Deep Learning with Augmentation for Accurate Coconut Leaf Disease Classification in Scenarios with Limited Data

Coconut cropping is a significant component of the economies of tropical agriculture, yet it is under severe threat from a variety of leaf diseases causing reduction in yield and crop health. Detection should be precise and early to deal with it effectively. Powerful deep learning algorithms for disease classification are usually built upon large labeled datasets, but in the real life of agricultural work, they do not exist. This paper proposes a solution with a deep learning (DL) method augmented by cycleGAN(generative adversarial Networks) to deal with the limited training data for coconut leaf disease classification. The model is optimized to learn strong spatial features and is also
regularized to avoid overfitting. The experimental result shows an impressive improvement in classification accuracy, precision, recall, and F1-score compared to custom CNN, ResNet50 and MobileNet learned from unaugmented data. The proposed method derives an overall classification accuracy of 93.1% for ResNet5, which proves its strength in handling small datasets and improving the performance of disease recognition. This study identifies the potential of data augmentation as a key enabler for the adoption of DL solutions in resource-constrained agri-environments.