Ensemble　empirical　mode　decomposition　(EEMD)　overcomes　the　mode　mixing　problem　of　EMD,　however　the　end　effect　still　exists.　Additionally,　two　key　parameters　in　EEMD:　the　amplitude　of　added　noise　and　the　number　of　ensemble　times　are　set　up　by　experience,　not　conducive　to　complete　the　signal　decomposition　fast　and　accurately.　To　solve　the　above　problems,　a　method　of　adaptive　KEEMD　was　proposed　in　this　paper.　First,　the　end　effect　was　restraimed　through　an　extrema　extension　based　on　the　kernel　extreme　learning　machine　(KELM)　and　mirror　extension,　then　it　was　applied　to　the　simulate　signal　decomposition　and　wheat　reflectance　spectrum　denoising,　and　the　effectiveness　of　restraining　end　effect　was　verified　by　the　result.　Second,　through　the　high　frequency　component　of　signal　obtained　by　above　step,　the　two　adaptive　parameters　in　KEEMD　was　obtained.　The　adaptive　method　was　used　to　cole　and　spinach　reflectance　spectrum　denoising.　Results　show　that　denoising　results　can　be　obtained　more　quickly　and　be　consistent　with　the　non-adaptive　method.　©　2016,　Beijing　University　of　Technology.　All　right　reserved.