AI-MedLeafX: a large-scale computer vision dataset for medicinal plant diagnosis.

Abstract

This study presents a large, meticulously curated and manually validated dataset aimed at classifying leaf quality into five critical categories: Healthy, Bacterial Spot, Shot Hole, Yellow, and Powdery Mildew. The dataset encompasses four distinct plant species-Cinnamomum Camphora (Camphor), Terminalia Chebula (Haritaki), Moringa Oleifera (Sojina), and Azadirachta Indica (Neem)-each represented across three or four disease categories, depending on observed symptoms and final number of classes is thirteen (13 classes). Data collection was conducted between November 1, 2024, and January 5, 2025, utilizing four different mobile cameras to ensure diversity in image resolution, lighting, and environmental conditions. The original dataset comprised 10,858 high-resolution images, which were subsequently expanded to 65,148 through the application of six comprehensive data augmentation techniques, including rotations (45°, 60°, and 90°), horizontal flipping, zooming and brightness adjustment. All images were standardized to 512×512 pixels to ensure uniformity and seamless compatibility with machine learning and computer vision models. This enriched dataset serves as a crucial resource for the development of automated plant disease detection systems and supports advancements in precision agriculture. It not only addresses the pressing need for scalable, high-quality data in agricultural research but also establishes a solid foundation for benchmarking novel deep learning architectures. By enabling more accurate and efficient leaf disease classification, the dataset contributes significantly to enhancing tree health monitoring, improving crop yield, and promoting sustainable agricultural practices.

Plant phenotyping relevance

植物葉の病害症状を画像で分類する大規模データセットであり、植物の病害状態を直接評価する再利用可能なベンチマーク資源が中心です。

Code and data availability

The paper is a Data in Brief article describing AI-MedLeafX, a public leaf-image dataset for medicinal plant disease classification, deposited on Mendeley Data with an explicit direct URL and DOI (10.17632/zz7r5y4dc6.1). This is the paper's own phenotyping image dataset (10,858 original images, 65,148 augmented, 13疾病/类