Nanosensors convert H2O2 to machine-learnable thermal signatures in plants

Abiotic stresses including drought and salt can severely inhibit plant growth and represent major threats to agricultural productivity. Hydrogen peroxide (H 2 O 2 ) serves as a key signaling mediator in plant stress response mechanisms, making the real-time monitoring of its spatiotemporal dynamics essential for research in plant phenology. This study presents an innovative detection approach that employs nanosensors to convert subtle endogenous H 2 O 2 fluctuations at sub-micromolar concentrations into machine-readable infrared thermal signals. These thermal imaging data are processed via an advanced deep learning framework, enabling non-invasive in situ monitoring of H 2 O 2 in plants under stress. Experimental results demonstrate that the average classification accuracy of multiple deep learning models reaches 98.8% and 99.6% on test sets under drought and salt stress conditions, respectively. Unlike conventional methods that focus primarily on data processing and model refinement, our method emphasizes the data source itself, significantly enhancing classification accuracy by capturing thermal data with distinctive features. The integration of interdisciplinary techniques offers a non-destructive, rapid, and accurate means for early detection of various plant stresses, thereby providing a new perspective for the study of plant stress characterization. • A novel method combines nanosensors, thermal imaging, and deep learning for early plant stress detection. • Our approach captures hidden stress signals before visible symptoms appear. • Deep learning models (e.g., ResNet50) achieve 100% accuracy in stress classification. • Compared to previous technologies, our technology is non-destructive and more accurate.