Asynchronous Data-driven Classification of Weapon Systems

This paper addresses real-time weapon classification by analysis of asynchronous acoustic data, collected from microphones on a sensor network. The weapon classification algorithm consists of two parts: (i) feature extraction from time-series data using Symbolic Dynamic Filtering (SDF), and (ii) pattern classification based on the extracted features using Language Measure (LM) and Support Vector Machine (SVM). The proposed algorithm has been tested on field data, generated by firing of two types of rifles. The results of analysis demonstrate high accuracy and fast execution of the pattern classification algorithm with low memory requirements. Potential applications include simultaneous shooter localization and weapon classification with soldier-wearable networked sensors.