Echo-waveform classification using model and model free techniques: experimental study results from central western continental shelf of India

Seafloor classifications and characterization using acoustic backscatter data from the central western continental shelf of India is presented in this work. We have acquired single beam sea-floor echo data using two frequencies: 33 kHz and 210 kHz along with the sediment samples for determining grain size to be used as ground truth. Analog echo output was digitized using a 1 mega sample per second A/D card (16 channels, 12 bit PCI-1712L). The study is initiated to observe the interaction effects of the sound signal with different sediment seafloor from off Goa shelf area, which covers finer clayey seafloor from inner shelf to coarser sandy seafloor from outer shelf. For classification of the seafloor, analysis conducted by determining the area experimental echo peak histograms and matching them with Rice pdf. We classify different seafloor with estimated model parameter /spl gamma/ (coherently reflected echo energy/incoherently scattered echo energy) for two different frequencies. The differences in estimated '/spl gamma/' parameter indicates variability in the roughness of the sedimentary layer structure from a same location. Analyses based on the certain features obtained from echo data acquired from seven data locations provide us insight about the complexity of the seafloor structures. Estimation of power law parameters using topographic data from 33 kHz and 240 kHz frequencies (from a operated shallow water multi-beam system) from transect was also used to find correlations with the estimated '/spl gamma/' parameter using echo backscatter data. Though, the critical analyses carried out by employing numerical modeling to bathymetric and echo-backscatter data is useful to understand the complex seafloor processes and characterization of continental shelf seafloor of India, but unable to provide a suitable means for seafloor classification. This paper also suggests a hybrid artificial neural network (ANN) architecture i.e., learning vector quantisation (LVQ) for seafloor classification. An analysis is presented to establish the efficient performance of the proposed network in terms of real time seafloor classification of the acoustic backscatter data.