Understanding Data Augmentation Saturation: A Comparative Study of CNNs and Vision Transformers for Pollen Grain Classification

• Júlio César da Silva Soares UFPI
• Kelson Romulo Teixeira Aires UFPI
• Juliana do Nascimento Bendini UFPI
• Luis Felipe Ferreira de Castro UFPI
• Rodrigo de Melo Souza Veras UFPI

DOI: https://doi.org/10.5753/eniac.2026.29639

Resumo

Automatic pollen grain classification has advanced with deep learning, but how architecture interacts with progressive data augmentation, especially regarding performance saturation, remains underexplored. We compare DenseNet201, DenseNet201+CBAM, and ViT-B16 on the Cretan Pollen Dataset (CPD1) under progressive augmentation (0–700 synthetic images/class, 10-fold cross-validation). Results show architecture-specific saturation at factors 200 (ViT-B16), 400 (DenseNet201), and 500 (DenseNet201+CBAM), with CNN-based models reaching a peak accuracy of 0.987. CBAM improves stability and delays saturation, offering practical guidance for data-efficient architecture selection in specialized domains such as palynology.

Referências

Astolfi, G., Gonçalves, A. B., Menezes, G. V., Borges, F. S. B., Astolfi, A. C. M. N., Matsubara, E. T., Alvarez, M., and Pistori, H. (2020). POLLEN73S: An image dataset for pollen grains classification. Ecological Informatics, 60:101165.

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al. (2020). An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929.

Flenley, J. (1968). The problem of pollen recognition. Problems in Picture Interpretation, pages 141–145.

Huang, G., Liu, Z., Van Der Maaten, L., and Weinberger, K. Q. (2017). Densely connected convolutional networks. Proceedings - 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, 2017-Janua:2261–2269.

Mahmood, T., Choi, J., and Park, K. R. (2023). Artificial intelligence-based classification of pollen grains using attention-guided pollen features aggregation network. Journal of King Saud University - Computer and Information Sciences, 35(2):740–756.

Sevillano, V. and Aznarte, J. L. (2018). Improving classification of pollen grain images of the POLEN23E dataset through three different applications of deep learning convolutional neural networks. PLOS ONE, 13(9):e0201807.

Tsiknakis, N., Savvidaki, E., Manikis, G. C., Gotsiou, P., Remoundou, I., Marias, K., Alissandrakis, E., and Vidakis, N. (2022). Pollen Grain Classification Based on Ensemble Transfer Learning on the Cretan Pollen Dataset. Plants, 11(7):919.

Woo, S., Park, J., Lee, J.-Y., and Kweon, I. S. (2018). Cbam: Convolutional block attention module. In Proceedings of the European conference on computer vision (ECCV), pages 3–19.

Zu, B., Cao, T., Li, Y., Li, J., Wang, H., and Wang, Q. (2025). RESwinT: enhanced pollen image classification with parallel window transformer and coordinate attention. The Visual Computer, 41(7):4975–4990.