Classification of underwater mines by means of the FRFT and SVM

Classification by acoustic echoes analysis of buried sea mines is the main focus of this paper. Both simulation and actual experiment data show that when the active sonar transmits chirp signal, the energy of buried mines and reverberation echoes will concentrate on different fractional domain, which results the very different properties between the fractional Fourier spectrum of target echoes and that of reverberation. Features are extracted by means of the FRFT (Fractional Fourier Transform) spectrum, and then the Karhunen-Loeve (K-L) transform is used to compress the features before sending to classification. A SVM (Support Vector Machine) classifier is trained and tested on the feature sets of both target and reverberation samples. Experiment results of the FRFT method under different elevations indicated good recognition and classification rates.