A Compressed Sensing Network for Acquiring Human Pressure Information

Abstract

In this article, sparse autoencoder in deep learning is integrated into the compressed sensing (CS) theory, and a reconstruction algorithm is designed based on the biological mechanism of human brain synaptic connections. The compressive sampling process is modeled as a neural network model. Then a biological mechanism-inspired stacked long short-term memory (LSTM) network model is proposed as a reconstruction algorithm of CS theory. Consequently, a CS network (ComsensNet) model is introduced, by integrating the compressive sampling process and reconstruction algorithm. ComsensNet can provide a bridge between sparse autoencoder in deep learning, synapses in human brain neurons and the CS theory. A deep neural network is designed based on the synaptic biological mechanism of human brain neurons, and then combine with the theory of CS. The effectiveness of ComsensNet is investigated by using acquired pressure data from the human body model. The experimental results demonstrate that the biological mechanism-inspired stacked LSTM network in ComsensNet can improve the reconstruction accuracy compared to other reconstruction algorithms.