In　this　paper,　An　improved　algorithm　for　the　extreme　learning　machine　is　proposed　and　applied　to　SAR　target　recognition.In　order　to　solve　the　influence　of　the　noise　and　spatial　distribution　of　the　training　samples　on　the　calculation　of　the　classification　plane,　different　penalty　factors　are　given　to　different　training　samples,and　according　to　this,　the　＂weighted　extreme　learning　machine＂　is　proposed.　And　then,the　kernel　function　is　introduced　into　the　＂extreme　learning　machine＂　to　improve　the　ability　of　nonlinear　function　approximation.　Considering　that　the　general　training　algorithm　of　the　weighted　extreme　learning　machine　is　slow　and　consumes　a　lot　of　computer　memory　when　the　number　of　training　samples　is　large,　a　training　method　based　on　conjugate　gradient　algorithm　is　proposed.　The　test　on　＂banana　benchmark　data＂　shows　that　the　weighted　extreme　learning　machine　based　on　the　conjugate　gradient　method　can　complete　the　convergence　in　the　number　of　iterations　far　less　than　the　number　of　samples,　and　the　calculation　speed　is　much　faster　than　the　traditional　algorithm.　Finally,　this　proposed　algorithm　is　applied　to　SAR　target　recognition.　The　test　on　MSTAR　data　set　shows　that　the　proposed　algorithm　is　not　only　extremely　fast　in　SAR　target　recognition,　but　also　has　better　recognition　performance　than　support　vector　machine.　general　limit　learning　machine.　BP　neural　network　and　other　algorithms.