Elderly Emotion Recognition System Based On Deep Residual Network

Mental diseases caused by mood swings are easily overlooked,which are threatening the physical and mental health of the elderly.Therefore,real-time monitoring of emotional state of the elderly is of great importance to prevent and timely resolve the mental illness caused by emotional fluctuations.At present,the most widely used method for emotion recognition is to extract body features.However,the extraction of such features is usually done by manual,and problems such as low levels of features are hard to avoid.The data collected,meanwhile,usually contains a lot of noise and insufficient quantity,which leads to poor recognition performance,emotion recognition system are usually too poor.Based on above backgrounds,this paper studies from following aspects: data collection and preprocessing,emotion recognition algorithm of the elderly,emotion recognition system design and implementation.Firstly,in view of at the chatacteristics of insufficient physiologocal signal data collection.An emotion induction experiment is designed to collects electrical skin signals,we propose a data preprocessing method that converts the time-series data to a two-dimensional picture after removing the baseline drift and then performs two-dimensional interpolation to constitute electrical skin data set.Secondly,aimed at the problem of low feature extraction level and large data noise,proposes a method which combines the asymmetry threshold mechanism with Res Net.By setting a threshold in advance,the amplitude of signal is reduced toward the threshold and will be set zero between threshold interval to improve the network’s ability to process high-noise data.At the same time,so as to avoid the complicated operation by hand,an automatic threshold learning module is embedded in the residual unit.The simulation is carried out with Python,compared with original Res Net.The analysis is carried out in four aspects : accuranct,recall,precision,F1-Score.the simulation results prove that this paper proposes that the Res Net combines with the asymmetry threshold mechanism performance better.Then,in order to make the software system more suitable for the elderly,conbined the investigates and analyzes the needs of the elderly regarding the emotion recognition app.Designed a shallow and broad human-computer interaction model.Designed the system modul structure,combined with the requirements of safety and ease of use,determined the specific functionl structure of the system.At last,In order to improve system response speed,reliability,and security,a dual buffer storage architecture is proposed.This architecture uses memory to improve query speed.It can modify the expiration time of hot data and intercept invalid fields to reduce the chance of cache avalanche,cache penetration,and cache breakdown in the system.Set up a lightweight RPC framework for service calls,based on netty network transmission,achieved a simple and efficient communication protocol.A time-series data storage structure is proposed,which combines a hash table with an ordered set to reduce the query time complexity from O(N)to O(1).Based on the Andriod platform,verify the effectiveness of each functional module of the system...