Rice Disease Classification Using Leaf Images with Mobilenetv2 Based on Mobile

Abstract

Rice is a major staple food in Indonesia, including in Merauke Regency, one of the largest rice-producing regions in South Papua Province. In 2023, Merauke Regency had a harvested rice area of ​​49,573.50 hectares, with a production of 236,500.33 tons. However, rice cultivation still faces challenges from pests and diseases that can reduce the quality and quantity of the harvest. Manual disease identification through direct observation or consultation with Field Agricultural Extension Workers (PPL) is time-consuming and risky due to the diverse types of diseases and leaf symptoms. This research aims to develop a leaf-image-based rice plant disease classification system using a Convolutional Neural Network (CNN) with the MobileNetV2 architecture. The system, developed as an Android application, can classify diseases through the camera or gallery, displaying symptoms and treatment solutions. The dataset consists of 7,200 images with six classes: Bacterial Leaf Blight, Brown Spot, Healthy, Leaf Blast, Leaf Scald, and Tungro. Testing uses accuracy, confusion matrix, Black Box, and User Acceptance Testing (UAT). The best results were obtained at epoch 50 and a learning rate of 0.001, with a training accuracy of 99.50%, a validation accuracy of 95.00%, a testing accuracy of 95.88%, and a real-world accuracy of 76.67%. Black Box showed all features were running as expected, while UAT achieved 83.2% for farmers and 92% for admins.