Early potato disease detection using spectral imaging and machine learning techniques: a review

Abstract

Potato is a globally important staple crop, requiring rapid and accurate disease detection to protect yield and quality. Phytopathological threats such as Root-Knot Nematodes, Early Blight, and Late Blight can spread quickly, often reaching damaging levels before visible symptoms appear. Conventional visual inspection methods are labor-intensive, prone to subjectivity, and ineffective for large-scale monitoring. to three broad spectral bands and cannot detect pre-symptomatic changes. Multispectral and hyperspectral imaging overcome these limitations by capturing dozens to hundreds of rich spectral signatures across the visible (VIS), near-infrared (NIR), and shortwave infrared (SWIR) regions, enabling the early detection of biochemical and structural changes in plant tissues. However, the high-dimensional and non-linear nature of spectral data demands advanced analytics. Computer vision and machine learning are effective at capturing complex patterns, handling large feature spaces, and adapting to variable conditions, enabling automated and accurate early-stage disease detection. This review provides a comprehensive synthesis of recent studies on potato disease detection based on spectral imaging and machine learning models, emphasizing emerging trends, ongoing challenges, and future opportunities for robust early-warning systems in precision agriculture.