An improved compressed sensing reconstruction algorithm based on artificial neural network

To meet both the precision and convergence rate requirement of reconstruction algorithm, an improved compressed sensing reconstruction algorithm based on artificial neural network (IANN-CS) is proposed in this paper. The approach applies Artificial Neural Network structure to compressed sensing (ANN-CS) to reconstruct sparse signal, and on this basis, a dynamic learning factor is obtained by using gradient descent method repeatedly to replace the one which is a constant in ANN-CS algorithm, this improved algorithm is also called IANN-CS in this paper. The experimental results show that, compared with ANN-CS algorithm, IANN-CS algorithm has greatly improved convergence rate with a little change in convergence precision. In addition, under the same reconstruction conditions, IANN-CS algorithm has a good compromise between reconstruction precision and convergence rate, what is more, the observation value needed in ANN CS and IANN-CS algorithm are less than which in the existing reconstruction algorithms.