基于改进FCN的肺炎图像分割方法
Pneumonia image segmentation method based on improved FCN
邹显迪 1何小利 1余谦 1龙源 1张博1
作者信息
- 1. 四川轻化工大学计算机科学与工程学院,四川宜宾 644000
- 折叠
摘要
针对胸部X射线影像中肺炎病灶识别工作量大,结果不够精准等问题,提出一种基于改进FCN的肺炎病灶图像分割方法.首先,构建Pascal数据集格式的健康肺部影像和感染肺炎影像的数据集.其次,对比不同ResNet网络和传统VGG网络训练损失的收敛速度.然后使用效果最好的ResNet50网络代替经典FCN算法中VGG网络作为主干网络,并提出一种多尺度特征提取模块,最后将改进的FCN网络与传统FCN网络、LR-ASPP、DEEPLAB-V3进行对比,改进的FCN网络较其他方法得到了更好的效果.实验结果表明,改进的FCN网络可以精准分割胸部X射线中各种形状和大小的肺炎病灶,分割效果良好,可以为临床的肺炎诊断提供可靠依据.
Abstract
Aiming at the problem that the recognition of pneumonia lesions in chest X-ray images is heavy workload and the results are not accurate enough,this study proposed an image segmentation method of pneumonia lesions based on improved FCN.First,data sets of healthy lung images and infected pneumonia images in Pascal dataset format were constructed.Secondly,the convergence rate of training loss between different ResNet networks and traditional VGG networks is compared.Then ResNet50 network with the best effect was used to replace the original VGG network in the classic FCN algorithm as the backbone network,and a multi-scale feature extraction module was proposed.Finally,the improved FCN network is compared with the traditional FCN network,LR-ASPP,DEEPLAB-V3.The experimental results show that the improved FCN network can accurately segment pneumonia lesions of various shapes and sizes in chest X-ray,and the segmentation effect is good,which can provide a reliable basis for clinical diagnosis of pneumonia.
关键词
FCN/图像处理/语义分割/膨胀卷积/多尺度提取/残差网络Key words
FCN/image processing/semantic segmentation/dilatational convolution/multi-scale extraction/residual network引用本文复制引用
基金项目
自贡市2022年重点科技计划项目(2022ZCYGY16)
四川轻化工大学研究生创新基金资助项目(Y2023120)
出版年
2024
NETL
NSTL
万方数据