Research On Effective Region Segmentation Method Of Infrared Image Of Photovoltaic Panel

With the development and application of infrared imaging technology,infrared imaging-based diagnostic technology is widely used in photovoltaic panel fault detection and diagnosis.In this case,there will be a large number of photovoltaic panel infrared images that need to be diagnosed,and the segmentation of the infrared images is an important prerequisite for photovoltaic panel fault detection and diagnosis.However,a large number of infrared images of photovoltaic panels obtained are usually manually segmented by professional technicians.This method is time-consuming and labor-intensive,and has very high requirements for personnel.Besides,infrared images have the characteristics of low contrast and fuzzy edges,all of which will lead to errors in the segmentation results of infrared images.Therefore,in order to improve the segmentation effect of the infrared image of photovoltaic panel,based on the existing image segmentation algorithm,this paper proposes a segmentation method based on fuzzy clustering,a segmentation method based on gray level co-occurrence matrix and region growing method.And Segmentation method based on improved U-Net network.The main research contents of this paper are as follows:1.In view of the low contrast and edge blur of infrared image of photovoltaic panel and the phenomenon of mis-segmentation by fuzzy clustering method,this paper proposes a kind of neighborhood spatial information based on fuzzy C-means method and fuzzy kernel C-means method.The improved fuzzy kernel C-means method is used to initialize the cluster center by K-means clustering method.The experimental results verify the effectiveness of the improved fuzzy kernel C-means segmentation method.2.Aiming at the complexity of the infrared image content of the photovoltaic panel and the interference of the surrounding environment,this paper proposes a segmentation method based on gray level co-occurrence matrix and region growing.The method preprocesses the image by entropy feature and gradient edge feature of the fused image,and then passes the improved region growing method performs segmentation and morphological processing on the pre-processed image to optimize segmentation results.The experimental results verify that the method can effectively segment the infrared image of photovoltaic panel.3.The above infrared image segmentation methods utilize the low-level semantic information of the image,such as the shape and texture of the image.In contrast,the image segmentation method based on deep learning network can extract deeper semantic information.Effectively overcome the external influence.This paper proposes a method based on improved U-Net network.The method changes the original U-Net network structure by introducing a residual structure and hole convolution.The experimental results show that the segmentation effect is more accurate and effective than the original U-Net network.In summary,the three image segmentation methods proposed in this paper can segment the photovoltaic panel area well and the infrared image segmentation effect of the three segmentation methods is getting better and better.And can provide an effective basis for photovoltaic panel infrared image fault detection and diagnosis.