Manuscript received March 12, 2026; revised April 14, 2026; accepted June 12, 2026; published October 19, 2026.

Abstract—Plant disease recognition remains an important challenge in precision agriculture due to variations in visual symptoms, environmental conditions, and computational constraints in real-world deployment scenarios. This study proposes a lightweight attention-enhanced ResNet18 framework that integrates a channel attention-based feature refinement strategy to improve plant leaf disease classification performance. The experimental evaluation was conducted using a PlantVillage-derived benchmark dataset comprising 20,638 annotated leaf images distributed across 38 plant disease categories. Adaptive data augmentation was employed to improve feature diversity and reduce overfitting risk during training. Experimental results demonstrated that the proposed framework achieved a classification accuracy of 88.38%, a precision of 91.67%, a recall of 88.38%, and an F1-Score of 88.77%. Further analysis using confusion matrix evaluation, Receiver Operating Characteristic (ROC) analysis with an Area Under the Curve (AUC) of 0.966, and Gradient-weighted Class Activation Mapping (Grad-CAM) visualization indicated that the model was capable of identifying disease-relevant regions while maintaining relatively stable discriminative performance across multiple classes. In addition, the proposed framework maintains a lightweight architecture with approximately 11.19 million trainable parameters, making it potentially suitable for resource-constrained agricultural monitoring applications.

Keywords—plant disease classification, convolutional neural networks, attention mechanism, ResNet18, agricultural image analysis
 

Cite: Junus Sinuraya, Wiwin Sry Adinda Banjarnahor, Hikmah Adwin Adam, Meri Sri Wahyuni, and Syarifah Fadillah Rezky, "Lightweight Attention-enhanced ResNet18 for Plant Disease Image Classification on Small-scale Agricultural Datasets," Journal of Image and Graphics, Vol. 14, No. 5, pp.849-858, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).