Gabor atomic decomposition of Green's functions for geoacoustic inversion and segmentation of a shallow water environment

This paper introduces a method that identifies acoustic transmission data with feature vectors for the segmentation of range-dependent shallow water areas. The measurement geometry consists of sonobuoy-like receivers drifting away from a fixed sound source. The data consist of the medium impulse response measured repeatedly along each buoy track (frequency-dependent Green's function). The feature vector entries are parameters obtained from a sparse, tree-based, matching pursuit decomposition of each response into Gabor atoms. The mutual distance of neighbor feature vectors is used to detect and localise main variations of the bottom geoacoustic properties and eventually isolate subareas of moderate variability. The method is tested with data from experiments of rapid environmental assessment (REA) ENVERSE 97-98.