Research And Implementation Of Body Sign Recognition Technology Based On Multi-physiological Signal Fusion

With the development of body area network(BAN)technology and communication technology,all kinds of wearable physiological sensors have been widely used,and the physiological signals that can be collected are more abundant.In the battlefield environment,we are not satisfied with the monitoring of soldiers' basic vital signs,but expect to further perceive the physical and emotional signs of soldiers.Based on the background of constructing intelligent detection system of individual soldier's physical signs in battlefield environment,this paper studies the recognition technology of human body's basic vital signs,hunger,exercise fatigue and fear.In order to realize the rapid and accurate recognition of human body signs,the multi-channel physiological signals which are universal and easy to be collected are selected to construct the recognition model of human body signs.The main research work of this paper is as follows:1.From the perspective of correlation and application scenarios,7 physiological signals including electrocardiogram(ECG),pulse wave(PPG),electrodermal activity(EDA),electromyography(EMG),skin temperature(SKT),blood oxygen saturation(Sa02)and blood glucose concentration(GLU)were investigated and selected.At the same time,the types of physiological signals that need to be collected and their identification processes are clarified.2.Construction and preprocessing of physical signs data set.The experimental platform was set up with MP150 as the main acquisition equipment,and the corresponding experiments were designed for the three signs of hunger,exercise fatigue and fear,and the physical signs data sets of 240,417 and 320 person-times were constructed respectively.Data preprocessing mainly carries out effective data interception,frequency reduction and denoising for signals in data set.The application of wavelet transform in physiological signal denoising is mainly studied.The corresponding wavelet basis,decomposition layer number and threshold coefficient are selected for different physiological signals,and the results are verified on the experimental data of nearly 1000 person-times.The results show that this method has a good denoising effect.3.Feature extraction and feature dimensionality reduction of physiological signals.Firstly,the characteristics of each physiological signal were extracted from two dimensions of time domain and frequency domain.In the time domain,the signal feature points are detected and the statistical index of the feature sequence is calculated as the time domain feature.The amplitude spectrum and power spectrum are used in frequency domain analysis,and their statistical indexes are calculated as frequency domain characteristics.In order to reduce the impact of irrelevant features and information redundancy,it is necessary to conduct dimensionality reduction treatment for high-dimensional features after feature extraction.Two methods of feature extraction based on principal component analysis(PCA)and feature selection based on improved genetic algorithm(GA)are mainly studied.PCA method does not rely on the feature recognition model,and the preliminary analysis of the feature sample dimension reduction results based on PCA method provides a basis for the subsequent model construction and improvement direction.4.Construction of physical signs recognition model.First,the feature data sets after feature dimension reduction were trained and recognized using SVM as a classifier,and the performance of PCA and improved GA feature selection methods was compared.The results showed that:for the two types of signs of hunger and exercise fatigue,the recognition effect was good,with an average recognition rate of 91.389%±2.307 and 90.519%±2.727,and the improved GA feature selection method was slightly better than PCA method.For fear signs,the recognition rate was poor,around 70%.According to the characteristics of short trigger and duration,the recognition model was reconstructed by using the learning network LSTM based on time series.Under this model,the recognition rate of fear signs was significantly improved,with an average recognition rate of 88.749%±1.855.5.Design and implementation of the sign recognition system.The system integrates the data processing method,feature extraction and dimension reduction algorithm,and the feature recognition model.In order to improve the performance of the system,the Protocol Buffers coding was optimized,the system structure was designed hierarchically and modularized,and the local client was developed from the point of view of convenient model parameter adjustment and data visualization.Finally,the system is deployed to the cloud to provide external sign identification services.