【摘要】 本研究旨在实现对中医红外热成像面部图像的精准分割。使用Resnet50代替传统U-Net网络的主干特征提取模块，移除特征融合中复制与剪切里的剪切操作。该方法能优化特征融合，避免梯度问题，并提高模型通用性。分割实验表明，与传统U-Net相比，该方法具有更高的平均交并比mIoU值和准确率，mIoU值达98.20%，准确率达99.03%。该方法为基于红外图像的中医辅助诊断和疗效评估提供了技术支持。更多还原

【Abstract】 This study aims to achieve accurate segmentation of facial images in TCM infrared thermography. Resnet50 is used to replace the backbone feature extraction module of the traditional U-Net network, and the cut operation is removed from copy and cut in feature fusion. This method optimizes feature fusion, avoids gradient problems, and improves model versatility. The segmentation results show that compared with the traditional U-Net, this method has a higher mean intersection over union(mIoU)value and accuracy, with mIoU reaching 98.20% and accuracy reaching 99.03%. This method provides technical support for infrared image-based TCM auxiliary diagnosis and treatment evaluation.更多还原