Research On Behavior Recognition Technology Based On Deep Neural Network And RFID

In recent years,with the development of Internet of Things technology,deep learning has been widely studied in the field of human-computer interaction.Radio Frequency Identification(RFID)technology is widely used in the field of behavior identification.RFID can provide a more reliable source of data,and the deep neural network in deep learning can obtain deeper abstract information from the data set.Therefore,in order to improve the accuracy of behavior recognition,this thesis combines deep learning algorithms with RFID technology.An outlier detection algorithm based on distance and density is proposed to preprocess RFID data.A fusion model combining Convolutional Neural Networks(CNN)and Long Short-Term Memory Neural Networks(LSTM)is used to train the processed data.The detection accuracy of the deep fusion model and the robustness of the algorithm are ensured by continuously adjusting the model parameters,so as to realize the recognition and classification of individual behaviors.The main tasks of this thesis are:(1)Aiming at the problem that the raw data collected by RFID tags contains a lot of noise information,this thesis uses k-Nearest Neighbor(kNN)-based outlier detection algorithm for data preprocessing.First,the kNN algorithm is introduced to divide the data set into different regions for outlier detection.Then,the hierarchical adjacency order is proposed to calculate the average sequence distance,thereby redefining the local outlier factor.Finally,the detection effect of the algorithm in this thesis is compared with the traditional algorithm,and the performance analysis shows that the proposed outlier detection algorithm can improve the accuracy of detecting outliers.(2)In order to improve the accuracy of behavior recognition,this thesis uses the hierarchical representation capabilities of deep convolutional neural networks to design a new behavior recognition fusion model to distinguish behaviors in the RFID environment.First of all,this thesis fuses CNN and LSTM.As two types of neural networks with excellent performance in deep learning algorithms,CNN can extract spatial features and LSTM can extract time series features.Then,the attention mechanism is introduced into the LSTM for weight distribution,and the recognition accuracy of the algorithm is improved through continuous training.Finally,the model in this thesis is compared with the training results of a single neural network.Based on performance indicators such as recall rate,accuracy rate and comprehensive evaluation,it shows that the fusion model in this thesis has good performance in recognizing confusing actions.(3)Based on the data preprocessing algorithm and behavior recognition model in the RFID environment,this thesis designs and implements an RFID behavior recognition prototype system.According to the data collected by RFID tags and the neural network fusion model to extract and classify the data features,the module functions of the system are designed and the overall frame structure of the system is constructed.Finally,a behavior recognition system based on RFID and deep learning models is established,and system tests show that the algorithm in this thesis is highly efficient in behavior recognition.