Abstract: Critical infrastructures,which usually have large flow and key position,are common in infrastructure networks (e.g．power transmission network,Internet)．The performance and reliability of the critical infrastructures directly influence the local abilities of the whole networks．To improve the ability and security-level of infrastructure networks,we proposed a novel method for critical infrastructures detection,which is mainly based on distribution similarity transfer．The aim of the proposed method is to automatically detect the critical routers in the current route．In the real application,due to several factors (e.g．network status,performance of routers),the behaviors of different routers withindifferent routes often belong to different distributions．Therefore,the proposed method models the problem as the distribution similarity transfer among different routes:First,the suspected routers are detected in the target domain (current route) by using spectral clustering；then,a newly proposed distribution similarity transfer classifier finally classifies the suspected routers obtained from the previous step．The proposed method was evaluated on the real dataset provided by Huawei Inst．The experimental results validate the proposed method can effectively detect the critical infrastructures．Meanwhile, it is demonstrated that the proposed method can successfully adopt the distribution similarity transfer to improve the classification results.

Key words: Spectral clustering,Transfer learning,Critical routers detection