Abstract: To address the problem of vibration signal denoising in hydropower units, this paper proposed a Feature Mode Decomposition (FMD) method.FMD decomposes signals into distinct modes through a set of predefined filters, which effectively identify the impulsive and periodic characteristics of noise-contaminated signals.First, the FIR filter bank was initialized using the Hann window to establish the initial framework for signal decomposition.Subsequently, a cyclostationary estimation and update process was employed to capture and isolate noise components.Finally, redundant and mixed modes were eliminated during the mode selection stage.For filter parameter optimization, this paper introduced the negentropy of the squared envelope spectrum as the fitness function and utilizes the Sparrow Search Algorithm (SSA) to iteratively determine the optimal FMD parameters.Simulation results and case studies demonstrated that compared to the Variational Mode Decomposition (VMD) algorithm and wavelet transform, the proposed FMD method achieved superior noise elimination in hydropower unit vibration signals.