Multimodal_Cactaceae_Dataset_25

A total of 109 healthy, fruit-bearing plants were selected according to inclusion criteria that prioritized structural integrity, absence of disease symptoms, and accessibility for imaging and in-field measurements. Data acquisition combined high-resolution multispectral drone imaging with ground-based spectrophotometry in the 400–700 nm range, enabling the capture of both canopy-level spatial information and precise point-based reflectance signatures.

Imaging was performed using a DJI Mavic 3M equipped with RGB and multispectral sensors. For each plant, one nadir RGB photograph and a corresponding set of multispectral captures were obtained. These data were processed to generate three complementary products: (i) original RGB images and baseline NDVI maps provided by the manufacturer’s software, (ii) spatially adjusted RGB frames cropped, scaled, and registered to NDVI outputs to ensure pixel-level alignment and allow region-of-interest (ROI) analysis, and (iii) recalculated NDVI images derived directly from NIR and Red bands to produce independent vegetation index maps.

Spectral measurements were acquired with a YS45 field spectrophotometer using a standardized white reference for reflectance calibration. All spectral data were exported in .csv/.txt format and subsequently processed in MATLAB, where reflectance curves were stored as .mat files and plotted as PNG graphs for visual inspection. Metadata for each sample include phenological stage, acquisition date and time, geographic coordinates, temperature, illumination conditions, and sensor configuration parameters.

Together, these components form a hierarchically organized dataset that facilitates temporal tracking of individual plants and robust integration of image-derived features with field-collected spectra. Preprocessing steps included reflectance normalization, geometric registration of images, and standardization of file naming conventions, resulting in structured dataset ready for computational analysis.