Abstract: Since the ship radiated noise contains the information of its tonnage, speed, propeller characteristics, etc, it has been the focus and difficulty of the study of ship recognition. So it has great significance to our country and livelihood. In this paper, for the ship radiated noise, we analysis the ship recognition under the Alpha stable distribution noise. Choose tangent non-linear transformation method to restrain the influence of the background noise. Then extract 1.5 order spectrum features and put them into the recognition machine based on particle swarm optimization method. Simulation results prove that when the mixed signal noise ratio is higher than -5dB, the ship radiated noise recognition accuracy reaches higher than 90%. This method is also effective for actual received ship radiated noise.