The new scheme of compressed sensing based on chaotic theory and deep learning

• 1、School of Cyberspace Security, Beijing University of Posts and Telecommunications, Beijing 100876

Abstract：In order to develop a new fast and accurate natural image compressed sensing scheme, we combine the advantages of two kinds of existing compressed sensing methods: the encryption effect of traditional iterative methods and the response speed of deep learning methods. Specifically, we propose a new compressed sensing model based on chaotic compressed sensing and deep neural network, which is called chaotic deep compressed sensing. We transform the iterative steps of traditional compressed sensing into the form of deep network and apply the relevant chaotic parameters to the measurement matrix generation and deep network training. All the parameters in the chaotic depth network are acquired by learning, and manual design is no longer required. The experimental results show that the reconstruction effect and running speed of chaotic deep compressed sensing are significantly better than other existing compressed sensing methods.

Keywords： compressed sensing deep learning neural network chaotic theory