Detecting Nitrogen Deficiency on Lettuce by Phenotype Observation Integrated with Machine-learning Technology

Objective An improved method over the conventional phenotype observation in detecting nitrogen deficiency (NO) on lettuce for facility agriculture was developed with the aid of machine-learning technology.

Method The purple-leaf lettuce, ‘Red Crinkle’, was hydroponically grown in a greenhouse with varied degrees of deficient N supply for the experiment. Multi-dimensional indicators on plant morphology, color, and texture of lettuce canopy were monitored. After standardization, collected data were screened by the principal component and correlation analyses to construct a mathematical model with the aid of various machine-learning methods.

Results The PCA analysis on the measured indicators showed a significant clustering on the increasing NO in lettuce. Ten morphological, color, and texture indicators on lettuce canopy were selected by a correlation analysis for data entry of the machine-learning with redundancy eliminated. Different machine-learning methods varied in performance in differentiating the normal from NO plants, during the entire growth cycle or at specific stage. The random forest model outperformed on comprehensive diagnosis and detecting NO at early growth stage of lettuce.

Conclusion A method of detecting NO in hydroponically grown Red Crinkle Lettuce by integrating the observation of phenotypic characteristics of lettuce canopy with a machine-learning method was developed. A mathematical model capable of reliably differentiating NO plants from normal ones was established for an improved facility agriculture operation of the vegetable.