Method for detecting hot spot fault of photovoltaic module
By improving the YOLOv8 model, using deformable convolution and depth separable convolution technology, combining infrared and visible light images, the problem of low thermal spot detection accuracy of photovoltaic modules in the existing technology is solved, achieving higher detection accuracy and Computational efficiency.
Patent Information
- Application Number
- CN202411317725.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-20
- Publication Date
- 2025-01-03
AI Technical Summary
When using complex images, existing photovoltaic module heat spot detection technology is difficult to accurately identify complex and varied hot spot morphology, resulting in low detection accuracy due to the inflexibility of fixed receptive fields and feature fusion paths.
Improved YOLOv8 model by replacing the C2f structure block with C2f_DCNV4 structure block and the FPNet feature fusion module as the AFPN-Faster feature fusion module, combining deformable convolution and deep separable convolution, adapting to the irregularity of photovoltaic hot spots, and through infrared and Multimodal fusion of visible light images improves detection accuracy.
It improves the accuracy and recognition capability of hot spot detection of photovoltaic modules, reduces computational complexity and memory usage, and is suitable for a variety of complex image processing tasks.
Smart Images
Figure CN119251174A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic component detection, and in particular to a method for detecting hot spot failure of a photovoltaic component. Background Art
[0002] Under certain conditions, a shaded PV module in a series branch will be treated as a load and consume the energy generated by other illuminated PV modules. The shaded PV module will generate heat at this time, which is called the hot spot effect. The hot spot effect may seriously damage solar cells. Therefore, drones are needed to conduct real-time inspection of PV modules.
[0003] In the prior art, in the images of photovoltaic modules taken by drones, hot spots may show different forms due to different shooting angles or lighting conditions. In traditional convolution operations, the fixed receptive field may cause some key hot spot features to be ignored. The main defects of ordinary convolution when extracting features when dealing with complex photovoltaic module hot spot identification tasks are its fixed receptive field, lack of adaptability to geometric deformation, and limitations in dealing with hot spots of different shapes and positions. Secondly, the feature fusion network PANet used in the existing YOLOv8 adopts a relatively fixed path and structure in the feature fusion process. Although this can effectively fuse features at different levels, it may not be flexible enough when facing complex and changing inputs. Compared with the adaptive feature fusion mechanism of AFPN, PANet may not be able to achieve the best feature selection and fusion when dealing with diverse hot spot forms. Summary of the invention
[0004] In view of this, an embodiment of the present invention provides a method for detecting hot spot failures of photovoltaic modules, the main purpose of which is to provide a method for detecting hot spot failures of photovoltaic modules that can improve the accuracy of detection and identification.
[0005] In order to achieve the above object, the present invention mainly provides the following technical solutions:
[0006] An embodiment of the present invention provides a method for detecting a hot spot failure of a photovoltaic module, the method comprising the following steps:
[0007] Acquire the image to be detected;
[0008] The existing YOLOv8 model is improved to obtain an improved YOLOv8 model, wherein the improvements include:
[0009] Replace the C2f structure block in the existing YOLOv8 model with the C2f_DCNV4 structure block to extract the shape and structure information of the target, and replace the FPNet feature fusion module in the existing YOLOv8 model with the AFPN-Faster feature fusion module;
[0010] Input the image to be detected into the improved YOLOv8 model for detection to determine the hot spot area of the photovoltaic module;
[0011] Output hot spot information of PV modules.
[0012] Further, infrared light image information and visible light image information of the same position of the photovoltaic module are obtained;
[0013] The infrared light image information and the visible light image information of the same position of the photovoltaic module are divided into regions and encoded.
[0014] Furthermore, the Bottleneck block in the C2f block is replaced with the DCNV4 block;
[0015] When the feature map is input to the DCNV4 block, it first passes through a convolution layer with a convolution kernel of 1×1 to expand the number of channels to n times the original number of channels, and then passes through two branches, one branch is not processed, and the other branch passes through a ReLU activation function. The results of the two branches are multiplied, and then pass through a convolution layer with a convolution kernel of 1×1 to reduce the number of channels to n times the original number of channels, and then the result is output.
[0016] Furthermore, the feature maps of photovoltaic infrared images extracted by Backbone are aligned to a uniform scale through upsampling and downsampling operations;
[0017] Use FasterNet blocks for depthwise separable convolution and partial convolution;
[0018] The FasterNet block includes multiple groups of convolution modules, separable convolution structures, skip connection enhancement modules and global pooling modules. The input feature map is separated by channels through multiple groups of convolution modules. Each group of channels performs independent convolution calculations. The separable convolution structure decomposes conventional convolution into channel-by-channel convolution and point-by-point convolution. The skip connection enhancement module establishes connections between convolution layers. The global pooling module aggregates the feature map in the spatial dimension.
[0019] Furthermore, the upsampling is a CAFARE block, wherein the upsampling is a CAFARE block, wherein the CAFARE block includes a Softmax block and a kernel generator, the Softmax block generates an adaptive convolution kernel of size 5×5, the kernel generator dynamically generates upsampling weights according to context information of the input feature map, and the adaptive convolution kernel reorganizes the input feature map.
[0020] Furthermore, the hot spot information includes at least one of the following: location information of the photovoltaic module, hot spot area of the photovoltaic module, and fault type of the photovoltaic module.
[0021] Compared with the prior art, the present invention has the following technical effects:
[0022] First, by replacing the C2f module in yolov8 with C2f_DCNV4, since deformable convolution can dynamically adjust the receptive field, the convolution kernel can more effectively capture the shape and structural information of the target, thereby improving the network's feature extraction ability for complex images. The receptive field is expanded through deformable convolution to adapt to the irregularity of photovoltaic hot spots and improve the accuracy.
[0023] Second, by replacing the FPNet feature fusion module with the AFPN-Faster feature fusion module, the number of parameters and the amount of calculation are significantly reduced while ensuring high performance. This allows our model to significantly reduce computational complexity and memory usage while maintaining high accuracy, making it suitable for a variety of complex image processing tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 A flow chart of a method for detecting hot spot failure of a photovoltaic module provided by an embodiment of the present invention;
[0025] Figure 2 A flow chart of another method for detecting hot spot failure of a photovoltaic module provided by an embodiment of the present invention;
[0026] Figure 3 This is an experimental effect diagram using the existing YOLOV8 (Baseline) model;
[0027] Figure 4 An experimental effect diagram of a method for detecting hot spot failure of a photovoltaic module provided by an embodiment of the present invention;
[0028] Figure 5 A structural diagram of an improved YOLOv8 model provided by an embodiment of the present invention;
[0029] Figure 6 A structural diagram of a C2f_DCNV4 structural block provided in an embodiment of the present invention;
[0030] Figure 7 A structural diagram of a C2f-Faster structural block provided in an embodiment of the present invention;
[0031] Figure 8 A structural diagram of a FasterNet block provided in an embodiment of the present invention;
[0032] Fig. 9 A structural diagram of an AFPN-Faster structural block provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0033] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.
[0034] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0035] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0036] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.
[0037] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.
[0038] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0039] like Figures 1 to 8 As shown, an embodiment of the present invention provides a method for detecting hot spot failure of a photovoltaic module, the method comprising the following steps:
[0040] Acquire the image to be detected;
[0041] The existing YOLOv8 model is improved to obtain an improved YOLOv8 model, wherein the improvements include:
[0042] Replace the C2f structure block in the existing YOLOv8 model with the C2f_DCNV4 structure block to extract the shape and structure information of the target, and replace the FPNet feature fusion module in the existing YOLOv8 model with the AFPN-Faster feature fusion module;
[0043] Input the image to be detected into the improved YOLOv8 model for detection to determine the hot spot area of the photovoltaic module;
[0044] Output hot spot information of PV modules.
[0045] Compared with the prior art, the present invention has the following technical effects:
[0046] First, by replacing the C2f module in yolov8 with C2f_DCNV4, since deformable convolution can dynamically adjust the receptive field, the convolution kernel can more effectively capture the shape and structural information of the target, thereby improving the network's feature extraction ability for complex images. The receptive field is expanded through deformable convolution to adapt to the irregularity of photovoltaic hot spots and improve the accuracy.
[0047] Second, by replacing the FPNet feature fusion module with the AFPN-Faster feature fusion module, the number of parameters and the amount of calculation are significantly reduced while ensuring high performance. This allows our model to significantly reduce computational complexity and memory usage while maintaining high accuracy, making it suitable for a variety of complex image processing tasks.
[0048] Example 1
[0049] like Figure 1 , Figures 3 to 8 As shown, an embodiment of the present invention provides a method for detecting hot spot failure of a photovoltaic module, the method comprising the following steps:
[0050] 101. Obtain an image to be detected.
[0051] The drone uses a camera to take pictures of photovoltaic modules that may have hot spots and upload the images. The photos taken include infrared images and visible light images. The images are uploaded to the cloud and then transferred to the detection model for detection.
[0052] 102. Improve the existing YOLOv8 model to obtain an improved YOLOv8 model, wherein the improvement includes:
[0053] The C2f structure block in the existing YOLOv8 model is replaced with the C2f_DCNV4 structure block to extract the shape and structure information of the target, and the FPNet feature fusion module in the existing YOLOv8 model is replaced with the AFPN-Faster feature fusion module.
[0054] The C2f structure block is replaced with the C2f_DCNV4 structure block. Since the deformable convolution can dynamically adjust the receptive field, the convolution kernel can more effectively capture the shape and structural information of the target, thereby improving the network's ability to extract features from complex images. This is particularly important in processing non-rigid targets or tasks with complex backgrounds. The receptive field is expanded through deformable convolution to adapt to the irregularity of photovoltaic hot spots and improve the accuracy. Then, the FPNet feature fusion module is replaced with the AFPN-Faster feature fusion module, that is, the FPNet feature fusion module is replaced with the ASFF_2 block and the ASFF_3 block. The number of parameters is reduced, but the convergence speed is accelerated, the recognition accuracy is also increased, the number of point-like spots recognized is improved, and the recognition accuracy of multiple spots on the same photovoltaic panel is also improved.
[0055] 103. Input the image to be detected into the improved YOLOv8 model for detection to determine the hot spot area of the photovoltaic module.
[0056] The improved YOLOv8 model detects the image to be detected, identifies the point-like light spots or multiple light spots, and then determines the hot spot area according to the identified range.
[0057] 104. Output hot spot information of the photovoltaic module.
[0058] The improved YOLOv8 model converts the identified hot spot areas into hot spot information, and then outputs it in the form of pictures.
[0059] Compared with the prior art, the present invention has the following technical effects:
[0060] First, by replacing the C2f module in yolov8 with C2f_DCNV4, since deformable convolution can dynamically adjust the receptive field, the convolution kernel can more effectively capture the shape and structural information of the target, thereby improving the network's feature extraction ability for complex images. The receptive field is expanded through deformable convolution to adapt to the irregularity of photovoltaic hot spots and improve the accuracy.
[0061] Second, by replacing the FPNet feature fusion module with the AFPN-Faster feature fusion module, the number of parameters and the amount of calculation are significantly reduced while ensuring high performance. This allows our model to significantly reduce computational complexity and memory usage while maintaining high accuracy, making it suitable for a variety of complex image processing tasks.
[0062] Example 2
[0063] like Figures 2 to 8 As shown, an embodiment of the present invention provides a method for detecting hot spot failure of a photovoltaic module, the method comprising the following steps:
[0064] 201. Obtain infrared light image information and visible light image information of the same position of a photovoltaic module.
[0065] The same position of the photovoltaic module is photographed by the camera of the drone, and the same position of the photovoltaic module is photographed by an infrared camera and an ordinary camera respectively, and then uploaded to the system.
[0066] 202. Divide the infrared light image information and the visible light image information of the same position of the photovoltaic module into regions and encode them.
[0067] The system divides the captured photos into regions and then encodes the divided regions for easy viewing and storage.
[0068] 203. Replace the Bottleneck block in the C2f block with the DCNV4 block.
[0069] When the feature map is input to the DCNV4 block, it first passes through a convolution layer with a convolution kernel of 1×1 to expand the number of channels to n times the original number of channels, and then passes through two branches, one branch is not processed, and the other branch passes through a ReLU activation function. The results of the two branches are multiplied, and then pass through a convolution layer with a convolution kernel of 1×1 to reduce the number of channels to n times the original number of channels, and then the result is output.
[0070] 204. The feature map of the photovoltaic infrared image extracted by Backbone is aligned to a uniform scale through upsampling and downsampling operations.
[0071] During upsampling, the upsampling is a CAFARE block, wherein the upsampling is a CAFARE block, wherein the CAFARE block includes a Softmax block and a kernel generator, the Softmax block generates an adaptive convolution kernel of size 5×5, the kernel generator dynamically generates upsampling weights according to context information of the input feature map, and the adaptive convolution kernel reorganizes the input feature map to achieve more delicate upsampling, so that the resolution of the output feature map is twice that of the input feature map, and has better detail retention capability.
[0072] 205. Use FasterNet blocks for depthwise separable convolution and partial convolution.
[0073] The FasterNet block includes multiple groups of convolution modules, separable convolution structures, skip connection enhancement modules and global pooling modules. The input feature map is separated by channels through multiple groups of convolution modules. Each group of channels performs independent convolution calculations, thereby reducing the amount of calculation. The separable convolution structure reduces the number of parameters and computational complexity by decomposing conventional convolutions into channel-by-channel convolutions and point-by-point convolutions. The skip connection enhancement module establishes connections between convolution layers to effectively prevent gradient disappearance and enhance feature learning capabilities. The global pooling module aggregates feature maps in the spatial dimension to reduce the computational burden of subsequent fully connected layers.
[0074] 206. Input the image to be detected into the improved YOLOv8 model for detection to determine the hot spot area of the photovoltaic module.
[0075] 207. Output hot spot information of the photovoltaic module.
[0076] The specific experimental data are shown in Table 1:
[0077]
[0078] Table 1
[0079] From Table 1, Figure 3 and Figure 4 It can be seen that the accuracy, recall and average precision of the YOLOV8 (Baseline) model are significantly lower than those of the C2f_DCNv4+AFPN+C2f_Faster model.
[0080] Compared with the prior art, the present invention has the following technical effects:
[0081] First, by replacing the C2f module in yolov8 with C2f_DCNV4, since deformable convolution can dynamically adjust the receptive field, the convolution kernel can more effectively capture the shape and structural information of the target, thereby improving the network's feature extraction ability for complex images. The receptive field is expanded through deformable convolution to adapt to the irregularity of photovoltaic hot spots and improve the accuracy.
[0082] Second, by replacing the FPNet feature fusion module with the AFPN-Faster feature fusion module, the number of parameters and the amount of calculation are significantly reduced while ensuring high performance. This allows our model to significantly reduce computational complexity and memory usage while maintaining high accuracy, making it suitable for a variety of complex image processing tasks.
[0083] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A method for detecting hot spot failure of a photovoltaic module, characterized in that: The following steps are involved: Acquire the image to be detected; The existing YOLOv8 model is improved to obtain an improved YOLOv8 model, wherein the improvements include: Replace the C2f structure block in the existing YOLOv8 model with the C2f_DCNV4 structure block to extract the shape and structure information of the target, and replace the FPNet feature fusion module in the existing YOLOv8 model with the AFPN-Faster feature fusion module; Input the image to be detected into the improved YOLOv8 model for detection to determine the hot spot area of the photovoltaic module; Output hot spot information of PV modules.
2. A method for detecting hot spot failure of a photovoltaic module according to claim 1, characterized in that: The acquiring of the image to be detected comprises: Acquire infrared light image information and visible light image information of the same position of a photovoltaic module; The infrared light image information and the visible light image information of the same position of the photovoltaic module are divided into regions and encoded.
3. A method for detecting hot spot failure of a photovoltaic module according to claim 1, characterized in that: The replacing of the C2f structure block in the existing YOLOv8 model with the C2f_DCNV4 structure block includes: Replace the Bottleneck block in the C2f block with the DCNV4 block; When the feature map is input to the DCNV4 block, it first passes through a convolution layer with a convolution kernel of 1×1 to expand the number of channels to n times the original number of channels, and then passes through two branches, one branch is not processed, and the other branch passes through a ReLU activation function. The results of the two branches are multiplied, and then pass through a convolution layer with a convolution kernel of 1×1 to reduce the number of channels to n times the original number of channels, and then the result is output.
4. A method for detecting hot spot failure of a photovoltaic module according to claim 1, characterized in that: Replacing the FPNet feature fusion module in the existing YOLOv8 model with the AFPN-Faster feature fusion module includes: The feature maps of photovoltaic infrared images extracted by Backbone are aligned to a uniform scale through upsampling and downsampling operations; Use FasterNet blocks for depthwise separable convolution and partial convolution; The FasterNet block includes multiple groups of convolution modules, separable convolution structures, skip connection enhancement modules and global pooling modules. The input feature map is separated by channels through multiple groups of convolution modules. Each group of channels performs independent convolution calculations. The separable convolution structure decomposes conventional convolution into channel-by-channel convolution and point-by-point convolution. The skip connection enhancement module establishes connections between convolution layers. The global pooling module aggregates the feature map in the spatial dimension.
5. A method for detecting hot spot failure of a photovoltaic module according to claim 4, characterized in that: The upsampling is a CAFARE block, wherein the CAFARE block includes a Softmax block and a kernel generator, the Softmax block generates an adaptive convolution kernel of size 5×5, the kernel generator dynamically generates an upsampling weight according to context information of an input feature map, and the adaptive convolution kernel reorganizes the input feature map.
6. A method for detecting hot spot failure of a photovoltaic module according to any one of claims 1 to 5, characterized in that: The hot spot information includes at least one of the following: location information of the photovoltaic module, the hot spot area of the photovoltaic module, and the fault type of the photovoltaic module.