Designing an adaptive approach for segmenting non-stationary signals

Signal analysis tools such as Fourier transform are often applicable on data with a limited length. Segmentation is an important pre-processing step in many signal processing applications. Statistical characteristics of the signal in a segment are often preferred to be similar. This characteristic, stationarity, improves performance of signal analysis technique. This article develops an adaptive segmentation method based on wavelet transform and fractal dimension from two aspects. One is to use discrete stationary wavelet transform in pre-processing step instead of using classical wavelet transform. The other is to choose the optimal parameters. Two parameters are needed to calculate the fractal dimension of a decomposed signal, window length and percentage overlapping of the successive windows, which affect the performance of the proposed approach. These parameters are optimally set using the particle swarm optimisation algorithm. Performance of the proposed method is compared with three other existing segmentation methods using both synthetic signal and real data. The results indicate the superiority of the proposed technique in terms of accuracy compared to existing methods.