Small Object Detection in Remote Sensing Based on Contextual Information and Attention

Abstract

Many small objects, for instance vehicles and small ships, are encountered in remotely sensed images. However, small object detection has been a challenging task in remote sensing because of the problem that small objects are easily missed and influenced by the background. To address this challenge, we propose a detection method based on contextual information and attention, divided into two main parts. Firstly, for purpose of further improve the backbone network features to derive more contextual information, a multi-branch feature enhancement module is constructed to fuse multiple sensory field features to improve the ability of the backbone network to extract feature information; secondly, a new effective channel attention mechanism is proposed to reduce problems such as information confusion caused by the feature fusion process, thus reducing the influence of the background. Compared with other methods, it effectively improves the detection of small object among remote sensing images.