基于残差U型网络与双判别网络的图像分割方法  

Image segmentation method based on residual U⁃network and double discriminant networks

作　　者：何佳阳 HE Jiayang(Xi’an Huawei Cloud Computing Technology Co.,Ltd.,Xi’an 710119,China)

机构地区：[1]西安华为云计算技术有限公司,陕西西安710119

出　　处：《电子设计工程》2025年第1期7-11,共5页Electronic Design Engineering

基　　金：国家自然科学基金青年项目(62101312)。

摘　　要：生成对抗网络在图像领域应用十分广泛,但在细胞核图像分割领域中的应用研究相对较少。由于细胞核的精准分割对于病理诊断工作有极大帮助,故提出了基于残差U型网络与生成对抗网络的图像分割方法。该方法以ResUNet网络作为生成网络,Image GAN为双判别网络,训练过程使用双损失函数和冻结策略进行优化。改进后网络在PanNuke数据集上评价指标MioU、Dice、Acc分别为80%、93%、80%,相比ResUNet网络的实验结果分别提升了2.3%、1.7%、2.1%。实验结果证明,改进后网络对细胞核分割具有较好准确率,可作为病理诊断工作的重要依据。Generative Adversarial Networks are widely used in the field of images,but relatively little research has been done on their application in the field of cell nucleus image segmentation.Since the accurate segmentation of cell nucleus is extremely helpful for pathological diagnosis work,an image segmentation method based on residual U⁃network and Generative Adversarial Network is proposed.The network uses ResUNet network as a generative network and Image GAN as double discriminant network,and the training process is optimized using a double loss function and freezing strategy.The evaluation metrics MioU,Dice and Acc of the improved network on the PanNuke dataset are 80%,93%and 80%,which are improved by 2.3%,1.7%and 2.1%,respectively,compared to the experimental results of the ResUNet network.The experimental results proved that the improved network has a good accuracy rate for cell nucleus segmentation,which can be used as an important basis for pathological diagnostic.

关 键 词：图像分割 细胞核图像 生成对抗网络 ResUNet 双判别网络 

分 类 号：TN919.8[电子电信—通信与信息系统]