SUGARCANE LEAF DISEASE CLASSIFICATION USING TENSORFLOW LITE API

The sugarcane industry is a vital source of various products like biofuels, energy, and sugar-based ethanol, with sugarcane being a significant biomass source for biofuels globally.The Philippines is among the top producers of sugarcane, but it faces challenges, including the devastating red rot disease that affects about 10% of sugarcane-growing areas in the country.The country's productivity also lags other Southeast Asian nations due to issues like poor irrigation, soil fertility, and climate change impacts.To address these problems, researchers developed a sugarcane disease detection and classification system using artificial intelligence to help farmers improve crop yield.The primary focus was on detecting diseases like red rot, brown leaf spot, pineapple disease, and leaf scorch using a mobile application compatible with Android OS.The study aimed to provide farmers with a mobile app that could detect and classify sugarcane diseases to enable timely prevention and intervention measures, contributing to increased productivity in the Philippine sugarcane industry.The researchers evaluated the performance of a model for diagnosing sugarcane leaf diseases based on precision, recall, F1-score, and accuracy.The study focused on red rot, brown leaf spot, pineapple disease, and leaf scorch detection, excluding other plant diseases and severity assessments.The evaluation of the model's performance revealed promising results.Precision values ranged from 81.6% to 91.4%, indicating high accuracy in identifying instances within each disease category.Recall values ranged from 87.5% to 92%, demonstrating the model's ability to correctly identify actual disease instances.F1 scores, combining precision and recall, were good, ranging from 0.866 to 0.937, indicating reliable classification.The accuracy of diagnosis varied between diseases, with the highest accuracy achieved for red rot and the lowest for leaf scorch.Brown leaf spot classification trials consistently achieved accuracy scores ranging from 95.7% to 99.9%.