Abstract
Sensing rice drought stress is crucial for agriculture, and chlorophyll a fluorescence (ChlF) is often used. However, existing techniques usually rely on defined feature points on the OJIP induction curve, which ignores the rich physiological information in the entire curve. Independent Component Analysis (ICA) can effectively preserve independent features, making it suitable for capturing drought-induced physiological changes. This study applies ICA and Support Vector Machine (SVM) to classify drought levels using the entire OJIP curve. The results show that the 20-dimensional ChlF features obtained by ICA provide superior classification performance, with Accuracy, Precision, Recall, F1-score, and Kappa coefficient improving by 18.15%, 0.18, 0.17, 0.17, and 0.22, respectively, compared to the entire curve. This work provides a rice drought stress levels determination method and highlights the importance of applying dimension reduction methods for ChlF analysis. This work is expected to enhance stress detection using ChlF.