Enhancing Morpho-Kinematic analysis for Plant Water Stress Classification through Leaf Movements

Abstract

Precise irrigation management requires robust classification of plant water stress. We expanded a morpho-kinematic (MK) framework that derives canopy-movement features from RGB time-lapse imaging evaluating how methodological refinements affect robustness and fine discrimination across four irrigation treatments representing distinct stress histories. The study tested both a biological (Agg) versus an isogonal (Unif) sectoring of the canopy image, within an additive scheme where to the baseline (i.e. flattened MK features, A0) were sequentially added non-linear descriptors (A1), irrigation-context variables (i.e. dry time, A2), and their interactions with baseline (A3). The multi-class problem was decomposed in biologically meaningful binary tasks, and the final classification confronted an adaptive - to the performance obtained in the out-of-fold predictions inside the leave-one-sample-out validation framework - linear opinion pooling (ALOP) ensemble, evaluated across its full parameter space, against hierarchical cascades (HCC). In our combined dataset from two sequential Lactuca sativa experiments (144 sample-days) ALOP median outperformed HCC in every configuration, while non-linear and contextual enrichments (A1-A2) produced consistent improvements in terms of prediction stability, variability (for ALOP), and balanced accuracy (BA). The highest balanced accuracy (median BA = 0.96) was reached under Unif scheme in A3, yet the Agg configuration in A2 achieved the best compromise between accuracy (median BA approx 0.91) and robustness. Concluding, this study identifies methodological pathways that strengthen resilience and transferability of movement-based water-stress classification, establishing a solid foundation for generalizable, low-cost phenotyping.