基于卷积神经网络的人体行为识别的开发与设计

Development and design of human behavior recognition basedon convolutional neural networkAbstract With the continuous innovation and breakthroughs in the computer field and theincreasing demand for computer applications in other fields,human-computer interaction has become aconcern of people.The emergence of deep learning has helped people achieve human-computerinteraction,and convolutional neural network is one of its representative algorithms,which is widelyused in the field of computer vision to realize picture recognition.Human behavior recognition refers tofeeding a series of data into a trained neural network,and the computer performs feature extraction onthe data and then recognizes and classifies the data,including data such as videos,image sequences,orsensor data.The network structure used in this paper includes 4 convolutional layers,1 fully connected layer,1maximum pooling layer,and 1 average pooling layer,using one-dimensional convolution,Reluactivation function,Softmax and Dropout technology.The data set used is the first version of the data set released by the WISDM laboratory.A total of36 people participated in the test.The acceleration sensor is used and the sampling frequency is 20HZ.The data set contains 6 types of behaviors:Downstairs,Jogging Jog,Sitting,Standing,Upstairs andWalking.The first part of this article analyzes the background and significance of research on humanbehavior recognition,and introduces the domestic and foreign research status.Next,we introduce ANNand two simple and representative network structures in detail,including related contents of neuronsand activation functions.Then elaborate the relevant theoretical knowledge of convolutional neuralnetworks,including the traditional convolutional neural network structure,and then focus on the CNNnetwork model structure used in this article and how to use this neural network model to achieve humanbehavior recognition on the WISDM data set.Finally,the performance of the model is analyzed,and conclusions and suggestions are made,andplanning prospects.Keywords:Convolutional neural network,human behavior recognition