This　paper　presents　a　daily　behavior　identification　algorithm　based　on　sEMG　to　improve　the　accuracy　of　behavior　identification.　In　the　preprocessing　stage,　the　original　sEMG　signal　is　effectively　denoised　by　the　combination　of　EMD　denoising　and　wavelet　denoising.　In　the　feature　extraction　stage,　the　characteristics　of　MAV　and　AR　model　are　extracted　by　time-frequency　domain　to　express　the　behavior　patterns.　In　the　behavior　classification　stage,　8　features　from　4　sEMG　channels　of　MAV　and　AR　model　are　use　an　input　neurons　of　the　BP　neural　network　to　improve　the　accuracy　of　behavior　classification　identification.　Through　the　learning　of　a　large　number　of　training　samples,　the　accuracy　of　the　behavioral　identification　on　the　test　samples　comes　to　91.02%　in　the　experiment,　which　indicates　that　the　daily　behavior　identification　based　on　sEMG　is　a　valuable　method.