A new multilayer LSTM method of reconstruction for compressed sensing in acquiring human pressure data

Abstract

According to the idea of deep learning, this paper designs a new multilayer long short-term memory (LSTM) network method, a data driven model for sequence modeling. We use this deep neural network to solve the reconstruction problem of Single Measurement Vector (SMV) in compressed sensing (CS) theory. We take the measurement vector of CS as the input of the multilayer LSTM network, and the data to be reconstructed as the output of the network. We investigate the effectiveness of the LSTM network by using acquired pressure data from human body model. Experimental results demonstrate that, in comparison with the state-of-the-art methods for reconstruction accuracy, our multilayer LSTM method approach can effectively improve the accuracy of recovery in acquiring the short measurement vector of human body.