Empirical　wavelet　transform　(EWT)　is　a　signal　decomposition　method　that　distinguishes　signals　from　the　frequency　domain.　When　processing　non-stationary　and　strong　noise　signals,　a　large　number　of　invalid　components　may　be　obtained,　or　modal　aliasing　may　occur.　The　biggest　contribution　of　sparsity-guided　multi-scale　empirical　wavelet　transform　(SMSEWT)　is　that　it　can　optimize　the　segmentation　method　and　extract　useful　frequency　band,　reduce　the　number　of　invalid　components,　and　suppress　modal　aliasing.　In　order　to　divide　frequency　bands　containing　similar　information　into　final　components,　Fourier　spectrum　will　be　divided　equally　and　used　to　calculate　kurtosis.　Frequency　bands　with　similar　kurtosis　are　considered　to　contain　the　same　kind　of　information,　which　will　be　combined　to　achieve　adaptive　segmentation　of　the　spectrum.　Subsequently,　empirical　wavelet　filters　will　be　constructed　and　the　time-domain　waveforms　of　each　frequency　band　can　be　obtained.　Using　sparsity　to　select　envelope　components　containing　abundant　periodic　pulses　can　diagnose　bearing　faults.