基于时/频复数卷积神经网络的高轨SAR超分辨成像研究

Deep learning is the important direction in the development of artificial intelligence, is also a hot spot. Using this powerful technology to solve difficult engineering problems is the research direction of Deep learning. Due to the limitations of the system, we cannot obtained the images of high-orbit SAR with high resolution. This project improves the resolution for the high-orbit SAR firstly with the Deep learning. Firstly, the improved frequency domain algorithm is used to focus the SAR data and a low resolution SAR image is obtained. Then, the image resolution is improved by using the super-resolution technique based on the Deep learning. In this project, high-resolution simulated data is used to generate high-resolution SAR image library, which leads to train deep neural network, and finally completes super-resolution recovery. The algorithm can realize the super-resolution recovery of high-orbit SAR image with high precision, so as to solve the bottleneck problem of low resolution of high-orbit SAR by using depth learning, and promote the cross fusion of computing intelligence and information science.

深度学习是当今人工智能发展的重要方向，也是热点方向。利用深度学习这一强有力的技术，解决现有技术上难以解决的工程问题，也是深度学习的研究方向。在现有技术条件下，高轨SAR由于系统限制，无法实现高分辨观测，导致其图像数据无法满足实际需求。本课题首次采用基于深度学习的超分辨技术，对于高轨SAR进行超分辨恢复。首先利用改进的频率域算法，实现SAR数据聚焦预处理，得到低分辨率SAR图像；然后利用基于深度学习的超分辨技术提高该图像分辨率。本课题采用高分辨的模拟数据，生成高分辨SAR图像库，进而产生深度神经网络，最终完成超分辨恢复。本算法将高精度的实现同步轨道SAR图像超分辨恢复，从而利用深度学习解决高轨SAR分辨率低的瓶颈问题，推动计算智能与信息科学的交叉融合。

深度学习是当今人工智能发展的重要方向，也是热点方向。利用深度学习这一强有力的技术，解决现有技术上难以解决的工程问题，也是深度学习的研究方向。在现有技术条件下，高轨SAR由于系统限制，无法实现高分辨观测，导致其图像数据无法满足实际需求。本课题首次采用基于深度学习的超分辨技术，对于高轨SAR进行超分辨恢复。首先利用改进的频率域算法，实现SAR数据聚焦预处理，得到低分辨率SAR图像；然后利用基于深度学习的超分辨技术提高该图像分辨率。本课题采用高分辨的模拟数据，生成高分辨SAR图像库，进而产生深度神经网络，最终完成超分辨恢复。本算法将高精度的实现同步轨道SAR图像超分辨恢复，从而利用深度学习解决高轨SAR分辨率低的瓶颈问题，推动计算智能与信息科学的交叉融合。