Parameter-Efficient Coffee Leaf Disease Classification Using Quaternion-Valued EfficientNets

• André Y. S. Okimoto Fundação Paulo Feitoza
• Arnaldo Matuchewski Fundação Paulo Feitoza
• Allan Junior Fundação Paulo Feitoza
• Hendrio Braganca Fundação Paulo Feitoza
• Italo Pelição Caliari Fundação Paulo Feitoza
• Antonio Fontenele Fundação Paulo Feitoza
• Wington L. Vital Fundação Paulo Feitoza

DOI: https://doi.org/10.5753/eniac.2026.31068

Resumo

Deep Learning has enabled the automated identification of biotic stresses in coffee crops; however, conventional Convolutional Neural Networks (CNNs) process RGB channels as independent feature maps, potentially overlooking cross-channel chromatic relationships. In this work, we investigate a quaternion-valued EfficientNet-B0 (Q-EfficientNet) for coffee leaf disease classification using the Symptom dataset derived from BRACOL. The proposed architecture represents color information in a unified hypercomplex domain and implements quaternion-valued depthwise operations through Kronecker-product formulations, allowing efficient integration within standard deep learning frameworks. Experimental results show that the proposed model achieves competitive classification performance while requiring only 1.7 million trainable parameters, compared to 5.9 million parameters for its real-valued counterpart. These findings indicate that quaternion-valued representations can provide a compact and effective alternative for modeling multidimensional color information in agricultural computer vision tasks, highlighting their potential for deployment in resource-constrained environments.

Referências

Cariow, A. and Cariowa, G. (2025). Reduced-complexity algorithms for tessarine neural networks. IEEE Transactions on Neural Networks and Learning Systems, 36(4):7751–7756.

Esgario, J. G., Krohling, R. A., and Ventura, J. A. (2020). Deep learning for classification and severity estimation of coffee leaf biotic stress. Computers and Electronics in Agriculture, 169:105162.

Grassucci, E., Zhang, A., and Comminiello, D. (2024). Phnns: Lightweight neural networks via parameterized hypercomplex convolutions. IEEE Transactions on Neural Networks and Learning Systems, 35(6):8293–8305.

Gupta, S. and Tan, M. (2019). Efficientnet-edgetpu: Creating accelerator-optimized neural networks with automl. Google AI Blog, 2(1).

Hu, J., Shen, L., and Sun, G. (2018). Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141.

Manso, G. L., Knidel, H., Krohling, R. A., and Ventura, J. A. (2019). A smartphone application to detection and classification of coffee leaf miner and coffee leaf rust. arXiv.

Neto, G. V. and Valle, M. E. (2025a). V-efficientnets: Vector-valued efficiently scaled convolutional neural network models. In 2025 International Joint Conference on Neural Networks (IJCNN), pages 1–8.

Neto, G. V. and Valle, M. E. (2025b). V-efficientnets: Vector-valued efficiently scaled convolutional neural network models. In 2025 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE.

Niemczynowicz, A. and Kycia, R. A. (2026). Fully tensorial approach to hypercomplex-valued neural networks. Information Sciences. In Press, Journal Pre-proof, Article 123796.

Parcollet, T., Morchid, M., and Linarès, G. (2020). A survey of quaternion neural networks. Artificial Intelligence Review, 53(4):2957–2982.

Parcollet, T., Ravanelli, M., Morchid, M., Linarès, G., Trabelsi, C., De Mori, R., and Bengio, Y. (2019). Quaternion convolutional neural networks for end-to-end automatic speech recognition. In ICASSP.

Rodrigues, C. and Valle, M. (2025). A study on convolutional vector-valued neural networks for color image classification. In Anais do XXII Encontro Nacional de Inteligência Artificial e Computacional, pages 1419–1430, Porto Alegre, RS, Brasil. SBC.

Sangwine, S. J. (1996). Quaternion neural networks for signal processing. IEEE Workshop on Neural Networks for Signal Processing.

Tan, M. and Le, Q. (2019). Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning, pages 6105–6114. PMLR.

Tan, M. and Le, Q. (2021). Efficientnetv2: Smaller models and faster training. In International conference on machine learning, pages 10096–10106. PMLR.

Vieira, G. and Valle, M. E. (2022). A general framework for hypercomplex-valued extreme learning machines. Journal of Computational Mathematics and Data Science, 3:100032.

Yang, Y. et al. (2019). Quaternion convolutional neural networks for color image classification. IEEE Access.