Research On Classification Technology Of Breathing Pattern Based On Continuous Wave Radar

With the aggravation of environmental pollution in China,more and more people suffer from the chronic respiratory diseases.Therefore,the diagnosis of respiratory diseases has become a hotspot in the field of clinical medicine and biomedical engineering.Traditional diagnostic methods of respiratory diseases often need to be operated by professional medical staff with many monitoring facilities,which cost amounts of human and material resources.Moreover,most of them need to contact the human body,even invade the human body,causing the patient's discomfort and cannot monitor the patients for a long time.However,non-contact monitoring techniques for human respiration overcome these drawbacks.One of these techniques named biotic radar technology has many advantages,including insensitivity to the environment and powerful penetration.Therefore,it is more suitable for monitoring of human respiration compared with other non-contact monitoring techniques,such as infrared detection technology and video tracking technology.This paper uses the continuous wave radar sensor to detect the respiratory signals,extracts the features of respiratory signals and then classifies them by the machine learning algorithms,which achieves the good pre-diagnosis of latent chronic respiratory diseases.And it can be applied in the field of clinical medicine assisted diagnosis and family telemedicine diagnosis.The application prospect is bright.The main work is as follows:1?The hardware platform of continuous wave radar system was introduced and the theoretical analysis for vital signs signal acquisition were explained.One FIR band-pass filter was designed to extract the respiratory signal from vital signs signals,and synchronize control experiment with the contact piezoelectric sensor was set up to verify the reliability of the continuous wave radar system to detect the respiratory signals.2?The physiological characteristics of tachypnea,radar echo signal characteristics and corresponding symptoms of the six different breathing patterns were described.The six different breathing patterns include Normal breathing,Cheyne-Stokes breathing,Cheyne-Stokes Variant breathing,Dysrhythmic breathing,Biot breathing and Tachypnea breathing.For the aim to realizing classification of different breathing patterns,the number of envelope peak,the variance of envelope peak,the variance of normalized short-time energy,the minimum value of instantaneous frequency,the variance of instantaneous frequency and the range of instantaneous frequency were extracted.In particular,for better use in clinical sleep samples,one peak threshold method for early screening and waveform matching method for secondary screening were provided to eliminate body dynamic interference.Moreover,one apnea-based threshold method was provided to optimize the feature extraction algorithm.3?The algorithm flowchart of basic principle of the breathing pattern classification algorithm based on machine learning was introduced.And the theories and over-fitting problem of the decision tree classification algorithm and the support vector machine(SVM)classification algorithm were studied.In addition,this paper introduces the evaluation method of classification model,laying a theoretical basis for the analysis of experimental results.4?The experiments were designed and the experimental results were analyzed.Firstly,f the experiment in laboratory environment was designed,and the "training sample libraries"were used to construct the classification model of decision tree and support vector machine,and the performance of the model were verified by the "verification sample libraries".We sought the best classification model by constantly adjusting parameters.The results showed that the optimal classification model was linear SVM model,and the total classification accuracy was 93.9%,which verified the effectiveness of the classification model.Secondly,the clinical sleep experiment was designed and the linear SVM model was tested with the patient's test sample.The experimental results showed that the classification accuracy of the whole night was 92.3%,and the feasibility of the breathing pattern classification algorithm was verified.