With　the　rapid　development　of　social　economy　and　information　technology,　human　physiological　characteristics　such　as　fingerprints,　face,　palm　print,　iris,　retina,　etc.　have　been　widely　used　in　the　field　of　commercial　biometrics.　In　recent　years,　the　dynamic　physiological　characteristics　of　human　body,　such　as　ECG,　heart　sound　and　voice,　have　been　proved　to　be　applicable　to　biometrics.　This　paper　mainly　studies　the　feature　extraction　and　classification　of　ECG　signals.　First,　the　ECG　signal　is　periodically　segmented　to　obtain　the　time-domain　feature　matrix,　and　the　periodic　signal　is　wavelet-transformed　to　obtain　the　frequency　domain　feature　matrix.　Then　PCA-ICA　is　used　to　perform　latitude　reduction　on　the　feature　matrix.　Finally,　the　parameters　of　the　fuzzy　decision　tree　for　modeling　are　intelligently　set　by　the　PSO　algorithm.　And　experimental　verification　on　the　MIT-Bill　standard　ECG　database.