Radar signal modulation mode identification method based on improved attention mechanism

The invention discloses a radar signal modulation mode identification method based on an improved attention mechanism, and the method comprises the steps: taking sampling data in a modulation period as a radar signal sample, marking each radar signal sample, and obtaining modulation type labels corresponding to the radar signal samples one by one; according to the method, I/Q complex signals of radar signals are directly adopted, various radar signal features do not need to be manually extracted, the method does not depend on priori knowledge of experts in the radar field, a four-head attention mechanism is adopted to independently carry out shallow feature learning on four types of data of a real part, an imaginary part, real-virtual combination and a module part of the I/Q complex signals, and then mutual fusion is carried out to obtain deep features; according to the method, a radar signal modulation mode identification model comprising a convolutional neural network, an improved multi-head attention mecha