Classification of SPVD resistance in 72 sweet potato accessions using machine learning and multispectral imagery

A total of 249 sweet potato clones, preliminarily classified by breeders into Resistant, Tolerant, and Susceptible groups based on their response to Sweet Potato Virus Disease (SPVD), were evaluated using two criteria: (i) expert visual assessments of SPVD severity and (ii) yield reduction of SPVD-infected clones relative to their corresponding control treatment. In parallel, features extracted from multispectral and thermal aerial imagery of the sweet potato canopy-acquired at solar noon to capture maximum plant stress-were used to train machine learning models aimed at classifying clones according to their resistance level to SPVD.

The results demonstrate the potential of image-based phenotyping to support and enhance virus-resistance screening in sweet potato breeding programs.