In　order　to　develop　an　automated　screening　system　for　diabetic　retinopathy　(DR),　a　grading　method　for　non-proliferative　diabetic　retinopathy　(NPDR)　images　based　on　microaneurysms　(MAs)　and　hemorrhages　(HAs)　automatic　detection　is　proposed　according　to　the　new　international　clinical　diabetic　retinopathy　classification　standard.　Firstly,　in　the　red　channel　of　the　retinal　image,　the　adaptive　thresholding,　morphological　technique　and　region-based　active　contour　model　are　adopted　to　accurately　locate　the　optic　disk　center.　Secondly,　after　pre-processing　in　the　green　channel,　a　novel　method　based　on　the　Radon　transform　is　presented　to　recognize　the　true　MAs　and　HAs　lesions.　The　Radon　based　method　effectively　suppresses　the　interference　of　blood　vessels　or　background　texture;　and　the　algorithm　neither　relies　on　the　blood　vessel　segmentation,　no　requires　the　multi-scale　matching　filter.　In　an　experiment,　120　high　resolution　color　images　were　used　to　evaluate　the　performance　of　the　developed　method.　The　experiment　results　show　that　the　proposed　system　achieves　high　sensitivity　and　specificity　for　the　detection　of　both　MAs　and　HAs,　and　the　grading　accuracy　reaches　above　85%　for　the　NPDR　images　without　a　missed　diagnosis.　The　proposed　algorithm　is　stable　and　has　high　practical　value.