A Novel Acoustic Pattern Recognition System for Wireless Sensor Networks

Wireless Sensor Networks (WSNs) have demonstrated their effectiveness in threat detection and localization. However, there are two critical issues related to signal processing in WSNs. First, energy-efficiency, memory-consumption, bandwidth-utilization, and computation-complexity are of main concern. Second, the application environments contain highly colored noise, multi-path echoes, and simultaneous emission sources. To address these issues, Intelligent Automation, Inc. (IAI) and its subcontractor, the University of Washington, propose a novel acoustic threat recognition system. The proposed system architecture is distributed and hierarchical. The functions of threat detection, classification and source localization are organized in multiple levels. At each level, the information processing task is performed in a distributed manner. On the other hand, the proposed system architecture allows cooperation among sensing nodes to collaboratively detect target signatures, reduce false alarms, classify target types, and estimate the acoustic source location. The system combines recent advances in Wavelet Analysis, intelligent learning and sensor fusion. In particular, the proposed Discrete Wavelet Packet Transform-based power-law detection algorithm is robust to environmental noise, yet computationally efficient. The advantages of the proposed threat recognition system include energy efficiency, reliable detection and classification, low detection and classification latency, reduced false alarms, efficient bandwidth utilization, and accurate source location estimation. BENEFITS: In addition to acoustic threat recognition, the proposed system along with the wireless sensor network architecture can be used in a widely range of other applications, such as moving vehicle tracking, speaker identification, firefighter/military personnel tracking, etc. The resulting technology can be readily used in law enforcement, border protection, etc. Moreover, the proposed classification and data fusion algorithms can also find other applications such as health monitoring of electro-mechanical systems, and intrusion detection in computer networks.