English
Non-destructive Detection Of Cucumber Damping-off Disease Using Hyperspectral Imaging And Machine Learning
Hyun_Ju Kim, Jiyoung Min ,Sun Ha Kim, and Sang_keun Oh*
Department of Applied Biology, Chungnam National University, Daejeon, Republic of Korea
*Email: sangkeun@cnu.ac.kr
Hyperspectral imaging (HSI) captures hundreds of contiguous spectral bands simultaneously, enabling non-destructive assessment of plant physiological and biochemical status invisible to the naked eye. This capability makes HSI particularly valuable for early plant disease diagnosis. Damping-off disease, primarily caused by Pythium ultimum, poses a serious threat to cucumber seedling production. However, inoculated plants are visually indistinguishable from healthy ones during early infection.This study developed a pre-symptomatic, non-destructive detection system by combining HSI with machine learning. Cucumber seedlings were inoculated with P. ultimum and imaged using a hyperspectral camera over three growing seasons (2023–2025). Regions of interest (ROI) were automatically extracted using YOLO-based object detection, and discriminative wavelengths were selected through ANOVA-based variance analysis, vegetation index (VI) analysis, and spectral difference analysis. A Random Forest (RF) classifier trained on 50 ANOVA-selected wavelengths achieved 89% accuracy, outperforming the full-spectra model (87%, 204 wavelengths) while substantially reducing model complexity. Consistent spectral discrimination was confirmed in the 770 nm and 966–997 nm regions across all three experimental years, identified as key diagnostic wavelengths. These results demonstrate that feature-optimized HSI combined with machine learning can reliably detect damping-off infection prior to symptom onset, offering a practical tool for early disease management in protected horticulture.
