Target Detection in Synthetic Aperture Radar Images Based on Improved YOLOv8

Synthetic Aperture Radar (SAR) is an advanced microwave remote sensing technology. With its all-weather and all-day working characteristics, it has demonstrated its outstanding capabilities in the field of ship detection and has become a critical part of ship detection systems. Ship detection is essential for coastal traffic management and territorial security. However, the detection of ships based on SAR images faces multiple challenges, such as a large amount of image noise, complex background clutter, and high training costs for relevant technical personnel. Therefore, in order to address these challenges, SAR ship detection combined with machine deep learning algorithms significantly improves detection accuracy and efficiency. However, as the demand for target detection accuracy grows, the original network’s accuracy becomes increasingly inadequate. To address these issues, the main contributions of this paper are as follows: 1. Integrating the backbone of YOLOv8 with RepViT 2. Introducing Slim-neck at the neck to achieve lightweight while ensuring model accuracy 3. Introducing a new attention mechanism called PSA to improve the ability to extract multi-scale information. The improved model was tested on the MSAR data sets, achieving a 0.6% increase in mAP to 91.5% and a 2.6% increase in precision to 93.4% compared to the original YOLOv8n, indicating that the improved YOLOv8 has better accuracy and robustness when detecting ships.