Diabetic Retinopathy Detection Algorithm Based on Improved YOLOv8

The existing methods of detecting Microaneurysms (MA) and Hemorrhage (HA) are not yet mature, and timely detection of Microaneurysms (MA) and Hemorrhage (HA) in color photos is helpful for the diagnosis of Diabetic Retinopathy (DR) and the prevention of visual impairment and even blindness. In view of this situation, an Efficient Channel Attention and RepNCSPELAN (ERep-YOLO) network based on YOLOv8 was proposed on the basis of YOLOv8. Firstly, the original Cross-stage Partial fusion (C2f) convolutional structure is replaced by a lightweight generalized high-efficiency layer aggregation network RepNCSPELAN, and a new backbone is formed with the attention mechanism Efficient Channel Attention (ECA) to improve the network's feature extraction ability for very small targets and enhance multi-scale learning. Secondly, the Generalized Intersection over Union (GIOU) loss is used to replace the Complete Intersection over Union (CIOU)loss to enhance the border regression performance during network training. Compared with YOLOv8, the improved Efficient Channel Attention and RepNCSPELAN (ERep-YOLO) network has a 12.3% increase in mAP, a training speed of 11 frames per second, and a 16% reduction in the number of network parameters. In summary, compared with the original YOLOv8 network, the proposed model reduces the number of model parameters and improves the detection accuracy and detection speed, which is more conducive to the detection of Diabetic Retinopathy (DR) lesions.