Ensemble Deep Learning Framework for Mango Leaf Disease Detection

Abstract

Mango cultivation is one of the most important agricultural activities in many tropical and subtropical regions, providing revenue for agricultural producers and contributing to food security. However, mango plants are often affected by several leaf diseases, including anthracnose, powdery mildew, bacterial leaf spot, sooty mold, and die-back disease. These diseases can significantly reduce crop yields, affect fruit quality, and lead to substantial economic losses if not identified and treated early. Traditionally, disease detection has relied on visual inspection by agricultural experts. Although effective, this approach can be time-consuming, subjective, and difficult for farmers in isolated regions to access. With the rapid development of deep learning and computer vision technologies, automated plant disease identification has emerged as a promising solution. This study presents an automated mango leaf disease detection system based on a deep learning ensemble framework. The proposed model combines three well-known convolutional neural network (CNN) architectures ResNet50, MobileNetV2, and EfficientNetB3 using a soft-voting ensemble technique. Leaf images are first preprocessed and enhanced via data augmentation before being analyzed by the ensemble models. The system then predicts the disease category and provides confidence scores for the classification results. To evaluate its performance, the proposed approach was tested on a dataset containing 4,500 mango leaf images representing seven disease classes and one healthy class. The experimental outcomes suggested that the ensemble model achieved an accuracy of 96.8%, outperforming the individual CNN models by approximately 3-5%. In addition, the precision, recall, and F1-score values demonstrated the effectiveness and reliability of the ensemble approach.