In　order　to　extract　the　early　weak　impact　fault　information　in　gearbox　vibration　signal　submerged　in　strong　background　noise,　a　new　sparse　signal　decomposition　method　using　dual-BP　is　proposed.　Firstly,　the　short-time　Fourier　transform　using　the　window　functions　with　different　lengths　is　adopted　to　perform　the　signal　sparse　decomposition,　and　the　initial　decomposition　coefficients　are　obtained;　secondly,　the　dual-BP　is　used　to　optimize　the　obtained　decomposition　coefficients;　lastly,　the　obtained　coefficients　are　reconstructed,　and　the　sustained　oscillation　component　and　the　failure　impact　component　of　the　signal　are　obtained;　further,　the　failure　impact　component　is　analyzed　and　the　diagnose　result　is　obtained.　The　analysis　results　of　simulation　signal　and　application　examples　confirm　the　feasibility　and　validity　of　this　method,　which　provides　a　new　idea　for　the　mechanical　fault　signal　extraction　in　strong　noise　environment.　©,　2014,　Science　Press.　All　right　reserved.