【作者】 蒋舒颖； 张俊然； 刘彦； 刘云飞； 孙昂；

【Author】 Jiang Shuying;ZhangJunran Liu;Yan Liu;Yunfei Sun;ang;College of Electrical Engineering,Sichuan University;

【机构】 四川大学电气工程学院；

【摘要】 为了解决目前白细胞图像数据集总体样本量少、有效样本不足的问题,提出了一种基于主动迭代优化的白细胞图像分类方法。本文在私有白细胞图像数据集的基础上,通过预处理对图像进行变换,以ResNet作为分类框架,加入迁移学习初始化网络参数。在训练过程中加入主动迭代优化方法,充分挖掘样本信息,在每轮迭代中提取对模型具有优化价值的样本加入后续训练,通过对总量有限的带标签样本进行训练,最大化提高模型对特征的判别能力和分类效果。实验结果表明,本文提出的方法在白细胞图像的分类上取得了较好的结果,准确率达到99.8%。更多还原

【Abstract】 In order to solve the problem that the current leukocyte image dataset has a small overall sample size and insufficient effective samples, a leukocyte image classification method based on active iterative optimization is proposed. Based on the private white blood cell image dataset, this paper transforms the image through preprocessing, uses ResNet as the classification framework, and adds transfer learning to initialize the network parameters. In the training process, the active iterative optimization method is added to fully mine the sample information. In each round of iteration, samples with optimization value for the model are extracted and added to the subsequent training. The discriminative ability and classification effect. The experimental results show that the method proposed in this paper has achieved good results in the classification of white blood cell images, with an accuracy rate of 99.8%.更多还原