Augmentation Pipeline: Multi-Crop Plant Disease Images for Classification

A curated dataset of approximately 45,000 original images of plant diseases and healthy plants, aggregated from multiple public repositories. It comprises 51 classes and was used to benchmark vision transformer and custom convolutional neural network models for resource-constrained deployment. The dataset was published by Md. Sifat Haque Zidan on figshare in April 2026.

Use Cases

• Benchmarking lightweight vision transformer architectures for image classification based on the 51 disease and healthy classes.
• Training custom convolutional neural networks for plant disease diagnosis based on the multi-crop image collection.
• Analyzing model decision regions using Grad-CAM visualizations based on the provided image data.
• Developing efficient models for deployment in practical agricultural settings based on the described computational constraints.

Strengths

• Comprises 51 distinct disease and healthy classes for multi-crop analysis.
• Includes approximately 45,000 original images, expanded through controlled augmentation.
• Models trained on this data achieved validation accuracies above 99%, as reported in the description.
• Released under a permissive CC-BY-4.0 license.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is very small (5.5 KB), suggesting the XLS file likely contains metadata or labels, not the images themselves.

Provenance

Source

Aggregated from multiple publicly available repositories.

Collection Method

Curated and expanded through controlled augmentation during training.

Time Range

null

Freshness

Last updated 2026-04-24 17:24:10; freshness should be verified.

Geography

null