Noise reduction of underwater acoustic signal is of great significance in improving signal recognition capabilities of sonar system. In this paper, as a new decomposition algorithm, modified uniform phase empirical mode decomposition (MUPEMD) is proposed to improve the calculation efficiency and alleviate mode aliasing problem. After being decomposed by MUPEMD, mode components can be classified adaptively through critical entropy value. Dynamic zero point interval filtering method based on genetic algorithm (GA-DZPIF) is proposed to filter single mode component. Structure similarity of phase diagram is established as a new evaluation index to avoid the subjective difference which can enhance the persuasiveness of experimental result. Finally, a new noise reduction method for ship radiated noise signal based on MUPEMD, GA-DZPIF and entropy cross curve is proposed. Good denoising performance has been achieved for measured ship radiated noise signal, and it has a wide range of practical application.