<![CDATA[Coffee leaf disease remains one of the most significant threats to global coffee production, particularly Coffee Leaf Rust (CLR) caused by Hemileia vastatrix. Early and accurate disease detection is essential for maintaining yield stability and ensuring sustainable coffee farming. This study proposes an Ensemble Convolutional Neural Network (CNN) architecture that combines MobileNetV2 and ResNet50 to enhance robustness and generalization in multi-class classification of coffee leaf diseases. The dataset consists of 1,664 images categorized into four classes: miner, nodisease, phoma, and rust, collected from both public repositories and real-field observations. Image preprocessing includes resizing, normalization, and augmentation to increase diversity and reduce overfitting. The ensemble model is trained using the Adam optimizer with a learning rate of 0.0001 and evaluated through accuracy, precision, recall, and F1-score metrics. Results demonstrate that the ensemble CNN outperforms single CNN architectures, achieving an accuracy of 95.6%, precision of 94.4%, and F1-score of 94.2%, even under challenging illumination and noise conditions. Compared to individual models, performance improvement ranges from 2%–4%. The model also maintains higher stability when tested under low-light and noisy images, confirming its robustness in real-world scenarios. This study concludes that ensemble CNN offers a reliable and efficient framework for real-time coffee leaf disease detection and can serve as a foundation for developing intelligent agricultural systems using edge computing.]]>
Copyrights © 2025
