电子学报

• • 上一篇    

基于语义与形态特征融合的语义分割网络

魏博文1, 全红艳2   

  1. 1.华东师范大学软件工程学院,上海 200062
    2.华东师范大学计算机学院,上海 200062
  • 收稿日期:2022-01-13 修回日期:2022-03-15
  • 作者简介:魏博文 男,1996年6月出生,安徽六安人.现为华东师范大学软件工程学院硕士研究生主要研究方向为计算机视觉、深度学习.E-mail:51194501070@stu.ecnu.edu.cn
    全红艳(通讯作者) 女,1968年6月出生,黑龙江哈尔滨人,华东师范大学计算机科学与技术学院副教授, 硕士生导师,研究方向为计算机视觉,人工智能.当前研究的兴趣领域为基于深度学习的医学影像分析、医学影像三维重建.

Semantic Segmentation Network Based on Semantic and Morphological Feature Fusion

WEI Bo-wen1, QUAN Hong-yan2   

  1. 1.Software Engineering Institute,East China Normal University,Shanghai 200062,China
    2.School of Computer Science,East China Normal University,Shanghai 200062,China
  • Received:2022-01-13 Revised:2022-03-15

摘要:

视网膜血管检测有助于医生诊断视网膜疾病,而以往基于特征融合的算法难以解决视网膜血管检测中出现的漏分割问题,且分割准确率较低.本文对特征融合方式做出进一步探索,并提出一种基于语义与形态特征融合的算法,通过挖掘输入特征中蕴含的语义与形态信息,建模特征间的相关关系.随后,使用特征融合模块实现多模态特征自适应地融合.在公开数据集DRIVE以及STARE上的实验结果表明,文章算法优于现有的语义分割模型,尤其在敏感性上,比传统U-Net网络提升了8.20%.

关键词: 语义分割, 视网膜血管检测, 形态特征, 语义特征, 卷积模块, 特征融合

Abstract:

Retinal blood vessel detection is helpful for doctors to diagnose retinal diseases, but the previous algorithm based on feature fusion is difficult to solve the problem of missed segmentation in retinal blood vessel detection, and the segmentation accuracy is low. This paper further explores the feature fusion method and proposes an algorithm based on the fusion of semantic and morphological features. It models the correlation between features by mining the semantic and morphological information contained in the input features. Then, the feature fusion module realizes the adaptive fusion of multi-modal features. The experimental results on the public datasets DRIVE and STARE show that, the article algorithm is better than the existing semantic segmentation model, especially in sensitivity, which is 8.20% higher than the traditional U-Net.

Key words: semantic segmentation, retinal vessel detection, morphological features, semantic features, convolution unit, feature fusion

中图分类号: