# EgyPLI: A Real-life Annotated Image Dataset for Egyptian Plant Leaf Identification

## TL;DR

EgyPLI is a new dataset of Egyptian plant leaf images designed to improve automated plant identification in real-world conditions.

## Contribution

The dataset introduces a geographically diverse, real-life annotated collection of leaf images from eight plant species in Egypt.

## Key findings

- EgyPLI includes 3,588 images of eight plant species with both healthy and diseased leaves.
- Custom CNN achieved 99.22% accuracy, outperforming ResNet50 and VGG16 on the dataset.
- The dataset supports robust model training under natural variability and noise.

## Abstract

The Egyptian Plant Leaf Image Dataset (EgyPLI) is the first comprehensive collection of plant leaf images curated in Egypt to support research in automated plant identification. It addresses the lack of locally representative datasets and the broader need for geographically diverse data to enable the development of generalized models. EgyPLI contains real-world leaf images captured under varying viewpoints, lighting conditions, and background clutter, reflecting realistic agricultural environments. Unlike laboratory-controlled datasets, it includes natural noise and variability, supporting the training of robust deep learning models suitable for real deployment. The dataset is carefully annotated and preprocessed to establish a consistent standard for plant identification tasks. EgyPLI comprises 3,588 images covering eight widely cultivated plant species: apple, berry, fig, guava, orange, plum, persimmon, and tomato, including both healthy and diseased leaves. This diversity supports classification, diagnosis, and health assessment applications. To demonstrate its effectiveness, the dataset was evaluated using ResNet50, VGG16, and a custom CNN, achieving accuracies of 61.67%, 96.81%, and 99.22%, respectively. As an available resource, EgyPLI fills a critical gap.