Crop Disease & Pest Detection Image Dataset

This is a synthetic dataset generated from high-quality expert-labelled seed data. All records are algorithmically derived — statistical distributions, inter-field correlations, and annotation characteristics faithfully replicate real-world patterns from the source data, while ensuring no real individual, organisation, or transaction can be identified or reconstructed.

The African Crop Disease & Pest Detection Image Dataset contains 80K+ high-resolution field photographs collected across Nigeria, Ghana, Cameroon, and Côte d'Ivoire — four countries that together account for a significant share of West and Central Africa's cassava, maize, and cocoa production. Images were captured under varying lighting and weather conditions using standardised smartphone protocols, ensuring realistic distribution of image quality for field-deployment models.

Each image is annotated by plant pathologists and certified agronomists with: a primary disease or pest class (29 distinct classes across the three crop types), a severity score on a 0–4 scale aligned with CABI severity standards, bounding-box or pixel-level segmentation masks (where available), and crop growth stage at time of capture. The class taxonomy covers fungal, bacterial, and viral diseases as well as major pest species including fall armyworm, cassava mealybug, and cocoa mirids.

The dataset is split into train / validation / test partitions stratified by crop type, country, and severity distribution. A held-out synthetic augmentation set generated via diffusion-model upsampling is included for regularisation experiments. All metadata fields are stored as JSON sidecar files compatible with standard annotation formats (COCO, YOLO, Pascal VOC).