In　order　to　enhance　the　adaptive　ability　of　Q　factor　wavelet　and　realize　the　multi-resolution　decomposition　of　signal　in　the　analysis　filter　bank,　a　sparse　feature　extraction　method　based　on　the　multi-resolution　decomposition　of　Q　factor　wavelet　is　proposed.　In　this　method,　the　multi-order　binary　analysis　filter　banks　are　firstly　constructed　by　using　the　Q　factor　wavelet,　and　then　the　optimal　sub-band　is　selected　by　optimizing　the　iterative　Q　factor.　Then,　the　shock　interval　of　the　optimal　sub-band　is　selected　as　the　atom,　and　the　atom　forms　a　complete　dictionary　through　toeplitz　extension　to　realize　the　sparse　decomposition　of　the　signal.　Finally,　the　sparse　signal　is　analyzed　by　envelope　demodulation,　and　the　fault　characteristic　frequency　can　be　extracted　effectively,　which　proves　that　the　sparse　signal　has　the　ability　to　express　fault　features.　The　simulation　and　experimental　results　show　that　this　method　can　effectively　extract　sparse　feature　of　signals　compared　with　DCT　and　DHT　dictionaries.　It　not　only　overcomes　the　weakness　of　adaptive　ability　of　traditional　complete　dictionaries,　but　also　can　effectively　express　sparsely.　©　2019　IEEE.