基于Swin-UNet血细胞分割方法研究

Research on Blood Cell Segmentation Method Based on Swin-UNet

血细胞分割结果是医生诊断病情的一项重要依据。医学检测血细胞方法容易受外界干扰且效率低下,传统图像分割模型精确度低,对背景杂乱的血细胞图像分割效果差。为提高血细胞分割效率与准确性,提出一种基于Swin-UNet改进的血细胞分割算法,首先通过迁移学习引入Swin-UNet在ImageNet上预训练模型参数作为特征提取前端,提高模型的泛化能力;其次根据Swin-UNet算法改进下采样模块归一化函数,提高模型训练速度。实验结果表明,所提方法在精确率、召回率和F1指标上有较大提升,其值分别是97%、98%和97%,相较于传统的UNet分割方法提高3%。

The result of blood cell segmentation is an important basis for doctors to diagnose patient's condition.Medical blood cell detection methods are susceptible to external interference and have low efficiency.Traditional image segmentation models have low accuracy and poor segmentation performance for blood cell images with cluttered backgrounds.To improve the efficiency and accuracy of blood cell segmentation,an improved blood cell segmentation algorithm based on Swin-UNet is proposed.Firstly,Swin-UNe is introduced through transfer learning to pre train model parameters on ImageNet as the feature extraction front-end,improving the model's generalization ability.Secondly,based on the Swin-UNet algorithm,the normalization function of the down-sampling module is improved to improve the training speed of the model.The experimental results show that the proposed method has significant improvements in accuracy,recall,and F1 index,with values of 97%,98%,and 97%,respectively,which is 3%higher than the traditional UNet segmentation method.