Research On Pulse Information Detection And Processing Method

Pulse diagnosis is a diagnostic method summarized by ancient Chinese medical scientists through long-term practice.In clinical practice,the cause and syndrome of the disease can be inferred,and it has a high historical status.With the development of science and technology,the objective research of pulse condition is an inevitable trend.This dissertation studies the pulse condition of TCM from three aspects: pulse condition information detection and processing,feature extraction and dimensionality reduction,and pattern classification.The following is the main work of this dissertation:The HK-2000 C sensor is used to collect the pulse signal,the wavelet denoising method is analyzed,and a new threshold function method is adopted.This method can reduce the deviation from the original signal while reducing the oscillation.Use the new threshold function to process the signal,and the signal-to-noise ratio reaches 31.0331.Aiming at the omission of peak and valley pair detection,a peak-valley pair detection algorithm combining amplitude threshold method and distance threshold method is adopted to accurately locate the position of peak and valley pair,which is beneficial to the extraction of average waveform.Aiming at the phenomenon of different waveforms of the collected pulse signals,this dissertation divides the pulse signals according to the number of first-order differential zero-crossing points,and adopts different methods to extract the characteristic points of various types of pulse waves.The pulse condition features are extracted from the time domain,frequency domain,and time-frequency domain respectively,and a total of50-dimensional feature quantities are extracted,and the corresponding feature data set is constructed.Aiming at the phenomenon of feature redundancy that is prone to feature redundancy,principal component analysis is used to reduce the feature dimension,and the final feature quantity is reduced to 12 dimensions,such as: h2/h1,w/T,K,α,SER5,R,etc.For the problem of small number of samples in the dataset,the K-fold cross-validation method was used to make full use of the data samples,and BP neural network was used to classify the pulse for different feature datasets,and the pulse classification accuracy was improved by 6.67% after feature dimensionality reduction.For the drawback of randomness of initial weights and thresholds of the neural network,genetic algorithm was adopted to optimize the BP neural network,and the classification accuracy after experimental validation reached 86.67%,which is 6.67% better than BP neural network.