Robot　cooperating　with　humans　is　a　hot　research　topic　recently.　And　the　first　challenge　is　how　to　specifically　identify　humans＇　intention.　In　this　paper,　we　propose　a　two-step　approach　based　on　LSTM　and　SSD　for　identifying　humans＇　specific　intention.　A　LSTM　with　attention　mechanism　is　firstly　presented　for　recognizing　humans＇　actions　based　3D　skeleton　data;　then,　the　hand-held　objects　are　detected　by　a　SSD　neural　network.　Finally,　a　human-robot　interaction　framework　are　constructed　based　on　the　proposed　intention　identification　model　and　DMP　model.　The　experiments　show　that　the　proposed　method　can　effectively　identifying　humans＇　intention.　And　the　robot　accomplishes　to　interact　with　the　human　with　the　built　motion　database　and　DMP.　©　2019　IEEE.