Intelligent Bioelectrical Sensing and Deep Learning Framework for Non-Invasive Monitoring of Plant Alkaline Stress

Alkaline stress disrupts ion balance and physiological homeostasis in plants, yet its timely assessment remains challenging because conventional phenotyping methods are often destructive, discontinuous, or delayed relative to the onset of stress symptoms. In this study, we developed a non-invasive plant electrophysiological sensing framework for the identification of alkaline stress in Clivia. Thin-film patch electrodes were used to record bioelectrical signals under five alkaline gradients (pH 7.0, 7.5, 8.0, 8.5, and 9.0) in a controlled environment. The acquired signals were subjected to wavelet denoising and normalization, and were then analyzed using a dedicated deep learning model, the Spatial Channel Alkaline Stress Network (SCANet). To provide a more rigorous evaluation of generalization, model performance was assessed using plant-wise five-fold cross-validation. Under this protocol, SCANet achieved 97.51% ± 0.77% accuracy, 97.55% ± 0.75% precision, 97.51% ± 0.77% recall, and 97.52% ± 0.77% F₁-score, outperforming representative convolutional and transformer-based baselines. Ablation experiments further showed that both the spatial reconstruction module and the channel reconstruction module contributed to performance improvement, and that a 30 s input window provided the best balance between signal completeness and discrimination. These results indicate that plant electrophysiological signals can support accurate, non-destructive identification of alkaline stress levels under controlled conditions, and that the proposed sensing-analysis framework may be useful for stress phenotyping and intelligent monitoring of plant status.