An Ensemble Convolutional Neural Network Framework for Automated Mango Leaf Disease Detection

Abstract

Abstract Mango diseases and pest infestations represent a major challenge to agricultural productivity, making early and accurate diagnosis crucial for reducing crop losses. This study presents a security-preserving ensemble convolutional neural network (CNN) framework for the automated identification and classification of mango leaf diseases using image-based analysis. The proposed system is designed to work with images captured under real field conditions, ensuring its suitability for practical agricultural applications. The dataset includes mango leaf images affected by various diseases and pests such as Gall Midge, Powdery Mildew, Sooty Mould, Die Back, Cutting Weevil, and Anthracnose, each characterized by distinct visual symptoms including discoloration, necrotic spots, fungal growth, leaf deformation, and edge damage. Traditional manual diagnosis of these conditions is often time-consuming, labor-intensive, and susceptible to human error. To overcome these limitations, the proposed framework employs an ensemble of transfer-learning-based CNN models to extract meaningful features related to texture, color distribution, shape, and lesion patterns. A security-preserving learning mechanism is integrated to ensure the safe handling of agricultural image data, minimizing data exposure risks while maintaining high model performance. Additionally, data augmentation techniques are utilized to improve model robustness, reduce overfitting, and address class imbalance commonly found in agricultural datasets. The system is capable of multi-class classification, reflecting real-world scenarios where multiple diseases may exhibit visually similar characteristics. Experimental results indicate that the ensemble CNN framework achieves high classification accuracy and demonstrates strong generalization across varying lighting conditions and complex backgrounds. By effectively capturing disease-specific visual features, the proposed approach enhances detection reliability in real-world field environments. Overall, this system offers a scalable, non-invasive, and security-aware solution for early mango leaf disease detection, contributing to precision agriculture and informed decision-making. The findings highlight the potential of deep learning and computer vision technologies in developing intelligent, secure, and efficient plant health monitoring systems.

Plant phenotyping relevance

マンゴー葉の病徴を画像から分類するCNNフレームワークの開発が研究の中心であり、植物の病害状態を直接推定する画像ベース表現型解析に該当する。

Code and data availability

The paper's plant-phenotyping input is the public Kaggle Mango Leaf Disease Dataset of mango leaf images (Gall Midge, Powdery Mildew, Sooty Mould, Die Back, Cutting Weevil, Anthracnose, Healthy), explicitly declared as publicly available with a link. No author code, models, or checkpoints are shared.