摘要
传统的视盘检测需要手动提取视盘几何、血管和自身属性等特征,该方法在一定程度上依靠经验,且耗时耗力。基于卷积神经网络(CNN),使用两种网络结构,在两种不同大小的数据集上,自动提取视网膜图像中视盘的特征,并依据CNN特征进行视网膜图像中视盘的自动检测。研究采用深度学习架构Caffe,在公开的眼底图像数据集上进行了实验验证。结果表明,此方法简单易行,准确率高达98.04%,超过了现有方法。同时,通过实验得出了一些有益的结论,为进一步的研究工作奠定了基础。
Traditional Optic Disk (OD) detection is based on manual extracting features, which include OD' s geometry, blood vessels and own properties. OD is detected by these features. Unfortunately, this method is based on experience and fortune to some degree, the worse is that it is time -consuming and complicated. To address this problem, a method based on Convolution Neural Network (CNN) is proposed in this paper, which can automatically obtain the features of OD in retinal image on two kinds of nets and two different size data sets, then to locate OD via CNN characteristics of retinal image. The experiments on public fundus image data show that the accuracy rate reach 98.04% by using deep learning architecture Caffe. At the same time, experimental results show that method proposed in this study is simpler and more effective than the previous ones. What' s more, some useful conclusions are obtained through the analysis and explanation of the experimental data, which lay a solid foundation for further research.
出处
《贵州师范学院学报》
2017年第3期27-32,共6页
Journal of Guizhou Education University
基金
贵州省科技厅联合基金(黔科合LH字[2014]7597号)
遵义医学院硕士启动基金(编号:F-641)
关键词
视网膜图像
CNN特征
视盘
自动检测
Retinal image
CNN feature
Optic disk
Automatic detection