• 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:窦 宇,陈宏远,谭华超,袁贵鸿,江彦博,刘 丹.基于改进UCTransNet的海洋微藻图像分割模型[J].软件工程,2024,(2):31-35.【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
基于改进UCTransNet的海洋微藻图像分割模型
窦 宇1, 陈宏远1, 谭华超1, 袁贵鸿1, 江彦博1, 刘 丹1,2
(1.大连海洋大学信息工程学院, 辽宁 大连 116023;
2.大连海洋大学设施渔业教育部重点实验室, 辽宁 大连 116023)
601096472@qq.com; 18473001460@163.com; tanhuachao_work@163.com; 858774793@qq.com; 1426101935@qq.com; liudan@dlou.edu.cn
摘 要: 海洋微藻是海洋生态系统的基石,对其进行分割识别可以监测海洋水质并防治藻华。UCTransNet是使用Transformer模块替代UNet中跳跃连接模块的分割模型,但UCTransNet过于重视图像的通道信息而忽略了图像的空间信息。针对此情况,提出一种将空间与通道融合的注意力机制,并将其加入UCTransNet中,得到CSAM-UCTransNet。该模型加强了编码器与译码器之间的联系。实验表明,CSAM-UCTransNet对海洋微藻样本的分割精度相较于UCTransNet提升了4.88%。与UNet、Attention-UNet、UNet++等分割算法相比,该模型分割精度更高,对细节的处理效果更好。
关键词: 海洋微藻;图像分割;UNet网络;UCTransNet网络;注意力机制
中图分类号: TP391    文献标识码: A
A Segmentation Model for Marine Microalgae Images Based on Improved UCTransNe
DOU Yu1, CHEN Hongyuan1, TAN Huanchao1, YUAN Guihong1, JIANG Yanbo1, LIU Dan1,2
(1.School of In f ormation Engineering, Dalian Ocean University, Dalian 116023, China;
2. Facilities Key Laboratory of Ministry of Fisheries and Education, Dalian Ocean University, Dalian 116023, China)
601096472@qq.com; 18473001460@163.com; tanhuachao_work@163.com; 858774793@qq.com; 1426101935@qq.com; liudan@dlou.edu.cn
Abstract: Marine microalgae are the cornerstone of marine ecosystems, and their segmentation and identification can monitor marine water quality and prevent algal blooms. UCTransNet is a segmentation model that replaces the skip connection module in UNet with the Transformer module. However, UCTransNet places so much emphasis on the channel information of images that it ignores the spatial information of images. Aiming at this problem, this paper proposes a Channel-Space Attention Module (CSAM), and it is added to UCTransNet to obtain CSAM-UCTransNet. This model strengthens the connection between the encoder and decoder. The experiment shows that CSAMUCTransNet improves the segmentation accuracy of marine microalgae samples by 4.88% compared to UCTransNet. Compared with other segmentation algorithms, such as UNet, Attention-UNet and UNet ++ , the proposed model has higher segmentation accuracy and better handling of details.
Keywords: marine microalgae; image segmentation; UNet network; UCTransNet network; attention mechanism


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