TeaLeaf-4: A curated tea leaf image dataset for classification of four leaf conditions

TeaLeaf-4 is a curated image dataset for computer-vision-based classification of tea leaf (Camellia sinensis) conditions. It contains 3,156 images collected under natural field conditions from two tea estates in Sylhet, Bangladesh, and organized into four classes: Healthy — 1,108 images: leaves with no visible disease or stress; uniform green appearance. RedSpider — 1,061 images: leaves showing discoloration and damage from red spider mite infestation. Skeletonization — 594 images: leaves with tissue loss and visible skeletal structure from pest damage. Sun_lightScorching — 393 images: leaves with burn-like symptoms from excessive sunlight exposure. Images were captured between 23–28 May 2025 at Tarapore Tea Estate (24°55′28.16″N, 91°51′50.43″E) and Star Tea Estate (24°55′01.06″N, 91°51′14.37″E) using four smartphone cameras (Honor X9B, OnePlus 7, Samsung Galaxy A55, Redmi Note 14 Pro+), giving a realistic range of resolution and lighting. All images were preprocessed for consistency (background removed and replaced with a uniform white background, resized to 3000 × 3000 pixels, brightness enhanced) and manually labeled and expert-validated. The images are arranged in a class-wise folder structure that loads directly with standard image-folder data loaders, making the dataset suitable for image classification, transfer learning, and benchmarking. As a baseline, a fine-tuned ResNet50 model reached 92.09% validation accuracy. The dataset supports research in agricultural image analysis and intelligent decision-support systems for tea cultivation.