Datasets:

ipartzix
/

Crop_Disease_Image_Dataset

license: cc-by-4.0
language:
  - en
tags:
  - agriculture
  - crop-disease
  - computer-vision
  - image-classification
  - deep-learning
  - cnn
  - smart-farming
  - iot
task_categories:
  - image-classification
task_ids:
  - multi-class-classification
pretty_name: Crop Disease Image Dataset (5 Crops, 19 Classes)
size_categories:
  - 10K<n<100K

Crop Disease Image Dataset (5 Crops, 19 Classes)

Dataset Summary

The Crop Disease Image Dataset is a curated, high-quality agricultural image dataset designed for computer vision, deep learning, and smart farming applications. It contains 22,169 RGB leaf images spanning 5 major crops across 19 distinct healthy and diseased classes.

This dataset was constructed by collecting, filtering, and standardizing images from multiple open-source agricultural repositories (such as the PlantVillage Dataset and Mendeley's 15 Crop and 45 Disease Dataset) to create a unified resource for training models like MobileNetV3, ResNet, EfficientNet, and Vision Transformers (ViT).

This dataset forms the core AI vision module for the IoT-Based Crop Disease Detection System for Smart Agriculture project.

Key Overview & Metrics

Property	Description / Value
Dataset Type	Image Classification
Domain	Agriculture / Smart Farming
Total Crops	5 (Corn, Potato, Rice, Tomato, Wheat)
Total Classes	19 (Healthy & Diseased)
Total Images	22,169
Image Format	RGB (JPG / JPEG)
Dataset Size	~1.36 GB

Crop & Class Distribution

🌽 Corn (Maize) — 6,617 images

Class Name	Image Count
Cercospora Leaf Spot (Gray Leaf Spot)	513
Common Rust	1,192
Healthy	1,662
Northern Leaf Blight	985
Leaf Spot	1,261
Maize Streak Virus	1,004

🥔 Potato — 2,968 images

Class Name	Image Count
Early Blight	1,000
Late Blight	1,000
Healthy	968

🌾 Rice — 3,637 images

Class Name	Image Count
Brown Spot	1,210
Leaf Blast	957
Healthy	1,470

🍅 Tomato — 6,627 images

Class Name	Image Count
Bacterial Spot	2,127
Early Blight	1,000
Late Blight	1,909
Healthy	1,591

🌾 Wheat — 2,820 images

Class Name	Image Count
Brown Rust	860
Yellow Rust	880
Healthy	1,080

Dataset Structure

The raw image archive is organized using standard class-folder hierarchy:

Crop_Disease_Image_Dataset/
│
├── Corn_(Maize)___Cercospora_Leaf_Spot_(Gray_Leaf_Spot)/
├── Corn_(Maize)___Common_Rust/
├── Corn_(Maize)___Healthy/
├── Corn_(Maize)___Northern_Leaf_Blight/
├── Corn_(Maize)___Leaf_Spot/
├── Corn_(Maize)___Maize_Streak_Virus/
│
├── Potato___Early_Blight/
├── Potato___Healthy/
├── Potato___Late_Blight/
│
├── Rice___Brown_Spot/
├── Rice___Healthy/
├── Rice___Leaf_Blast/
│
├── Tomato___Bacterial_Spot/
├── Tomato___Early_Blight/
├── Tomato___Healthy/
├── Tomato___Late_Blight/
│
├── Wheat___Brown_Rust/
├── Wheat___Healthy/
└── Wheat___Yellow_Rust/

Dataset Creation Pipeline

Multiple Public Datasets (PlantVillage, Mendeley, etc.)
                      │
                      ▼
               Data Collection
                      │
                      ▼
            Class Standardization
                      │
                      ▼
               Duplicate Removal
                      │
                      ▼
            Dataset Consolidation
                      │
                      ▼
  Final Cleaned Dataset (22,169 Images, 19 Classes)

Recommended Usage & Preprocessing

Data Splitting Strategy

For robust evaluation and avoiding data leakage across classes, a stratified split is strongly recommended:

Train: 70%
Validation: 15%
Test: 15%

Data Augmentation & Pipelines

When loading data into frameworks like PyTorch (torchvision) or TensorFlow (tf.keras), standard spatial transformations improve model generalization:

Resize & Center Crop (e.g., 224x224 or 256x256)
Pixel Normalization (ImageNet mean & std)
Random Flips, Rotations, and Color Jittering

Project Context: Dual-Module IoT System

This vision dataset forms Module 1 of a broader smart agriculture framework:

Vision Module (This Dataset): Image-based leaf diagnosis using Deep Learning models deployed on local/edge devices.
Sensor Module: Environmental monitoring using IoT sensors (Temperature, Humidity, Soil Moisture) fed into tabular ML models to assess environmental risk.

Source Datasets & Attributions

This aggregated dataset builds upon key open-access research contributions:

PlantVillage Dataset
- Paper: Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). Using deep learning for image-based plant disease detection. Frontiers in Plant Science, 7, 1419. DOI: 10.3389/fpls.2016.01419
- Repository: GitHub - PlantVillage Dataset
15 Crop and 45 Disease and Healthy Dataset
- Citation: Jain, A. (2026). 15 Crop and 45 Disease and Healthy dataset (Version 1). Mendeley Data. DOI: 10.17632/8fr7grr73p.1

Limitations

Lighting & Occlusion: Images originate from controlled and semi-controlled settings; real field conditions (direct glare, severe shadows, partial leaves) might require targeted fine-tuning.
Class Imbalance: Minor imbalance exists between classes (e.g., Potato healthy vs. Tomato Bacterial spot). Class-weighted loss functions are recommended.

Citation & License

This dataset is shared under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

If you use this dataset in your research or project, please cite it as:

@dataset{crop_disease_dataset_2026,
  title     = {Crop Disease Image Dataset (5 Crops, 19 Classes)},
  author    = {IoT Based Crop Disease Detection System Project},
  year      = {2026},
  publisher = {Hugging Face},
  note      = {Created by combining PlantVillage and Mendeley datasets}
}