Manuscript received April 21, 2025; revised April 28, 2025; accepted August 15, 2025; published April 28, 2026.

Abstract—In this study, we explore the effectiveness of using a few-shot learning approach integrated with deep learning models to classify sugarcane diseases from a limited dataset. Specifically, we evaluated four models: MobileNet_V2 combined with a Prototypical Network, MobileNet_V2, ResNet50, and VGG. Our main objective was to demonstrate the potential of the Prototypical Network, paired with a lightweight model like MobileNet_V2, in addressing the challenge of limited data availability. A dataset consisting of six sugarcane disease types was used to train and test these models. The results show that the MobileNet_V2 with Prototypical Network outperformed the other models regarding accuracy, precision, recall, and overall stability, making it a strong candidate for real-world applications in resource-constrained environments. This approach demonstrated superior performance in handling small datasets and exhibited the potential for deployment in mobile or edge devices for real-time agricultural disease diagnosis. Future work will explore less significant downscaling, multi-scale processing to enhance pattern recognition, and focal loss to improve the classification of challenging samples. Additionally, incorporating explainability techniques could reveal the focus areas in model predictions, guiding feature extraction improvements for classes with subtle distinctions.

Keywords—sugarcane disease classification, few-shot learning, prototypical network, MobileNet_V2, deep learning, agricultural disease diagnosis

Cite: John P. Q. Tomas, Miguel A. A. Agno, Paul J. M. Vale, and Marineil C. Gomez, "Sugarcane Disease Classification Using Few-Shot-Prototypical Deep Learning Techniques," Journal of Image and Graphics, Vol. 14, No. 2, pp. 353-365, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.