Custom Dataset for "Plant Species and Disease Detection using Deep Learning on Plant Images Captured in Realistic Natural Field Conditions."

This dataset (Plant Species and Disease Detection using Deep
Learning on Plant Images Captured in Realistic
Natural Field Conditions) presents a robust and efficient deep
learning-based approach for plant species disease identification
using images captured in realistic, natural field conditions. With the rising global population and increasing demand for food
production, crop losses due to plant diseases pose a significant
threat to food security. Traditional disease detection methods,
which often rely on human expertise, are time-consuming, laborintensive, and prone to bias. Deep learning offers a promising alternative, achieving high accuracy in plant disease detection tasks. However, many existing models are trained on datasets collected under controlled conditions, which fail to capture realworld complexities such as variable lighting, occlusions, and diverse natural backgrounds.
To address this gap, a custom dataset comprising 9,469 images
was created, containing four plant species—grapes, groundnuts,
guava, and soybean—captured directly in natural field environments.
A total of seven convolutional neural network (CNN)
architectures were trained: Basic CNN, ResNet18, ResNet50,
GoogleNet, InceptionV3, Xception65p, and EfficientNet V2 S.
Among these, ResNet18, ResNet50, GoogLeNet, InceptionV3,
and Xception65p achieved notably high accuracy. The proposed
system enhances robustness in species classification and disease
diagnosis under diverse environmental conditions, contributing
to sustainable and scalable agricultural practices.
Keywords— Precision Agriculture, Deep Learning, Artificial
Intelligence, Plant Disease Detection, Species Classification, Computer
Vision.

A new dataset has been custom-built for this research
project. The dataset comprises of 9,469 images, containing
four plant species grapes, groundnuts, guava, and soybean
captured directly in natural field environments. All images
captured in actual field environments under varying natural
conditions.

1. Total Images: 9469
2. Plant Species Covered: Grapes, Groundnut, Guava, and
Soybean
3. Original Dataset Size: Approximately 48 GB
4. Dataset Size after resizing: 5.04 GB
5. New Image Resolution after resizing: 2000 x 1126
6. Environmental Variations: Images have been collected
under diverse lighting conditions (e.g., afternoon,
evening) and weather situations (e.g., cloudy, rainy, and
windy), making the dataset more representative of realworld
scenarios.