GMDR-Net: A Lightweight OBB-Based SAR Ship Detection Model Based on Gaussian Mixture Data Augmentation and Distance Rotation IOU Loss

Abstract

Aiming at the problems of oriented-bounding-box-based (OBB-based) synthetic aperture radar (SAR) ship detection, which include the large model volume, the imbalance of positive and negative samples, and imprecise ship detection and positioning due to dense arrangement and directional diversity, we propose a lightweight OBB-based SAR ship detection model, GMDR-Net, which is based on Gaussian mixture data augmentation (GMDA) and distance rotation Intersection over Union (IoU) loss. First, based on the Oriented RepPoints and ConvNeXt model, we propose a lightweight baseline model. Second, we propose a novel GMDA method, which effectively enhances the richness of samples and alleviates the problem of imbalanced positive and negative samples by combining the Gaussian mixture model and SAR image dataset labels. Finally, we propose a distance rotation IoU loss function, which introduces a distance penalty term based on generalized IoU, to achieve more accurate regression of the bounding box. Experimental results show that GMDR-Net achieved an average precision of 77.14% and 87.56% for the high-resolution SAR images dataset and the SAR ship detection dataset, respectively. In addition, at 800 × 800 and 640 × 640 input resolutions, its floating-point operations are only 11.78 G and 7.54 G, and frames per second reaches 29.62 and 44.76, striking a balance between accuracy and efficiency and exhibiting wonderful insights in remote sensing field.

Keywords

• Data augmentation (DA) method
• intersection over union (IoU) loss function
• oriented bounding box (OBB)
• synthetic aperture radar (SAR) ship detection