The invention belongs to the technical field of radio signal identification, and particularly relates to an ultrashort wave specific signal identification method based on a convolutional neural network. The method comprises the following steps: performing short-time Fourier transform for a specific signal in a sample library, and acquiring a signal time-frequency spectrum, wherein the specific signal is a signal containing a frame synchronization code in a signal transmission data frame structure; training a convolutional neural network model with the time-frequency spectrum; and identifying the specific signal in ultrashort wave communication through the trained convolutional neural network model. In the method provided by the invention, firstly, visual characteristics, presented on the time-frequency spectrum, of the specific signal are analyzed, and training is executed through the convolutional neural network model, thus, identification of the ultrashort wave specific signal is realized, and signal identification rate is improved; and finally, through a simulation experiment, influence of aliasing interference on an ultrashort wave channel is reduced effectively, ultrashort wave specific signal identification under low signal-to-noise rate is realized, moreover, anti-interference performance can be improved through optimizing network structures and increasing the number ofnetwork layers, so the method provided by the invention has relatively strong practical application value.