The　existing　glaucoma　diagnosis　methods　heavily　rely　on　the　scale　of　training　datasets.　To　solve　this　problem,　in　this　paper,　we　propose　a　deep-transfer　learning　network　to　automatically　diagnose　glaucoma.　This　network　can　better　capture　the　discriminant　glaucoma-related　features　under　limited　supervision.　Particularly,　we　adopt　a　convolutional　neural　network　(CNN)　to　transfer　the　general　features　from　similar　ophthalmic　datasets　and　then　use　the　maximum　mean　discrepancy　to　reduce　the　feature　gap　and　refine　the　specific　features.　To　verify　the　effectiveness　of　the　proposed　method,　we　conducted　experiments　on　real-world　datasets.　Compared　with　other　models,　our　method　achieved　better　classification　performance,　with　an　accuracy,　sensitivity,　and　specificity　of　91.　15%,　90.　13%,　and　92.　25%,　respectively,　having　certain　medical　values　and　importance　for　the　early　screening　of　glaucoma.　©　2023　Editorial　Board　of　Journal　of　Harbin　Engineering.　All　rights　reserved.