A Real-World Hibiscus and Tea Leaf Image Dataset for Classification

Published: 8 October 2026| Version 6 | DOI: 10.17632/5bzy89brkv.6
Contributors:
,
,
,

Description

The Combined Hibiscus and Tea Leaf Image Dataset comprises 1,412 original leaf images collected under natural outdoor conditions in Bangladesh using a SONY α7 II DSLR camera and OnePlus 7T smartphone. The dataset contains 13 classes, including 8 Hibiscus classes and 5 Tea classes, covering healthy, diseased, damaged, and senescent leaf conditions. Based on the original dataset, 984 images (70%) are allocated for training, 222 images (15%) for testing, and 206 images (15%) for validation. To address class imbalance, data augmentation techniques including brightness/contrast adjustment, rotation, flipping, scaling, cropping, shifting, zooming, and Gaussian noise were applied. Following augmentation, each of the 13 classes was expanded to 1,000 images, resulting in a balanced augmented training dataset of 13,000 images. The class-wise original and augmented distributions are provided in the dataset organization, with all images stored in JPG format and arranged in class-specific folders. This balanced dataset provides a suitable benchmark for plant disease classification, agricultural image analysis, and computer vision research.

Files

Institutions

Categories

Computer Vision, Machine Learning, Agricultural Health, Plant Diseases, Plant Health, Convolutional Neural Network, Deep Learning, Agriculture

Licence