A method and device for detecting hot spot defects in photovoltaic panels used in photovoltaic power stations

Abstract

The present invention discloses a method and device for detecting hot spot defects of photovoltaic panels in photovoltaic power stations. The method comprises acquiring a low-altitude infrared image of the photovoltaic panel; inputting the low-altitude infrared image of the photovoltaic panel into a pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image; reading the color mask in a binarized manner based on OpenCV, and adding the color mask to the low-altitude infrared image of the photovoltaic panel pixel by pixel to obtain a segmented photovoltaic panel infrared image with a pure black background; inputting the segmented photovoltaic panel infrared image into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection and obtain a detection result. The present invention adopts a two-step method of first segmentation and then detection, which effectively eliminates the influence of environmental interference heat sources on the performance of the neural network and effectively improves the detection accuracy.

Description

Technical Field

[0001] The present invention relates to a method and device for detecting hot spot defects of photovoltaic panels used in photovoltaic power stations, belonging to the technical field of photovoltaic panel detection. Background Art

[0002] Reducing carbon emissions has long been a consensus. Photovoltaic power generation, as a clean energy source, significantly reduces carbon emissions. Furthermore, it offers numerous advantages, including easy and low-cost equipment construction and high-quality energy. China's photovoltaic sector started relatively late, but thanks to its vast land area and abundant solar resources in its western region, the industry has developed rapidly. Currently, China's installed photovoltaic capacity ranks first in the world, and continues to grow rapidly.

[0003] The continuous increase in photovoltaic installed capacity has also brought some problems. Most of China's photovoltaic power plants are located in remote areas and are unmanned. If photovoltaic faults go undetected, they can cause fires and other accidents, threatening the safety of the power plant. Traditional methods for detecting photovoltaic panel defects mainly include manual inspection and measuring the electrical parameters of the photovoltaic panels. These two methods are mainly inefficient and costly.

[0004] With the rapid development of image processing technology, deep learning technology, and the civilian drone industry, practitioners began using drones to capture infrared images of photovoltaic panels and apply deep learning image detection methods to detect defects. However, most object detection networks, whether single-stage or two-stage, have performed poorly. This is because infrared images often contain complex backgrounds, including not only photovoltaic panels but also trees, weeds, rivers, animals, and other objects. These heat sources significantly affect the performance of the detection network.

[0005] Therefore, it is of great significance to invent a method for detecting hot spot defects of photovoltaic panels used in photovoltaic power stations, which can eliminate the influence of interfering heat sources and improve the accuracy of defect detection. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and device for detecting hot spot defects in photovoltaic panels of photovoltaic power stations, which can eliminate the interference of background heat sources in the infrared images of photovoltaic panels and greatly improve the accuracy of hot spot defect detection in photovoltaic panels.

[0007] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0008] In a first aspect, the present invention provides a method and apparatus for detecting hot spot defects in photovoltaic panels of a photovoltaic power station, comprising:

[0009] Acquire low-altitude infrared images of photovoltaic panels;

[0010] Input the photovoltaic panel low-altitude infrared image into the pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image;

[0011] Based on OpenCV, the color mask is read in a binary manner, and the image is added pixel by pixel to the low-altitude infrared image of the photovoltaic panel to obtain a segmented infrared image of the photovoltaic panel with a pure black background;

[0012] The segmented photovoltaic panel infrared image is input into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection and obtain a detection result.

[0013] Furthermore, the method for constructing the Deeplabv3+ photovoltaic panel semantic segmentation model and the YOLOv5s photovoltaic panel hot spot defect detection model includes:

[0014] Acquire a low-altitude infrared image of a photovoltaic panel, wherein the low-altitude infrared image includes an image of the photovoltaic panel and an image of a hot spot defect;

[0015] Using the photovoltaic panel image to train a Deeplabv3+ network to obtain a Deeplabv3+ photovoltaic panel semantic segmentation model;

[0016] The YOLOv5s network is trained using hot spot defect images to obtain the YOLOv5s photovoltaic panel hot spot defect detection model.

[0017] Furthermore, the photovoltaic panel image is used to train the Deeplabv3+ network to obtain a Deeplabv3+ photovoltaic panel semantic segmentation model, including:

[0018] Using the Labelme tool to perform pixel-level annotation on the low-altitude infrared image to create a photovoltaic panel infrared image dataset;

[0019] A Deeplabv3+ semantic segmentation network was built based on a deep learning framework, and the network was trained using the photovoltaic panel infrared image dataset to obtain a Deeplabv3+ photovoltaic panel semantic segmentation model.

[0020] Furthermore, the hot spot defect image is used to train the YOLOv5s network to obtain the YOLOv5s photovoltaic panel hot spot defect detection model, including:

[0021] Inputting the low-altitude infrared image into the Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image;

[0022] Based on OpenCV, the color mask is read in a binary manner and added pixel by pixel to the original infrared image to obtain the segmented infrared image of the photovoltaic panel with a pure black background.

[0023] Using the LabelImg tool to label the hot spot defect areas in the segmented photovoltaic panel infrared image to create a photovoltaic panel hot spot defect image dataset;

[0024] A YOLOv5s target detection network was built based on a deep learning framework, and the network was trained using the photovoltaic panel hot spot defect image dataset to obtain a YOLOv5s photovoltaic panel hot spot defect detection model.

[0025] Furthermore, the photovoltaic panel low-altitude infrared image is input into a pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image, including:

[0026] Feed an infrared image of a photovoltaic panel into the Deeplabv3+ photovoltaic panel semantic segmentation model;

[0027] The infrared image of the photovoltaic panel is passed through the feature extraction network to obtain a low-level feature map;

[0028] The low-level feature map is further subjected to multi-scale feature extraction through a dilated spatial convolutional pooling pyramid layer to obtain a high-level feature map;

[0029] The high-level feature map is upsampled by 4 times bilinear interpolation and then fused with the low-level feature map, and then restored to the original image resolution by upsampling by 4 times bilinear interpolation;

[0030] The processed feature map is passed through the Softmax classification function layer to obtain the final corresponding color mask.

[0031] Furthermore, for photovoltaic panels with hot spot defects, the positioning and attitude data carried by the infrared image can be used to calculate the longitude and latitude information of the defective photovoltaic panel to achieve positioning and alarm.

[0032] Furthermore, the step of inputting the segmented photovoltaic panel infrared image into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection includes:

[0033] Feed a segmented photovoltaic panel infrared image into the YOLOv5s photovoltaic panel hot spot defect detection model;

[0034] The photovoltaic panel infrared image is passed through the backbone network to extract image feature information at different image granularity levels;

[0035] The feature maps output from the three different stages of the backbone network are fed into the neck network with a path aggregation network structure for multi-scale feature fusion;

[0036] The three fused feature maps are fed into three prediction head networks for prediction box regression and category regression;

[0037] The non-maximum suppression algorithm is used to screen the prediction frame obtained in the previous step, and the optimal prediction frame is returned to complete the final photovoltaic panel hot spot defect detection.

[0038] In a second aspect, the present invention provides a device for detecting hot spot defects in photovoltaic panels of a photovoltaic power station, comprising:

[0039] An acquisition unit, used for acquiring low-altitude infrared images of photovoltaic panels;

[0040] A color mask acquisition unit is used to input the photovoltaic panel low-altitude infrared image into a pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image;

[0041] A segmentation unit is used to read the color mask in a binary manner based on OpenCV, and add the color mask to the photovoltaic panel low-altitude infrared image pixel by pixel to obtain a segmented photovoltaic panel infrared image with a pure black background;

[0042] The detection unit is used to input the segmented photovoltaic panel infrared image into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection and obtain a detection result.

[0043] In a third aspect, the present invention provides a device for detecting hot spot defects in photovoltaic panels of a photovoltaic power station, comprising a processor and a storage medium;

[0044] The storage medium is used to store instructions;

[0045] The processor is configured to operate according to the instructions to execute the steps of any of the aforementioned methods.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the aforementioned methods when executed by a processor.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention provides a method and device for detecting hot spot defects in photovoltaic panels for photovoltaic power stations. For the detection of hot spot defects in photovoltaic panels, a two-step method of first segmentation and then detection is adopted, which effectively reduces the impact of environmental interference heat sources on the performance of the neural network and effectively improves the detection accuracy.

[0049] 2. The present invention provides a method and device for detecting hot spot defects in photovoltaic panels used in photovoltaic power stations. It uses a trained deep convolutional neural network model to automatically extract image features and automatically detect surface defects in photovoltaic panels, avoiding the complex work of manual feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a method and apparatus for detecting hot spot defects in photovoltaic panels of a photovoltaic power station provided by an embodiment of the present invention;

[0051] Figure 2 Schematic diagram of Deeplabv3+ photovoltaic panel semantic segmentation model;

[0052] Figure 3 This is a schematic diagram of adding the original infrared image of the photovoltaic panel and the corresponding pixels of the mask;

[0053] Figure 4 Schematic diagram of the YOLOv5s photovoltaic panel hot spot defect detection model. DETAILED DESCRIPTION

[0054] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0055] Example 1

[0056] This embodiment introduces a method and apparatus for detecting hot spot defects in photovoltaic panels of a photovoltaic power station, including:

[0057] Acquire low-altitude infrared images of photovoltaic panels;

[0058] Input the photovoltaic panel low-altitude infrared image into the pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image;

[0059] Based on OpenCV, the color mask is read in a binary manner, and the image is added pixel by pixel to the low-altitude infrared image of the photovoltaic panel to obtain a segmented infrared image of the photovoltaic panel with a pure black background;

[0060] The segmented photovoltaic panel infrared image is input into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection and obtain a detection result.

[0061] like Figure 1 As shown, the application process of the method and device for detecting hot spot defects of photovoltaic panels in photovoltaic power stations provided in this embodiment specifically involves the following steps:

[0062] (1) Collect low-altitude infrared images of photovoltaic panels. The collection methods include using drones equipped with infrared cameras to take photos and using web crawlers to obtain images. In the embodiment of this application, 1,306 low-altitude infrared images of photovoltaic panels were collected, of which 771 images contained hot spot defects.

[0063] (2) Use the photovoltaic panel images to train the Deeplabv3+ network and obtain the Deeplabv3+ photovoltaic panel semantic segmentation model. The model network framework structure diagram is as follows: Figure 2 As shown, the specific process includes:

[0064] (21) Use the Labelme tool to perform pixel-level annotation on the infrared images collected in step (1) to create a photovoltaic panel infrared image dataset. Write a Python script to randomly extract 80% of the images from the dataset as a training set and the remaining 20% ​​as a test set.

[0065] (22) A Deeplabv3+ semantic segmentation network was built based on the Pytorch deep learning framework, and the photovoltaic panel infrared image dataset obtained in step (21) was used to train the network to obtain a Deeplabv3+ photovoltaic panel semantic segmentation model. In this embodiment, the server's graphics processor was NVIDIA's GeForce 1080ti, the operating system was Ubuntu 16.04, and the ResNet50 network was selected as the feature extraction network of Deeplabv3+. The SGD algorithm was used to learn the model parameters, with the learning rate set to 0.001, the network momentum parameter set to 0.9, the weight decay set to 0.0001, the batch size set to 16, the first stage of training was set to 4000 batches, and the second stage of training was set to 6000 batches.

[0066] (3) The YOLOv5s network is trained with the hot spot defect image to obtain the YOLOv5s photovoltaic panel hot spot defect detection model. The model network framework structure diagram is shown in the figure. Figure 4 As shown, the specific process includes:

[0067] (31) feeding the photovoltaic panel infrared image collected in step (1) into the Deeplabv3+ photovoltaic panel semantic segmentation model obtained in step (2) to obtain a color mask corresponding to the photovoltaic panel infrared image;

[0068] (32) Based on OpenCV, the color mask is read in a binary manner and added pixel by pixel to the original infrared image to obtain an infrared image of the photovoltaic panel with a pure black background. The schematic diagram of this process is shown in the figure below. Figure 3 As shown;

[0069] (33) Use the LabelImg tool to label the hot spot defect areas in the segmented photovoltaic panel infrared image obtained in step (32) to create a photovoltaic panel hot spot defect image dataset. Write a Python script to randomly extract 80% of the images from the dataset as a training set and the remaining 20% ​​as a test set;

[0070] (34) A YOLOv5s target detection network was built based on the Pytorch deep learning framework. The network was trained using the photovoltaic panel hot spot defect image dataset obtained in step (33) to obtain a YOLOv5s photovoltaic panel hot spot defect detection model. In this embodiment, the CSPDarketNet53 network was selected as the feature extraction network of YOLOv5s. The SGD algorithm was used to learn the model parameters. The learning rate was set to 0.001, the network momentum parameter was set to 0.9, the weight decay was set to 0.0001, the batch size was set to 16, the first stage of training was set to 8000 batches, and the second stage of training was set to 12000 batches.

[0071] (4) UAV captures low-altitude infrared images of photovoltaic panels. The drone used in this example is a DJI Mavic 2 Advanced Edition aircraft equipped with a 640*512 thermal imaging camera. The images were taken between 2:00 and 4:00 p.m. in autumn on a clear day, at an altitude of 28 m.

[0072] (5) Segmenting the photovoltaic panels in the infrared image, specifically including:

[0073] (51) The low-altitude infrared image of the photovoltaic panel collected by the drone in step (4) is fed into the Deeplabv3+ photovoltaic panel semantic segmentation model obtained in step (2) to obtain a color mask corresponding to the infrared image of the photovoltaic panel. The process specifically includes:

[0074] (511) An infrared image of a photovoltaic panel is fed into the Deeplabv3+ photovoltaic panel semantic segmentation model;

[0075] (512) The infrared image of the photovoltaic panel is passed through the feature extraction network to obtain a low-level feature map;

[0076] (513) The low-level feature map obtained in the previous step is further used to extract multi-scale features through the dilated spatial convolutional pooling pyramid layer to obtain a high-level feature map;

[0077] (514) The high-level feature map obtained in the previous step is fused with the low-level feature map in step (512) after being up-sampled by 4 times bilinear interpolation, and then restored to the original image resolution after being up-sampled by 4 times bilinear interpolation;

[0078] (515) The feature map obtained in the previous step is passed through the Softmax classification function layer to obtain the final corresponding color mask;

[0079] (52) Based on OpenCV, the color mask is read in a binary manner and added pixel by pixel to the original infrared image to obtain the segmented infrared image of the photovoltaic panel with a pure black background;

[0080] (6) Detection and location of hot spot defects in photovoltaic panels, including:

[0081] (61) The segmented photovoltaic panel infrared image obtained in step (52) is fed into the YOLOv5s photovoltaic panel hot spot defect detection model obtained in step (3) to perform photovoltaic panel hot spot defect detection. The process specifically includes:

[0082] (611) A segmented photovoltaic panel infrared image is fed into a YOLOv5s photovoltaic panel hot spot defect detection model;

[0083] (612) Photovoltaic panel infrared images are extracted through the backbone network at different image granularity levels;

[0084] (613) The feature maps output from the three different stages of the backbone network are fed into the neck network with a path aggregation network structure for multi-scale feature fusion;

[0085] (614) The three feature maps obtained in the previous step are fed into three prediction head networks for prediction box regression and category regression;

[0086] (615) Using the non-maximum suppression algorithm to filter the prediction frame obtained in the previous step, returning the optimal prediction frame, and completing the final photovoltaic panel hot spot defect detection;

[0087] (62) For photovoltaic panels with hot spot defects, the positioning and attitude data carried by the infrared image are used to calculate the longitude and latitude information of the defective photovoltaic panel to achieve positioning and alarm.

[0088] Example 2

[0089] This embodiment provides a photovoltaic panel hot spot defect detection device for a photovoltaic power station, comprising:

[0090] An acquisition unit, used for acquiring low-altitude infrared images of photovoltaic panels;

[0091] A color mask acquisition unit is used to input the photovoltaic panel low-altitude infrared image into a pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image;

[0092] A segmentation unit is used to read the color mask in a binary manner based on OpenCV, and add the color mask to the photovoltaic panel low-altitude infrared image pixel by pixel to obtain a segmented photovoltaic panel infrared image with a pure black background;

[0093] The detection unit is used to input the segmented photovoltaic panel infrared image into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection and obtain a detection result.

[0094] Example 3

[0095] This embodiment provides a photovoltaic panel hot spot defect detection device for a photovoltaic power station, including a processor and a storage medium;

[0096] The storage medium is used to store instructions;

[0097] The processor is configured to operate according to the instructions to execute the steps of the method according to any one of the first embodiments.

[0098] Example 4

[0099] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of any one of the methods described in Embodiment 1 are implemented.

[0100] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting hot spot defects in photovoltaic panels of photovoltaic power stations, characterized in that: include: Acquire low-altitude infrared images of photovoltaic panels; Input the photovoltaic panel low-altitude infrared image into the pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain the color mask corresponding to the photovoltaic panel infrared image; including: Feed an infrared image of a photovoltaic panel into the Deeplabv3+ photovoltaic panel semantic segmentation model; The infrared image of the photovoltaic panel is passed through the feature extraction network to obtain a low-level feature map; The low-level feature map is further subjected to multi-scale feature extraction through a dilated spatial convolutional pooling pyramid layer to obtain a high-level feature map; The high-level feature map is upsampled by 4 times bilinear interpolation and then fused with the low-level feature map, and then restored to the original image resolution by upsampling by 4 times bilinear interpolation; Passing the processed feature map through the Softmax classification function layer to obtain the final corresponding color mask; Based on OpenCV, the color mask is read in a binary manner, and the image is added pixel by pixel to the low-altitude infrared image of the photovoltaic panel to obtain a segmented infrared image of the photovoltaic panel with a pure black background; Input the segmented photovoltaic panel infrared image into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection and obtain detection results; for photovoltaic panels with hot spot defects, use the positioning and attitude data carried by the infrared image to calculate the latitude and longitude information of the defective photovoltaic panel to achieve positioning and alarm; The method for constructing the Deeplabv3+ photovoltaic panel semantic segmentation model and the YOLOv5s photovoltaic panel hot spot defect detection model includes: Acquire a low-altitude infrared image of a photovoltaic panel, wherein the low-altitude infrared image includes an image of the photovoltaic panel and an image of a hot spot defect; Using the photovoltaic panel image to train a Deeplabv3+ network to obtain a Deeplabv3+ photovoltaic panel semantic segmentation model; The YOLOv5s network is trained using hot spot defect images to obtain the YOLOv5s photovoltaic panel hot spot defect detection model; including: Inputting the low-altitude infrared image into the Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image; Based on OpenCV, the color mask is read in a binary manner and added pixel by pixel to the original infrared image to obtain the segmented infrared image of the photovoltaic panel with a pure black background. Using the LabelImg tool to label the hot spot defect areas in the segmented photovoltaic panel infrared image to create a photovoltaic panel hot spot defect image dataset; A YOLOv5s target detection network was built based on a deep learning framework, and the network was trained using the photovoltaic panel hot spot defect image dataset to obtain a YOLOv5s photovoltaic panel hot spot defect detection model.

2. The method for detecting hot spot defects of photovoltaic panels used in photovoltaic power plants according to claim 1, characterized in that: The photovoltaic panel image is used to train the Deeplabv3+ network to obtain a Deeplabv3+ photovoltaic panel semantic segmentation model, including: Using the Labelme tool to perform pixel-level annotation on the low-altitude infrared image to create a photovoltaic panel infrared image dataset; A Deeplabv3+ semantic segmentation network was built based on a deep learning framework, and the network was trained using the photovoltaic panel infrared image dataset to obtain a Deeplabv3+ photovoltaic panel semantic segmentation model.

3. The method for detecting hot spot defects of photovoltaic panels used in photovoltaic power stations according to claim 1, characterized in that: Inputting the segmented photovoltaic panel infrared image into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection includes: Feed a segmented photovoltaic panel infrared image into the YOLOv5s photovoltaic panel hot spot defect detection model; The photovoltaic panel infrared image is passed through the backbone network to extract image feature information at different image granularity levels; The feature maps output from the three different stages of the backbone network are fed into the neck network with a path aggregation network structure for multi-scale feature fusion; The three fused feature maps are fed into three prediction head networks for prediction box regression and category regression; The non-maximum suppression algorithm is used to screen the prediction frame obtained in the previous step, and the optimal prediction frame is returned to complete the final photovoltaic panel hot spot defect detection.

4. A photovoltaic panel hot spot defect detection device for a photovoltaic power station, using the photovoltaic panel hot spot defect detection method for a photovoltaic power station according to claim 1, characterized in that: include: An acquisition unit, used for acquiring low-altitude infrared images of photovoltaic panels; A color mask acquisition unit is used to input the photovoltaic panel low-altitude infrared image into a pre-built Deeplabv3+ photovoltaic panel semantic segmentation model to obtain a color mask corresponding to the photovoltaic panel infrared image; A segmentation unit is used to read the color mask in a binary manner based on OpenCV, and add the color mask to the photovoltaic panel low-altitude infrared image pixel by pixel to obtain a segmented photovoltaic panel infrared image with a pure black background; The detection unit is used to input the segmented photovoltaic panel infrared image into a pre-built YOLOv5s photovoltaic panel hot spot defect detection model to perform photovoltaic panel hot spot defect detection and obtain a detection result.

5. A photovoltaic panel hot spot defect detection device for a photovoltaic power station, characterized by: including processors and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images