脉象特征提取方法分析-系统工程专业毕业论文.docxVIP

II II ABSTRACT Wrist pulse contains important information about the health status of a person so that the organ pathological change could be perceived by pulse-feeling which has been popular for thousands of years. However, the traditional Chinese theory usually portrays the pulse types in a vague and general language, and the diagnoses from physicians often diverge greatly due to their subjective experience. Thus, the objectification of pulse diagnosis is imperative under the modern computer technology circumstance. The paper has collected pulse signals of 150 healthy persons and 289 patients using the pulse collecting system with fusion of the photoelectric sensor and pressure sensor. Signals are sampled, processed and stored in the computer. The patients belong to different diseases, such as gastrosis, diabetes, angiocardiopathy etc. The pulse database is then established for later analysis. The pulse signal is a weak biomedical signal and subject to the influence of surrounding circumstance. The paper draws the help from digital signal processing so as to optimize and preprocess the original signal. First, soft threshold wavelet is applied to remove burr noise and digital linear filter to filter 50Hz power electric high frequency noise. Second, the paper repairs the baseline wander issue from breathing or body movement with cubic spline interpolation approach. Then, the optimized pulse waveform is segmented and normalized. The pulse feature extraction step is the core work of the project. The paper makes improvements of the time-domain morphological features and Gaussian model features. Finally, k-nearest neighbor algorithm, linear discriminant analysis and support vector machine are used to classify pulse waves. Experiments show the improvement of feature extraction algorithm contributes around 3% increase of accuracy. The combination of photoelectric signal and pressure signal reveals a better classification with a correct rate of 73.33%. It also shows a good