This　paper　presents　the　implementation　of　an　autonomous　aerial　manipulation　using　a　hexacopter　equipped　with　a　two　DOF　robotic　arm.　The　kinematic　and　dynamic　models　are　developed　by　considering　the　dynamic　characteristics　of　the　combined　manipulation　platform.　A　novel　adaptive　sliding　mode　controller　is　proposed　for　both　position　and　velocity　control.　By　building　SSD　(Single　Shot　Detection)　deep　neural　network　based　on　deep　learning,　an　object　detection　solution　is　developed.　The　three　dimensional　coordinates　of　the　target　object　relative　to　the　multirotor　are　obtained　via　combining　the　vertical　plane　position　obtained　from　object　detection　law　with　the　depth　value　from　stereo　camera.　Finally,　with　the　proposed　controller　and　object　detection　law,　an　autonomous　flight　experiment　is　accomplished　including　approaching,　grabbing　and　delivering　the　target　object.　The　proposed　approaches　are　demonstrated　with　effectiveness　and　could　be　utilized　in　various　manipulation　applications.　©　2018　IEEE.