Study Of The Analysis System And Recognition Based On ANN For The Manifestation Of The Pulse

The pulse-diagnosis is a part of the motherland medical four sciences. It is indispensable objective basis of dialectical management. But the qualities and subjectivity of pulse-diagnosis very affected its accuracy and feasibility;this became the limitation factor for the Chinese medicine pulse-diagnosis' application, develbpment and communication. Thus, impersonal research of the pulse-diagnosis has important meaning for inheriting and expanding of our country Chinese medicine. This paper research the pulse analysis and identified on the base of research and exploiture of the Chinese medicine pulse artificial intelligence identifies system.First, designing the system for collecting pulse signal is introduced. The signal collection system procedure is established by VB in this paper, and the VB has the stronger advantage in the sketch customer interface development, but the software of MATLAB is made use of when processing, analyzing and identifying the signal, so it introduced the commutating the data between MATLAB and VB. Second, making use of the spectral analysis method, carried on the spectrum analysis to the time domain pulse signal, and get the frequency chart curve. Then by the characteristic analysis of the frequency chart curve distilled the spectral feature. At the same time, applying the wavelet analysis theory in the processing and analysis, withdrew a new token pulse feature parameter that is transform by wavelet analysis at different dimensions of energy and carried out the pulse classification. This is based on that wavelet analysis has the good time-spectral synchronously localizing speciality. Third, as an active mathematics branch in nonlinear subject, the fractal theory is another means that describes the chaotic signal. This paper attempts to distill boxing dimension character basing on fractal theory, and the result testify that boxing dimension can more easily become a undee characteristic, but is not the typical model characteristic of identify the pulse.Lastly, applies the network of BP in the classification for four kinds of pulse signal, and compares with the difference between the spectral characteristic and the energy of on the different dimensions by wavelet transformation and the boxing dimension and the spectral characteristic as the neural network input. Though the training sample is limited in the text, the emulational result indicates: for some particular characteristic (such as a the energy of on the different dimensions by wavelet transformation, spectral feature) of the pulse signal, we make use of the neural network to identify pulse is a viable and effective, and it has obviously superiority compared with the conventional pattern recognition method in terms of self adapting and self learning ability.