Extraction and Analysis of Fine Spectrum Features of Underwater Objects

Abstract

In underwater object detection, the distance between the object and the receiving platform, sea conditions, and other factors seriously affect the strength of the object, while the pitch and tone of the object signal are almost unaffected by the those factors, which can reflect the difference in the essential attributes of different objects together. This paper studies the extraction and analysis of fine spectral features of underwater objects, and extracts the auditory perception features from the power spectrum and high-order diagonal slice spectrum of the object signal to achieve a fine description of the object spectral features. The information contained in the 1.5-dimensional and 2.5-dimensional spectra of linear frequency modulation (LFM) pulses is derived. The separability of object echo, noise and reverberation in the spectral feature space is discussed. Radial basis function support vector machines are used. Classify the extracted auditory features. The data processing results verify the feasibility and effectiveness of extracting auditory perception as a classification feature. The 2.5-dimensional spectral feature can obtain better recognition results than the second-order statistical feature.