Data Augmentation for Reagent-less Spectroscopy Point-of-Measurement in Hydroponics

Nutrient quantification is crucial for hydroponic systems. However, reagent-less spectral quantification of nitrogen (N), phosphate (P), and potassium (K) encounters challenges in identifying information-rich spectral signals and separating interference from each component. We introduce the concept of 'information specificity, ’ as opposed to 'chemical specificity, ’ which requires physical or chemical reactions to isolate information from a specific nutrient in a complex hydroponic mixture. , the optical configuration used transmittance fiber optics with six illumination fibers and a center collection fiber. To study spectral interference and specificity of information of NPK in Hoagland nutrient solutions, a UV–Vis Deuterium/Halogen light source was used, emitting in the 200–550 nm range. 'Information Specificity' in spectral quantification is paramount to understand if the constituent is being quantified by its unique information and not by a spurious relationship present in the knowledgebase. This is especially relevant when resilient systems are necessary for application in harsh hydroponic conditions, where contamination that leads to spectral interference occurs. Herein, we show that nutrient Hoagland solutions exhibit spectral information specificity for N, P, and K, and this information is preserved in more complex samples obtained from uncontrolled fertirrigation production sites, allowing maximization of information about NPK, minimizing information about interferents, and allowing better predictions under extrapolation. We also present a method for overcoming extrapolation difficulties by data augmentation of the knowledge base by combining the spectral features of a small group of outlier samples and Hoagland nutrient solutions, greatly increasing the prediction accuracy of unknown samples under blind testing.