EUVIMEDEUVIMED entdecken
Crossref

Data augmentation strategy through key category prioritization for imbalanced image classification

Liyuan Lin, Linyuan Shi, Chaofan Wang, Aolin Wen, Shuxian Zhao, Jian Chen, Weibin Zhou, Yuan Zhou

Measurement Science and Technology · 13.03.2026 · 37 (10) · 105005

DOI: 10.1088/1361-6501/ae4e00 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 16.03.2026. Verfasst von Liyuan Lin und 7 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Class imbalance in engineering applications poses significant challenges to data-driven models, particularly when majority categories dominate the dataset. The scarcity of minority-class samples hinders effective learning of discriminative features, resulting in degraded classification accuracy for underrepresented categories. To address this challenge, we propose an effective data augmentation (DA) strategy. First, we propose an innovative measurement method that focuses on key categories within minority groups. By adjusting the sample sizes of critical categories, we achieve targeted DA, improving overall performance while enhancing classification efficiency and sample utilization. Second, we develop an imbalanced random sampling method to systematically generate datasets with different imbalance levels, which facilitates rigorous evaluation of the proposed strategy and generalizes well to real-world distribution scenarios. In addition, our strategy remains robust across different enhancement methods and classification networks. Extensive experiments conducted on multiple datasets fully demonstrate the effectiveness of our method and provide valuable guidance for future research on imbalanced data challenges. Specifically, on the highly imbalanced APDDD dataset, classification accuracy improves from 98.075% to 99.519% (+1.444%). More importantly, recall for the key categories identified by our measurement method increases from 48% to 96%, achieving a 48-percentage-point gain.

Zitieren

APA

Lin, L., Shi, L., Wang, C., Wen, A., Zhao, S., Chen, J., Zhou, W., & Zhou, Y. (2026). Data augmentation strategy through key category prioritization for imbalanced image classification. Measurement Science and Technology, 37(10), 105005. https://doi.org/10.1088/1361-6501/ae4e00

Vancouver

Lin L, Shi L, Wang C, Wen A, Zhao S, Chen J, et al. Data augmentation strategy through key category prioritization for imbalanced image classification. Measurement Science and Technology. 2026;37(10):105005. doi:10.1088/1361-6501/ae4e00

BibTeX
@article{Lin2026,
  author = {Lin, Liyuan and Shi, Linyuan and Wang, Chaofan and Wen, Aolin and Zhao, Shuxian and Chen, Jian and Zhou, Weibin and Zhou, Yuan},
  title = {Data augmentation strategy through key category prioritization for imbalanced image classification},
  journal = {Measurement Science and Technology},
  year = {2026},
  volume = {37},
  number = {10},
  pages = {105005},
  doi = {10.1088/1361-6501/ae4e00}
}
Kostenlos starten →