Curated Binary Tea Leaf Image Dataset Derived from teaLeafBD and CS-D Assam Tea Leaf Dataset
Description
This dataset is a curated and preprocessed derivative of the publicly available teaLeafBD and CS-D Assam Tea Leaf datasets. Both datasets were combined to create a unified binary tea leaf disease dataset for deep learning research. To improve data quality, duplicate and near-duplicate images were detected using perceptual hashing (pHash) with Hamming distance, while a BK-tree was used for efficient similarity search. Images with a similarity threshold between 10 and 14 were removed to eliminate redundant samples, reduce the risk of data leakage between dataset splits and improve the reliability of model evaluation. Removing duplicate and near-duplicate images also helps ensure a more rigorous and unbiased model training and evaluation process. In addition, every image was manually reviewed to verify its quality, preprocessing outcome and class label. The image backgrounds were removed using the rembg library, and each tea leaf was extracted, cropped and placed on a uniform white background to reduce background variability. All images were resized to 224 × 224 pixels while preserving their original aspect ratio. All disease categories from the original multiclass datasets were merged into a single Disease class, while the Healthy class was retained, resulting in a binary classification dataset. The final dataset is clean, standardized and suitable for developing and evaluating deep learning models for binary tea leaf disease detection. As this is a derivative dataset, users should cite both the original teaLeafBD and CS-D Assam Tea Leaf dataset along with this curated derivative dataset when using these data.