针对高分辨率合成孔径雷达图像设计了一种舰船目标几何特征提取算法.通过视觉注意机制检测目标区域的算法, 通过频谱残差视觉显著计算模型求取显著图, 完成显著区域的检测以实现舰船目标的初步定位, 基于获得的视觉显著图采用最大熵算法完成阈值分割筛选出舰船区域.在提取的舰船切片的基础上, 采用针对几何特征的提取算法, 经图像预处理、方位角估计、旋转获取最佳表征舰船目标几何轮廓的外接矩形, 相对有效准确地提取几何特征; 最后, 采用典型的TerraSAR-X数据进行仿真实验.结果表明, 与传统方法相比, 本文提出的频谱残差视觉模型完成合成孔径雷达图像舰船切片的区域分割能够有效降低虚警率, 舰船目标的检测速度提高了25%~50%.该方法能够快速稳定地提取舰船目标的几何特征, 也更加符合实际高分辨率图像舰船目标检测的应用需求.

A new method for geometric features extraction of ship target in high-resolution Synthetic Aperture Radar (SAR) image was proposed, After detecting and locating ship targets from high-resolution SAR images. The algorithm continues to acquire target slices to construct the process of the ship target geometric feature extraction. Firstly, the algorithm obtained a saliency map, completed the detection and positioning of ship targets, and obtained the ship target slices. Secondly, the algorithm extracted the geometric features based on the resulting ship slices. The slices was estimated by the azimuth to obtain the exact minimum bounding rectangle, then effective and accurate extraction of geometric features can be completed. Finally, the algorithm appied to SAR image target detection, which is efficient proved by experimental results. The experiments on TerraSAR-X and a large number of satellite data demonstrate that the proposed algorithm can extract the geometric features with high accuracy and good stability. Unlike traditional methods, the use of improved spectral residual visual significant computational models to locate and segment ship targets can effectively reduce the false alarm rates, and the detection speed increased by 25% to 50%. And it is suitable for practical requirements of ship target detection in high-resolution images.