sairiamu/maize_desease_dataset

AgroSense Maize Crop Health Dataset

Dataset Summary

The AgroSense Maize Crop Health Dataset is a curated computer vision dataset designed for multi-class image classification. The collection focuses on identifying maize (Zea mays) health conditions, encompassing key agronomic challenges such as nutrient deficiencies, plant diseases, pest damage, and healthy control plants.

The primary objective of this dataset is to support automated plant pathology, computer vision models, and precision agriculture applications.

---

Dataset Structure

Data Splits

The dataset consists of 18,031 total images divided into three canonical splits:

Split	Number of Examples	Size
Train	13,113	~82.08 MB
Validation	86	~26.73 MB
Test	4,832	~193.93 MB

Data Fields

• image (PIL.Image): Image object containing the visual sample of the maize leaf, stem, or ear.
• label (ClassLabel): Integer corresponding to the target health state or condition.

Category Overview

The visual samples cover 12 categories organized into four primary domain groups:

1. Nutrient Deficiencies:
   • Nitrogen Deficiency
   • Phosphorus Deficiency
   • Potassium Deficiency
   • Combined Zinc & Nitrogen Deficiency
2. Pathological Diseases:
   • Leaf Blight (Exserohilum turcicum)
   • Common Rust (Puccinia sorghi)
   • Ear Rot (Fusarium / Aspergillus spp.)
   • Gray Leaf Spot (Cercospora zeae-maydis)
3. Pest Infestations:
   • Fall Armyworm (Spodoptera frugiperda) damage
   • Maize Stem Borer (Busseola fusca) damage
4. Healthy Control:
   • Healthy foliage and stalks without visible pathology or nutrient stress.

---

Data Curation & Quality Assurance

To ensure robust benchmark reliability, the image set underwent standardized quality control processes prior to dataset publishing:

• Exact Duplicate Removal: MD5 hash validation was applied across all raw images to eliminate identical files.
• Label Disambiguation: Instances where identical image files contained conflicting category labels were filtered out.
• Resolution & Color Standardization: Images were unified into the RGB color space, low-quality/corrupted thumbnails were removed, and higher-resolution images were standardized to a maximum dimension of 640 pixels while preserving aspect ratios.
• Data Leakage Mitigation: Groupings based on perceptual hashing were utilized during partition creation to isolate visual near-duplicates into identical splits, ensuring strict separation between training, validation, and test sets.

---

Considerations for Dataset Usage

• Class Imbalance: Specific categories—such as stem borer and ear rot—have lower total sample counts compared to foliage-based nutrient deficiencies. Users may need to apply class-weighting or resampling strategies depending on their task.
• Environmental Variability: Visual samples include varied field conditions, lighting environments, and background textures. Evaluation on real-world field photography is encouraged for field deployment tasks.

---

Additional Information

Licensing

The original image licensing terms were not specified in the initial dataset metadata release. Please contact the dataset maintainers for commercial usage permissions.

Citation

If you utilize this dataset in your work, please cite the Hugging Face repository reference:

@misc{agrosense_maize_dataset,
  title={AgroSense Maize Crop Health Dataset},
  author={Sairiamu Namuti},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/sairiamu/maize_desease_dataset}}
}