一种改进的I-Unet网络的皮肤病图像分割算法  

An improved skin disease image segmentation algorithm based on I-Unet network

作　　者：蒋宏达 叶西宁[1] JIANG Hongda;YE Xining(College of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学信息科学与工程学院

出　　处：《现代电子技术》2019年第12期52-56,共5页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(60974066)~~

摘　　要：黑色素瘤是常见的皮肤癌,皮肤病图像分割在皮肤癌诊断过程中起到至关重要的作用。为了利用I.Unet深度神经网络强大的编码解码功能来自动分割出皮肤病病灶区域,文中提出一种改进的I.Unet网络的皮肤病图像分割算法。该方法采用空洞卷积扩大卷积感受野,利用类Inception和循环神经网络(RCNN)分别提取图像不同尺度的特征,并进行多尺度特征融合,运用全连接条件随机场(CRF)进行图像后处理。结果表明,所提算法在皮肤病图像分割中取得了良好的效果,算法的Jaccard系数达到了0.780,Dice系数稳定在0.871;与同类最佳研究结果相比,Jaccard系数及Dice系数分别提高了1.5%,2.2%,表明该方法有效提升了网络图像分割的性能。Melanoma is the most common form of cancer,and the skin disease image segmentation plays an important role in the diagnosis process of the skin cancer. Therefore,an improved skin disease image segmentation algorithm based on the IUnet network is proposed in this paper,so as to automatically segment out the focus area of the skin disease by utilizing the powerful encoding and decoding function of the I-Unet deep neural network. In the method,the dilated convolution is used to expand the receptive field of the convolution. The classification Inception and recurrent convolutional neural network(RCNN)are used to respectively extract different scales of image features and conduct multi-scale feature fusion. The fully-connected conditional random field(CRF)is used for image post-processing. The results show that the proposed algorithm can achieve a good effect in image segmentation of the skin disease,with its Jaccard coefficient reaching 0.780,and Dice coefficient stabilizing at 0.871,respectively improved by 1.5% and 2.2% in comparison with the best-in-class research results,which indicates that the method can effectively improve the performance of network image segmentation.

关 键 词：皮肤病 I-Unet网络 图像分割 空洞卷积 特征融合 全连接条件随机场 

分 类 号：TN711.34[电子电信—电路与系统] TP391[自动化与计算机技术—计算机应用技术]