Activity Recognition Based On Channel State Information

The intellectualization of coal preparation plants puts forward new requirements for the construction of coal preparation plants.In terms of personnel supervision,it is also necessary to use intelligent technologies to improve supervision efficiency.In the process of intelligent construction of the coal preparation plant,Wi Fi signals are used to perceive the various dangerous activities of employees,which can conduct seasonable and accurate supervision of employees at low cost and high efficiency,thereby protect the lives of employees,and provides guarantee for safe production of coal preparation plants.Primarily a channel state information(CSI)path component extraction algorithm is proposed to obtain CSI path components.The path component signal can describe the state of channels more fine-grained and sensitively.The thesis explains the whole process of the CSI path component extraction algorithm.The algorithm first uses the CSI signal data of different subcarriers at the same time to construct the Hankel matrix.Then constructs an equation to convert the component extraction problem into a mathematical problem.The roots of the equation are several CSI signal components on the propagation paths.The human body detection experiment verifies the sensitivity of the CSI path component when detecting human motion on non-line-of-sight(NLOS).Then a classification network model with more comprehensive feature extraction capabilities was designed for activity classification and recognition.The network model uses the autonomous learning ability of deep learning to extract the spatial and temporal features from input data.To a certain extent,this solves the problem of incomplete description of human movements when using artificial features.The model adopts a dual-stream structure.The 2DCNN network module and the Conv LSTM network module extract primary spatial features and primary temporal features on the twostream branches respectively.Finally the activity recognition results are outputted according to these features.This network model is used to conduct activity recognition experiments.The model has achieved good classification and recognition accuracy when classifying human activities occur on NLOS,which improves the sensitivity of the human activity recognition system.The paper has 35 pictures,9 tables,and 83 references.