AI-Driven 3D Electrophysiological Sensor for Non-Invasive Crop Health Monitoring

Smart agriculture that integrates real-time plant health monitoring sensors and artificial intelligence (AI) offers a promising solution for global food security and environmental sustainability.¹ Plant health is closely impacted by irrigation, as water deficiency leads to stunted plant growth while excess watering can impair root respiration and waste water resources. Plant electrophysiology (EP) signal monitoring can provide a direct and real-time reflection of plant water status accurately and effectively.² Conventional methods to measure plant EP signals are either invasive (intracellular) or non-invasive (extracellular). The invasive approach, though effective, is limited by short monitoring periods due to the potential diffusion of electrolyte into plant cells to cause inaccurate measurements and inevitable physical damages to plant tissues. The non-invasive methods, while avoiding tissue damage, suffer from inconsistent electrode contact to plant organs, especially with hairy plant surfaces such as tomato. Moreover, the interpretation of the collected EP signals is challenging due to the non-stationary and nonlinear nature of the signals.
Here, we present a universally applicable, non-invasive, conformable, and effective 3D-printed EP sensor system for real-time monitoring of tomato plant health under various irrigation conditions.³ The system comprises a 3D EP sensor and a portable Faraday cage for data acquisition, and a customized convolution neural network (CNN) for data analysis. Our findings demonstrate the 3D EP sensor system, with a signal resolution of 0.0122 mV, have lower and more precise contact resistance with hairy tomato leaves than flat thin-film EP sensor (2.10±0.52 MW versus 2.96±1.45 MW), offering highly constant, reliable, and sensitive tool for monitoring plant physical status. For the first time, we applied continuous wavelet transform to convert plant EP signals into 2D scalogram images to extract detailed time-frequency characteristics such as frequency ranges, signal intensity, morphology and continuity. This approach clearly differentiates between optimal irrigation and water stress conditions, achieving a classification accuracy of 86.91%, comparable to the widely studied RGB image-based method (86.37%). The high reliability, sensitivity, and accuracy of our non-invasive 3D EP sensor system make it an ideal tool for long-term real-time monitoring of plant health in smart agriculture. The discovery of plant EP signal variations in frequency characteristics also offers a powerful data analysis method for gaining deep insights into the mechanisms of plant signal transportation.
References
1. a) Science, 2010, 327, 828. b) Biosens. Bioelectron. 2023, 222, 115005.
2. a) Adv. Mater. 2021, 33, 2207764. b) Small, 2021, 17, 2104482.
3. Integrating AI-Driven 3D Electrophysiological Sensing for Real-Time, Non-Invasive Crop Health Monitoring and Intelligent Irrigation Management (submitted)