ARFNet: adaptive receptive field network for detecting insulator self-explosion defects

Insulators are one of the critical components of high-altitude transmission lines. Insulator defects can lead to the failure of the power transmission system and even more severe consequences. Therefore, accurately locating and identifying insulator defects are particularly important. To address the problem of insulator information loss after insulation self-explosion and the large gap in insulator size, in this paper, we propose an effective and innovative module called adaptive receptive field network (ARFNet) to get proper context information for insulator self-explosion defects. ARFNet is an effective component that can be used in different networks to give the networks the ability to adapt the size of the receptive field through the attention mechanism. Besides, to further reduce the false detection rate, we also build a novel insulator dataset, including two categories of the whole insulator and the insulator self-explosion area. In addition, experiments show that our method can effectively improve detection accuracy and reduce the false detection rate.