Research On Indoor Intrusion Detection Algorithm Based On WLAN

The intrusion detection refers to the anomaly detection of the environment by perceiving the impact of the target on the radio wave without carrying the signal transceiver.It has potential application value in smart home,police security,enterprise management,national defense and military and other fields.Traditional vision-based and image-based detection techniques are influenced by light and line-of-sight propagation,and the sensor-based and radar-based detection technology requires special hardware equipment,which limits the promotion of intrusion detection technology.The intrusion detection technology based on wireless local area networks(WLAN)makes use of wireless network widely deployed in indoor environment,which has aroused concern in recent years because of its low-cost,good concealment and strong environmental adaptability.Existing intrusion detection methods contain the following characteristics: Firstly,normal links and abnormal links are not distinguished in data preprocessing phase.The signal of the link may be abnormal because of co-frequency interference,multipath effect and other factors,while the abnormal signal will affect the detection performance of the system.Secondly,the type of characteristics is less.The existing detection methods mainly use time-domain statistical characteristics,but use less frequency-domain transform characteristics.However,frequencydomain characteristics are often more effective than time-domain characteristics in representing signal changes.Third,it is difficult for the existing detection methods to take account of both training time cost and system detection performance.Based upon the conditions described above,this thesis presents an indoor intrusion detection algorithm based on WLAN.The main contents include:Firstly,this thesis starts to research the abnormal link detection algorithms.This thesis places the emphasis on obtaining the probability distribution function of each data stream in silent state by using kernel density estimation method.The Kullback-Leibler divergence is used to measure the distribution differences between links.Finally,anomalous links are detected according to the distribution differences.Secondly,the algorithm of effective characteristic subset selection is studied.To solve the problem of single characteristics type,the sliding window function is used to extract statistical characteristics in time domain and transform characteristics in frequency domain to describe different environmental states.At the same time,in order to avoid the redundancy between characteristics and the inability of the same characteristic combination to adapt to different experimental scenarios,unsupervised characteristic selection method based on mutual information is used to select the effective characteristic combination.Thirdly,the research on intrusion detection algorithm is carried out.The one-class classification,that is support vector data description(SVDD),is applied to construct intrusion detection system based on silent signal samples.The parameters of SVDD are obtained by combining cross validation with particle swarm optimization.And the final intrusion detection is judged by joint multi-link decision results.The results in the complex indoor environment and open corridor environment show that the proposed algorithm can effectively improve the detection performance of the system while guaranteeing low training overhead.