Estimation of tissue nitrogen content in hydroponic basil using low-cost RGB imaging and deep learning-based computer vision

Abstract Nitrogen is a critical macronutrient for plant development, yet conventional monitoring methods are often invasive, labour-intensive and costly. This study evaluates affordable, non-invasive RGB-stereo imaging to estimate nitrogen content in hydroponically grown basil in a vertical farm. Daily top-down images were used to derive plant height, projected leaf area, growth rate and the Normalised Green-Red Difference Index, which were related to chemically measured nitrogen. Plant height and growth rate emerged as the most reliable predictors, explaining over 65.7 % of the variance in nitrogen content in the full dataset ( <m:math xmlns:m="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <m:msubsup> <m:mrow> <m:mi>R</m:mi> </m:mrow> <m:mrow> <m:mtext>adj</m:mtext> </m:mrow> <m:mrow> <m:mn>2</m:mn> </m:mrow> </m:msubsup> <m:mo>=</m:mo> <m:mn>0.613</m:mn> </m:math> ${R}_{\text{adj}}^{2}=0.613$ , n = 106) and over 76.4 % in a reduced dataset ( <m:math xmlns:m="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <m:msubsup> <m:mrow> <m:mi>R</m:mi> </m:mrow> <m:mrow> <m:mtext>adj</m:mtext> </m:mrow> <m:mrow> <m:mn>2</m:mn> </m:mrow> </m:msubsup> <m:mo>=</m:mo> <m:mn>0.714</m:mn> </m:math> ${R}_{\text{adj}}^{2}=0.714$ , n = 73). The findings demonstrate that low-cost imaging can support data-driven, site-specific nutrient diagnostics in vertical farming while reducing destructive sampling.