This　paper　presents　a　LSTM　neural　network　gesture　recognition　method　based　on　Doppler　radar　technology.　Firstly,　the　velocity　and　distance　of　the　gesture　target　are　obtained　by　2D-FFT,　and　the　antenna　array　expansion　is　realized　by　using　the　high-order　cumulant　to　estimate　the　azimuth　of　the　recognition　target.　Then,　the　relationship　among　distance,　velocity,　azimuth　and　time　is　arranged　to　form　a　data　set.　At　the　same　time,　a　multi-LSTM　fusion　network　model　is　proposed.　Finally,　in　order　to　improve　the　accuracy　of　recognition,　the　influence　of　initial　learning　rate　and　the　number　of　hidden　nodes　on　the　performance　of　multi-LSTM　fusion　network　model　is　analyzed.　The　experimental　data　show　that　when　the　initial　learning　rate　is　0.005　and　the　hidden　layer　node　is　128,　the　average　correct　rate　of　the　six　gestures　can　reach　96　%.　©　2021　IEEE.