Waveform classification based on wavelet transform and K-means clustering

Waveform classification is a primitive step of many signal processing problems, such as ECG processing and the analysis of power bus signal. In order to judge whether the power bus signal works well, comparison needs to be made between real waveforms and standard waveforms. General methods derived from time domain or frequency domain lapse according to different amplitudes and sampling rates of the electric waveforms. Due to different sampling rates, similar waveforms in time domain appear distinctly in frequency domain. Because of the superposition and incongruous time sequence, similar waveforms in frequency domain may vary temporally. We extract wavelet features and classify real electronic waveforms by K-means clustering. The classification results on proposed waveforms that labeled artificially have been demonstrated using leave-one-out cross validation.