Research On OSAHS Snoring Recognition Technology Based On Machine Learning

Snoring is a common phenomenon in people’s daily life,and the proportion of people suffering from snoring symptoms in the population has reached 20% to 40%.Snoring not only causes trouble for the patients and affects their companions,but also threatens the patients’ health.Obstructive sleep apnea syndrome is a respiratory disease,and snoring is one of its main symptoms.Patients get tired easily during the day and it can cause cardiovascular disease.Currently,the main technical method for diagnosing and analyzing the disease is polysomnogram(PSG),but it requires the patient to stay in the sleep laboratory all night and be connected to a large number of physiological electrodes.The acoustic analysis method of snoring signal has attracted many scholars’ attention and research due to its "non-invasive",low-cost and practical characteristics,and has great potential for development.Therefore,it is necessary to conduct acoustic analysis of OSAHS pathological snoring in order to provide a convenient,practical and low-cost detection method for OSAHS patients.In this article,the hardware platform is used in conjunction with PSG device marking for data collection.After preprocessing the snoring signal,the snoring samples of the experimental subjects were acoustically analyzed.The snoring signal contained and carried the structural information and characteristics of the human respiratory tract.The snoring samples of a number of simple snorers and OSAHS patients were analyzed and studied.It’s found that compared with normal people,the respiratory passages of OSAHS patients have undergone structural changes.Therefore,the acoustic analysis method of snoring signal can determine whether the person under test has OSAHS.The study found that the snoring sound of normal people and OSAHS patients have great differences in acoustic characteristics.In this experiment,a total of five characteristics were extracted,including the Linear Prediction Cepstrum Coefficient(LPCC)of the snoring sound sample,and the Mel-scale Frequency Cepstral Coefficient(MFCC),fusion LPCC-MFCC,and spectrum centroid.For the large-dimensional LPCC and MFCC features,the feature fusion of the two methods is used on the basis of the Fisher ratio.The fusion feature dimension remains unchanged and the recognition effect is better.At the same time,the anti-noise performance is also improved.Finally,the PSG equipment was used to mark the patient’s snoring samples,and the snoring in the respiratory disorder event was divided into two types,namely,the snoring of simple snorers and the pathological snoring of OSAHS.The experiments in this paper first verify the influence of different kernel functions of support vector machine on the classification of snoring samples,and classification experiments are performed on the optimal kernel function according to different acoustic analysis features.The results show that the Gaussian radial basis kernel function has the best overall recognition accuracy for the second-class snoring signal,and the fusion feature scheme has better classification performance,and the overall recognition accuracy reaches 95.8%.In addition,through the method of decision tree model,the effect of XGBoost,random forest,and decision tree model on the classification of snoring samples has been verified.The experimental results indicate that the XGBoost algorithm has a better overall effect on the classification of snoring samples,with the accuracy of snoring recognition up to 96.7%,which can be used as the best classification method for snoring samples.The experimental results demonstrate the feasibility of this method in assisting the diagnosis of OSAHS,which can provide certain help for clinical trials.