Multi-rotor unmanned aerial vehicles (UAVs) are widely used in both military and civil fields, which poses a serious threat to the military bases and civilian facilities. If the behavior mode of UAVs can be accurately identified, the intention of UAVs will be figured out, which can provide a basis for formulating accurate response measures as early as possible. In this paper, the behavior modes of multi-rotor UAVs are modeled, and the echo characteristics under different modes are analyzed. Then the machine learning dataset based on the time-frequency distribution matrix of the echo is constructed, and the convolutional neural network (CNN) is used to classify and distinguish the flight modes. Finally, the dynamic echoes of electromagnetic computation are simulated and verify the proposed method.