Surveillance Of Apnea Syndrome Based On Snoring Signal Recognition

Obstructive sleep apnea hypopnea syndrome(OSAHS)patients with sleep disordered breathing due to blockage of the upper airway pharyngeal structure and respiratory pathological behavior,including hypoventilation and apnea symptoms.And this disease is a cause of a series of cardiovascular diseases,which seriously endangers the health of patients.For a long time,polysomnography(PSG)has been used by hospitals to detect symptoms of respiratory disorders,but it has the disadvantages of complicated instrument monitoring and low audience rate.This article analyzes pathological snoring from the perspective of snoring signal analysis.The purpose is to provide a convenient and low-cost monitoring method for OSAHS patients.In this paper,the hardware platform combined with PSG device labeling is used for data collection,and the endpoint detection algorithm for the sleep snoring signals with complex and diverse environmental noise in the home environment is studied.After the signal pre-processing process,for the conventional algorithm in low signal-to-noise ratio(SNR)environment,the detection rate is low,the sound leakage and misjudgment phenomenon,the multi-window spectral spectrum reduction and noise reduction combined with the subband energy entropy ratio method of endpoint detection composite algorithm.The experimental results show that under the complex noise environment,the endpoint detection results of the snoring signals of different subjects are accurate.Furthermore,the noise of pure noises was added to different types of noise in the Noise-92 standard noise database,and the overall recognition rate of the noised samples was still96%.The algorithm in this paper achieves accurate interception of the target snoring segment in the overnight snoring signal.For the collected snoring samples,a patient information table and snoring samples of various categories are used to construct a snoring sample database to facilitate subsequent data query and management.Secondly,the snoring samples of the subjects were analyzed acoustically to explore the differences in the acoustic characteristics of snoring between patients with ordinary snoring and OSAHS.In this paper,Mel cepstrum coefficient(MFCC),linear predictive cepstrum coefficient(LPCC),formant,fundamental frequency,spectral entropy,PR500,spectral centroid and spectral flatness are extracted.For large-scale MFCC and LPCC features,the feature fusion of the two schemes based on Fisher's ratio is performed.The fused feature dimensions are unchanged and the recognition effect is better.Finally,combined with the PSG device to mark the snoring samples of patients,the snoringduring the respiratory disturbance event was divided into four categories: normal,low ventilation,and before and after the respiratory disturbance event.Firstly,the effect of different kernel functions of support vector machine on the classification of snoring samples is verified,and the optimal kernel function is classified based on different acoustic analysis features.The experiments show that the radial basis kernel function has the best effect on the overall recognition rate of the four types of snoring signals,and the classification performance of the fusion feature scheme is better,reaching an overall recognition rate of 81.6% and an AUC of 0.923.And through the ensemble learning method,the adjusted XGBoost algorithm has a better overall effect on the classification of snoring samples,with an AUC of 0.943,which can be used as an optimal classification method to classify snoring samples.The experimental results verify the feasibility of the method in this paper for assisting the monitoring of OSAHS symptoms and can provide some reference in clinical.