跨模态域自适应SAR图像舰船检测与识别  

Ship detection and recognition in SAR images with cross-modality domain adaption

作　　者：宋玉成[1] 李景润 田甜[1] 田金文[1] SONG Yucheng;LI Jingrun;TIAN Tian;TIAN Jinwen(School of Artificial Intelligence and Automation,Huazhong University of Science and Technology,Wuhan 430074,China;National Key Laboratory of Science and Technology on Multi-spectral Information Processing,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学人工智能与自动化学院,湖北武汉430074 [2]华中科技大学多谱信息处理技术国家级重点实验室,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2022年第11期107-113,共7页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(42071339)。

摘　　要：针对合成孔径雷达(SAR)数据成像质量差和标注不足的问题,提出一种跨模态域自适应端到端的SAR图像舰船目标检测与识别方法.首先,针对SAR舰船检测中目标特征辨识度低和背景信息复杂度高的问题,设计了融合多尺度特征和局部上下文信息及背景抑制来改进兴趣区域变换网络(RoITransformer)的特征表达能力;然后,针对SAR数据属性标签标注困难及样本缺乏的问题,设计了域适应模块在全局和实例级别实现对齐特征,将属性知识从光学数据跨模态迁移到SAR数据上.自建了近岸区域高分辨率SAR舰船数据集(HRSSRD),结果表明:所提方法在HRSSRD上实现了88.5%的检测精度,在舰船检测和军民识别任务上具有良好的性能.Aiming at the problems of poor imaging quality and insufficient labeling of synthetic aperture radar(SAR) data,an endto-end method for ship target detection and recognition in SAR images with cross-modality domain adaption was proposed.First,in view of the problems of low discrimination of target features and high complexity of background information in SAR ship target detection,the fusions of multi-scales feature and local context information,and background suppression were designed to improve the feature expression ability of region of interest-Transformer(RoITransformer). Then,aiming at the difficulty of labeling attributes in SAR data and lack of labeled samples,a domain adaptation module was designed to implement features alignment at the global and instance level to transfer attributes knowledge from optical data to SAR data across modalities.High resolution SAR ship recognition dataset(HRSSRD) of the nearshore area was built up. Results show that the proposed method could realize an mean average precision of 88.5% on the self-built HRSSRD,and has good performance in ship detection and military-civilian recognition task.

关 键 词：舰船检测 军民舰船识别 合成孔径雷达(SAR)图像 迁移学习 域自适应 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]