AI identifies tomato diseases in field images with 99% accuracy

AI model trained on 8,934 tomato leaf images detects seven disease types with 99% accuracy in field conditions

Researchers from Charles Darwin University and University of Peradeniya trained an AI model using 8,934 augmented tomato leaf images (from 890 raw photos) to identify seven disease types. The Inverted Residual Convolutional Block Attention Module (IR-CBAM) architecture achieves field deployment accuracy exceeding 99% for spotting bacterial spot, early blight, and other diseases. Published in Neural Computing and Applications (2026), the system targets resource-constrained environments like mobile phones to enable early disease detection in low-income farming contexts.

This approach could reduce pesticide use and crop losses by allowing farmers to intervene before diseases spread. If scaled to tomato-growing regions, it might lower food costs and health risks from chemical exposure—particularly relevant for areas where Sri Lanka’s 2018 average yield of 18.9 metric tons per hectare was recorded.

The impact hinges on real-world deployment scale, which remains unspecified. Accuracy applies specifically to field conditions with background clutter, and the yield figure reflects historical data that may not align with current agricultural practices. Researchers note the system’s effectiveness depends on local field conditions and mobile device accessibility.

Source: Phys.org