宽幅SAR海上大型运动舰船目标数据集构建及识别性能分析

以TopSAR和ScanSAR成像模式为代表的宽幅合成孔径雷达(SAR)可以实现更大范围的海洋场景观测。但实现宽测绘带的同时降低了成像分辨率，因此宽幅SAR图像中的舰船目标缺乏清晰的结构特征，给大范围海上舰船目标识别带来了极大的挑战。同时，由于缺乏海上运动航母、两栖舰等大型关键目标的宽幅SAR样本数据，使得海上运动大型关键舰船目标识别更加困难。针对该问题，该文构建了宽幅SAR海上大型运动舰船目标数据集，共包含2291个大型舰船目标样本，类别划分为大型军事舰船，长度为≥250 m的大型民用舰船和长度为150～250 m的大型民用舰船。其构建过程为:首先，通过先验知识辅助获取港口区域大型军事舰船目标样本；其次，利用属性信息对OpenSARShip数据集进行长度筛选获取大型民用舰船目标样本；最后，在样本的距离-多普勒域添加二次相位误差进行运动仿真来模拟海上运动舰船目标的成像结果。此外，该文使用经典识别算法和深度学习方法对构建的数据集与运动仿真处理后的数据进行了识别性能分析，结果表明在低分辨率情况下采用SAR图像复数信息可以在一定程度上提高算法的识别率；并且运动舰船目标的散焦问题对识别准确率具有较大影响。 

Construction and Recognition Performance Analysis of Wide-swath SAR Maritime Large Moving Ships Dataset

Wide-swath Synthetic Aperture Radar (SAR), represented by TopSAR and ScanSAR acquisition modes, can observe a vast area of ocean scenes. However, achieving wide-swath reduces the quality of imaging resolution, which causes the ships captured in wide-swath SAR images to not have clear structural characteristics. This phenomenon brings a great challenge to the identification of large maritime ships. Further, the lack of wide-swath SAR sample data of large critical ships, such as moving aircraft carriers and amphibious ships, makes the identification of maritime moving ships difficult. To solve this problem, we construct a wide-swath SAR large maritime moving ships dataset, which includes 2291 samples. The dataset is divided into the following categories: large military ships, large civilian ships of lengths greater than 250 m, and large civilian ships of lengths between 150～250 m. The construction process of the dataset is as follows: first, the sample data of large military ships in the port area are obtained from prior knowledge; second, the sample data of large civilian ships are obtained via the length screening of OpenSARShip dataset with attribute information; finally, the imaging results of moving ships at sea are simulated by adding quadratic phase error in a range-Doppler domain. This study also analyzes the recognition performance of the constructed dataset and motion simulation of the processed data using classical recognition algorithms and deep learning methods. Experimental results show that using SAR image complex information at low resolution can improve the recognition rate of the algorithm to a certain extent, and the defocusing problem of the moving ship target has a considerable impact on the recognition accuracy.