基于DenseNet迁移学习的年龄相关性黄斑变性图像的分类方法研究

Research on the Classification Method of Age-related Macular Degeneration Images Based on DenseNet Transfer Learning

针对年龄相关性黄斑变性图像的分类研究,提出采用DenseNet迁移学习的图像分类方法.对原始图像进行归一化、限制对比度自适应直方图均衡化等预处理方法,采用旋转、剪裁等数据增强方法扩增数据.在DenseNet网络模型基础上,采用数据集ImageNet首先对DenseNet网络模型进行预训练,然后将训练后得到的网络模型予以迁移,在做增强后的目标数据集上进行微调训练.结果表明:采用迁移学习方法的DenseNet网络模型不仅可以快速收敛,而且可以达到99.31%的分类准确率,整体性能优于对DenseNet直接训练方法.

Aiming at the research on classification of age-related macular degeneration images, an image classification method based on DenseNet transfer learning is proposed in this paper.The original image is preprocessed by normalization, limited contrast adaptive histogram equalization and other preprocessing methods, and the data is amplified by data enhancement methods such as rotation and clipping.Based on the DenseNet network model, firstly, the data set ImageNet is used to pre train the DenseNet network model.Then, the trained network model is migrated, and fine-tuning training is carried out on the enhanced target data set.The results show that the DenseNet network model using transfer learning method can not only converge quickly, but also achieve 99.31% classification accuracy, and the overall performance is better than the direct training method on DenseNet.