Research On Feature Extraction And Classification For Underwater Target

The target identification technique is the precondition of detecting the target and stealthy striking under the novel technical environment. It is also drawn much attention by many countries since it is the key technique to build intelligent weapon system among torpedo and naval mine. This paper focuses on research the application of cyclic autocorrelation demodulation, the GFCC (Gammatone Frequency Cepstral Cofficients) and artificial neural network in extraction and classification of the ship radiated noise. The efficiency of the algorithm is verified by processing the simulated and real trial data. And a software is designed to simulate and classify the ship radiated noise.1. Explore the mechanism of ship radiated noise, on the basis of analysing the typical feature of ship radiated noise, the paper studies the mathematics model and implement method of line spectrum, broadband constant spectrum and noise modulation. Design a software to simulate the target noise.2. Elaborate the cyclostation signal and extract the DEMON spectrum by using the cyclic autocorrelation demodulation technique. Compare the effect of cyclic autocorrelation demodulation technique and Hilbert demodulation technique.3. This paper studies ship radiated noise features extraction based on GFCC and MFCC and elaborates the concept of Mel frequency cepstrum coefficients and Gammatone frequency cepstrum cofficients. Improve the GFCC by changing the compression method.4. The paper elaborates underwater acoustic target classifier design method and research the RBF neural network. The paper verifies the effectiveness of feature extraction and classification by the processing and analysis of experimental data and classification results.