In　this　paper,　an　automatic　detection　for　time-frequency　map　of　Morse　signal　is　proposed　base　on　machine　learning.　Firstly,　a　preprocessing　method　based　on　energy　accumulation　is　proposed,　and　the　signal　region　is　determined　by　nonlinear　transformation.　Secondly,　the　feature　extraction　of　different　types　of　signal　time-frequency　maps　is　carried　out　based　on　the　graphics.　Finally,　a　signal　detection　classifier　is　built　based　on　the　feature　matrix.　Experiments　show　that　the　classifier　constructed　in　this　paper　has　the　generalization　ability　and　can　detect　the　Morse　signal　in　the　broadband　shortwave　channel,　which　improve　the　accuracy　of　Morse　signal　detection.