Ensemble method using real images, metadata and synthetic images for control of class imbalance in classification.
Ensemble method using real images, metadata and synthetic images for control of class imbalance in classification.
复制标题
DOI：
10.1007/s10015-022-00781-8
复制
发表时间：
2022
影响因子：
0.9
通讯作者：
中科院分区：
文献类型：
--
作者：
二进制分类和异常检测面临着数据集类别不平衡的问题。本文的贡献是提供一个集成模型，提高图像二进制分类，减少少数和多数类之间的类不平衡的数据集。集成模型是真实的图像、合成图像和与真实的图像相关联的元数据的分类器。首先，我们应用生成模型从真实的图像数据集中合成少数类图像。其次，我们训练的集成模型与少数民族类的合成图像，真实的图像，和元数据。最后，我们使用敏感性度量来评估模型的性能，以观察类别不平衡调整所导致的分类差异。通过增加多数类的一半大小来改善少数类的不平衡，我们观察到RESNET 50和DENSENet121的基准预训练模型的分类器灵敏度分别提高了12%和24%。
Binary classification and anomaly detection face the problem of class imbalance in data sets. The contribution of this paper is to provide an ensemble model that improves image binary classification by reducing the class imbalance between the minority and majority classes in a data set. The ensemble model is a classifier of real images, synthetic images, and metadata associated with the real images. First, we apply a generative model to synthesize images of the minority class from the real image data set. Secondly, we train the ensemble model jointly with synthesized images of the minority class, real images, and metadata. Finally, we evaluate the model performance using a sensitivity metric to observe the difference in classification resulting from the adjustment of class imbalance. Improving the imbalance of the minority class by adding half the size of the majority class we observe an improvement in the classifier’s sensitivity by 12% and 24% for the benchmark pre-trained models of RESNET50 and DENSENet121 respectively.
登录
查看更多内容
影响因子：
2.8
作者：
Zunair H;Hamza AB
通讯作者：
Hamza AB
影响因子：
3.3
作者：
Ningrum DNA;Yuan SP;Kung WM;Wu CC;Tzeng IS;Huang CY;Li JY;Wang YC
通讯作者：
Wang YC
影响因子：
10.6
作者：
Shin HC;Roth HR;Gao M;Lu L;Xu Z;Nogues I;Yao J;Mollura D;Summers RM
通讯作者：
Summers RM
影响因子：
10.6
作者：
Yu, Lequan;Chen, Hao;Heng, Pheng-Ann
通讯作者：
Heng, Pheng-Ann
影响因子：
15.8
作者：
Zech JR;Badgeley MA;Liu M;Costa AB;Titano JJ;Oermann EK
通讯作者：
Oermann EK