Image Based Banana Leaf Disease Detection with Machine Learning Technique

Abstract

Today, international trade has grown significantly in many countries. Many fruit products are imported from other countries, such as bananas and apples. Devastating economic losses and production losses are incurred all over the world due to fruit diseases. Little study has been done throughout the years for fruit disease detection to assist remote farmers technically. The bulk of these farmers require proper cultivation support, but little research has been done for this system. The use of one's eyes to visually inspect fruits and vegetables allows trained professionals to identify imperfect produce; however, this paper presents a lightweight machine-learning framework that integrates K-means clustering for image segmentation and a Random Forest classifier for disease recognition. A locally collected dataset of 422 banana leaf images from farms in Bangladesh was used, covering four classes: Cordana, Sigatoka, Pestalotiopsis, and Healthy. Images were preprocessed using RGB–to–Lab color conversion and resized for consistency before feature extraction based on color and texture descriptors. Among several tested classifiers, the Random Forest achieved the best performance with 96.25 % accuracy, 93.56 % precision, and 97.85 % specificity, outperforming the Decision Tree (95.28 %) and Naïve Bayes (89.27 %). Owing to its low computational cost and use of regionally collected images, this framework is suitable for real-time mobile or IoT- based agricultural systems, supporting smart and sustainable farming in resource-limited environments.