Research And Implementation Of Android-based Health Monitoring System

In our country, the situation of population aging is growing increasingly severe and health care has become the topic of most concern for the whole society. Meanwhile,the maldistribution of medical resources also leads to the the difficulty of getting medical service for the elderly. A lot of diseases caused a lot of regret due to the delayed detection and treatment. Thanks to the development of mobile Internet and medical, The use of convenient and wearable equipment for routine health monitoring is more and more widespread. Monitor our physical condition by using the wearable equipment and data analysis methods for early diagnosis and early warning will become a major trend in mobile medical equipment.Among the main diseases that affect the health of the old people, Parkinson's disease is a kind of degenerative disorder of the nervous system that can't be detected easily. In early stage of the disease, there will be regular tremor of upper limb during somatic rest,which can be an important basis for a doctor to diagnose a patient as PD in their early stages.The thesis is based on the deeply understanding of deep learning, achieved the recognition algorithm based on the BP neural network of Parkinson, However, the recognition rate is not high.So we focus on the study of CNN and RNN's characteristics of neural network and apply it to Parkinson's tremor recognition, The main findings are as follows:(1) According to the excellent effect of CNN in the field of image recognition, the acceleration data of the patients were converted into images.The generated spectrograms are used as the input data of CNN and trained.After a number of experiments to adjust the model parameters, the final recognition effect is better than the existed BP neural network Parkinson recognition algorithm.The recognition rate of the algorithm reached 84.4%.(2) By using the advantage of RNN in dealing with the data of time series, the acceleration data set is cut into fragments and than we put them into RNN for training.After a number of experiments to adjust the model parameters, the final recognition rate is between the traditional algorithm and the CNN recognition algorithm, The recognition rate of the algorithm reached 74.8%.(3) The thesis transplanted the best algorithm of the Parkinson tremor recognition to the Android mobile terminal. so we design a CNN-based Parkinson's Disease surveillance system on Android. We use wearable sensors to collect user acceleration data, and it will be transmitted to the client through the BLE, We use API of TensorFlow through JNI in client to realize the recognition of tremor and give the possibility of users suffered from Parkinson. Then the thesis introduce the system development from design to implementation in detail.Finally,we test the system to verify the reliability and applicability of its functions.