Vol. 2, Issue 2, Part A (2025)

Background: Aeroponic systems offer highly controlled root-zone environments that can substantially enhance crop growth and yield, but their full potential depends on precise, real-time monitoring of crop developmental stages. Multispectral imaging combined with artificial intelligence (AI) provides a promising, non-destructive approach for automated growth-stage classification and decision support in soilless production.
Objectives: This study aimed to 
•    Acquire a longitudinal multispectral image dataset of aeroponically grown crops across key phenological stages, 
•    Develop and compare predictive AI algorithms for growth-stage classification, 
•    Identify the most informative spectral features for stage discrimination, and 
•    Evaluate the relationship between predicted stages and final biomass and yield.
Methods: A total of 180 plants were cultivated in a controlled-environment aeroponic facility and imaged over 12 weeks, yielding 2, 160 quality-controlled multispectral image sets. Four growth stages early vegetative, late vegetative, early reproductive and late reproductive were annotated by expert agronomists. Canopy-level spectral features (raw bands, vegetation indices and simple texture metrics) were extracted and used to train conventional classifiers (support vector machine, random forest, gradient boosting) and deep learning models (1D-CNN, 2D-CNN). Model performance was evaluated using accuracy, macro-averaged precision/recall/F1-score and Cohen’s kappa. Temporal stability of predicted stage trajectories and associations with final biomass and yield were assessed through time-series analysis and regression modelling.
Results: The 2D-CNN achieved the highest overall accuracy (93.5%) and macro F1-score (0.92), significantly outperforming traditional machine-learning models. Confusion was largely confined to transitions between late vegetative and early reproductive stages, while early and late stages were classified with very high precision. Feature importance analysis highlighted red-edge and NIR-based indices as dominant predictors. Predicted early-stage metrics, particularly the timing of the late vegetative-early reproductive transition, explained up to 71% of the variance in final biomass and 68% of yield. Plants with earlier predicted transitions exhibited significantly higher yields than those with delayed transitions.
Conclusions: Predictive AI algorithms applied to multispectral imaging can accurately and stably classify growth stages in aeroponic crops, while providing early indicators of biomass and yield potential. Integrating these models into aeroponic management systems can enable stage-aware fertigation, lighting and harvest decisions, supporting more efficient, predictive and autonomous precision horticulture.