Abstract: For problems of single data source, reliance on manual experience to judge danger existed in current early warning system, perception technology of Internet of Things was applied to the early warning system of dangerous sources to establish a cluster-tree wireless sensor network and hardware structures of sensor nodes and sink node were designed. Algorithm of information fusion and processing based on Relief algorithm and BP neural network was studied and simulated. The experiment results show that the application of the perception technology of Internet of Things in early warning system of dangerous sources can improve efficiency of data acquisition and coverage of environmental monitoring, and reduce false alarm rate and undetected rate.