Abstract
Dampers can prevent wires from breaking due to vibration, so it plays an important role in ensuring the stable operation of transmission lines. But the accuracy of popular object detection algorithm is still very low, because the occlusion phenomenon of dampers is serious due to the variable shooting angle. In this paper, we introduce our Supervised Attention RCNN (SA RCNN) to improve the detection effect of occluded dampers. SA RCNN mainly consists of two components: Box Supervised Attention Module (BSAM) and One Proposal Multiple Predictions (OPMP). In order to solve the problem of false detection caused by similarity of features, BSAM enhances features by using ground truths as supervised information to guide the generation of attention. In order to solve the problem of missed detection due to the NMS strategy, OPMP predicts a set of instances for one proposal rather than a single one. Experiments show that SA RCNN obtains 4.2% AP improvements on the dataset of occluded dampers compared to FPN.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
