A Robust Efficient Dictionary Learning Algorithm for Compressive Data Gathering in Wireless Sensor Networks

To enhance the accuracy and improve the efficiency of dictionary learning algorithm for compressive data gathering (CDG) in wireless sensor networks (WSNs), a robust efficient dictionary learning (REDL) algorithm is proposed. REDL algorithm separates outliers from training data by sparse coding them on identity matrix, and double-sparsity dictionary structure is introduced into the algorithm to provide efficient operators. REDL algorithm iterates among a sparse coding step, an outliers separation step and a dictionary update step alternately. The convergence of proposed algorithm is proved. Simulation results show that REDL algorithm can suppress outliers in training data better and decrease computational complexity compared to online robust dictionary learning (ORDL) algorithm and error source decomposition dictionary learning (ESDDL) algorithm. And REDL-CDG has a better reconstruction accuracy than ORDL-CDG and ESDDL-CDG in WSNs.