Pulse Identification Base On Image Sequence

The pulse diagnosis is important because the pulses can transfer the message of physiology and pathology in each part of body, it is the window to look into the change of body function,and also can provide important basis for diagnosis of disease. To detect the pulse objectively and analyze the pulse image, not only for extensive and accurate data collection, but also can overcome the interference of subjective psychological factors,summarize the clinical experience, and evaluate the result exactly. Therefore, the objectification and standardization of Chinese medicine diagnosis will become an important field of increasing investment in the industry. Currently, from various reports we can see that the sensors in use of pulse signal detection are mostly single point or composite points pressure sensor, and single point of pulse signal or simply several points of pulse signal can not reflect the particular change of each point in surface space of pulse-taking skin, it cannot fully describe the information of pulse image, so there will be less physiological information we can get. For the traditional model of sample source, because of the subjectivity and asynchronization in diagnosis of pulse sample, the sampling will have certain differences.In the method of analyzing pulse image, we find that one-dimensional pulse image analysis in time domain, frequency domain and time-frequency has a full and various development, but there almost have no sight of two-dimensional pulse image analysis.To solve these problems, this paper is mainly study for how to use the pulse image sensor to get the parameters of multi-dimensional pulse feature, and with these parameters combining the artificial neural network to identify pulse.The major work and innovation (chapter 3,4) are summarized as follows:This paper first introduces the research status of the characteristics of pulse signal and pulse sensor, pulse analysis and pulse recognition; outlines the problems to be solved and the main contribution,the research content of this paper.The second part first introduces the design, working principle and structure of the imagelized collection device of pulse in detail.According to the shortage in practical work, we improve the sensor to make the whole system more convenient and practical. Then we explain the basic structure and working principle of the pulse simulation system,and give the flow chart of experiment.In the third part, the dynamic pulse image sequences of bionic hand are collected by pulse image sensor, which is self-made based on CCD monocular camera. According to the change of grid area in each frame, we can get the three-dimension displacement variation of multiple points from pulse image surface. The data of 8 grid points selected in pulse length and 441 frames selected in time axis are formed into two-dimensional matrix,then it is transformed into two-dimensional gray scale image. Using three-dimensional variation of the displacement to reconstruct the pulse in three-dimension.Extract the feature of each pulse in spatial domain and frequency domain of two-dimensional pulse image and in spatial domain of three-dimensional pulse image, then according to the feature extracted and combining with BP artificial neural network for recognition of pulse.Finally, the respectively using PNN,SOM,BP, Elman artificial neural network of four comprehensive classification of the four pulse and recognition.Pulse using two-dimensional map from the space domain and frequency domain extraction pulse pulse rate, total energy, high frequency, low frequency component and three-dimensional reconstruction maps pulse width extracted to form feature 11 vectors. A pulse for each type of inference rules to develop and set parameters on the pulse of each sample was trained and tested, and finally on the various neural networks and the identification of pulse performance are compared and analyzed.The results show that two-dimensional map from the pulse spatial and frequency domain and three-dimensional space domain characteristic parameters extraction combined with artificial neural networks to identify pulse rate has been greatly improved for pulse diagnosis research has provided new ideas and Methods.