<![CDATA[Plant leaf diseases can reduce crop quality and productivity, making accurate disease identification important for supporting plant management. This study evaluates the performance of EfficientNetV2-B0 with and without the Convolutional Block Attention Module (CBAM) for tea leaf disease classification. The Tea Sickness Dataset was examined and preprocessed to address class imbalance. After preprocessing, six classes were retained with 90 images per class, resulting in 540 images. The dataset was divided into training, validation, and testing sets using a stratified split of 70%, 10%, and 20%, respectively. Two model configurations were evaluated under the same training and testing conditions. EfficientNetV2-B0 achieved a test accuracy of 92.59%, macro precision of 93.85%, macro recall of 92.59%, and macro F1-score of 92.61%. In comparison, EfficientNetV2-B0 + CBAM achieved 89.81% accuracy, 90.93% macro precision, 89.81% macro recall, and 89.68% macro F1-score. The results show that adding CBAM did not improve classification performance on the evaluated dataset. Furthermore, CBAM increased the number of model parameters without improving test performance. These findings indicate that the effectiveness of an attention mechanism depends on the characteristics of the dataset and model architecture. Under the experimental conditions of this study, EfficientNetV2-B0 without CBAM provided the best classification performance.]]>
Copyrights © 2026
