2019_programme: DETECTION AND CLASSIFICATION OF BOATS IN COASTAL ZONE BY SUPERVISED LEARNING



  • Session: 06. Underwater Noise - Modelling and Measurements
    Organiser(s): Gavrilov Alexander, Skarsoulis Emmanuel, Taroudakis Michael
  • Lecture: DETECTION AND CLASSIFICATION OF BOATS IN COASTAL ZONE BY SUPERVISED LEARNING
    Paper ID: 781
    Author(s): Troussard Corentin, Magnier Caroline, Gervaise (Chorus) Cedric
    Presenter: Troussard Corentin
    Presentation type: oral
    Abstract: The detection and the classification of ships became of major interest during the Second World War and the Cold War and gave rise to the first studies on the radiated noise of ships [Ross_1976]. First reserved for the military application, these questions on the acoustic classification of boats are transposed in the deep-sea civil domain which has been growing rapidly since the 1990s [Tournadre_2014], mainly for large commercial vessels [Lourens_1987, Das_2011, Arveson_2000]. With the growth of coastal maritime activity and coastal activities, new threats appear: small crafts, AUVs. This leads to a need in a discreet surveillance of areas such as natural protected areas, harbours, borders and aquacultures [Sutin_2013, Filinger_2011, Chung_2011]. The classification of small crafts, practically non-existent in the literature, represents an important and complex issue. Indeed, very few studies exist, due to the difficulties to obtain information on maritime traffic in the absence of AIS data and non-suitability of classification tools.\nTo answer this problem, RTSYS has developed a software suite based on Machine Learning method. These methods exploiting a large acoustic database recorded in coastal environment by RTSYS acoustic recorders: more than 21000 acoustic sequences among multimodal database acoustic and photography of 5980 sequences of individual identification of the vessel. These algorithms allow a detection of proximity and distant small vessel with goods performances: the percentage of no detection (PND) is 1.2 % and the percentage of false-alarms (PFA) is 1% for proximity detection.\nPND is 5.3 % and PFA is 3 % for distant vessel. Specific adaptation of machine learning method for the small vessels’ classification issues shows interesting and encouraging results. Terms, these algorithms will be embedded on the RTSYS system such as acoustic recorders (RESEA, EASDA14, EASDA416), buoys (REMHY and RUBHY) and AUVs.
  • Corresponding author: Mr Troussard Corentin
    Affiliation: RTSYS
    Country: France
    e-mail: