Automatic signal segmentation based on singular spectrum analysis and imperialist competitive algorithm

Electroencephalogram (EEG) is generally known as a non-stationary signal. Dividing a signal into the epochs within which the signals can be considered stationary, segmentation, is very important in many signal processing applications. Noise often influences the performance of an automatic signal segmentation system. In this article, a new approach for segmentation of the EEG signals based on singular spectrum analysis (SSA) and imperialist competitive algorithm (ICA) is proposed. As the first step, SSA is employed to reduce the effect of various noise sources. Then, fractal dimension (FD) of the signal is estimated and used as a feature extraction for automatic segmentation of the EEG. In order to select two acceptable parameters related to the FD, ICA that is a more powerful evolutionary algorithm than traditional ones is applied. By using synthetic and real EEG signals, the proposed method is compared with original approach (i.e. without using SSA and ICA). The simulation results show that the speed of SSA is much better than that of the discrete wavelet transform (DWT) which has been one of the most popular preprocessing filters for signal segmentation. Also, the simulation results indicate the performance superiority of the proposed method.