Adaptive Fine-Grained Feature Mining and RoI Feature Interaction Network for Small Object Detection in Aerial Images

IEEE International Conference on Acoustics Speech and Signal Processing, pp. 1–5

Abstract

Object detection in drone view images remains challenging due to small-scale objects distributed non-uniformly and exhibiting weak semantic features. Additionally, external factors such as lighting conditions, viewing angles, and background clutter further complicate the detection of small objects. To address these issues, we propose a small object detector called Adaptive Fine-grained Feature Mining Network (AFMNet) based on an anchor-free framework. Firstly, an Adaptive Fine-grained Feature Mining (AFFM) module is designed to generate Regions of Interest (RoI) on high-resolution feature maps and mine fine-grained information that benefits the representation of small objects. Secondly, an Inter-RoI Feature Interaction (IRFI) block is developed to efficiently integrate complementary cues— namely, deep layer semantic information and shallow layer spatial details—and enhance RoI features. Experiments on the VisDrone dataset demonstrate that our method achieves state-of-the-art performance. The source code of our AFMNet is available at https://github.com/xiefangmin/AFMNet.

Authors 3

1. Zhan Xiong Aachen Institute of Advanced Computer Science

   Leiden University

   Affiliation as printed

   Leiden University,Institute of Advanced Computer Science,Netherlands

2. Zhuoyue Wang

   University of California, Berkeley

   Affiliation as printed

   University of California, Berkeley,Department of Electrical Engineering and Computer Sciences,Berkeley,USA

3. Guoqiang Xiao

   Southwest University

   Affiliation as printed

   Southwest University,College of Computer and Information Science,China