual mask to enhance the shallow feature map of the photovoltaic panel thermal infrared distribution, and determines the hot spot identification and detection result of the photovoltaic panel to be analyzed based on the semantic residual mask enhanced shallow feature map of the photovoltaic panel thermal infrared distribution. Compared with the traditional photovoltaic panel infrared thermal image processing method in the prior art, the photovoltaic panel hot spot identification and detection method based on deep learning provided by the present disclosure can more accurately identify the hot spot area, reduce the false positive rate, realize the automation of hot spot identification and detection, reduce the need for manual intervention, make the hot spot analysis process more efficient and faster, improve the detection efficiency, accuracy and robustness, and can be applied to large-scale photovoltaic panel arrays.

[0056] One embodiment of the present disclosure provides a method for identifying and detecting photovoltaic panel hot spots, and its process is as Figure 1 shown, and its data flow is as Figure 2 shown.

[0057] As Figure 1 shown, and in combination with Figure 2, the method for identifying and detecting hot spots on a photovoltaic panel provided by this embodiment includes the following steps:

[0058] Step S110, collect the infrared thermal image of the photovoltaic panel to be analyzed through a camera installed above the photovoltaic panel array.

[0059] Specifically, the position and angle of the camera need to be able to comprehensively capture the infrared thermal image of the photovoltaic panel to be analyzed, ensuring the adequacy of the image quality and coverage. By using a camera installed above the photovoltaic panel array to collect the infrared thermal image of the photovoltaic panel to be analyzed, the monitoring of the entire photovoltaic panel array can be realized, thereby providing more comprehensive photovoltaic panel hot spot data.

[0060] Step S120, extract the shallow features and deep features of the infrared thermal image respectively to obtain the corresponding shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel.

[0061] Specifically, in step S120, appropriate feature extraction algorithms can be used to extract the shallow features and deep features of the infrared thermal image respectively. Among them, the shallow features of the infrared thermal image can characterize information such as the size, range, and position of the thermal distribution to preliminarily judge the thermal distribution characteristics of the photovoltaic panel. The deep features of the infrared thermal image can characterize the change trend and overall distribution texture of the thermal distribution, capturing more advanced semantic information in the infrared thermal image, such as the existence of hot spots and the position and shape of the hot spots.

[0062] Step S130, fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain a semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel.

[0063] Specifically, in step S130, appropriate fusion algorithms can be used to fuse the shallow features and deep features of the infrared thermal image by using residual connections or attention mechanisms to enhance the semantic information of the shallow features. By fusing the shallow features and deep features of the infrared thermal image, different levels of information can be comprehensively utilized to improve the representation ability of the thermal infrared distribution of the photovoltaic panel and enhance the visibility and recognition accuracy of hot spots.

[0064] Step S140, based on the semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel, determine the hot spot identification and detection result of the photovoltaic panel to be analyzed.

[0065] Specifically, in step S140, an appropriate classification or detection algorithm, such as a Support Vector Machine (SVM) or an object detection algorithm, can be used to identify and detect hot spots based on the enhanced shallow feature map, i.e., the semantic residual mask enhanced shallow feature map of the thermal infrared distribution of the photovoltaic panel. By identifying hot spots based on the shallow feature map enhanced by the semantic residual mask, the accuracy and robustness of hot spot detection can be effectively improved, and the automation of photovoltaic panel hot spot detection and identification can be realized.

[0066] The method for identifying and detecting hot spots of a photovoltaic panel provided by the embodiments of the present disclosure, compared with the prior art, uses a camera to collect the infrared thermal image of the photovoltaic panel to be analyzed, extracts the shallow features and deep features of the infrared thermal image, fuses the shallow features and deep features to obtain an enhanced shallow feature map, and performs hot spot identification and detection based on the enhanced shallow feature map, effectively realizing the automation of photovoltaic panel hot spot detection and identification, and improving the detection accuracy and robustness.

[0067] Exemplarily, in step S120, a convolutional neural network can also be used to extract the shallow features and deep features of the infrared thermal image respectively. At this time, step S120 includes: passing the infrared thermal image through a shallow feature extractor of the temperature distribution based on the first convolutional neural network model to obtain a shallow feature map of the thermal infrared distribution of the photovoltaic panel; passing the shallow feature map of the thermal infrared distribution of the photovoltaic panel through a deep feature extractor of the temperature distribution based on the second convolutional neural network model to obtain a deep feature map of the thermal infrared distribution of the photovoltaic panel.

[0068] Specifically, in step S120, passing the infrared thermal image through a shallow feature extractor of the temperature distribution based on the first convolutional neural network model to obtain a shallow feature map of the thermal infrared distribution of the photovoltaic panel includes: using each layer of the shallow feature extractor of the temperature distribution based on the first convolutional neural network model to perform convolution processing, pooling processing, and non-linear activation processing on the infrared thermal image respectively in the forward pass of the layer, and extracting the shallow feature map of the thermal infrared distribution of the photovoltaic panel from the shallow layer of the shallow feature extractor of the temperature distribution based on the first convolutional neural network model.

[0069] Among them, by using a convolutional neural network model, multiple levels of features can be extracted from infrared thermal images. The shallow feature extractor can capture the low-level visual features of the image, such as edges and textures, while the deep feature extractor can learn more abstract and semantic features, such as shapes and structures. Such multi-level feature representations help to more comprehensively describe the thermal infrared distribution of photovoltaic panels. The deep learning-based feature extractor can extract rich information from infrared thermal images. The shallow features can capture the basic features of the thermal infrared distribution of photovoltaic panels, while the deep features can capture more complex patterns and structures. By fusing shallow and deep features, different levels of information can be comprehensively utilized to improve the representation ability of hot spots. The deep learning-based feature extractor has good robustness and can adapt to the changes in infrared thermal images of different photovoltaic panels, which means that even under different lighting conditions, different weather conditions, or different photovoltaic panel materials, the feature extractor can still extract effective features, thereby achieving stable recognition and detection of hot spots.

[0070] Therefore, the shallow feature extractor of the temperature distribution based on the first convolutional neural network model and the deep feature extractor of the temperature distribution based on the second convolutional neural network model can extract rich and multi-level feature representations from infrared thermal images, thereby enhancing the understanding and analysis ability of the thermal infrared distribution of photovoltaic panels, and further achieving accurate recognition and detection of hot spots.

[0071] Exemplarily, step S130 includes: using a residual information enhancement fusion module to fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain a semantic residual mask to enhance the shallow feature map of the thermal infrared distribution of the photovoltaic panel.

[0072] Specifically, the residual information enhancement fusion module can enhance the information between different layers, that is, fuse the shallow features and deep features in the infrared thermal image, so as to retain the target feature distribution as much as possible.

[0073] Exemplarily, in step S130, using a residual information enhancement fusion module to fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain a semantic residual mask to enhance the shallow feature map of the thermal infrared distribution of the photovoltaic panel includes: performing upsampling and convolutional processing on the deep feature map of the thermal infrared distribution of the photovoltaic panel to obtain a reconstructed deep feature map of the thermal infrared distribution of the photovoltaic panel; calculating the position-wise difference between the reconstructed deep feature map of the thermal infrared distribution of the photovoltaic panel and the shallow feature map of the thermal infrared distribution of the photovoltaic panel to obtain a difference feature map; performing non-linear activation processing on the difference feature map based on the Sigmoid function to obtain a mask feature map; performing element-wise multiplication on the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the mask feature map to obtain a fused feature map; performing an attention-based PMA pooling operation on the fused feature map to obtain a semantic mask to enhance the shallow feature map of the thermal infrared distribution of the photovoltaic panel.

[0074] Specifically, the residual information enhancement fusion module can utilize the residual information between deep features and shallow features to enhance semantic information. By fusing the residual information into the shallow feature map, the semantic expression ability of the feature map can be improved, enabling it to more accurately reflect the important features in the thermal infrared distribution of the photovoltaic panel. Fusing the residual information can enhance the key features in the shallow feature map, thereby improving the discriminability of the features, which helps to better distinguish hot spots and background regions and improve the detection accuracy of hot spots. Traditional feature fusion methods may lead to information loss or blurring, while in this embodiment, the residual information enhancement fusion module is used to fuse the deep features and shallow features of the infrared thermal image, which can retain the information of the original features to the greatest extent, reduce information loss, and can better maintain the details and locality of the features. The residual information enhancement fusion module can provide an adaptive feature fusion mechanism, making the model more robust to changes in input data. Whether it is changes in lighting conditions, differences in photovoltaic panel materials, or other factor changes, the model can better adapt and maintain good performance.

[0075] By using the residual information enhancement fusion module to fuse the shallow feature map of the thermal infrared distribution of the photovoltaic panel and the deep feature map of the thermal infrared distribution of the photovoltaic panel, the semantic information, discriminability, and robustness of the features can be enhanced, thereby improving the recognition and detection performance of hot spots on the photovoltaic panel.

[0076] Exemplarily, step S140 includes: performing dimensionality reduction processing on the shallow feature map of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement to obtain a shallow feature matrix of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement; performing feature distribution correction on the shallow feature matrix of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement to obtain a corrected shallow feature matrix of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement; using the Softmax classification function to perform class prediction on each pixel in the corrected shallow feature matrix of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement to obtain the hot spot recognition and detection result.

[0077] Specifically, by performing class prediction on each pixel and using the probability value of each pixel belonging to a hot spot as the hot spot recognition and detection result of the photovoltaic panel to be analyzed, accurate and detailed hot spot recognition and detection can be effectively achieved.

[0078] Exemplarily, in step S140, performing dimensionality reduction processing on the shallow feature map of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement to obtain a shallow feature matrix of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement includes: performing pooling processing on the shallow feature map of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement along the channel dimension to obtain a shallow feature matrix of the thermal infrared distribution of the photovoltaic panel with semantic residual mask reinforcement.

[0079] Specifically, by performing dimensionality reduction on the shallow feature map of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask, the dimension of the features can be reduced, thereby reducing the computational complexity of subsequent processing, which is very beneficial for real-time applications or large-scale data processing.

[0080] Through feature distribution correction, the feature distribution of the shallow feature matrix of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask can be further optimized, which helps to highlight the features of hot spots, reduce the influence of background noise, and improve the expression ability and discrimination of features. By using the Softmax classification function to predict the category of each pixel in the corrected feature matrix, the hot spots and background areas of the photovoltaic panel can be effectively distinguished, so as to effectively realize the accurate identification and detection of the hot spots of the photovoltaic panel and improve the accuracy and reliability of classification.

[0081] In summary, in the method for identifying and detecting hot spots of photovoltaic panels provided in the above embodiments of the present disclosure, the shallow feature map of the photovoltaic panel thermal infrared distribution and the deep feature map of the photovoltaic panel thermal infrared distribution respectively express the shallow image semantic features and deep image semantic features under different depth convolutional encodings of the infrared thermal image of the photovoltaic panel to be analyzed. And, considering that the deep feature map of the photovoltaic panel thermal infrared distribution is obtained by continuously extracting the local association features of the image semantic based on the deep image semantic local association scale on the basis of the shallow feature map of the photovoltaic panel thermal infrared distribution, therefore, after using the residual information enhancement fusion module to fuse the shallow feature map of the photovoltaic panel thermal infrared distribution and the channel-saliency deep feature map of the photovoltaic panel thermal infrared distribution, that is, the deep feature map of the photovoltaic panel thermal infrared distribution, the obtained shallow feature map of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask not only contains the shallow image semantic features and deep image semantic features at different scales, but also includes the inter-layer residual image semantic features based on the residual information enhancement fusion, so that the shallow feature map of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask has a multi-scale and multi-depth image semantic association feature distribution in the multi-dimensional semantic space.

[0082] Therefore, since the shallow feature map of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask has the properties of multi-dimensional, multi-scale and multi-depth image semantic association feature distribution from the perspective of the semantic space as a whole, when using the Softmax classification function to perform classification regression on the global pooling feature matrix of the shallow feature map of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask, that is, the shallow feature matrix of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask, it is necessary to improve the efficiency of classification regression.

[0083] Therefore, when using the Softmax classification function to perform classification and regression on the shallow feature matrix of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel, the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel after unfolding the shallow feature matrix of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel is optimized position by position, which is specifically expressed as: using the following optimization formula to optimize the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel after unfolding the shallow feature matrix of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel, and obtaining the corrected shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel after unfolding the corrected shallow feature matrix of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel. The optimization formula is as follows:

[0084]

[0085] where, v i is the eigenvalue at the i-th position of the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel after unfolding the shallow feature matrix of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel, is the global mean of all eigenvalues of the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel, v max is the maximum eigenvalue of the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel, v′ i is the eigenvalue at the i-th position of the corrected shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel after unfolding the corrected shallow feature matrix of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel, and exp[·] represents calculating the value of the natural exponential function with the numerical value as the power.

[0086] That is to say, through the concept of the regularization functor of the global distribution parameters, the above optimization is based on the parametric vector representation of the global distribution of the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel, and uses the regular expression of the regression probability to simulate the cost function, so as to model the point-by-point regression characteristics of the classification function for the feature manifold representation of the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel in the high-dimensional feature space, and capture the parametric smoothing optimization trajectory of the shallow feature vector of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel to be decoded in the scenario geometry of the high-dimensional feature manifold through the parameter space of the classification function, so as to improve the training efficiency of the shallow feature matrix of the semantic residual mask enhanced thermal infrared distribution of the photovoltaic panel under the class probability regression of the Softmax classification function.

[0087] Another embodiment of the present disclosure relates to a photovoltaic panel hot spot identification and detection system 200, as Figure 3As shown in the figure, it includes an infrared thermal image acquisition module 210, a shallow feature and deep feature extraction module 220, a feature map fusion module 230, and a photovoltaic panel hot spot recognition detection result determination module 240.

[0088] The infrared thermal image acquisition module 210 is used to acquire the infrared thermal image of the photovoltaic panel to be analyzed through a camera installed above the photovoltaic panel array.

[0089] The shallow feature and deep feature extraction module 220 is used to extract the shallow features and deep features of the infrared thermal image respectively, and obtain the corresponding shallow feature map of the photovoltaic panel thermal infrared distribution and the deep feature map of the photovoltaic panel thermal infrared distribution.

[0090] The feature map fusion module 230 is used to fuse the shallow feature map of the photovoltaic panel thermal infrared distribution and the deep feature map of the photovoltaic panel thermal infrared distribution, and obtain the shallow feature map of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask.

[0091] The photovoltaic panel hot spot recognition detection result determination module 240 is used to determine the hot spot recognition detection result of the photovoltaic panel to be analyzed based on the shallow feature map of the photovoltaic panel thermal infrared distribution enhanced by the semantic residual mask.

[0092] Exemplarily, the shallow feature and deep feature extraction module 220 includes a shallow feature extraction unit and a deep feature extraction unit.

[0093] The shallow feature extraction unit is used to pass the infrared thermal image through a temperature distribution shallow feature extractor based on the first convolutional neural network model to obtain the shallow feature map of the photovoltaic panel thermal infrared distribution.

[0094] The deep feature extraction unit is used to pass the shallow feature map of the photovoltaic panel thermal infrared distribution through a temperature distribution deep feature extractor based on the second convolutional neural network model to obtain the deep feature map of the photovoltaic panel thermal infrared distribution.

[0095] Exemplarily, the shallow feature extraction unit is specifically used for: using each layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model to perform convolution processing, pooling processing, and non-linear activation processing on the infrared thermal image respectively in the forward pass of the layer, and extracting the shallow feature map of the photovoltaic panel thermal infrared distribution from the shallow layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model.

[0096] For the specific implementation method of the photovoltaic panel hot spot recognition detection system provided by the embodiments of the present disclosure, reference can be made to the photovoltaic panel hot spot recognition detection method provided by the embodiments of the present disclosure, which will not be elaborated here.

[0097] The photovoltaic panel hot spot identification and detection system provided by the embodiments of the present disclosure, compared with the prior art, uses a camera to collect the infrared thermal image of the photovoltaic panel to be analyzed, extracts the shallow features and deep features of the infrared thermal image, fuses the shallow features and deep features to obtain an enhanced shallow feature map, and performs hot spot identification and detection based on the enhanced shallow feature map, effectively realizing the automation of photovoltaic panel hot spot detection and identification, and improving the detection accuracy and robustness.

[0098] As described above, the photovoltaic panel hot spot identification and detection system 200 provided by the embodiments of the present disclosure can be implemented in various terminal devices, such as a server for photovoltaic panel hot spot identification and detection. In one example, the photovoltaic panel hot spot identification and detection system 200 provided by the embodiments of the present disclosure can be integrated into the terminal device as a software module and / or a hardware module. For example, the photovoltaic panel hot spot identification and detection system 200 can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the photovoltaic panel hot spot identification and detection system 200 can also be one of the many hardware modules of the terminal device.

[0099] Alternatively, in another example, the photovoltaic panel hot spot identification and detection system 200 and the terminal device can also be separate devices, and the photovoltaic panel hot spot identification and detection system 200 can be connected to the terminal device through a wired and / or wireless network and transmit and interact information according to a predefined data format.

[0100] Figure 4 Shows the application scenarios of the photovoltaic panel hot spot identification and detection method and the photovoltaic panel hot spot identification and detection system provided by the embodiments of the present disclosure. As Figure 4 shown, in this application scenario, first, use step S110 or the infrared thermal image acquisition module 210 to collect the infrared thermal image of the photovoltaic panel to be analyzed through a camera installed above the photovoltaic panel array, and the infrared thermal image can be represented as Figure 4 C as shown in; then, input the collected infrared thermal image into a server deployed with a photovoltaic panel hot spot identification and detection algorithm, where the server can be represented as Figure 4 S as shown in, and the server can process the infrared thermal image based on the photovoltaic panel hot spot identification and detection algorithm corresponding to steps S120 to S140 to determine the hot spot identification and detection result of the photovoltaic panel to be analyzed, or the server can also process the infrared thermal image based on the photovoltaic panel hot spot identification and detection algorithm corresponding to the shallow feature and deep feature extraction module 220, the feature map fusion module 230, and the photovoltaic panel hot spot identification and detection result determination module 240 to determine the hot spot identification and detection result of the photovoltaic panel to be analyzed.

[0101] Those of ordinary skill in the art will understand that the above-described embodiments are specific implementations of the present disclosure, and in actual applications, various changes may be made to them in form and detail without departing from the spirit and scope of the present disclosure.

Claims

1. A method for identifying and detecting hot spots in photovoltaic panels, characterized in that: The method comprises: The infrared thermal image of the photovoltaic panel to be analyzed is collected by a camera installed above the photovoltaic panel array; Extracting shallow features and deep features of the infrared thermal image respectively to obtain a corresponding shallow feature map of thermal infrared distribution of photovoltaic panels and a deep feature map of thermal infrared distribution of photovoltaic panels; The shallow feature map of thermal infrared distribution of photovoltaic panels and the deep feature map of thermal infrared distribution of photovoltaic panels are integrated to obtain the shallow feature map of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask; The shallow feature map of thermal infrared distribution of photovoltaic panels is enhanced based on the semantic residual mask to determine the hot spot recognition detection result of the analyzed photovoltaic panel.

2. The method according to claim 1, characterized in that The shallow features and deep features of the infrared thermal image are extracted respectively to obtain the corresponding shallow feature map of thermal infrared distribution of photovoltaic panels and the deep feature map of thermal infrared distribution of photovoltaic panels, including: The infrared thermal image is passed through a temperature distribution shallow feature extractor based on a first convolutional neural network model to obtain a thermal infrared distribution shallow feature map of the photovoltaic panel; The shallow feature map of thermal infrared distribution of the photovoltaic panel is passed through a deep feature extractor of temperature distribution based on a second convolutional neural network model to obtain a deep feature map of thermal infrared distribution of the photovoltaic panel.

3. The method according to claim 2, characterized in that The infrared thermal image is passed through a temperature distribution shallow feature extractor based on a first convolutional neural network model to obtain a shallow feature map of thermal infrared distribution of the photovoltaic panel, including: Use each layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model to perform convolution processing, pooling processing and nonlinear activation processing on the infrared thermal image in the forward transfer of the layer, and extract the shallow feature map of the thermal infrared distribution of the photovoltaic panel from the shallow layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model.

4. The method according to claim 1, characterized in that: The shallow feature map of thermal infrared distribution of photovoltaic panels and the deep feature map of thermal infrared distribution of photovoltaic panels are integrated to obtain the shallow feature map of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask, including: The residual information enhancement fusion module is used to fuse the shallow feature map of thermal infrared distribution of photovoltaic panels and the deep feature map of thermal infrared distribution of photovoltaic panels to obtain the semantic residual mask enhanced shallow feature map of thermal infrared distribution of photovoltaic panels.

5. The method according to claim 4, characterized in that The residual information enhancement fusion module is used to fuse the shallow feature map of thermal infrared distribution of photovoltaic panels and the deep feature map of thermal infrared distribution of photovoltaic panels to obtain the shallow feature map of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask, including: Upsampling and convolution processing are performed on the deep feature map of thermal infrared distribution of the photovoltaic panel to obtain a reconstructed deep feature map of thermal infrared distribution of the photovoltaic panel; Calculating the position difference between the reconstructed photovoltaic panel thermal infrared distribution deep feature map and the photovoltaic panel thermal infrared distribution shallow feature map to obtain a difference feature map; The difference feature map is subjected to nonlinear activation processing based on a Sigmoid function to obtain a mask feature map; Performing a dot multiplication on the shallow feature map of thermal infrared distribution of the photovoltaic panel and the mask feature map to obtain a fused feature map; An attention-based PMA pooling operation is performed on the fused feature map to obtain the semantic mask-enhanced shallow feature map of thermal infrared distribution of photovoltaic panels.

6. The method according to any one of claims 1 to 5, characterized in that: Based on the semantic residual mask, the shallow feature map of thermal infrared distribution of the photovoltaic panel is enhanced to determine the hot spot recognition detection result of the analyzed photovoltaic panel, including: Performing dimensionality reduction processing on the shallow feature map of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask, so as to obtain a shallow feature matrix of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask; Performing feature distribution correction on the shallow feature matrix of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask, so as to obtain a corrected shallow feature matrix of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask; The Softmax classification function is used to perform category prediction on each pixel in the shallow feature matrix of the thermal infrared distribution of the modified semantic residual mask enhanced photovoltaic panel to obtain the hot spot recognition detection result.

7. The method according to claim 6, characterized in that The semantic residual mask enhanced photovoltaic panel thermal infrared distribution shallow feature map is subjected to dimensionality reduction processing to obtain the semantic residual mask enhanced photovoltaic panel thermal infrared distribution shallow feature matrix, including: The shallow feature map of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask is pooled along the channel dimension to obtain the shallow feature matrix of thermal infrared distribution of photovoltaic panels enhanced by semantic residual mask.

8. A photovoltaic panel hot spot recognition and detection system, characterized in that: The system comprises: An infrared thermal image acquisition module, used to acquire infrared thermal images of the photovoltaic panels to be analyzed through a camera installed above the photovoltaic panel array; A shallow feature extraction module and a deep feature extraction module are used to extract shallow features and deep features of the infrared thermal image, respectively, to obtain a corresponding shallow feature map of thermal infrared distribution of photovoltaic panels and a deep feature map of thermal infrared distribution of photovoltaic panels; A feature map fusion module is used to fuse the shallow feature map of thermal infrared distribution of photovoltaic panels and the deep feature map of thermal infrared distribution of photovoltaic panels to obtain a semantic residual mask-enhanced shallow feature map of thermal infrared distribution of photovoltaic panels; The photovoltaic panel hot spot recognition detection result determination module is used to enhance the shallow feature map of the thermal infrared distribution of the photovoltaic panel based on the semantic residual mask to determine the hot spot recognition detection result of the analyzed photovoltaic panel.

9. The system according to claim 8, characterized in that The shallow feature and deep feature extraction module includes: A shallow feature extraction unit, used for obtaining a shallow feature map of thermal infrared distribution of the photovoltaic panel by passing the infrared thermal image through a temperature distribution shallow feature extractor based on a first convolutional neural network model; The deep feature extraction unit is used to obtain the deep feature map of thermal infrared distribution of the photovoltaic panel through the shallow feature map of thermal infrared distribution of the photovoltaic panel through the deep feature extractor of temperature distribution based on the second convolutional neural network model.

10. The system according to claim 8, characterized in that The shallow feature extraction unit is specifically used for: Use each layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model to perform convolution processing, pooling processing and nonlinear activation processing on the infrared thermal image in the forward transfer of the layer, and extract the shallow feature map of the thermal infrared distribution of the photovoltaic panel from the shallow layer of the temperature distribution shallow feature extractor based on the first convolutional neural network model.

Images