Research On The Key Technology Of Respiratory Abnormality Recognition Based On FPGA

With the rapid development of China's economy and the accelerating pace of people's life,the air pollution is getting worse,and the incidence of obesity and hypertension is increasing.So respiratory diseases have become a hindrance to people's normal life.The detection and diagnosis of respiratory diseases has become the hotspots of clinical medical research.Conventional measurement methods of respiratory disease require professional operation,which is prone to cause discomfort and cannot be used to monitor a patient for a long time.Non-contact measurement based on RF sensor can overcome the above shortcomings.Compared to other non-contact measurement methods,it is less susceptible to environmental interference and more convenient to monitor respiratory diseases at home,and it has higher precision.In this paper,the non-contact method based on RF sensor is used to monitor the respiratory signal,and then the extraction of respiratory signals and respiratory characteristics is realized based on FPGA platform.Finally,the normal breath and abnormal breath is distinguished by the machine learning classification models based on FPGA and DSP development platform,and it comes out with good results.The main work of the thesis is as follows:1.The identification system of respiratory abnormality based on radio frequency sensor was introduced in this thesis.Then Xilinx Virtex 6 FPGA and TI 6678 DSP development platform used in this system are introduced.2.FPGA implementation of respiratory signal preprocessing and feature extraction was designed.The digital intermediate frequency module and preprocessing module were designed and implemented based on Xilinx Virtex 6 platform,and the respiratory signals were extracted from the radar echoes.Then the features in time domain and frequency domain of the respiratory signals were studied,and respiratory signal feature extraction was implemented on Xilinx Virtex 6 platform.Data transmission between Xilinx Virtex 6 and TI 6678 was realized through ethernet.3.The identification based on machine learning of respiratory abnormality and its realization were studied.The k-nearest neighbor classification algorithm,linear support vector machine algorithm and random forest algorithm were studied.The performance of three algorithms was analyzed by Matlab.The k-nearest neighbor classification algorithm was implemented based on Xilinx Virtex 6.The linear support vector machine algorithm and random forest algorithm were trained in Matlab and implemented based on TI 6678.4.The experiments were designed and the experimental results were analyzed.The human experiment in laboratory environment was designed to construct the respiratory pattern database.The performance of k-nearest neighbor classification algorithm,linear support vector machine algorithm and random forest algorithm was analyzed.The experimental results show that the classification accuracy of k-nearest neighbor classification algorithm is 73%,and the classification accuracy rate of linear support vector machine algorithm is 75.6%,and the classification accuracy rate of random forest algorithm is 76.8%.Then,the decision-level fusion algorithm was adopted for k-nearest neighbor algorithm,linear support vector machine algorithm and random forest algorithm.The accuracy rate after fusion was 78.2%.Finally,to decrease the confusion rate of Cheyne-Stokes breathing,Cheyne-Stokes Variant breathing and Dysrhythmic breathing,the apnea detecting algorithm was adopted in extraction algorithm of respiratory signal's feature in frequency domain,and according to the actual experimental requirements,body motion rejection was adopted in extraction algorithm of respiratory signal's feature in time domain.The total accuracy rate after optimization can reach 83.2%.