Accurate Object Detection of Electric Power Equipment Based on CapsNet Framework
Wang Li, Zhou Fang · 2019
The automatic monitoring of the electric power equipment through computer vision is the critical part of intelligence of the power industry. While the accurate object detection of electric power equipment is the premise of automatic inspection. The best-performing methods are complex ensemble systems that typically combine multiple low-level image features with high-level context. In this paper, we propose a detection algorithm that solve the multiple object detection of electric power equipment once with higher mean average precision (mAP). Our approach's key insight is: we apply CapsNet framework to classification of the regions proposal in order to localize and segment different objects. Instead of CNN, we use capsules with dynamic routing to classify the features extracted compare to RCNN. We find that this method outperforms the RCNN with samples not large and retain more direction relationships between the different parts of the objects.