Moving vehicle classification using wireless acoustic sensor networks

Acoustic signal processing over wireless acoustic sensor networks (WASN) currently constitutes a topic receiving extensive attention by the scientific community. However, the majority of the related literature fails to consider several types of phenomena directly affecting the smooth operation of such networks, such as sensor faults, aging phenomena, environmental changes, etc. This paper provides as a basis a problem formulation systematizing such obstacles and on top of that builds a sound classification system that is able to consider the appearance of (multiple) sensor faults and environmental noise. The cornerstone of the proposed classifier is an echo state network operating at the sensor level, while the decisions are combined at a higher level via a correlation-based dependence graph. We carried out a thorough experimental campaign utilizing data coming from a WASN composed of 23 sensors aiming at the acoustic classification moving vehicles.