Clinical　research　shows　that　glaucoma　is　primarily　caused　by　pathological　changes　in　the　optic　nerve　structure,　which　may　bring　about　irreversible　damage　of　sight.　In　relevant　literature　research,　the　cup-to-disc　ratio　(CDR)　is　mainly　used　as　an　important　indicator　for　glaucoma　detection,　which　needs　to　segment　optic　disc　(OD)　and　optic　cup　(OC)　region　clearly　and　accurately.　However,　due　to　the　low　contrast　image　of　boundary,　the　segmentation　of　OD　and　OC　is　still　a　challenging　problem.　In　this　paper,　a　novel　hybrid　model　based　on　the　ensemble　random-forest　deep-neural-network　(RF-DNN)　is　proposed　for　OD　and　OC　segmentation,　which　can　calculate　more　accurately　for　glaucoma　detection.　At　the　same　time,　the　advantages　of　DNN　and　ensemble　RF　are　combined　to　carry　out　the　corresponding　feature　extraction　and　classification,　which　employs　the　winner-takes-all　strategically　to　the　segmentation　and　classification　for　glaucoma　detection.　Finally,　the　experiment　result　shows　that　the　proposed　method　has　reached　the　best　evaluation　level　in　terms　of　OD　and　OC　segmentation　results　on　ORIGA　dataset　and　SCES　dataset,　which　achieves　highest　diagnosis　accuracy　with　AUC　of　0.96　and　0.98　on　ORIGA　dataset　and　SCES　dataset,　respectively.　©　2020　IEEE.