The current baggage and article security inspection system uses human eyes to judge, the detection efficiency is low, and there are problems of missed inspections. An improved faster-rcnn cross-layer detection network is proposed, which connects the network layers across layers, collects dangerous object information from multiple angles, and combines the low-level features and high-level features extracted from the network structure to train the network Structure to increase the complexity of the network. Research shows that the method is trained and tested on the data set constructed in this paper, and the results show that the detection effect is good.