AE-FPN: adaptive enhance feature learning for detecting wire defects

Wire defects usually occur in high-altitude transmission lines, leading to line transmission failures and even the possibility of large-scale power outages. Therefore, timely and accurate locating wire defects detection is a key technology for power transmission. However, there are still challenges for wire defect objects with large aspect ratios, arbitrary orientations, and complex backgrounds. In this paper, we design a novel Adaptive Enhancement Feature Pyramid Network (AE-FPN) to focus on the wire defect features through an attention mechanism during feature fusion and extraction. AE-FPN is a plug-and-play component that can be used in different networks. Using AE-FPN in a basic Faster R-CNN system, our method achieves a 3.2% AP gain at a very marginal extra cost. In addition, a multi-scenario multi-object dataset of wire defects is established that provides the baseline for detecting wire defects in transmission lines.