AI-Based Detection of Leaf Diseases in Durian (Durio zibethinus) Using Convolutional Neural Networks: Model Development and Performance Evaluation

Abstract

Early detection of durian leaf diseases using artificial intelligence (AI) plays an important role in supporting effective plant disease management within digital agriculture systems. This study aimed to develop an AI-based system for the preliminary detection and classification of common durian leaf diseases using image analysis. Diseased leaf samples were collected from durian orchards in Pa Phayom District, Phatthalung Province, and Chang Klang District, Nakhon Si Thammarat Province, southern Thailand. Associated microorganisms were isolated using the tissue transplanting technique, and morphological characteristics indicated the presence of Colletotrichum spp. in leaf spot samples and Fusarium spp. in leaf blight samples. In contrast, the algal leaf spot pathogen could not be cultured on standard fungal media. A Convolutional Neural Network (CNN) was employed to classify three disease categories: leaf spot, leaf blight, and algal leaf spot. The image dataset was divided into training, validation, and testing sets, and the model was trained and evaluated accordingly. The results demonstrated that the proposed AI model effectively classified durian leaf disease images, particularly leaf spot and leaf blight. However, this study used a relatively limited dataset, and the proposed system should therefore be considered a preliminary disease-detection tool. The developed AI-based system shows potential for deployment on mobile devices or online platforms to support early disease recognition and decision-making by farmers, thereby improving disease management strategies in durian cultivation.

Plant phenotyping relevance

葉の病徴画像から植物の病害状態を分類するCNNシステムの開発・評価が中心であり、植物病害フェノタイピング手法に該当する。

Code and data availability

The supplied article blocks describe a cross-sectional ergonomics study of musculoskeletal disorders among rubber tappers in Thailand, not the durian leaf-disease CNN phenotyping paper named in the metadata. The text contains no public phenotype/trait datasets, plant images, sensor/3D/annotation inputs, author analysis

No evidence-backed public reproduction asset is currently recorded.