# A comprehensive annotated image dataset for deep learning analysis of eggplant leaf diseases

## TL;DR

This paper introduces a comprehensive dataset and a high-performing model for identifying eggplant leaf diseases, aiding precision agriculture and sustainable farming.

## Contribution

The paper presents a new annotated eggplant leaf disease dataset and a novel CBAM–EfficientNetB0 model with 98.70% accuracy for disease classification.

## Key findings

- The CBAM–EfficientNetB0 model achieved 98.70% classification accuracy, outperforming other models like ResNet50 and VGG.
- The dataset includes 10,000 images across 10 disease classes, collected from real-world agricultural conditions in Bangladesh.
- The proposed model and dataset enable AI-powered early disease detection and automated monitoring in agriculture.

## Abstract

The Eggplant Leaf Disease Dataset was meticulously developed to address challenges in accurately identifying diseases that threaten eggplant crops, a vital agricultural resource worldwide. This dataset includes 3116 high-resolution images captured between March and May 2024 from two major agricultural regions in Bangladesh, representing real-world conditions. It comprises 10 distinct disease classes—Aphids, Cercospora Leaf Spot, Defect Eggplant, Flea Beetles, Fresh Eggplant, Fresh Eggplant Leaf, Leaf Wilt, Phytophthora Blight, Powdery Mildew, and Tobacco Mosaic Virus—making it the most comprehensive dataset for eggplant diseases to date. To enhance its utility, rigorous data augmentation techniques, including flipping, rotating, shearing, shifting, noise addition, and brightness adjustment, were applied. This expanded the dataset to 10,000 images, ensuring its robustness for machine learning applications. Expert annotations further enhance its quality, providing critical insights for precise disease classification.

Our Proposed CBAM–EfficientNetB0 model had an amazing classification accuracy of 98.70 %, which was much better than the baseline architectures. ResNet50 only got 32.60 %, VGG16 got 73.00 %, and VGG19 got 68.00 %. The proposed model's better performance shows that combining channel and spatial attention through CBAM with EfficientNetB0′s feature extraction abilities works well. This architecture does a good job of picking out the distinguishing features in eggplant leaf images, which makes it possible to accurately identify diseases. The dataset and model work together to make AI-powered early disease detection, automated monitoring, and decision support in precision agriculture possible. These tools help farmers use sustainable farming methods by making timely interventions, reducing the need for manual inspection, and increasing crop productivity and food security.