Research On Human Pulse Signals Recognition Algorithm Based On Cepstrum Features

Pulse is called the language of life in Traditional Chinese Medicine (TCM), it contains a wealth of information about the status of human health. However, due to pulse at the teaching of Chinese medicine is very difficult, when the doctors have lots of difference in clinic, so the experience of pulse diagnosis can not be communicated. The qualitative and subjective pulse diagnosis greatly affects its accuracy and feasibility, and restricts the application, developing and communication of Traditional Chinese Medicine pulse diagnosis. With the development of sensor and computer technology, people hope to apply modern technology to human pulse diagnosis to reveal the essence and feature of pulse phenomena scientifically, which is the main research aspect in this paper.Higher-order Statistics (HOS) is the primary analysis tool in analyzing non-Gaussian and non-linear signal. It possesses plenty of advantages in signal detection, feature extraction and harmonic retrieval. Based on the Bicepstrum, the features of Human Pulse Signal are extracted and classified in the paper.There have been a lot of reports about extracting features in time domain by analyzing the characteristics of human pulse. There are also a lot of works focused on how to use neural network to classify and recognize the pulse manifestations. But these techniques are far from perfect. In this paper, the following works are reported.Firstly, based on comparing different objective algorithms of human pulse manifestations considering that most traditional methods are apt to obtain the information only from time-domain and can not preserve all the original information of pulse, varieties of approaches feature extraction of human pulse signal based on Bicepstrum are developed, which include the value of zero diagonal slice of Bicepstrum, the value of diagonal slice of Bicepstrum when m = 1, n=1, the kurtosis of Bicepstrum within a specific range and the value of third-order Cepstrum entropy within a specific range. Although these methods can achieve respective effects, larger operations and complex identification can exist.Secondly, Pattern Recognition technique is simple, effective and reliable. Especially, it can greatly decrease the computational load and memory. This technique obtains ideal results with high precision of recognition and rapid response rate. By analyzing the pulse signal, the recognition systems based on Bicepstrum and Mahalanobis distance have been constructed. Using the system, the pulse signals of the normal and drug addicts have been successful recognized, and the average recognition rate is about 87.5%. By comparing the result, the system optimization is introduced for clinical experiment in future.The experiment results show that the extracted parameter values based on the above methods as the feature vector using Mahalanobis distance is effective and correct.As the algorithm implemented in this paper is in accordance with characteristics acquisition and recognition of Traditional Chinese Medicine syndrome types, it could be used as reference in objective research and real clinics. It also has much significance to develop the study of Traditional Chinese Medicine.