Signal strength–based localization is commonly employed in wireless sensor networks due to its low complexity and simplicity. However, in non-line-of-sight (NLOS) environments with unknown transmit power, effective and efficient multi-target localization is a challenging task. In this paper, a fast multi-target localization based on a neural network (FMLNN) is proposed. The received signal strength difference (RSSD) is employed and NLOS bias is considered. Determining the maximum likelihood (ML) estimator is a complex and highly non-convex problem, so it is solved indirectly using a neural network. First, prior data composed of known target information and RSSD values are used in offline training to learn the nonlinear relationship. Then, the locations of multiple targets are estimated online using the trained network. Results are presented which show the proposed method provides fast and efficient localization of multiple targets, and has greater robustness to NLOS bias than conventional state-of-the-art methods.

In this paper, a novel dynamic position control (PC) approach for mobile nodes (MNs) is proposed for ocean sensor networks (OSNs) which directly utilizes a neural network to represent a PC strategy. The calculation of position estimation no longer needs to be carried out in the proposed scheme, so the localization error is eliminated. In addition, reinforcement learning is used to train the PC strategy, so that the MN can learn a more highly accurate and fast response control strategy. Moreover, to verify its applicability to the real-world environment, we conducted field experiment deployment in OSNs consisting of a MN designed by us and some fixed nodes. The experimental results demonstrate the effectiveness of our proposed control scheme with impressive improvements on PC accuracy by more than 53% and response speed by more than 15%.