Optimizing　the　state　of　charge　(SOC)　estimation　for　lithium-ion　batteries　contributes　to　more　efficient　energy　management　and　battery　maintenance　strategies.　Ultrasound　has　been　proven　feasible　for　SOC　estimation,　but　current　methods　often　rely　on　a　single　feature　or　model,　which　results　in　incomplete　capture　of　acoustic　signals　and　limited　generalization　under　varying　conditions,　thereby　constraining　their　potential　in　SOC　estimation.　Aiming　at　this　problem,　this　article　proposes　a　multifeature　SOC　estimation　method　using　model　stacking　for　ultrasonic　reflected　waves.　First,　an　experimental　platform　was　developed　to　perform　ultrasonic　inspection　on　a　soft-packed　lithium-ion　battery,　extracting　multifeature　from　the　bottom　surface　wave　and　full　wave　power　spectrum　based　on　waveform　analysis.　Next,　a　multimodel　fusion　approach　is　employed　to　enhance　the　model＇s　estimation　performance.　Initial　estimations　were　performed　using　a　backpropagation　neural　network　and　a　long　short-term　memory　network.　For　secondary　estimation,　the　initial　model　results　and　the　extracted　features　are　fed　into　random　forest　regressor.　Finally,　we　conduct　a　series　of　experiment　to　verify　the　method＇s　superiority.　The　method　demonstrates　an　accurate　SOC　estimation,　achieving　a　maximum　mean　absolute　error　of　1.593%　and　a　maximum　root　mean　square　error　of　1.898%　on　the　test　set,　while　also　exhibiting　strong　robustness　under　complex　conditions.　　©　IEEE.　1982-2012　IEEE.