摘要
针对眼成像数据的解释需要大量的经验和时间成本,迫切需要自动算法辅助实现眼疾识别和检测的问题,提出基于卷积神经网络的眼疾识别算法。主要使用了AlexNet,VGG和GoogLeNet三种卷积神经模型对患者眼底视网膜图像进行重症近视识别,对3种模型的损失函数值、模型参数量和准确率进行对比实验分析。实验结果表明,卷积神经网络对重症近视具有较好的识别能力,模型在一定程度上可以辅助医生判断,具有一定的临床应用价值。
The interpretation of eye imaging data requires a lot of experience and time,and automatic algorithm is urgently needed to realize eye disease recognition and detection.An eye disease recognition algorithm based on convolutional neural network is proposed.Three convolutional neural models are used to identify patients with severe myopia on the fundus retinal images.The three models are AlexNet,VGG and GoogLeNet.The loss function value,model parameters and accuracy of the three models are compared and analyzed.The experiments results show that the convolution neural network has good recognition ability for severe myopia,and the model can assist doctors to judge to a certain extent,which has certain clinical application value.
作者
娄茹珍
徐丽
蒋正乾
申林山
LOU Ruzhen;XU Li;JIANG Zhengqian;SHEN Linshan(College of Computer Science and Technology,Harbin Engineering University,Harbin 150001,China)
出处
《无线电工程》
北大核心
2021年第11期1202-1207,共6页
Radio Engineering
基金
黑龙江省自然科学基金项目(F2018006)
网络威胁深度分析软件项目(KY10800210013)
2020年工业互联网创新发展工程项目子项目——工业企业网络安全综合防护平台项目(KY10600210025)。
关键词
机器学习
图像分类
卷积神经网络
眼疾识别
machine learning
image classification
convolutional neural network
eye disease recognition
