Waterlogging stress, as one of the major abiotic stress factors, imposes significant constraints on the early growth process and yield potential of rapeseed, representing a critical challenge in rapeseed production. Traditional methods for assessing waterlogging stress level (WSL) rely on manual measurement of morphological and physiological traits, which are time-consuming and labor-intensive. In this study, the morphological, physiological and photosynthetic traits were quantitatively analyzed under different waterlogging durations to establish four WSL: normal (0 days), mild (2–4 days), moderate (6–8 days) and severe (> 10 days). Then, a dataset of 4800 low-cost RGB images were classified the WSL based on this criterion, and all these images were input into five deep learning models for training and testing. In addition, five input sizes (224 × 224, 299 × 299, 320 × 320, 384 × 384 and 448 × 448) were tested for selecting the optimal performance. The results showed that the ResNetRS-50 model consistently exhibited superior performance across all tested image sizes, achieving the highest classification accuracy of 92.08% with an input size of 448 × 448. Furthermore, a graphical user interface was developed using PyQt6 based on the optimal model (ResNetRS-50), enabling visual identification and display of WSL. This study demonstrated the effectiveness of low-cost RGB images and deep learning models in WSL classification of rapeseed seedlings, and provided a practical framework for large-scale waterlogging stress monitoring and precise field management.
Ji et al. (Mon,) studied this question.