Preference of echo features for classification of seafloor sediments using neural networks

Selection of a set of dominant echo features to classify seafloor sediments using a multilayer perceptron neural network is investigated at two acoustic frequencies (33 and 210 kHz). Several sets of inputs with different combinations of two, three, four, five, and six echo features are exploited with three-layer neural networks. The performances of the networks are analyzed to assess the most discriminating set of echo features for classification of seafloor sediments. The results of the overall average performances reveal that backscatter strength and time spread are the two most important echo features at 33 kHz, whereas backscatter strength has higher discriminating characteristics at 210 kHz for seafloor sediment classification. In addition, a set of four echo features consisting of backscatter strength, time-spread, statistical skewness, and Hausdroff dimension gives the highest success at both the acoustic frequencies.