A Lightweight Improved YOLOv8 for SAR-based Ship Detection in ICRS LEO Satellite Systems

  • Lee, Younggyu
  • Park, Jaewoo
  • Im, Gyeongrae
  • Kang, Jinho
Citations

SCOPUS

0

초록

Recently, integrated communication and remote sensing (ICRS) based LEO satellite systems have received great attention. To overcome the limited resources such as restricted battery capacity in the LEO satellite systems, lightweight deep neural network (DNN) models for object detection are essential. In this paper, we develop an improved YOLOv8s for Synthetic Aperture Radar (SAR)-based ship detection to enhance the computational efficiency in the LEO satellite systems, by replacing specific convolution layers with Spatial-Channel Decoupled Downsampling (SCDown) modules. On the SSDD dataset, it reduces GFLOPs and parameters and improves inference speed by 28.7 FPS while maintaining detection performance compared to the original YOLOv8s. © 2025 IEEE.

키워드

communicationdeep neural networkLEO satelliteobject detectionremote sensingSAR shipYOLOv8
제목
A Lightweight Improved YOLOv8 for SAR-based Ship Detection in ICRS LEO Satellite Systems
저자
Lee, YounggyuPark, JaewooIm, GyeongraeKang, Jinho
DOI
10.1109/ICTC66702.2025.11388047
발행일
2026-02
유형
Conference paper
저널명
International Conference on ICT Convergence
페이지
2036 ~ 2037