Multispectral reflectance database of Cannabis sativa
Description
Our research hypothesis postulates that different strains of Cannabis sativa grown for medicinal purposes exhibit distinctive multispectral reflectance signatures in the 410 nm to 890 nm range that can be used for nondestructive and low-cost classification. The accompanying database contains 800 spectral signatures, each with 2442 lambda reflectance values, providing detailed resolution of the optical data. These data were collected from a medical cannabis crop by randomly sampling 30 plants of two varieties and 20 plants of two other varieties, spanning three distinct phenological stages, from which both high and low leaf measurements were taken for each sample. Notable findings from this data collection will focus on identifying subtle spectral patterns and differences between varieties, as well as within-variety variability as a function of phenological stage and leaf position. The data can be interpreted to develop machine learning models capable of discriminating between medical cannabis varieties efficiently and without damaging the plant, representing a valuable alternative to current identification methods that are often destructive and costly, opening the door to applications in quality control, traceability and crop optimization in the medical cannabis industry.
Files
Steps to reproduce
The collection of the multispectral reflectance signature database of Cannabis sativa was meticulously carried out to ensure consistency and reproducibility. A RED TIDE USB650 spectroradiometer from Ocean Optics was used to acquire the spectra. This instrument was connected to an Ocean Optics fiber optic probe for direct measurement of reflectance on plant leaves. Illumination during signature capture came from ambient light in the greenhouse, where conditions (temperature, humidity, illumination and nutrients) were kept strictly controlled for all varieties: EMBERA CBD, YOCOTO CBD, LIMON CBD and VALLE DE LOS UMBRAS. To ensure the accuracy of reflectance measurements, a calibration was performed on each captured signature using a target reflectance reference panel, which allowed normalizing the data and obtaining absolute reflectance values. The sampling procedure was based on a random selection of 30 plants of the varieties EMBERA CBD and VALLE DE LOS UMBRAS, and 20 plants of LIMÓN CBD and YOCOTO CBD, which were marked at the beginning of the crop. Measurements were taken at three different phenological stages: early vegetative, early flowering and mid-flowering. For each leaf sample (both high and low leaves), three measurements were taken and averaged to obtain a single representative spectral signature. The instrument probe was held at a distance of approximately 1 cm from the leaf surface for each reading. Spectroscopy OceanView software was used for direct acquisition of spectral data from the spectroradiometer. Once acquired, the raw data were exported in .txt format. Subsequently, a simple normalization of the data was performed to correct for possible variations in light intensity and to facilitate comparison between signatures. For the analysis and processing of these data, the R and JASP software environments were used, allowing the 800 spectral signatures to be organized and prepared for use in training machine learning models.
Institutions
- Universidad Catolica de Manizales
Categories
Funders
- Cubikan Group