Graph Laplacian Regularization With Procrustes Analysis for Sensor Node Localization

In this paper, we investigate the problem of sensor node localization and propose a non-linear semi supervised noise minimization algorithm through iterative manifold learning. The method works on the observation that noisy data lie on a higher dimension space even though the actual data are embedded in a low dimensional manifold. The collective labeled and unlabeled data are represented as a weighted graph. A prediction function is created based on the available labeled data along with manifold learning to exploit the intrinsic geometry. On top of prediction function, iterative feedback mechanism is used, which incrementally flattens the higher dimensional manifold. This reduces the error boundary in each stage for every data point. Result found to converge after a few iterations. This is followed by localized Procrustes analysis to further reduce the error. Experiment using TelosB motes and simulation with labeled and unlabeled data show that the proposed technique is able to reduce the noise, on an average, by around 70%. Results also show that the mechanism is able to localize the sensor nodes with high accuracy and outperforms the baseline method and LapRLS in different conditions.