Development of a Machine Learning technique for the detection of late blight in potato leaves

Sanchez-Sanchez J.; Minguillo-Rubio C.; Arcila-Diaz J. · Journal of Agriculture and Food Research (2026) · DOI: 10.1016/j.jafr.2026.103109

Resumen

This research developed a machine learning technique for accurate detection of late blight in potato leaves, considering variations in image conditions and different cultivation environments. A model based on the ResNet50 architecture was implemented through transfer learning and fine-tuning, using a dataset of 2059 images captured with a conventional mobile device in Caserío Sapalí, Cajamarca, Peru. The method included image preprocessing (resizing to 600×600 pixels, normalization), data augmentation techniques (rotations, translations, zoom, horizontal flip), and dataset splitting into training (70%), validation (20%), and testing (10%) sets. Training was conducted in two phases: transfer learning with frozen convolutional layers for 30 epochs, followed by fine-tuning of the last 50 layers for an additional 46 epochs, employing the Adam optimizer and regularization callbacks. The model achieved outstanding performance metrics: 93% accuracy, 94% precision, 92% recall, 93% F1-score, and an AUC of 0.9852 on the test set. It is concluded that the developed technique constitutes an effective tool for automated diagnosis of late blight, without requiring advanced capture technology. Gradient-weighted Class Activation Mapping (Grad-CAM) analysis confirmed that the model's attention was consistently directed toward biologically relevant leaf tissue — chlorotic and necrotic areas — rather than background artifacts, validating the diagnostic reliability of the proposed system. The research demonstrates that the use of transfer learning with conventional mobile devices represents a viable and accessible alternative for implementing early disease detection systems in rural contexts with limited resources, contributing significantly to sustainable precision agriculture.

Instituciones: Universidad Señor de Sipán