压缩感知作为信号采集与处理的新方法，能够从少量观测数据中准确恢复信号，极大推动成像、编码和医疗诊断等技术的发展。本项目旨在创新图像先验信息的学习策略和压缩采样方案，探索用模型和数据协同驱动的思想增强基于学习的重构算法的鲁棒性。针对目前压缩感知重构算法，尤其是基于深度学习的算法仅适用特定的采样模式和采样率，以及抗噪声能力弱的问题，（1）在非局部字典学习中，将多本外部字典作为边信息引入到字典更新中，实现内外部信息协同的更稳健的图像重构；（2）结合数据驱动的深度学习网络与模型优化方法，充分挖掘观测模型和先验模型的潜在有用信息，利于实现重构中复杂噪声的拟合和估计，增强算法的鲁棒性；（3）构建采样与重构同步优化的方案，引入群智能优化算法寻找最佳的傅里叶测量值组合，并迭代优化采样模式与深度学习网络，以实现最优采样与重构。研究内容为压缩感知技术提供了创新方案，具有重要理论意义和实用价值。

Compressive sensing is a recently emerging technique in signal sampling and processing, which can achieve accurate reconstruction of a signal from a small number of measurements, and has the potential of significantly improving the efficiency in many applications, such as imaging，coding and medical diagnosis. The project aims to propose learning strategies of prior information in image reconstruction model and compressive sampling schemes, and explore the use of model-driven and data-driven to enhance the robustness of learning-based reconstruction algorithms. In view of current compressed sensing reconstruction algorithms, especially the algorithms based on deep learning are only suitable for specific sampling mask and sampling rate, as well as the weak anti-noise ability, (1) multiple external dictionaries are introduced in non-local dictionary learning as side information into dictionary updating to achieve internal and external information fusion and more stable image reconstruction; (2) the combination of data driven deep learning network and model optimization method is employed to fully exploit the potential useful information of observation model and prior model and be beneficial to fit and estimate the complex noise in reconstruction, for enhancing the robustness; (3) to construct the synchronous optimization scheme of subsampling pattern and reconstruction model, swarm intelligence optimization algorithm is introduced to find the best combination of Fourier measurements in sampling, and then iterative optimization of sampling mask and deep learning network is employed to achieve optimal sampling and reconstruction. The research contents provide innovative schemes for compressed sensing technology, which have important theoretical significance and practical value.