TeaLeafBD: Evaluation of a Proposed Dataset for Enhanced Tea Leaf Disease Identification Using Baseline Models

Tea leaf diseases significantly impact crop yield and the livelihoods of tea-growing communities, particularly in countries like Bangladesh. Existing datasets for tea leaf disease identification often suffer from limitations such as class imbalance, environmental inconsistency, and lack of diversity, which reduce the reliability and generalizability of machine learning models. To address these issues, we propose TeaLeafBD, a curated and well-balanced image dataset specifically focused on common tea leaf diseases found in Bangladesh. The dataset is designed to reduce bias by ensuring class uniformity, diverse environmental conditions, and consistent image quality suitable for computer vision tasks. To assess the dataset's reliability, we evaluated several baseline deep learning models, including lightweight convolutional networks and pre-trained architectures. The models demonstrated consistently strong performance across all classes, confirming the dataset's suitability for building reliable detection systems. TeaLeafBD is intended as a public resource to support further research in agricultural computer vision and to enable practical solutions for early disease diagnosis in tea plantations.