Our nation has a vast territory, which holds variate geology and climate condition. In order to ensure homeland security, it is usually necessary for the relevant personnel to carry out routine inspections, which will consume a lot of manpower and resources. To this end, this paper proposes an aeronautical surveillance method based on deep learning, which uses drones to collect images from high altitude, and uses convolutional neural networks to classify and judge the collected images to monitor the scene. The purpose is to use artificial intelligence to replace the manual inspection by drones, thereby improving the efficiency of homeland security monitoring. To this end, this thesis establishes a database of top-down perspectives containing 10 different scenarios. Through the convolutional neural network model, the image features of different scenes are learned, so that the model can distinguish different scenes. In order to verify the feasibility of this method, this paper carried out experiments on 10 kinds of space-based perspective databases, and the results showed that the classification accuracy reached 97%. This shows that this method can meet the needs of security monitoring and provides ideas for intelligent monitoring.