Impact of Image Augmentation Techniques on CNN-Based Lemongrass Leaf Classification

This study investigates the impact of image augmentation techniques on the performance of Convolutional Neural Networks (CNNs) for classifying lemongrass leaves into three categories: Healthy, Unhealthy, and Dried. A dataset of 10,042 images was utilized, and five pre-trained CNN architectures—InceptionV3, Xception, MobileNetV2, ResNet152V2, and DenseNet201—were evaluated. The proposed pipeline applies geometric augmentations, including rotation, shift, zoom, shear, and horizontal flipping, to enhance model generalization. All models were trained with consistent hyperparameters and validated using accuracy, precision, recall, and F1-score metrics. Experimental results show significant performance gains due to augmentation. DenseNet201 outperformed all other models, achieving 98.81% accuracy, with nearly perfect classification across all classes. These findings confirm that integrating augmentation with deep CNN architectures improves classification robustness, offering practical solutions for real-time plant health monitoring in agriculture.