Lightweight AI Spots Potato Blight With 99% Accuracy and Explains Its

Researchers in India and the United States have built SpudNet5-R3, a compact AI model designed to identify potato late blight and early blight from a single leaf photo with near-perfect performance. Described in Discover Artificial Intelligence, the system uses a custom five-layer convolutional neural network with a residual connection placed at the third convolutional block to preserve fine spatial details and avoid gradient issues. Trained and tested on two public datasets—PlantVillage (2,152 images) and a more varied Potato Leaf Disease dataset (4,072 images)—the model was evaluated after resizing images to 224×224 and applying preprocessing, with CLAHE selected. On PlantVillage, SpudNet5-R3 achieved 99.07% test accuracy and a 99% macro F1-score, misclassifying just 2 of 216 test images.