Feature extraction for acoustic classification of small aircraft

Abstract

Standard approach to aircraft classification in acoustic sensor networks is based on propeller blades rotation harmonics and fundamental frequency analysis. However, propeller blades induced sound is not the only sound producing mechanism. As a result, aircraft of different classes (small plane, helicopter, and ultralight aircraft) often provide similar acoustic signatures in the spectral domain. We present new feature extraction methods beyond the spectral peaks-based ones. First, general feature extraction considerations are reviewed. Then features are described including harmonic-based feature, a feature based on Lloyd's mirror multi-path propagation effect, and a feature based on sound spikes caused by propeller blades interaction with flow vortices. Experiments are provided to justify the efficiency of the proposed features for target classification.