Identification of Soft Rot using Plant Signature in Tuber Crops

Ginger is a high-value tuber crop with significant economic and medicinal importance, yet its productivity is severely threatened by soft rot disease, which often causes devastating yield losses. Conventional detection relies on visual symptoms that emerge only at advanced stages of infection, limiting timely intervention. To address this gap, in this paper we propose a phenology aware, multi-scale detection framework integrating ground-based spectral data with satellite remote sensing for early and regional-level identification of soft rot. Spectral signatures of healthy and diseased plants were collected using a handheld spectroradiometer and stratified across three growth stages: early (30–45 days), mid (60–90 days), and late (?120 days). Sentinel-2 and MODIS imagery were employed to extend detection capabilities to landscape scales. Preprocessing included noise reduction, atmospheric correction, and the derivation of disease-sensitive vegetation indices. Machine learning models (Random Forest, SVM) and deep learning architectures (CNN, U-Net) were evaluated for both plant- and regional-level classification. Results showed that Random Forest achieved 94.2% accuracy at the plant scale, while U-Net segmentation produced regional maps with a Kappa coefficient of 0.87. The study demonstrates that growth-stage awareness enhances detection robustness. Future work will focus on expanding datasets across agro-climatic zones and integrating UAV hyperspectral imaging for improved scalability.