多视角智能观测SAR目标识别理论与方法研究

Multiview intelligent observing SAR target recognition is a novel radar target recognition technology. It can plan the efficient observing path of the platform online and select the optimal imaging views. Thus, it rapidly obtains the multiple SAR images of the target, and recognizes the high-value target accurately with the scattering information from different views, which is of great value in the fields of disaster rescue, border control, military reconnaissance, and so on. This project will focus on both the basic theoretical and technical problems of the multiview intelligent observing SAR target recognition. Detection in imaging (DII) of multiview SAR, online intelligent planning of the observing path and micro-growth deep network target recognition will be studied in this project. And we will break through some key technologies such as the optimal target detection in imaging, fast prediction of the optimal observing views, and the growth and metabolism of the deep network. At the same time, simulations and experiments will be carried out, and the proprietary intellectual property rights will be obtained further. This project is of important scientific significance for improving the basic research level and the ability of independent innovation in intelligent SAR target recognition fields of our country.

多视角智能观测SAR目标识别是一种新型的雷达目标识别技术。它能在线规划出高效的平台观测路径，选择最优的成像视角，快速获取目标的多幅SAR图像，并利用不同视角的散射信息对高价值目标进行精准识别，在灾难救援、边境管制、军事侦察等领域具有重要的应用价值。本项目将围绕多视角智能观测SAR目标识别中的基础理论和技术问题，开展多视角SAR“成像中检测”、观测路径智能在线规划和微生长深度网络目标识别等研究工作，解决成像过程中目标最优检测、最优观测视角快速预测、深度识别网络可生长与代谢等关键科学问题，完成系统仿真与实验验证，形成具有自主知识产权的技术体系。本项目的研究可以提高我国在智能SAR目标识别领域的基础研究水平和自主创新能力，具有重要的科学意义。

作为一种新型的雷达目标识别技术，多视角智能观测SAR能在线规划出高效的平台观测路径，选择最优的成像视角，快速获取目标的多幅SAR图像，并利用不同视角的散射信息对高价值目标进行精准识别，在灾难救援、边境管制、军事侦察等领域具有重要的应用价值。多视角SAR“成像中检测”理论、观测路径智能在线规划、微生长深度网络目标识别是实现该技术的核心关键。本项目针对这些问题进行了理论分析、方法研究、仿真验证等工作，主要内容如下：（1）建立了多视角SAR“成像中检测”处理框架，研究了渐变孔径目标快速成像方法，实现了成像过程中目标的快速准确检测。（2）建立了多视角SAR观测路径在线规划模型，研究了观测路径自主规划方法，实现了目标观测视角和路径的优化选取。（3）建立了多视角微生长深度识别网络架构，形成了面向不同视角输入的识别网络结构自调整，实现了SAR目标的高效准确识别。以上模型与方法已通过理论分析、计算机仿真和实测数据进行了验证，结果表明，上述方法可实现多视角智能观测SAR对目标的有效识别。