Font Size: a A A

Exercise Heart Rate Measurement System Based On Long Short-term Memory Networks

Posted on:2020-09-12Degree:MasterType:Thesis
Country:ChinaCandidate:F W RongFull Text:PDF
GTID:2392330596978908Subject:Biomedical engineering
Abstract/Summary:PDF Full Text Request
The heart is an important organ of the human body.It transmits blood to various tissues and organs of the human body through pulsation,and realizes blood circulation of the human body.The heart rate information can reflect the health status of the body,and the heart rate information of the human body can be detected under the state of motion,which can help people adjust the amount of exercise and prevent physical damage caused by strenuous exercise.The existing heart rate detecting device usually uses a photoelectric sensor to acquire a pulse wave,and the ECG waveform collected under the motion state has serious artifacts,which makes it difficult for the wearable device to extract an accurate heart rate value from the pulse wave signal with severe artifacts.In this paper,the photoelectric volume pulse wave tracing(PPG)in heart rate signal detection is combined with artificial neural network to design a heart rate signal detection system based on nRF52832 controller.The main control chip of the heart rate signal detection system,nRF52832,is selected from the NORDIC Bluetooth chip,and the main control chip processes the collected PPG data in real time and transmits the acquired data through the embedded 2.4 GHz transceiver.The ECG data is collected in real time through the ECG detector for training of machine learning algorithms.This paper uses two machine learning algorithms and compares their effects.One is the Stacked Auto-Encoders(SAE)network model,and the other is the LSTM neural network model,which are used to construct the encoding module and decoding module.Based on the above machine learning algorithm,a large number of collected PPG signals and ECG signals are used to train the network model and parameters,and finally the accurate prediction of the PPG signal center rate is realized.This paper combines human body physiological signal detection,electronic information and machine learning and other fields of knowledge,and has completed the following work in the development of human exercise heart rate measurement system:1.Describe systematically of the research background of heart rate information detection system and the development status at home and abroad.In this paper,the specific requirements analysis of the pulse detection device is given,and the overall design scheme is given.The Convolutional Neural Network(CNN),Recurrent Neural Networks(RNN)and SAE machines are studied.Learning algorithm,compare andpropose a suitable algorithm model for pulse information detection;2.Realize the power supply circuit,reset circuit,motion attitude detection circuit,PPG signal acquisition circuit and data wireless transceiver circuit in the pulse measurement system,and explain the principle and function of the circuit;3.Introduce two machine learning frameworks,TensorFlow and TFLearn.Combined with the actual network model of machine learning and the corresponding structural framework scheme,the paper analyzes the data preprocessing,the implementation of network structure and data training prediction and the core program structure.4.Present the SAE and LSTM network data training and testing process.Experiments show that the wireless heart rate detection system and artificial neural network model based on nRF52832 can realize the demand of PPG signal center rate information extraction under severe motion,thus helping people realize science.Reasonable exercise and fitness also have a positive effect on health monitoring during human exercise.
Keywords/Search Tags:Heart rate, PPG signal, TensorFlow, TFLearn, LSTM, SAE
PDF Full Text Request
Related items