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

Area of Science:

• Plant physiology
• Bioelectrical sensing
• Machine learning for agriculture

Background:

• Alkaline stress disrupts plant homeostasis, but early detection is difficult with current methods.
• Conventional plant phenotyping is often destructive, discontinuous, or delayed.
• Accurate and non-invasive stress assessment is crucial for plant health management.

Purpose of the Study:

• To develop a non-invasive framework for identifying alkaline stress in Clivia using electrophysiological signals.
• To analyze plant bioelectrical responses under varying alkaline conditions (pH 7.0-9.0).
• To evaluate the performance of a deep learning model (SCANet) for stress detection.

Main Methods:

• Utilized thin-film patch electrodes to record plant bioelectrical signals.
• Applied wavelet denoising and normalization to signal data.
• Developed and validated the Spatial Channel Alkaline Stress Network (SCANet) deep learning model.
• Employed plant-wise five-fold cross-validation for rigorous performance assessment.

Main Results:

• SCANet achieved high accuracy (97.51%), precision (97.55%), recall (97.51%), and F1-score (97.52%).
• The model outperformed baseline convolutional and transformer networks.
• Ablation studies confirmed the contribution of spatial and channel reconstruction modules.
• A 30-second input window optimized signal analysis.

Conclusions:

• Plant electrophysiological signals enable accurate, non-destructive identification of alkaline stress levels.
• The SCANet framework offers a promising approach for plant stress phenotyping.
• This technology can aid in intelligent monitoring of plant physiological status.