New method for feature extraction in ship noise

Feature extraction is the key to classification and recognition of ship noise in frequency domain. A multi-fractal theory was applied to extract fractal feature in the power spectrum of ship noise. The wavelet based multi-fractal formalism was used to compute the generalized dimension of multi-fractal measure in triadic Cantor set. The computed results agree well with theoretical values. To apply the formalism to power spectrum of ship noise, a definition of singular measure was proposed first. The singular measure was used to describe the multi-fractal features of ship noise and reflect completely the properties of target in frequency domain. The results of multi-fractal analysis for three classes of ship noise samples have shown that the multi-fractal features in power spectrum are very effective for classification.