摘要
深度学习网络在医学图像分割领域应用广泛,传统语义分割模型只考虑局部像素点,在小目标的医学图像语义分割中检测精度不高。提出了基于Unet的双任务图像语义分割模型,对传统的Unet语义分割进行改进,编码阶段采用经过预训练的Resnet34作为框架进行特征提取,设计了SCSE模块对图像特征信息进行修正,从空间和通道两个方向获取图像的全局信息,损失函数采用“分类”和“分割”融合的多任务策略进行学习,对气胸医学图像进行语义分割。为进一步提高网络模型的泛化能力,对数据集图像进行随机水平翻转、垂直翻转等图像增强处理。实验表明该语义分割方法比传统的Unet语义分割方法在分割精度上提高5%以上。
Deep learning network is widely used in the field of medical image segmentation.The traditional semantic segmentation model only considers local pixels,the detection accuracy is not high for the semantic segmentation of small target medical image.We propose a two task image semantic segmentation model based on UNET,which improves the traditional UNET semantic segmentation.In the encoding stage,we use Resnet34 backbone for feature extraction,design SCSE module to modify the image feature information,and obtain the global information of the image from the two directions of space and channel.The loss function uses the fusion of“classification”and“segmentation”.In order to improve the generalization ability of the network model,the dataset image is enhanced by random horizontal flipping,vertical flipping and other image enhancement processing.Semantic segmentation is performed on the pneumothorax images.The results show that the semantic segmentation method improves the segmentation accuracy by more than 5%compared with the traditional UNET semantic segmentation method.
作者
沈旭东
楼平
吴湘莲
朱立妙
雷英栋
SHEN Xudong;LOU Ping;WU Xianglian;ZHU Limiao;LEI Yingdong(College of Intelligent Manufacturing,Jiaxing Vocational&Technical College,Jiaxing Zhejiang 314036,China;Department of Mechanical and Automotive Engineering,Tongji Zhejiang College,Jiaxing Zhejiang 314051,China)
出处
《电子器件》
CAS
北大核心
2022年第3期618-622,共5页
Chinese Journal of Electron Devices
基金
嘉兴市科技计划项目(2020AD10027,2018AY11012)。
关键词
语义分割
气胸医学图像
多任务
semantic segmentation
medical image of pneumothorax
multi-task
