RoseLeafSet: Real-World Leaf Image Dataset for AI-Based Agricultural Solutions
Description
This RoseLeafSet dataset has been meticulously curated collected images (Primary 3,113, and Expanded 10,000) to support research in machine learning (ML), deep learning (DL), and computer vision (CV) for plant health assessment and disease classification. Collected from Amin Model Town, Khagan, Ashuliya, Dhaka, the dataset is a reflection of real-world agricultural conditions, capturing a variety of rose leaf health states under natural environmental lighting and backgrounds. The dataset consists of 3,113 annotated images, systematically categorized into four major classes: Primary Images (3,113): 1. Healthy Leaf – 818 images 2. Black Spot – 1,288 images 3. Leaf Hole – 683 images 4. Dry Leaf – 324 images Preprocessing Steps: 1. Background Remove 2. Add White Background 3. Cropping Image 4. Brightness Enhance 5. Contrast Enhance 6. Resize into 2024 * 2024 Augmented Images (10,000): 1. Healthy Leaf – 2,500 images 2. Black Spot – 2,500 images 3. Leaf Hole – 2,500 images 4. Dry Leaf – 2,500 images The classification reflects some of the most common rose plant conditions seen in Bangladeshi gardens and nurseries. Each class represents unique visual features—such as color changes, textural differences, and structural anomalies—that are vital for building accurate and robust classification models. Image Collection Methodology: Images were captured using high-resolution smartphone cameras at various angles and distances to ensure diversity. Different times of the day were considered to include a range of natural lighting conditions (e.g., direct sunlight, overcast, and shade), making the dataset suitable for real-world deployment scenarios. The data was further reviewed manually to eliminate duplicates, ensure class consistency, and maintain quality. Use Cases and Research Scope, and dataset is highly applicable for: - Supervised learning models: Training CNNs (e.g., DenseNet201, VGG16, InceptionV3, ResNet50) to classify leaf diseases. - Transfer learning: Fine-tuning pre-trained architectures on this domain-specific dataset. - Data augmentation studies: Investigating the impact of rotation, flipping, zooming, and brightness normalization to improve model generalization. - Explainable AI (XAI): Developing interpretable models to assist agronomists and farmers in understanding model predictions. - Mobile and edge computing deployment: Enabling real-time disease detection in low-resource settings using lightweight DL models. Agricultural and Societal Impact: Rose cultivation is economically and culturally significant in many parts of Bangladesh. Early detection of leaf diseases can substantially reduce the use of harmful pesticides, improve plant health, and boost productivity. This dataset serves as a foundation for precision agriculture systems and intelligent farming applications, promoting sustainability, cost-effectiveness, and technological integration in agriculture.
Files
Institutions
- Daffodil International University