무인기 기반 초분광 영상을 이용한 보리 습해 조기 탐지

초록

Barley is a cereal crop that is highly susceptible to waterlogging stress, making early diagnosis and rapid response essential for minimizing yield loss. This study developed machine learning classification models to diagnose waterlogging stress in barley using UAV-based hyperspectral imagery acquired at seven time points at 2-day intervals from March 21 to April 3, following waterlogging treatment initiated on March 19, 2026. The potential for early detection was also compared with the conventional SPAD measurement method. First, representative bands in the Blue, Green, Red, Red-edge, and near-infrared (NIR) regions were selected using two-sample t-tests on canopy reflectance between the waterlogged and well-watered treatments for both the entire time-series dataset and each individual time point. Vegetation indices calculated from these representative bands were then progressively reduced from five to one, and machine learning classification models were developed for each combination. For the entire time-series dataset, the K-nearest neighbor (KNN) model using four vegetation indices was identified as the optimal model, providing stable classification performance while minimizing misclassification. For the time-series-specific models, the KNN model consistently maintained reliable classification performance using only two vegetation indices, regardless of the image acquisition date, including the early stage immediately after waterlogging treatment on March 21. These results suggest that UAV-based hyperspectral imagery, combined with a small number of vegetation indices derived from representative spectral bands, has strong potential for the practical early detection of waterlogging stress in barley.

키워드