The joint time-frequency analysis is the effective tool in analyzing the time-varying and non-stationary signals. Its application in signal processing includes wavelet-based denoising, harmonic extraction and parameteric time-frequency, and so on. The passage characteristics of ship-radiated noise is the typical time-varying non-stationary noise. In the paper, the ship-radiated signal is denoised with time-frequency analysis, then the target noise features are extracted, used to target classification. The dissertation can be divided into two parts.The first part is about the features extraction, including the estimation of source level of the ship-radiated noise and the line extraction of the rotation frequency of propeller noise spectrum. The source level of the ship-radiated noise is estimated by analyzing the measurement records from passage ship noise. To extract the rotation frequency of propeller, the ship-radiated noise is transformed with complex analytical wavelet, then the noise envelope is obtained and denoised by the matching pursuits algorithm. By the means of FFT, analyzing the noise envelope yields the envelope spectrum, including the line of propeller rotation frequency.The radial-basis-function neural network (RBFNN) is introduced in the second part, and a classifier is designed to classify the ships. The experimental results denote that the RBFNN can get better classification result than B-P neural network. |