Design Of Adaptive Pillow Based On Deep Learning

Sleep duration accounts for about one-third of the total life span of humans,and good sleep quality is essential to human health.At present,about one-third of adults have different degrees of sleep problems.These problems are often related to inappropriate pillow height.A large number of studies have shown that the optimal pillow height when a person lies on his side should be greater than the optimal pillow height when lying on his back.However,for a long time,the height of pillows has been fixed and not adjustable.With the emergence and development of smart homes,pillows need to be "smarter".There is an urgent need to design a pillow that can automatically adjust its height according to different sleeping postures.And how to recognize two sleeping postures effectively is the key to solving this problem.Based on the current research status of sleeping posture recognition,the existing sleeping posture recognition methods were summarized.On this basis,two effective sleeping posture recognition schemes were proposed.Data collection platforms were set up to collect corresponding sleeping posture data.Different deep learning network models were designed to recognize sleeping posture.In the end,the best solution was selected to make a pillow product with automatic height adjustment.The main work and research contents of this paper are as follows:(1)Based on the characteristics of the difference in head pressure distribution in different sleeping postures,this thesis proposed a sleeping posture recognition scheme based on head pressure distribution.Firstly,a small area of flexible array piezoresistive sensor was used to collect the sleeper's head pressure distribution data on the pillow.Then convolutional neural networks were used to extract and process the characteristics of the pressure distribution data,and complete the recognition of the sleeping posture.The feasibility of the scheme was proved through experiments.(2)Through the study of the time series model,the down-lying and side-lying sleeping postures that appear during sleeping were divided into 6 states.And a weight-pressure-based sleeping posture recognition scheme was designed.The weighing sensor and the air pressure sensor arranged inside the pillow were used to collect the weight of the user's head and the air pressure of the airbag in the pillow respectively,and a weight-air pressure data frame was generated.A network model composed of a one-dimensional convolutional neural network and GRU was used to identify six sleeping states.At the same time,this paper used different network models to conduct comparative experiments,which proved the superiority of the network model.(3)Based on the principles of comfort and effectiveness,reasonable design ideas for pillow height adjustment and pillow structure were proposed.And based on the above two sleeping posture recognition schemes,corresponding hardware circuits,and data acquisition software were designed and produced respectively,and the corresponding sleeping posture data were collected for experiments.The network module of the weight-pressure-based sleeping posture recognition scheme with better comprehensive performance was transplanted and applied to embedded devices to complete the production of a self-adaptive human sleep pillow.And the feasibility and effectiveness of the pillow were proved through tests.