SAR Aircraft Image Detection Network Based on Scene Feature Enhancement and Multi-Field Interaction

In recent years, with the rapid development of deep learning technology, SAR aircraft target detection research has shown new vitality. SAR aircraft image has the characteristics of strong background noise interference and multi-scale, which makes aircraft image detection very challenging. This paper presents a novel SAR aircraft detection network, ASFF-NET, designed to enhance detection accuracy while remaining lightweight. It features an advanced scene feature fusion module (ASFF) to mitigate background noise, a multi-receptive field interaction module (MRFISPPF) for handling multi-scale targets, and a multi-channel hybrid downsampling module (AFdown). Experimental results demonstrate that ASFF-NET outperforms the original YOLOv8 by 2.3% in mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">75</inf> and 0.8% in mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50:95</inf>, with minimal increases in parameters and computation, proving its improved accuracy and generalization while maintaining a lightweight structure.