Hyperspectral image analysis: Utilizing sweetpotato potyviruses for image-based pathogen detection

Hyperspectral (HS) imaging, leveraged by machine learning (ML), offers a promising solution for addressing growing monetary and labor constraints in plant disease monitoring, but the lack of accessible open-source pipelines has limited adoption by plant pathologists. In sweetpotato, global production exceeds 90 million tons annually, but yields are threatened by a litany of pathogens, including the potyvirus sweet potato feathery mottle virus (SPFMV). SPFMV can reduce sweetpotato yields by 25-40%, yet detection is hindered by asymptomatic infections and labor-intensive detection methods. In this study we aimed to address limitations in detection methods by developing a high-throughout image based detection system. An HS image library of SPFMV-negative and SPFMV-positive plants was established, consisting of sweetpotato cultivars 'Beauregard' and 'Orleans', as well as the common indicator plant Ipomoea setosa. A custom python pipeline was developed, featuring scripts for preprocessing, masking, ML training and evaluation. Using this pipeline, a binary classification model achieved greater than 85% accuracy, with further evaluation identifying spectral regions linked to SPFMV infections, particularly those associated with chlorophyll-based vegetative indices. This study fills a critical gap by providing a simple, flexible pipeline tailored to plant pathologists, making HS image analysis and ML model training more accessible while also confirming its efficacy as a tool for virus detection in sweetpotato.