Tea Leaf Disease Image Dataset for Automated Classification of Tea Garden Diseases
Description
This dataset contains 2,500 high-quality images of tea leaves collected under natural field conditions in tea gardens across Habiganj, Sylhet, Bangladesh. It is designed to support machine learning and computer vision research for the automatic detection and classification of tea leaf diseases, contributing to innovations in precision agriculture and AI-based crop monitoring. Data Collection Details: ⦁ Collection Period: April 7, 2025 – June 23, 2025 ⦁ Total Duration: Approximately 2.5 months ⦁ Collection Locations: Chāndpur Tea Garden, Deundi Tea Garden, Nalua Tea Estate, and Amo Tea Estate (Chunarughat, Habiganj, Sylhet, Bangladesh) ⦁ Environmental Conditions: Images were captured outdoors under natural daylight, during various weather conditions (sunny, cloudy, and humid) to ensure dataset diversity. ⦁ Equipment Used: All photographs were taken using an iPhone 12 Pro Max, featuring a 12 MP triple-camera system (Ultra Wide, Wide, and Telephoto) with image resolutions ranging from 1080×1080 to 3024×3024 pixels. ⦁ Data Validation: All images were manually reviewed, cleaned, and labeled by agricultural experts and plant pathologists to ensure class consistency and labeling accuracy. Dataset Composition: The dataset is organized into four distinct classes based on disease symptoms and healthy leaf conditions. ⦁ Blight (555 images): Leaves showing symptoms of fungal blight infection such as brown lesions and decayed edges. ⦁ Healthy Leaf (650 images): Fresh, disease-free green tea leaves. ⦁ Red Rust (620 images): Leaves infected with red-orange fungal rust spots. ⦁ Helopeltis (675 images): Leaves damaged by the Helopeltis insect pest, characterized by small dark punctures and yellowing areas around bites. Annotation and Preprocessing: ⦁ All images were manually inspected and cleaned to remove duplicates, blurred samples, and inconsistent labels. ⦁ Labeling was carried out under expert supervision, following standard plant disease identification guidelines. ⦁ No artificial data augmentation was applied, preserving real-world variability in lighting, leaf texture, and background conditions. Applications: This dataset can be effectively used for: ⦁ Image classification and disease recognition tasks ⦁ Deep learning model development (CNNs, transfer learning, etc.) ⦁ Comparative benchmarking for plant pathology research ⦁ AI-based agricultural monitoring and tea yield optimization File Information ⦁ Total Images: 2,500 ⦁ File Format: JPEG (.jpg) ⦁ Resolution Range: 1080×1080 – 3024×3024 pixels ⦁ Average File Size: 0.67 – 7.8 MB per image ⦁ Folder Structure: Each class stored in a separate labeled directory
Files
Steps to reproduce
1.Download the dataset: Access the dataset files from this Mendeley Data repository and extract the contents. 2. Dataset structure: TeaLeaf_Dataset/ ├── Blight/ │ ├── Brown_Blight_1.jpg │ ├── ... ├── Healthy_Leaf/ │ ├── Healthy_Leaf_1.jpg │ ├── ... ├── Red_Rust/ │ ├── Red_Rust_1.jpg │ ├── ... ├── Helopeltis/ ├── Helopeltis_1.jpg ├── ... Each subfolder contains images corresponding to one disease class. 3. Data preprocessing: Resize all images to a fixed dimension (e.g., 224×224 or 299×299). Normalize pixel values to [0, 1]. Split into training (70%), validation (15%), and testing (15%) sets. 4. Model training example (optional): Load data using TensorFlow or PyTorch image data generator. Apply data augmentation (rotation, flip, zoom) to improve generalization. Train a model such as MobileNetV3, ResNet50, or a custom CNN. 5. Evaluation: Evaluate model accuracy, precision, recall, and F1-score. Visualize confusion matrix to check misclassification among disease types.
Institutions
- Daffodil International University