Doctor of Philosophy (Ph.D.)
Wireless sensor networks for human health monitoring, military surveillance, and disaster warning all have stringent accuracy requirements for detecting and classifying events while maximizing system lifetime. to meet high accuracy requirements and maximize system lifetime, we must address sensing diversity: sensing capability differences among both heterogeneous and homogeneous sensors in a specific deployment. Existing approaches either ignore sensing diversity entirely and assume all sensors have similar capabilities or attempt to overcome sensing diversity through calibration. Instead, we use machine learning to take advantage of sensing differences among heterogeneous sensors to provide high accuracy and energy savings for performance critical applications.;In this dissertation, we provide five major contributions that exploit the nuances of specific sensor deployments to increase application performance. First, we demonstrate that by using machine learning for event detection, we can explore the sensing capability of a specific deployment and use only the most capable sensors to meet user accuracy requirements. Second, we expand our diversity exploiting approach to detect multiple events using a distributed manner. Third, we address sensing diversity in body sensor networks, providing a practical, user friendly solution for activity recognition. Fourth, we further increase accuracy and energy savings in body sensor networks by sharing sensing resources among neighboring body sensor networks. Lastly, we provide a learning-based approach for forwarding event detection decisions to data sinks in an environment with mobile sensor nodes.
© The Author
Keally, Matthew Alexander, "A Learning-based Approach to Exploiting Sensing Diversity in Performance Critical Sensor Networks" (2012). Dissertations, Theses, and Masters Projects. Paper 1539623357.