Efficient Ensemble of CNN and Transformer Models for Cassava Leaf Disease Classification

• Kaique O. A. dos Santos UEA
• Raimundo Correa de Oliveira UEA

DOI: https://doi.org/10.5753/eniac.2025.12386

Abstract

This work proposes an ensemble approach for the multiclass classification of cassava leaf diseases, integrating two architectures with complementary characteristics in representational capacity and computational efficiency: CropNet, based on MobileNetV3, and Swin Transformer. The images used were collected in the field, reflecting real-world variations in lighting, background, and positioning. The models were evaluated through stratified K = 5 fold cross-validation, using the F1-score macro as the main metric, given the imbalanced class distribution. Data augmentation techniques were dynamically applied during training to improve generalization capacity. The ensemble outperformed the individual models, achieving a mean F1-score macro of 0.8329 and an accuracy of 0.9087. These results demonstrate predictive robustness and indicate potential for accurate visual diagnoses in agricultural contexts with computational constraints.

References

Albuquerque, L. and Guedes, E. (2023). Um comparativo de abordagens com redes neurais artificiais para detecção inteligente de patologias na folha do café. In Anais do XIV Workshop de Computação Aplicada à Gestão do Meio Ambiente e Recursos Naturais, pages 131–140, Porto Alegre, RS, Brasil. SBC.

Batista, L. M. S. and de Paiva, M. S. B. (2019). Sistemas agroflorestais no município de capitão poço: motivações de implementação na comunidade do barro vermelho. Orientador: Prof. Dr. José Sebastião Romano de Oliveira; Coorientadora: Prof.ª Msc. Ana Paula Dias Costa.

Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. (2009). Imagenet: A large-scale hierarchical image database. [link]. CVPR 2009.

Embrapa Mandioca e Fruticultura (2023). Mandioca. Acesso em: 5 maio 2025.

Fathima, M. and Bondili, H. S. S. (2025). A comprehensive survey on cassava disease detection and classification using deep learning models. In SCT Proceedings in Interdisciplinary Insights and Innovations, volume 3, pages 377–393. SCT.

Google AI and TensorFlow Hub Team (2021). Cropnet: Cassava disease classification model. [link]. Model available at TensorFlow Hub. Accessed: 2025-05-25.

Howard, A., Sandler, M., Chu, G., Chen, L.-C., Chen, B., Tan, M., Le, Q. V., and Adam, H. (2019). Searching for mobilenetv3. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 1314–1324.

John, A. A. (2022). Identification of diseases in cassava leaves using convolutional neural network. In 2022 Fifth International Conference on Computational Intelligence and Communication Technologies (CCICT), pages 1–6.

Junior, A. C., da Silva, F., and Rios, R. (2024). Deep learning-based transfer learning for classification of cassava disease. In Anais do XXI Encontro Nacional de Inteligência Artificial e Computacional, pages 364–375, Porto Alegre, RS, Brasil. SBC.

Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., and Guo, B. (2021). Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 10012–10022.

Loshchilov, I. and Hutter, F. (2019). Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101.

Ministério da Agricultura e Pecuária (MAPA) and Embrapa (2023). Projeções do agronegócio: Brasil 2022/23 a 2032/33. [link]. Acesso em: 5 maio 2025.

Müller, R., Kornblith, S., and Hinton, G. (2020). When does label smoothing help? In Advances in Neural Information Processing Systems, volume 33, pages 4694–4703.

Mwebaze, E., Mostipak, J., Joyce, Elliott, J., and Dane, S. (2020). Cassava leaf disease classification. [link]. Kaggle Competition Dataset.

Sadaiyandi, J., Arumugam, P., Sangaiah, A. K., and Zhang, C. (2023). Stratified sampling-based deep learning approach to increase prediction accuracy of unbalanced dataset. Electronics, 12(21):4423.

Shorten, C. and Khoshgoftaar, T. M. (2019). A survey on image data augmentation for deep learning. Journal of Big Data, 6(1):1–48.

Silva, W., Soares, B., Almeida, V., Viana, L., Pastori, P., Magalhães, D., and Rocha, A. (2024). Detecção da praga spodoptera frugiperda no cultivo de milho usando armadilhas inteligentes e visão computacional. In Anais do XV Workshop de Computação Aplicada à Gestão do Meio Ambiente e Recursos Naturais, pages 61–70, Porto Alegre, RS, Brasil. SBC.