A Multi-Class Real-Field Eggplant Leaf Disease Dataset for Computer Vision and Deep Learning Research

Description

This dataset presents a comprehensive real-field collection of eggplant (Solanum melongena L.) leaf images designed to support research in computer vision, deep learning, and precision agriculture. The dataset was developed through extensive field surveys conducted in agricultural regions of Bangladesh between December 2025 and January 2026 under natural environmental conditions. The dataset comprises 1,500 high-resolution RGB images categorized into three classes: Healthy leaves, Fungal-infected leaves, and Spider Mite-infected leaves, with 500 images per class. All images were captured directly from eggplant cultivation fields using smartphone cameras under diverse real-world conditions, including variations in illumination, background complexity, viewing angles, and plant growth stages. Such diversity enhances the robustness and generalization capability of machine learning models trained on this dataset. The collected images were manually inspected, annotated, and organized into class-specific directories to facilitate supervised learning tasks. The dataset can be utilized for a wide range of agricultural artificial intelligence applications, including image classification, disease detection, transfer learning, feature extraction, and model benchmarking. By providing a real-field, annotated, and diverse image collection, this dataset aims to serve as a valuable benchmark resource for developing intelligent and automated eggplant disease diagnosis systems, thereby contributing to sustainable agriculture and precision farming practices.

Steps to reproduce

1. Eggplant leaf images were collected directly from agricultural fields in Bangladesh between December 2025 and January 2026. 2. Images were captured under natural environmental conditions using smartphone cameras at an approximate distance of 30–50 cm from the leaves. 3. Data collection was performed under varying lighting conditions, backgrounds, and plant growth stages to ensure real-world diversity. 4. Collected images were manually inspected and categorized into three classes: Healthy, Fungal, and Spider Mite based on visible disease symptoms. 5. Low-quality, blurred, and duplicate images were removed during the data cleaning process. 6. The finalized images were organized into separate class-specific folders. 7. The dataset was then used for deep learning experiments involving image classification and transfer learning-based disease recognition.

Institutions

• Daffodil International University

  Dhaka Division, Dhaka

Categories

Licence