Semantic Understanding Of Indoor Behavior Based On Multi-task Learning Network Of Radio Frequency Signals

With the widespread popularity of intelligent devices and wireless networks,more and more network sensing technologies have been applied to human-computer interaction and scene sensing.However,action semantics are often closely related to the scene in which the action takes place,and single action recognition cannot meet the semantic understanding requirements of the action in a specific scene.Focusing on the behavioral semantic understanding of cross-domain actions,which is a key technology in the field of intelligent perception,this paper designs Wisenet,a multitask learning network combining attention mechanism,and proposes an architecture of scene understanding system based on indoor wireless signals.The research contents and contributions of this paper include the following three aspects:Firstly,the attention mechanism based on convolutional neural network is designed.In the sub-tasks of domain and action recognition,the attention mechanism based on minimum pooling and maximum pooling is used to focus the static signal and dynamic signal respectively,so that the weight of the same signal in different tasks is different.Secondly,the multi-task learning network Wisenet,which is composed of three subtask networks,is designed to realize behavior recognition in multiple scenarios using multitask neural network.The gradients from the three tasks of domain,position and action provide discrimination for Wisenet training at the same time,realize the mining of hidden information,and realize the parameter sharing among networks through the hard sharing mechanism.At the same time,an advanced training termination algorithm(PTA)is designed to solve the unbalanced allocation of training resources caused by the large difference in loss scales of multiple tasks.Finally,data collection is carried out in the real environment,data sets are constructed,and data are trained and tested using the network proposed in the paper.The accuracy of training for understanding behavioral semantics by using Wisenet is over 95%,and the accuracy of testing is over 83%.