Ship Radiated Noise Feature Extraction And Classification And Identification Based On Higher-order Statistics

Higher-order Statistics (HOS) is the primary analysis tool in analyzing non-Guassian signal and nonlinear system. It possesses plenty of advantages in signal detection, feature extraction and harmonic retrieval. Based on the HOS, the features of ship-radiated noise are extracted and the three types of ship are classified in my paper. By analyzing the properties of the ship-radiated noise, Varieties of approaches feature extraction of ship-radiated noise based on HOS are developed, Which include Bi-spectrum and tn-spectrum feature,1(1/2)-spectrum feature, 2(1/2)-spectrum phase-coupling feature, and so on. A new method-High-order Cyclic Statistics (HOCS) is presented to extract the feature of modulated ship-radiated noise in the paper. The periodical components in ship-radiated noise are drawn out successfully by the techniques of three-order cyclic cumulant. The feature vectors for the classification are consisted of the features extracted from the above approaches. A B-P neural network classifier is designed to the classification of three types of ship by using the feature vectors. The average correct recognition rate of three types ship reaches to 91.5%. The result illustrates that the above techniques based on HOS are valid. As the modulated ship-radiated noise include much information of the ship (such as velocity, main-engine type and work conditions), the way that the modulated ship-radiated noise is analyzed by HOCS has some superiorities: first, it can restrain random Guassian noise; then it can separate non-stationary signal from stationary signal etc. So a good result of recognition and classification maybe obtained. In view of heavy task in the paper, it is worthy to research more in eliminating the background noise and choosing features for better classification result in the future work.