Diabetic　retinopathy　is　one　of　the　primary　sources　of　blindness　in　the　developed　society　of　over　50　percent.　The　early　diabetic　diagnosis　of　retinopathy　may　help　prevent　severe　risks　of　vision　loss.　A　great　deal　of　the　analysis　already　done　focuses　on　the　manual　retrieval　of　properties.　Still,　this　essay　attempts　to　diagnose　diabetes　retinopathy　automatically　in　their　multiple　groups　to　include　a　new　approach　focused　on　a　deep-neuronal　network　of　Gray　Wolf　optimization　(GWO).　The　Gaussian　space　theory　and　generic　data　structures　improve　the　accuracy　and　quantity　of　the　dataset　to　overcome　unbalanced　dataset　problems.　Original　and　added　data　sets　are　used　to　train　and　validate　the　proposed　model.　The　model　precision　of　the　convolutional　neural　network　in　this　paper　for　an　increased　data　collection　is　0.9691,　which　is　ideal　compared　to　the　original　data　and　other　procedures.　©　2021　IEEE.