Fig. 6: ML-powered plant stress status monitoring.
a Schematic of the ML architecture for signal preprocessing, feature extraction, supervised learning, stress identification, and species differentiation. b Heat maps of fluorescence changes in nanosensors monitoring different stress types and plant species over an hour. The intensity levels represent the changes in F/Fo of nanosensors in lettuce leaves under different stress conditions, including no stress, wound, heat, and flg22 treatment (top panel), as well as in response to mechanical damage across different plant species, including tobacco, lettuce, spinach, and pepper leaves (bottom panel). a Created or b partially created in BioRender. Hu, H. (2025) https://BioRender.com/jtldpq7. c, d t-distributed stochastic neighbor embedding plots for the no stress and stress datasets (c) and different species stress datasets (d). Visual clustering results showed the feature separation in two-dimensional space. e F1 scores (a metric of accuracy combining precision and recall) of different ML models for stress classification. f F1 scores of different ML models across an increasing number of ranges categorized by stress identification. The ranges of stress classification include 2 ranges (no stress/stress), 3 ranges (no stress/biotic stress/abiotic stress), and 4 ranges (no stress/flg22/wound/heat). g Confusion matrices of an XGBoost model displaying the classification accuracy for predicting each type of stress and different plant species.
