Research And Implementation Of Image Deep Compressed Sensing Algorithm

With the development and popularization of image acquisition equipment,the emergence of a large number of digital images brings challenges to traditional image processing technology.The compressed sensing(CS)algorithm compresses the signal at the same time as the signal acquisition,accurately reconstructs the signal under Nyquist sampling,and reduces the excessive occupation of computer resources during the image processing process.The compressed sensing algorithm based on deep learning solves the problems that traditional algorithms have existed,such as high time complexity and poor reconstruction quality at low sampling rates.However,the improvement of reconstruction quality of existing deep compressed sensing algorithms is still limited by the design and improvement of measurement matrix and reconstruction network.Therefore,this paper studies the deep compressed sensing algorithm,and the main research contents include:(1)Aiming at the problem of loss of original image details in the under-sampling measurement process of existing deep compressed sensing algorithms,an image compressed sensing algorithm based on residual sampling enhanced network is proposed in this paper.In the sampling phase,the original sampling network and the residual sampling network are trained separately to generate and combine two kinds of measurements,so that the residual information can be used to supplement the missing details information.During reconstruction,a deep reconstruction network is used to simulate the traditional compressed sensing reconstruction algorithm to realize the mapping of low-dimensional measurements to high-dimensional reconstructed images.The experimental results show that the reconstruction quality of the proposed algorithm is significantly improved compared with the existing advanced deep compressed sensing methods.(2)Aiming at the problem of over-smoothing and loss of detailed texture in reconstructed images generated by deep compressed sensing algorithm,an image compressed sensing algorithm based on a parallel enhanced network is proposed to learn and restore the detailed texture of the reconstructed image.In this network,the basic reconstruction network is used for the initial reconstruction and further refinement of the image.The enhanced reconstruction network is connected with each sub-module of the basic reconstruction network to gradually obtain the detailed information of each module,which is used for reconstruction of image detail texture.The reconstructed image of the parallel enhancement network is obtained by accumulating the outputs of each network.Experimental results show that the network has high reconstruction quality.(3)In order to improve the imaging speed of MRI,an MRI system based on deep compressed sensing algorithm is designed and implemented.The deep compressed sensing algorithm proposed in this paper is applied to MRI imaging.The MRI system uses real MRI data to verify the effectiveness of our algorithm,and improves the reconstructed speed while maintaining high reconstruction quality.