Multi region dataset improves crop disease recognition across regions
Multi-Crop Leaf Disease Recognition: A Unified Benchmark and Cross-Region Study
Computer Vision and Pattern Recognition
Summary
Crop disease recognition using AI often looks very accurate when tested on limited data collected under lab conditions. The authors combined six leaf disease datasets from different parts of the world into one large, diverse collection called MLD to study how well AI models generalize across regions. They found that models trained in one region perform poorly when tested on others due to domain differences. Training on the combined dataset helps improve performance across regions, and a hierarchical method that considers crop type before disease also provides small improvements. Overall, better data diversity matters more than complex model design for this task.
What this means in practice
- •For agriculture technology developers: Develop and evaluate AI tools for identifying crop diseases that perform robustly across multiple regions by using the unified MLD dataset and protocols.
- •For precision farming equipment makers: Improve disease detection features in farm equipment by training models that generalize better to real-world cross-region crop leaf conditions using the MLD dataset.
Authors
Rosemary Nalwanga, Sebastian Bunda, Godliver Owomugisha, Luuk Spreeuwers, Estefania Talavera Martinez
Abstract
Deep learning models for crop leaf disease recognition routinely report near-perfect accuracy yet are typically trained and evaluated on a single dataset collected under controlled laboratory conditions, leaving their behavior under realistic cross-region domain shift poorly understood. We introduce MLD (Multi-crop Leaf Disease) dataset, a unified multi-region benchmark that combines six public crop-disease datasets from the USA, Asia, and Africa into a shared hierarchical taxonomy spanning 18 crops, 56 crop-disease classes (including one healthy class per crop) making 167,427 images. We define standardized single-source and pooled multi-source evaluation protocols that explicitly probe cross-region generalization. We also investigate whether exploiting the inherent crop-to-disease dependency via a hierarchical formulation (HiLeaD) that conditions disease prediction on the predicted crop improves recognition under cross-region shift. Under the HiLeaD, the model trained on PlantVillage achieves 99.07% in-domain disease F1 but collapses to 12.88% when tested on PlantDoc, exposing a severe cross-region domain gap. The model trained on the pooled MLD dataset partly recovers cross-region disease F1 from 12.88% to 39.64% on PlantDoc (HiLeaD), achieving a 26.76 percentage point improvement. The hierarchical formulation provides a consistent additional gain, ranging from 1.71 to 4.88 percentage points in disease F1 over the flat baseline under the MLD dataset indicating that progress in this area is currently limited more by data coverage and diversity than by model design.