An Optimized Classification Model for Coffea Liberica Disease using Deep Convolutional Neural Networks
Francis Jesmar P. Montalbo, Alexander A. Hernandez
Abstract
The Philippines gave coffee a unique identity and taste in the form of Barako coffee, a variant that came from the species of Coffea liberica. However, the crop yielded faces a challenge in coping up with several widespread diseases leading it to lose in quantity and quality. Barako became a less prioritized crop compared to Excelsa, Robusta, and Arabica. The troublesome method of identifying diseases like rust, spots, and insect infestation caused many losses for farmers due to improper diagnosis and treatment. This research aims to apply deep learning methods to alleviate the problem. A deep convolutional neural network was optimized to perform the difficult task of classifying diseases to assist farmers in applying an appropriate treatment. With several experiments, this study develops several models to distinguish the best possible model that can yield the most accurate results. This study acquired 3958 high-resolution images to train the desired model. The result achieved a 100 percent accuracy rate while other models misclassified due to overfitting problems. Proper tuning of hyperparameters can address overfitting together with an appropriate choice of optimizers resulting in an efficient classifier. Moreover, this research identified potential future research works to apply the model for real-life scenarios.