Learning patterns in wireless sensor networks based on wavelet neural networks

In this paper it is demonstrated how some of the algorithms developed within the artificial neural-networks tradition can be simply adopted to wireless sensor network platforms and still meet most of the requirements for sensor networks. Neural-networks clustering algorithms also provide dimensionality reduction which further leads to lower communication costs and thus bigger energy savings. Two different data aggregation architectures are presented. They both utilize algorithms which apply wavelets for initial data-processing of the sensory inputs at different resolutions. Artificial neural-networks which make use of unsupervised learning methods are used for categorization of the sensory inputs. These architectures are tested on a data obtained from a set of several motes, equipped with several sensors each. Results from simulations of intentionally made defective sensors demonstrate the data robustness of these architectures