A comprehensive combined dataset on Hibiscus and Tea plant leaf disease images for classifications

TL;DR

This paper introduces a combined leaf disease dataset for Hibiscus and Tea plants, using high-resolution images and a deep learning model to achieve high classification accuracy.

Contribution

The novel contribution is a publicly available combined leaf disease dataset and a fine-tuned ConvNextTiny model for multi-species plant disease classification.

Findings

A dataset with 1,413 original and 13,000 augmented images of Hibiscus and Tea leaf diseases was created.

The fine-tuned ConvNextTiny model achieved 96% accuracy in classifying leaf conditions across both species.

Data augmentation techniques effectively addressed class imbalances and improved model performance.

Abstract

In this study, we present a combined image dataset created from two distinct plant species: Hibiscus and Tea leaf. The dataset consists of high-resolution images of leaves from both species, captured using a SONY α7 II DSLR camera and a OnePlus 7T lubricant Tea Leaf dataset includes images categorized into five disease classes: Algal Leaf Spot, Brown Blight, Grey Blight, Red Leaf Spot, and Healthy, while the Hibiscus Leaf dataset includes images labeled across eight conditions, including citrus spot, fungal infection, mild edge damage, and healthy foliage. To ensure balanced representation and address class imbalances, extensive data augmentation techniques—such as flipping, rotation, zooming, shifting, noise addition, and brightness adjustment—were applied, resulting in a total of 1,413 combined original images and 13,000 augmented images. The ConvNextTiny deep learning model was fine-tuned on this combined dataset to classify the various leaf conditions, achieving an overall accuracy of 96%. This demonstrates the model's robust performance and high discriminatory power across the diverse set of leaf diseases and conditions. This experiment highlights the utility of combining multiple plant species into a single dataset and utilizing a lightweight yet effective model like ConvNextTiny for plant disease classification. The resulting dataset, along with the model and training scripts, is publicly available to facilitate further research in plant pathology, computer vision, and smart farming applications, enabling more accurate and efficient early-stage disease detection for both Hibiscus and Tea plants.

Figures