基于改进U-Net模型的咽拭子图像分割研究

Research on Pharyngeal Swab Image Segmentation Based on an Improved U-Net Model

摘要: 为提高咽拭子图像中口腔M区域的分割效果，本文提出一种改进U-Net模型。针对传统U-Net的浅层特征冗余与边缘敏感性不足的问题，引入循环残差模块增强边缘特征传递能力，并结合注意力机制抑制背景噪声，优化多尺度特征融合策略。实验表明，相比先前研究所采用的U-Net方法，改进模型的交并比(IoU)、Dice相似系数(DSC)和像素精确率(PA)分别提高3.8%、2.1%和3.5%，可精准分割复杂背景下低对比度的口腔M目标区域，为咽拭子采样机器人提供高效的视觉支持，降低人工采样风险。

Abstract: This paper proposes an improved U-Net model to enhance the segmentation performance of the oral M region in pharyngeal swab images. To address the issues of redundant shallow features and insufficient edge sensitivity in traditional U-Net models, a recurrent residual module is introduced to enhance edge feature propagation. Additionally, an attention mechanism is incorporated to suppress background noise and optimize the multi-scale feature fusion strategy. Experimental results demonstrate that compared to the U-Net method used in previous studies, the improved model achieves increases of 3.8%, 2.1%, and 3.5% in Intersection over Union (IoU), Dice Similarity Coefficient (DSC), and Pixel Accuracy (PA), respectively, which can enable precise segmentation of the low-contrast oral M region in complex backgrounds, providing efficient visual support for pharyngeal swab sampling robots and reducing the risks associated with manual sampling.