Feature extraction of underwater targets using generalized S-transform
Xu Xi · Journal of Computer Applications · 2012
A method based on singular value of Generalized S-Transform module time-frequency matrix(GSTM)for feature extraction of underwater targets was studied in the paper because of the problem that feature vector is not easily extracted accurately in underwater target recognition.Firstly,the target signal was processed by using generalized S-transform.Then,singular value of GSTM was extracted as feature vector for target recognition.Lastly,target recognition was accomplished by probabilistic neural network.The generalized S-transform is improved from S-transform,which has higher frequency resolution or time resolution for signal analysis by changing window width factor.It can realize time-frequency analysis for non-stationary and nonlinear signal in underwater targets according to signal analysis need.Experimental result shows that the singular value of GSTM is suitable for target recognition and the method has different recognition result by selecting different window width factor.