Detection of Diabetic Retinopathy Based on Target Detection Algorithm

Aiming at the problems of slow detection speed and low accuracy of detection in diabetic retinopathy, this paper first developed a LadderNet vascular segmentation model by using vascular segmentation data. Then the model was used to segment the diabetic retinopathy classification dataset. Then the blood vessels were sharpened and combined with the background of the original fundus image. Then, a VGG19 network was developed using the obtained vascular sharpening data to complete the classification of fundus images. Finally, the lesion area was highlighted by Grad-CAM algorithm and marked in the corresponding position in the original image. Experimental results show that the accuracy of the vascular segmentation model used in this paper on the vascular segmentation dataset can reach 0.9592, and can realize the accurate segmentation of fundus images of blood vessels. At the same time, the accuracy of model classification was improved by 0.1 through the processing of blood vessel segmentation and blood vessel sharpening of fundus images in this paper. This result indicates that the dataset processing method in this paper is of great help to improve the accuracy of the model, and has a good application prospect in improving the accuracy and efficiency of the detection of diabetic retinopathy.