Recognizing specific human pulse signal based on clustering analysis

In this paper, we propose a pattern classification approach to learn and recognize human pulse signal. Different from previous work, the approach introduced in this paper of processing signals is oriented by the perspective of system analysis, and is aimed to recognize the pulse signal of pregnant objects from that of unfertilized female adults. Firstly, we apply homomorphic deconvolution model to get two types of human pulse signal curves in cepstrum domain and extract its Mel-Frequency Cepstrum Coefficient. This step is the learning process which learns characteristic parameters of human pulse signal, and besides, frequency characteristics and formant parameters of human pulse transmission system. Secondly, the Mel-Frequency Cepstrum Coefficient is processed via Dynamic Time Warping and Fuzzy C-Means Clustering, thus detecting parameter ranges of human pulse signal and utilizing them as the classifier in the subsequent recognition process. Instead of optimizing by solely applying Dynamic Time Warping, our approach, which combines Fuzzy C-Means Clustering and Dynamic Time Warping, tends to optimize the recognition rate significantly due to its advantage of searching for the globally optimal solution.