Research On Deep Learning Algorithm For Compressed Sensing Of Images

Compressed sensing(CS)is a new type of signal sampling technology that can reconstruct the original image from random linear measurement data well below the Nyquist sampling frequency.Among them,the efficient reconstruction process is the key to the realization of compressed sensing.However,the traditional reconstruction algorithms use an iterative optimization method and the process is very time consuming.In addition,at low sampling rates,the reconstruction quality of the image may be poor.Deep learning has excellent learning ability and can effectively express the reconstruction map from low-dimensional random measurement data to original high-dimensional image.To this end,this dissertation focuses on deep learning models and algorithms for compressed sensing image reconstruction,and learns compressed sensing image reconstruction mapping in a data-driven manner.New measurements can be quickly reconstructed by simple forward calculation,which greatly reduces the time complexity of reconstruction process.This dissertation proposes various types of compressed sensing image reconstruction networks and makes most use of different network structure advantages to improve image reconstruction quality.Specific research results include:We propose a novel sub-pixel convolutional generative adversarial network(GAN).The generator constructs the sub-pixel convolutional network to learn the explicit mapping from the low-dimensional measurement vector to the high-dimensional reconstruction,in which a compound loss,including reconstruction loss,measurement loss and adversarial loss,is designed to guide the network learning.The test image can be fast reconstructed by simply passing the low-dimensional measurement vector through the well-trained generator network.By means of the adversarial training with discriminator,the generator can learn the inherent image distribution and improve the reconstruction quality.We propose a new dual-path attention network for compressed sensing image reconstruction.Motivated by the structure and texture representation mode of the image,the network is composed of a structure path and a texture path.The structure path aims to reconstruct the dominant structure component of the unknown image,and the texture path targets at recovering the remaining texture details.To better bridge the information between two paths,the texture attention module is designed to deliver the useful structure information to the texture path for predicting the texture region,thereby facilitating the recovery of texture details.Two paths are jointly optimized with a unified loss function.In the testing phase,given the measurement vector of a new image,it can be well reconstructed by carrying out thewell trained dual path attention network and integrating the outputs of the structure path and the texture path.The design of this network is beneficial to reconstruct the texture details of the image,and at the same time it can achieve the decomposition of the image structure and texture components in the compressed sensing domain.We propose a cascading network for compressed sensing of images with progressive reconstruction.Specifically,we decompose the complex reconstruction mapping into the cascade of incremental detail reconstruction(IDR)modules and measurement residual updating(MRU)modules.The IDR module is designed to reconstruct the remaining details from the residual measurement vector,and MRU is employed to update the residual measurement vector and feed it into the next IDR module.The contextual memory(CM)module is introduced to facilitate the information interaction among all the IDR modules,therefore augmenting the capacity of IDR modules.The final reconstruction is calculated by accumulating the outputs of all the IDR modules.The cascade reconstruction mode can effectively reduce the difficulty of network learning and improve the quality of the image reconstruction.