The Pulse Signals Extraction And Recognition

The pulse is a comprehensive phenomenon about the bit, number, shape, and potential of pulsating. The pulse formation not only relates to the heart, blood and veins, but also closely relates to other body organs functional activity. The nature of the pulse diagnosis plays the very important role in judging the state of an illness and the identification of a disease. Therefore, pulse diagnosis has an important significance in condition judgment and disease identification.For the realization of the signal quantization of common clinical pulse, we extracted features of pulse signal from the time domain and frequency domain. Firstly, the time domain used the threshold method, and combined the time domain feature parameters with modal energy ratio to get a valid time-domain feature parameters that can distinguish the pulse more effectively. Then, the frequency domain used FFT to get frequency-domain feature parameters fo, ho, SER10and x. By analyzing200cases of pulse signals, we get some typical feature parameters in clinical, which are h3/h1, h4/h1, t1, t, K, R,f0,h0and x.An improved fuzzy C-means, BP neural networks and fuzzy neural network respectively used on six pulse signal classification and recognition. The recognition results showed that membership functions, fuzzy inference rules and the cluster center could affect the improved fuzzy C-means, the recognition accuracy rate was76%; BP neural network recognition accuracy of the pulse signal was84%, the recognition rate increased to92%after training, but the training time was longer due to the hidden layer; Fuzzy neural network recognition rate was92%for six pulses, the accuracy rate increased to99%after training, and improved training speed. Fuzzy neural network recognized pulse signal well.This thesis analyzed the time domain and frequency domain characteristics of six pulses in clinic:flat vein, wiry pulse, sunken pulse, fine pulse, slippery pulse and slow pulse. The selected feature parameters were chosen as input recognition. Fuzzy neural network was used to classify and recognize these feature parameters, its recognition accuracy was92%. It achieves the classification and recognition of six pulse signal.