| 基于多尺度融合注意力网络的视网膜血管分割方法 |
 点此下载全文 |
| 引用本文：黎翠1,2，张旭刚1,2.基于多尺度融合注意力网络的视网膜血管分割方法[J].计算技术与自动化,2026,(2):86-93 |
| 摘要点击次数: 47 |
| 全文下载次数: 20 |
|
|
| 中文摘要:视网膜血管分割对眼底疾病的诊断和治疗至关重要，早期的检测与干预可以减轻视力障碍，甚至避免失明。然而，由于特征提取不足会导致血管信息丢失，准确的分割血管仍是一个巨大挑战。基于此，提出了一种基于多尺度融合注意力网络的视网膜血管分割方法。首先，设计了一个多分支残差模块，使得低层特征与高层特征能更好聚合，提高网络的特征提取能力。其次，设计了一个并行扩张卷积模块，并将其嵌入到解码端与编码端中，增强模型感受野同时获取更丰富的上下文信息。最后，设计了一个多尺度融合注意力模块来加强特征提取。模型在DRIVE、STARE和CHASEDB1三个公开的眼底图像数据集上进行实验，实验结果表明，与基线网络相比，模型的分割准确率均有提高，其分割灵敏度分别提升了0.28%，1.29%和1.3%。 |
| 中文关键词:视网膜血管分割 眼底图像 多尺度融合注意力 |
| |
| Retinal Vessel Segmentation Method Based on Multi-scale Fusion Attention Network |
|
|
| Abstract:Retinal vessel segmentation is crucial for the diagnosis and treatment of fundus disorders, and early detection and intervention can alleviate visual impairment and even prevent blindness. However, Since insufficient feature extraction can lead to the loss of vascular information,accurately segmenting blood vessels remains a significant challente. Based on this, this paper proposes a retinal vessel segmentation method based on multi-scale fusion attention network. First, we design a multi-branch residual module, which enables better aggregation of low-level features with high-level features and improves the feature extraction capability of the network. Next, a parallel dilation convolution module is designed and embedded into the decoding and encoding ends to enhance the modeling sensory field while acquiring richer contextual information. Finally, a multi-scale fusion attention module is designed to enhance feature extraction. The model in this paper is experimented on three publicly available fundus image datasets, DRIVE, STARE, and CHASEDB1,compared to the baseline network, the experimental results show that the model’s segmentation accuracy has improved across the board, its segmentation sensitivity has increased by 0.28%,1.29% and 1.3%, respectively. |
| keywords:retinal vessel segmentation fundus image multi-scale fusion attention |
| 查看全文 查看/发表评论 下载pdf阅读器 |
|
|
|
