NIR-hyperspectral imaging and machine learning for non-invasive chemotype classification in Cannabis sativa L

Abstract

The current public acceptance rate towards medical cannabis feasibility has led to a worldwide increase in this plant species production. Nevertheless, the currently transforming legal framework does not prevent the originally unlawful knowledge around cannabis breeding, which lacks quality control regulations or standards for correct manufacturing processes, a fact that could subsequently lead to uncontrolled and even harmful crop products. In this line, the objective of this work was to develop a non-invasive methodology for cannabis chemotype classification in different cultivars during the plant cultivation process, in order to keep undoubtful production control over cannabis crops. Hence, hyperspectral imaging (HSI), coupled with various multivariate data analysis approaches, such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA), enabled the non-invasive in-situ analysis of the plants. Hence, two PLS-DA classification models were trained with the plant spectral data for three chemotypes, based on the cannabinoid content of the plant inflorescences, with the difference between both approaches being the regard of the stem part of the plant as a bias. Thus, obtained sensitivity and specificity values in the inflorescences were 0.845/0.845 for Chemotype I, 0.954/0.920 for Chemotype II, and 0.888/0.925 for Chemotype III. At last, a hierarchical PLS-DA, which considered the stem as a bias, presented an overall 94.7 % trueness in the external validation of 57 different plant individuals, divided as 92.3 % trueness for chemotype I, 100.0 % trueness for chemotype II and 88.9 % trueness for chemotype III. Based on these results, the proof of concept for comprehensive agricultural control of cannabis crops through a non-invasive analytical technique was demonstrated, a previously unproven fact. Therefore, this work could further pave the way for non-invasive technology development for horticultural quality control in medical cannabis productions, as this emerging industry will require strict control over the cannabis chemotypes, with the strong advantage of avoiding destructive and time-consuming analytical techniques such as chromatography.