Human Behavior Recognition Model Based On Sensor Data Of Deep Learning

With the improvement of people's living standard,people pay more and more attention to health.On the other hand,with the development of science and technology,the functional smart phones and other electronic products have become very popular.New smart phones have been added with micro sensors,which can provide detailed mobile phone acceleration data.Therefore,by using statistical knowledge to analyze the sensor data,we can identify the behavior actions that users are doing,help users complete the record of behaviors,and then assist medical diagnosis.This application scenario makes the recognition method of human behavior based on mobile phone sensor data more and more attention.In this paper,a CNN LSTM model is constructed to recognize human behavior data in mobile phone motion sensor by using deep learning method.Although the technology of sensor data behavior recognition based on machine learning has developed greatly in recent years,there are still some problems such as low accuracy and relying on artificial prior construction of feature indicators.Because of the large individual differences of human behavior data,the generalization performance of the model is poor.In view of these problems,this paper proposes a multi-layer CNN LSTM model.Compared with the traditional machine learning method,the model can automatically extract features,eliminate the steps of artificial prior feature extraction,and reduce the computational complexity.The model is also compared with the simple structure single layer CNN model and single layer LSTM model.Because of the less parameters of single layer network,feature extraction is not sufficient,and the increased network depth provides more parameter networks,which improves the learning ability and feature extraction ability of the model,so as to learn more details in the data.And the generalization ability of the model is improved by the advantage of the strong adaptive ability of the deep learning method.Finally,the recognition accuracy of CNN LSTM model has improved significantly.CNN is a kind of feature extraction network with excellent effect.Therefore,the model uses double layer CNN as feature extractor.After two convolution operations,CNN fully recognizes data features,not only realizes automatic data extraction,but also greatly reduces the data time step and reduces the computation time required for the subsequent LSTM layer.Sensor data in the field of human behavior recognition is a typical time series data.As an improved layer of cyclic neural network,LSTM is the most suitable for processing time series data.It can summarize all the time step information in the time series data.Therefore,this paper selects multi-layer LSTM as the component of the model.Finally,the model maps the output of LSTM layer into the final classification result through the full connection layer,and identifies the human being's ongoing behavior.In this paper,the UCI har data set and wisdm data set are used to verify the effect of CNN LSTM model.In UCI har data set,two methods are used to segment data sets,which are called full segmentation and half segmentation.The multi-layer CNN LSTM model achieves 92.94%accuracy under the full segmentation method,which is better than the traditional machine learning model and the single layer LSTM model and single layer CNN model used for comparative experiments.It also proves that the model has strong generalization ability.The multi-layer CNN LSTM model has achieved 98.38% accuracy under the semi segmentation method,which proves that the model has strong adaptive ability.On the data set wisdm,the model effect is tested again by semi segmentation method,and the precision rate,recall rate,F1 value and other indexes are introduced to evaluate the model effect comprehensively.It is proved that the model reaches 95.21% accuracy and there is no fitting problem.The results of the two data sets show that CNN LSTM model can recognize common human behavior with high accuracy,and then realize activity recording.