• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:郁华鑫,何利文.基于UNet的图像分割研究[J].软件工程,2024,27(10):50-53.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
基于UNet的图像分割研究
郁华鑫, 何利文
(南京邮电大学物联网学院, 江苏 南京 210003)
1221077026@njupt.edu.cn; helw@njupt.edu.cn
摘 要: 为了提高CT图像中病灶分割的准确性,辅助医生快速诊断和制订治疗方案,文章采用基于UNet网络结构的深度学习算法,提出了一种在跳跃连接中融入通道域和空间域注意力机制的方法。此方法增强了高层次特征对低层次特征的指导,以达到对小目标病灶的关注。同时,为了提高模型性能,提出了CrossEntropyLoss和DiceLoss的混合损失函数。实验结果表明,改进后的UNet模型分割平均准确率达到90.86%,相较于传统的UNet和SegNet模型,分别提升了3.01百分点和2.38百分点,表现出更高的像素准确率及更快的收敛速度。
关键词: 图像分割;UNet;注意力模块;损失函数
中图分类号: TP399    文献标识码: A
Research on Image Segmentation Based on UNet
YU Huaxin, HE Liwen
(School of Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing 210003, China)
1221077026@njupt.edu.cn; helw@njupt.edu.cn
Abstract: In order to improve the accuracy of lesion segmentation in CT images to assist doctors in rapid diagnosis and treatment decision-making, this paper employs a deep learning algorithm based on the UNet network structure and proposes a method that incorporates channel and spatial attention mechanisms into the skip connections. This method enhances the guidance of high-level features on low-level features, allowing for better attention to small target lesions. Additionally, to improve model performance, a hybrid loss function combining Cross Entropy Loss and Dice Loss is introduced. Experimental results show that the modified UNet model achieves an average segmentation accuracy of 90.86% , which is 3.01 percentage points and 2.38 percentage points higher than traditional UNet and SegNet, respectively, showing higher pixel accuracy and faster convergence speed.
Keywords: image segmentation; UNet; attention module; loss function


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫