| Obstructive sleep apnea/hypopnea syndrome(OSAHS)is a sleep-related respiratory disease with a prevalence of 3%to 7%in adult males and 2%to 5%in adult females over the world.The scientific research which focus on snoring acoustic analysis to reveal the pathological characteristics of patients with OS AHS has been more than ten years of history,but few studies are related to the determination of the source of snoring(Soft palate,epiglottis,tongue base and other vibration parts).Drug-induced sleep endoscopy allows clinicians to observe patterns of upper airway obstruction in patients with sleep disorders,however the sound recordings of snoring are inevitably contaminated with environmental noise.To solve these problems,based on the National Natural Science Foundation of China(61271410):"The recognition of snoring signal resource and upper airway obstruction site from snoring crowd based on the acoustic analysis",we propose a cascade generalized sidelobe canceller and convolutional neural network approach to establish a snoring signal processing system.A generalized sidelobe canceller-based adaptive enhancement technique is used to suppress various ambient noise from medical staffs and equipments in the operating room for providing a more pure snoring signal and improving recognition performance.Convolutional neural network is a deep learning method which relys on a large amount of marker data,independent of the known experience.This method makes it possible to automatically learn the feature representation and classifier which are most suitable for the task.Finally,the performance evaluation of the whole system shows that the optimal network model has 89%accuracy for identifying 5 different sources of snoring data,and the recognition accuracy of each category is similar.The accuracy of the measured array data and the denoised data is also only a difference of 6%,indicating that the noise reduction process retains important snoring signal characteristics. |
PDF Full Text Request