Abstract

Several diseases threaten the coconut sector by impacting the well- being and production of coconut trees. This affects the quality and sustainability of coconut cultivation in India. It is vital to promptly and precisely identify diseases to implement practical management approaches. In this study, we used an unbalanced coconut tree disease dataset from Mendley and applied data augmentation techniques to minimize the imbalance and improve the model’s robustness. To classify the diseases accurately, we explored, customized, and trained five deep learning models-VGG19, InceptionV3, DenseNet201, Xception, and ResNet50. Among these, InceptionV3 achieved the highest performance across all metrics, with an accuracy of 99% and a Cohen’s Kappa value of 0.96. Furthermore, we compared the performance of our best model against existing methods, demonstrating that our approach outperforms previous works.

Authors

P.S. Pravitha, V. Kumar
Central University of Kerala, India

Keywords

Deep Learning, Coconut Disease Detection, Pre-Trained Models, Transfer Learning