Damaged Insulator Detection Based on Matching Network

Abstract

Traditional transmission line inspection of the power system is mostly manual inspection, but with the emergence of higher voltage, higher power, longer distance transmission lines, and more complicated geographical environment which the line through, the application of helicopters or UAVs to complete the circuit inspection task has become the need of the development of the times. We use a neural network to process the images collected by the equipment and mark the transformers, circuit breakers, knife switches, transformers, power cables, insulators and other parts in the images. However, due to the unobvious rules of demaged parts, the hard-to-achieve manual labeling task, the lack of a large number of damaged parts of the image data, it is difficult to train an effective neural network for the screening of damaged parts. Moreover, gradients may disappear in high-level networks because of the scene is complexity and the components to be detected are likely to be occluded. In this paper, we use the idea of small sample learning matching network and match the semantic information of the image with the double attention model to propose a detection scheme that takes the detection of damaged insulators as an example.