IA-CIOU: An Improved IOU Bounding Box Loss Function for SAR Ship Target Detection Methods

Abstract

Ship detection in synthetic aperture radar (SAR) images is crucial in both civilian and military fields, offering extensive application prospects. Nonetheless, owing to the distinctive characteristics of SAR imaging, this task confronts numerous challenges. Specifically, ships with high aspect ratios, dense arrangements and small sizes in complex environments frequently yield in suboptimal positioning effects, consequently impacting detection performance. In response to the challenges in ship target detection, this article introduces a novel approach, termed Inner-alpha-CIOU (IA-CIOU), that relies on an enhanced intersection over union (IOU). Primarily, the method introduces Inner IOU, which effectively regulates generation of auxiliary bounding boxes through scale factor r. This ensures a better fit for dimensions of ship target frames, thereby enhancing target detection performance as well as expediting model convergence. Subsequently, this method introduces Alpha IOU, enhancing robustness of small-size ship targets in complex backgrounds by adjusting α. This allows the detector to achieve greater flexibility in ship regression accuracy. Following numerous experimental validations, proposed algorithm consistently outperforms on both SAR-Ship-Dataset, MSAR-1.0 dataset, and SAR ship detection dataset (SSDD) dataset. This groundbreaking innovation not only possesses immeasurable practical worth, but also introduces a fresh perspective together with enlightening insights for future research efforts.

Keywords

• Intersection over union (IOU) loss function
• synthetic aperture radar (SAR) images
• ship detection
• YOLOv5