Super　Resolution　Convolutional　Neural　Network　(SRCNN)　solves　the　problems　of　poor　robustness　and　complex　calculation　of　traditional　image　super-resolution　reconstruction　algorithm,　but　its　training　data　set　and　the　number　of　layers　of　neural　network　is　relatively　small,　and　the　edge　and　texture　detail　information　are　not　handled　well.　For　the　above　problems,　the　Maxout　activation　function　is　adopted　in　this　paper　to　avoid　the　problems　encountered　by　traditional　activation　functions　such　as　gradient　disappearance　or　overflow.　Then　the　combination　of　Maxout　and　Dropout　can　train　large　data　set　and　deepen　neural　network.　Experimental　results　show　that,　compared　with　the　classical　algorithm,　the　algorithm　proposed　in　this　paper　can　train　a　large　amount　of　data,　improve　the　quality　of　reconstructed　images　and　the　generalization　ability　of　the　network　model,　and　can　enhance　the　robustness　of　the　model.　©　2019　IEEE.