Horizontal Feature Pyramid Network for Object Detection in UAV Images
Weiqian Tang, Jian Qiao Sun, Gang Wang · 2021 China Automation Congress (CAC) · 2021
Feature extraction for object detection in UAV images is of great importance. In this context, a method named Horizontal Feature Pyramid Network (HFPN) is proposed, aiming at generating abundant features from the original Feature Pyramid Network (FPN). Specifically, we enhance the entire feature hierarchy with multiple stages of simple convolution and channel-wise addition operations, which enrich the classification and location information, and reduce the amount of computation. The detection accuracy on UAV image dataset VisDrone is improved with the proposed method.