Research On Snoring Recognition Algorithm And System Design In Complex Environment

Snoring is the main symptom of sleep apnea.It also has a certain relationship with high blood pressure and cardiovascular disease.It not only reflects the physical condition of the individual,but also affects the quality of sleep,life and work of others.Therefore,the identification,analysis and diagnosis of snoring are very important.Nowadays,the research on snoring mainly focuses on the distinction between normal snoring and pathological snoring,the diagnosis of snoring obstruction and the recognition of snoring in simple environment.The work of this paper studies the algorithm of snoring signal recognition in complex environment with multiple kinds of complex sounds,as well as the software and hardware implementation of the algorithm.This method is suitable for real complex environment,and has the advantages of high accuracy and real-time,and has strong application value.The main contents of this paper are as follows:Firstly,the existing methods of snoring recognition were introduced,after the sample data were preprocessed,the unsupervised snoring recognition algorithm based on K-means clustering and the supervised snoring recognition algorithm based on linear SVM were tested.The accuracy of snoring recognition algorithm based on K-means was 73.8%,and the accuracy of snoring recognition algorithm based on linear SVM was 89%.Then a snoring recognition algorithm based on DTW(Dynamic Time Warping)was proposed,which combined template matching and supervised learning.The method calculated the three-dimensional matching distance between audio data and three templates by DTW.The matching distance was manually labeled,and the classification model was obtained by training the labeled data through decision tree.The accuracy can reach 92.5%.Further,in view of the unsatisfactory recognition effect of snore recognition algorithm based on DTW under low signal-to-noise ratio,a snore recognition algorithm based on VMD(variational mode decomposition)was proposed,the average accuracy was 94%.Finally,the snoring recognition system based on DTW was completed on STM32F407-VGT6 hardware platform,and the VMD-based snoring recognition system PC software was completed on the computer side.Using two systems to test the sample data,the accuracy of the DTW-based snoring recognition system was 85.5%,and the accuracy of the VMD-based snoring recognition system can reach 88%,which verify the effectiveness of the proposed method.