Infrared Image Detection of Transmission Line Insulators Utilizing a CNN Combination Implemented With Small Sample Data Augmentation and Transfer Learning

The abnormal heat identification of transmission line insulators has always been a necessary link in power inspection, and the improvement of its intelligence is an inevitable development trend. Infrared image recognition of insulators often requires a large number of infrared image data, which is one of the biggest resistance obstacles at present. In this paper, we use the small sample set for data augmentation and introduce transfer learning at the same time. First of all, the original set of 150 small samples is expanded to nearly 2000, and then the image recognition is carried out through the pre-trained VGG16 network, and the final recognition accuracy can reach 92.65%, which has certain advantages compared with the current infrared recognition methods of various small samples.