Feature Extraction and Classification Based on Bispectrum for Underwater Targets

According to the symmetric characteristics of bispectrum, a novel feature extraction scheme, which includes the summation-at-every-column feature vector, the summation-at-every-row feature vector and their combination in a triangle area, one of the 12 symmetric areas of bispectrum, is proposed. By using One-against-One (OAO) method of multi classification of Support Vector Machine (SVM), the mean classification accuracy for the radiated noise of underwater targets in three types is steadily above 98% for the summation-at-every-column feature vector and the combination feature vector respectively. The summation-at-every-row feature vector as a supplementary feature improves the classification performance but burdens the computation load of classification.