A decentralized GP learning, fusion, and planning (RESIN) algorithm for a mobile sensor network to actively learn the motion pattern of multiple moving targets, and thus planning for each sensor to pursue the targets based on the information entropy was proposed. RESIN is combined with a decentralized GP fusion method which is robust to rumor propagation and computational efficient by using the weighted exponential product based on Chernoff information, and an information-driven path planning (IPP) method that is able to generate the most information sensitive path for the mobile sensor network by using sequential planning and fusing each sensor with its predecessors' planning information. Various numerical simulations were done to show that RESIN is effective and could achieve near-optimal performance for the sensor network. Also, RESIN shows more applicability while in the situation that the number of sensors is less than number of targets.