A non-destructive method to quantify the nutritional status of Cannabis sativa L. using in situ hyperspectral imaging in combination with chemometrics

Abstract

The cultivation of medicinal cannabis (Cannabis sativa L.) indoors is mainly aimed at the homogeneity of the chemical profile produced in flowers and leaves. Nutrient supply in different stages of development highly affects the chemical profile and yield. This study describes a novel approach using hyperspectral imaging as an in situ monitoring system for the nutritional status of cannabis plants under greenhouse conditions. Specifically, the quantification of nitrogen (N), phosphorus (P), and potassium (K) in individual leaves and at canopy level was investigated. Hyperspectral images of whole plants and individual leaves of a phytocannabinoid-rich genotype, grown under different environmental conditions, were acquired in the wavelength range of 400–1000 nm at different growth stages under natural daylight conditions with supplementary greenhouse light. A wide variation of nutrient levels was created by the use of different growth substrates, fertilizer levels, and vegetation lengths for plant cultivation. Pixels of single leaves were filtered out from the background using the normalized difference red edge index, while pixels of plant canopy were separated from the background using a supervised classification algorithm. Mean reflectance per leaf/plant was calculated. Partial least squares regression (PLSR) was able to predict N concentration with a root mean square error (RMSE) of 0.47 % (R² = 0.9) and 0.6 % (R² = 0.86) in plant canopy and single leaves, respectively. Satisfactory models could be obtained for P with RMSE of 0.07 % (R² = 0.74) and 0.08 % (R² = 0.73) in plant canopy and single leaves, respectively. An R² of 0.57 (RMSE = 0.48) was obtained for the prediction of K concentration at the plant canopy level, presumably based on the high correlation between K and N. In contrast, the prediction accuracy was insufficient for K at single-leaf level (R² = 0.13). After wavelength selection, PLSR achieved better model performance than full spectrum PLSR in all cases and significantly reduced the model size for N concentration in plant canopy. This study is a proof of concept, that hyperspectral imaging can be used as real-time sensing technique for nutrient quantification in cannabis greenhouse production, without interfering with growing conditions.

Plant phenotyping relevance

植物の栄養状態（N・P・K）を、ハイパースペクトル画像と化学計量学で非破壊推定する手法の開発・実証が中心であり、植物表現型計測の方法論に該当する。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.