HYPERSPECTRAL SPECTRAL LIBRARIES FOR CROP GENOTYPE AND VARIETY DISCRIMINATION: LINKING SPECTRAL PHENOTYPING WITH GENETIC DIVERSITY

Abstract

Crop breeding depends on characterizing genetic diversity, but phenotyping remains a bottleneck relative to genotyping, motivating interest in hyperspectral imaging, which captures continuous reflectance curves shaped by leaf pigments, water, nitrogen and structure, together forming a "spectral phenotype." Genotype-oriented spectral libraries linking this spectral phenotype rigorously to genetic diversity remain scarce, and remote sensing and genetics literatures have developed largely in isolation. This review addresses that gap by synthesizing evidence on the biological pathway from genotype to spectrum, spectral library design and metadata standards, preprocessing and feature-selection methods, machine learning and validation practices, and cross-scale sensing from field to UAV to satellite, drawing on studies across rice, wheat, maize, soybean, cotton and other crops. Findings show genotype and variety discrimination is achievable, particularly at leaf and seed level under controlled conditions, but reported accuracies are frequently confounded by growth stage and genotype-by environment effects and inflated by scan-level data leakage rather than genuine genetic signal. Discriminating wavelengths cluster in red-edge and pigment-sensitive visible regions for canopy studies and extend into shortwave infrared for composition-driven seed studies, while standardized, multi-environment libraries with grouped validation remain rare and genomic integration is concentrated in few groups and crops. A standardized, metadata-rich library framework coupling spectral phenotypes with genomic data could enable cost-effective germplasm screening, genomic prediction and improved breeding decisions, provided rigorous external validation and explicit biological interpretation accompany future work before spectral discrimination is trusted as a genetic proxy.