Feature extraction for acoustic signatures of small boats

Small boats represent a particular challenge to maritime surveillance due to their ubiquity, low radar cross-section, and absence of AIS transmission. We present a review of previously developed acoustic classification methods and demonstrate the importance of both the engine, and propeller as the two sources of small vessel noise. The features used for classification are extracted using various signal processing methods including signal spectra, spectra of Detection of Envelope Modulation on Noise (DEMON), and Cepstrum. The normalization of the features allowing to remove the dependence of the parameters on the distance to a boat and its speed was conducted. Acoustic signatures of six small boats including a panga, a jetski, an electrical boat, a go fast boat and two speed boats of the type common in the United States were collected by Stevens in a large glacial lake in NJ. The sound sources of various tonal components in the signal spectra are discussed. The program for automated extraction of classification features is presented and the identification of each of the 6 small boats with these features is demonstrated. [This work was supported by DHS’s S&T Directorate.]