The incipient fault characteristics of planetary gearbox are weak and difficult to effectively identify under strong background noise interference and variable working conditions. To address these issues， an improved feature mode decomposition （IFMD） algorithm is proposed to extract the weak fault characteristics of planetary gearbox under time-varying speed conditions. Firstly， for the key input parameters of the FMD algorithm， such as the number of decomposition mode n， the number of filter K， and the length of filter L， which need to be set manually and lack adaptability， an adaptive scale space spectrum segmentation method is proposed to determine the required number of decomposition modes n. On this basis， the Spectral Gini Index （SGI） is used as the objective function， and particle swarm optimization algorithm is used to automatically determine the optimal filter number K and filter length L. Subsequently， the IFMD is applied to perform optimal modal decomposition on the fault signal under the optimal parameter combination， and the decomposed component with the highest SGI value is selected as the sensitive modal component. Finally， significant fault feature orders are extracted from the envelope order spectrum of sensitive component to accurately diagnose the fault type and location of planetary gearbox. The analysis results of variable speed simulation signals and engineering experimental data indicate that compared to the PSO-VMD method， MED method， SGMD method， and fast spectral kurtosis method， the proposed method can extract weak fault information more clearly and comprehensively， thereby improving the characterization ability and diagnostic accuracy of early fault features of planetary gearbox under time-varying speed conditions.