基于改进Faster R-CNN的SAR船舶目标检测方法

合成孔径雷达（Synthetic Aperture Radar, SAR）船舶检测在海洋交通监控中发挥着重要作用，传统SAR目标检测算法一般利用目标与背景杂波之间的对比度差异进行检测，在近岸海域等复杂场景下检测效果较差。为了提高在复杂场景下的检测性能，本文提出一种基于改进Faster R-CNN的船舶检测方法，在分析不同特征分辨率对检测性能影响的基础上，结合VGG的思想与扩张卷积设计一个适用于SAR船舶目标检测的特征提取网络，以提升对小型船舶目标的检测能力。另外，根据sentinel-1A数据集中目标尺寸分布选取小尺寸anchor，并通过去除冗余anchor，将检测速度提升了一倍。在sentinel-1A数据集上的实验证明本文提出的算法能够快速、有效地从复杂场景SAR图像中检测出船舶目标。

A SAR Ship Detection Method Based on Improved Faster R-CNN

SAR ship detection plays an important role in marine traffic monitoring. Traditional SAR target detection algorithms are mostly based on contrast difference between target and background clutter, whose performance is limited especially in complex scenes, for instance coastal areas. In order to improve the detection performance in complex scenes, a SAR ship detection algorithm based on Faster R-CNN is proposed in this paper. After analyzing the influence of feature resolution on detection performance, a feature extraction network suitable for SAR ship target detection is designed based on the idea of VGG and dilated convolution to improve the detection capability of small ship targets. In addition, a small size anchor is selected according to the target size distribution in the sentinel-1A dataset. And by removing the redundant anchor, the detection speed is improved. Experiments on the sentinel-1A dataset demonstrate that the proposed algorithm can detect ship targets in SAR images of complex scenes with high speed and accuracy.