Toward Smart Agriculture: Deep Learning Approaches for Rice Salinity Stress Monitoring

Rice is one of the most widely consumed food crops globally. However, high salinity levels, especially during the seedling stage, can significantly affect rice productivity. Early identification and development of salinity-tolerant rice varieties are therefore crucial to prevent yield loss. Traditionally, the classification of salinity stress in rice seedlings relies on visual inspection and the Standard Evaluation System (SES), which requires human expertise. This manual approach is not only time-consuming but also prone to errors, leading to inaccurate classifications. This study presents a deep learning approach to classify salinity stress in rice seedlings using leaf images. Four pre- trained models—Inception V3, Visual Geometry Group 16 (VGG16), Visual Geometry Group 19 (VGG19), and Residual Neural Network (ResNet)—were trained, with VGG16 showing the best performance. The model classifies the rice seedling images according to the scores as Grade1, Grade3, Grade5, Grade7, Grade9 based on physical conditions, where Grade 1 indicates healthy seedlings, and higher grades (3, 5, 7, 9) represent increasing levels of salinity stress. By employing deep learning over traditional methods, the need for human expertise in salinity stress assessment is reduced, minimizing errors and saving time. Key evaluation metrics such as accuracy, precision, recall, and F1-score demonstrate the effectiveness of this approach. This research highlights the potential of automated systems in agricultural monitoring, enabling more efficient and accurate salinity stress assessment in rice cultivation.