GLER-Unet: An ensemble network for hard exudates segmentation

The detection of hard exudation in diabetic retinopathy is a hot topic medical image segmentation. Aiming at the irregular shape and different size lesion area Hard Exudates segmentation task common few-shot learning challenge task, Global-Local Ensemble Robust U-Net proposed. network consists Global Contour Extraction for extracting long-range semantics exudates contour which use complete training, Local Refined Feature Segmentation local refined rules patch Revise fusing features extracted by first two networks generating binary masks. proposed method obtains DICE, TPR PPV 0.8741, 0.8752, 0.8730 0.8960, 0.8964, 0.8956 respectively on E-Ophtha IDRiD. At same time, methods shows strong robustness cross dataset testing, better than other baseline models.