基于深度学习的机载线阵3D SAR超分辨成像方法研究

With the advantage of overcoming layover and shadow effects caused by terrain and targets occlusion, the airborne downward looking linear array three dimensional SAR (LA-3D-SAR) performs high precision surveying and mapping in high topographic regions and has become a hit in microwave imaging research. Based on the LA-3D-SAR imaging theory, this project focuses on the key problems of degraded imaging quality and low cross-track resolution caused by short sparse array. The data acquisition model under the condition of single-flight is to be created and high efficiency high-precision 3D imaging algorithm is to be developed; the possibility of applying deep learning technology to super-resolve the 3D imagery is investigated and the super-resolution evaluation method is to be proposed; the numerical simulations and laboratory experiments are to be conducted. This project will provide a novel approach to the LA-3D-SAR super-resolution imaging and will provide model and method support for real system development.

机载阵列下视三维 SAR（LA-3D-SAR）能大幅度降低地形起伏、目标相互遮挡所引起的叠掩、阴影等影响，高起伏山地和城市等测绘困难区的高精度测绘，成为近年来微波成像领域研究的热点。本项目在LA-3D-SAR三维成像理论研究的基础上，围绕实际机载平台条件下阵列尺寸过短引起的成像分辨率较低这一关键问题开展研究：建立高效的单航过机载LA-3D-SAR数据获取模型，发展高效高精度三维高分辨率成像算法；研究深度学习技术在LA-3D-SAR超分辨成像中的有效性和适用性，提出基于深度卷积网络的超分辨成像方法，构建相应的超分辨成像质量评估体系，分析超分辨处理方法的性能、精度和适用范围；通过仿真及实验验证数据获取机制、三维成像算法和超分辨处理方法的正确性和有效性；为机载LA-3D-SAR超分辨成像提供新的技术手段，为未来该型雷达的研制和应用提供理论和方法支撑。

本项目针对阵列下视三维 SAR（LA-3D-SAR）在实际应用时需要对观测目标更精细特征描述以获取更多的目标细节信息，增强对LA-3D-SAR图像中目标的理解和判读解译能力的目标，开展了高效数据获取与成像模型、超分辨深度卷积神经网络和仿真实验验证，通过项目研究，建立了高效的单航过机载LA-3D-SAR数据获取模型，发展了高效高精度三维高分辨率成像算法；提出了基于深度卷积网络的超分辨成像方法，构建了相应的超分辨成像质量评估体系；通过仿真实验验证了相应的数据获取机制、三维成像算法和超分辨处理方法的正确性和有效性，为未来超分辨成像机载LA-3D-SAR的研制和应用提供理论和方法支撑。