动态可配置的压缩感知成像系统

Compressive sensing imaging technique, which has more advantages than traditional digital imaging, combine imaging and compression into one process. Because of the advantages, it has many potential applications in the future, and will cause a great change in the digital image and video industry even the entire information industry.But in the current research of compressive sensing imaging, only digital images captured by traditional imaging devices are used in the most experiments, which is not able to test and verify the measurement and recovery algorithm in the real environment. To meet the needs of the research of compressive sensing imaging, in the project, we will develop a dynamically configurable imaging system based on compressive sensing to provide a general test and verification platform for the research of compressed sensing imaging in the real world. In the capture subsystem, the measurement matrix, in which the elements can be assigned to any value between 0 and 1, is realized by liquid crystal materials. So the imaging system can verify the effect of any measurement matrix in the optical imaging. In the recovery subsystem, the reconstruction algorithm adopts parallel hardware acceleration to achieve image acquisition, reconstruction and display in real time. The dynamically configurable imaging system based on compressive sensing will vigorously promote the development of compressive sensing imaging studies, and accelerate the compressed sensing imaging more and more practical. It will help the our researchers to catch up with the international advanced level in compressive sensing imaging field even the whole compressive sensing field as an efficient scientific equipment.

压缩感知成像技术将成像和压缩过程合二为一，与传统数字成像相比，具有显著优势，有极其广阔的应用前景，将引起数字影像产业乃至整个信息产业发生革命性变革。但是当前的研究，基本上都是利用传统成像设备采集的图像为模拟测试的对象，难以在真实压缩感知成像环境中对编码测量和恢复算法的研究成果进行实际测试与验证。本项目面向压缩感知成像研究的实际需求，研制动态可配置的压缩感知成像系统。该设备可以提供任意值测量矩阵的配置，用以验证不同的测量矩阵在光学成像中的实际效果；实现图像采集，恢复，呈现一体化，能够让相关研究人员高效快捷地验证成像效果，为压缩感知成像研究提供通用的实际环境下实验测试与验证平台。该设备将有力推动压缩感知成像研究的发展，促进压缩感知成像技术向实用化方向迈进，为我国科研人员在压缩感知成像乃至整个压缩感知研究领域赶超国际先进水平提供有力的科研仪器装备。

压缩感知成像技术将成像和压缩过程合二为一，与传统数字成像相比，具有显著优势，有极其广阔的应用前景，本项目旨在面向压缩感知成像研究的实际需求，提供任意值测量矩阵的配置， 用以验证不同的测量矩阵在光学成像中的实际效果； 实现图像采集， 恢复， 呈现一体化， 让研究人员高效快捷地验证自己的恢复算法和成像效果，为压缩感知成像研究提供通用的实际环境下实验测试与验证平台。经过4年的研发，本项目取得如下成果：..1. 实现了动态可配置的压缩感知成像装置，该装置支持多种采集与成像方式，采集方式包括单像素成像和阵面成像；成像方式支持无透镜成像和有透镜成像；采集对象支持图像和视频；测量矩阵实现包括液晶与DMD；采集频谱包括红外与可见光。整套系统实现了采集恢复流程一体化，给研究人员提供高效快捷的压缩感知实验平台。..2. 对压缩感知图像重建算法方面进行了深入的研究，提出了全局测量局部稀疏的压缩感知图像恢复算法，基于非局部相似度模型的压缩感知图像恢复算法，基于秩极小化理论的压缩感知图像恢复算法等多项压缩感知图像恢复算法，提高压缩感知图像重建质量10%~20%。..3. 提出一种基于残差网络的压缩感知图像恢复网络,实现快速非迭代的压缩感知图像重建。将最优化问题求解所必须的迭代计算，转变为不需迭代的深度网络重建问题，保持对传统算法100倍速度优势的情况下，平均PSNR提高了3dB左右。..4. 在硬件加速方面，实现了GPU并行的压缩感知图像恢复算法，与CPU相比，恢复速度提高了100倍以上。基于Xilinx V7 2000T，实现了一套完整的硬件化压缩感知数据采集重建架构，实现从数据采集到重建全过程的硬件化，数据同步和控制精度达到纳秒级，对外数据带宽达到10Gbps，实现高速采集和重建结果高速回传。..项目组与国内外研究机构紧密合作，在Optics Letters, IEEE Transactions on Multimedia, IEEE Geoscience and Remote Sensing Letters等顶级期刊和会议上发表论文28篇，其中在SCI收录22篇。培养博士生10名，硕士生14名，申请12项国家发明专利，其中2项已授权。2名项目主要研究人员在本项目进行期间获得国家杰出青年基金支持。