A New Signal Segmentation Approach Based on Singular Value Decomposition and Intelligent Savitzky-Golay Filter

Signal segmentation, dividing non-stationary signals into semistationary parts that each has rather equal statistical characteristics is necessary in many signal analysis approaches. In this article, a novel signal segmentation approach based on the modified singular value decomposition (SVD) and intelligent Savitzky-Golay filter is proposed. First, Savitzky-Golay filter is used to minimize the least-squares error in fitting a polynomial to frames of noisy data. There are two parameters in this filter adjusted by many trials. In this paper we propose to use new particle swarm optimization (NPSO) for appropriate selecting of these parameters. Then, we employ two approaches based on the modified SVD to attain the boundaries of each segment. The proposed methods are applied in the both comprehensive synthetic signal and real EEG data. The results of using the proposed methods compared with three well-known algorithms, demonstrate the superiority of the proposed method.