Research On Wireless Gesture Recognition Based On Deep Learning

Device-Free Gesture Recognition(DFGR)based on wireless is a emerging technology in recent years.The technology is a new target state perception method without any equipment carried by the human.It only uses the masking effect of the target on the wireless network to analyze the target state information contained in the wireless signal affected by the target,so as to realize the recognition of human gesture.With the continuous development of wireless technology,Wi Fi,5G,mm Wave are all around us,which provide basis of researching and apply.Meanwhile,DFGR does not involve privacy leakage,and could work under dark or obstructed environment in a device-free manner.The technique has broad application prospects in human–machine interaction,smart home,intelligent space,etc.Device-Free gesture recognition is essentially a pattern recognition problem,and thus extracting discriminative features from affected wireless signals is the key to solve DFGR problem.Due to its powerful feature extraction capability,deep learning is expected to be used to solve the DFGR problem.In this article,a deep learning-based wireless gesture recognition method is proposed,which uses deep network to extract deep features from constructed RF images,so as to achieve high-performance recognition of human gestures.The construction of wireless gesture recognition model based on deep learning often requires a large number of training samples.However,collecting training samples is a timeconsuming and laborious process,and it is even difficult to collect gesture samples in some practical application scenarios.Motivated by the excellent ability of the generative adversarial network in synthesizing samples,in this article,we propose a DFGR algorithm based on generative adversarial network.Specifically,we first design a single scenario network to generate a large number of virtual samples using a few number of real samples.Then,we further develop a scenario transferring network to generate virtual samples by utilizing the real samples not only from the current scenario but also from another available scenario as well,which could improve the quality of synthesized samples with the extra knowledge learned from another scenario.In the wireless gesture recognition method,the same human gesture will have different effects on the wireless signal in different scenarios.Therefore,due to the feature shift incurred by different scenarios,a DFGR system typically achieves poor performance in a new application scenario.In this article,we propose a DFGR algorithm based on domain adaptation,which uses self-adaptive adversarial learning strategy to achieve the local alignment of the feature distribution of the target scenario and the source scenario,so as to achieve highperformance cross-scenario DFGR by unsupervised learning.In this thesis,a large number of experiments are designed to verify the feasibility and superiority of the two new DFGR algorithms.The experimental results show that the recognition accuracy of the proposed wireless gesture recognition algorithm based on generative adversarial network can reach at least 92.5% under the condition of small samples,and the recognition accuracy of the proposed wireless gesture recognition algorithm based on domain adaptation can reach at least 91.4% under the condition of cross scenario.