Empirical wavelet transform （EWT） is inadequate in the early weak diagnosis fault of planetary gearboxes under strong noise background， mainly due to the improper segmentation of the signal spectrum， which cannot effectively determine the number of modal components. Consequently， an early fault diagnosis method of planetary gearbox is proposed， which combines the modified empirical wavelet transform （MEWT） and adaptive sparse coding shrinkage （ASCS） algorithms. According to the scale space representation of signal frequency spectrum， the original fault signal is adaptively decomposed into a series of narrowband intrinsic mode components. Envelope spectrum kurtosis （ESK） value is used to select the sensitive component. To further highlight fault information， the ASCS algorithm is applied to sparse noise reduction for sensitive component， thus the obvious characteristic frequency information can be extracted from its envelope spectrum. Through the analysis of numerical simulation and actual data， this method can adaptively decompose the planetary gearbox fault signal and enhance weak fault impulse characteristics. In addition， compared with EWT， EWT-ASCS， and ASCS， the proposed method can effectively extract the component that contains abundant fault information， and then the signal fault characteristics are highlighted after ASCS processing， thereby realizing the accurate identification of early weak fault of planetary gearbox.