基于改进U-Net网络的视杯视盘联合分割方法  

Joint Segmentation Method for Optic Cup and Optic Disc Based on Improved U-Net Network

作　　者：于海波 刘振宇[1] YU Haibo;LIU Zhenyu(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]沈阳工业大学信息科学与工程学院,沈阳110870

出　　处：《微处理机》2025年第2期56-60,共5页Microprocessors

摘　　要：为更好实现青光眼早期筛查诊断中视杯视盘的联合精准分割,针对杯盘比(CDR)评估,提出一种改进的U-Net卷积神经网络Seg-UNet。该网络使用ResNet50作为编码层,增强图像的特征提取能力;加入模拟人类视觉的融合感受野模块,增大网络对全局信息的感知能力。借鉴高分辨率网络(HRNET)的思想设计了多尺度特征融合模块来融合上下文语义信息。使用公开数据集REFUGE对所提出方法进行性能验证和比较。结果表明,本方法在REFUGE数据集上分割视杯和视盘的效果优于现有分割方法。To achieve more accurate joint segmentation of the optic cup and optic disc in the early screening and diagnosis of glaucoma,an improved U-Net convolutional neural network,Seg-UNet,is proposed for the evaluation of the cup-to-disc ratio(CDR).The network employs ResNet50 as the encoding layer to enhance the feature extraction capability of images.A simulated human visual fusion receptive field module is incorporated to increase the network's perception of global information.Drawing on the concept of the High-Resolution Network(HRNet),a multi-scale feature fusion module is designed to inte-grate contextual semantic information.The proposed method is validated and compared using the publicly available REFUGE dataset.The results demonstrate that this method outperforms existing segmentation methods in segmenting the optic cup and optic disc on the REFUGE dataset.

关 键 词：视杯视盘联合分割 U-Net网络 语义分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]