Indoor intrusion detection based on deep signal feature fusion and minimized-MKMMD transfer learning

Abstract

Indoor intrusion detection based on Wireless Local Area Network (WLAN) has been widely used for security monitoring, smart homes, entertainment games, and many other fields in the Beyond 5G (B5G) wireless communication network environment. Most existing indoor intrusion detection methods directly exploit the Received Signal Strength (RSS) data collected by Monitor Points (MPs) and pay little attention to the instability of WLAN signals in complex indoor environments. In response to this problem, this paper proposes a novel WLAN indoor intrusion detection method based on deep signal feature fusion and Minimized Multiple Kernel Maximum Mean Discrepancy (Minimized-MKMMD). Firstly, the multi-branch deep convolutional neural network is used to conduct the dimensionality reduction and feature fusion of the RSS data, and the tags are obtained according to the features of the offline and online RSS fusion features that are corresponding to the silence and intrusion states, and then based on this, the source domain and target domain are constructed respectively. Secondly, the optimal transfer matrix is constructed by minimizing MKMMD. Thirdly, the transferred RSS data in the source domain is utilized for training the classifiers that are applying in getting the classification of the RSS fusion features in the target domain in the same shared subspace. Finally, the intrusion detection of the target environment is realized by iteratively updating the process above until the algorithm converges. The experimental results show that the proposed method can effectively improve the accuracy and robustness of the intrusion detection system.