Tissue-Region-Resolved Visible–Near-Infrared Hyperspectral Estimation of Nutritional and Structural Traits in Dendrocalamus brandisii Shoots

Bamboo shoot quality is determined by sugars, proteins, water status, and cell-wall components that are conventionally quantified by destructive, low-throughput assays. Here we tested whether visible–near-infrared (VIS–NIR) hyperspectral imaging over 400–1000 nm can estimate five nutritional and structural traits—soluble sugar, protein, moisture, lignin, and cellulose—in Dendrocalamus brandisii shoots, and whether the tissue region from which spectra are taken matters. Spectra were acquired from 102 shoots, and partial least squares regression (PLSR), Ridge regression, and principal component analysis–Ridge regression (PCA–Ridge) were compared under repeated held-out validation stratified by shoot length class. Under this internal validation, PCA–Ridge gave the highest mean test-set R2 for soluble sugar (R2 = 0.64 ± 0.04) and lignin (R2 = 0.51 ± 0.06), PLSR for protein (R2 = 0.62 ± 0.05), and Ridge for moisture (R2 = 0.52 ± 0.04). Cellulose was not reliably predictable (R2 ≈ 0.04). Modeling the outer and inner tissue surfaces separately showed that the most informative region is trait-dependent: the inner tissue surface provided stronger predictive information for protein and moisture, while the outer surface showed a slight advantage for soluble sugar; for lignin, the regional difference was small. Within length classes, soluble sugar was positively correlated with hardness in the 30 cm class (r = 0.728, BH-adjusted p = 0.005). VIS–NIR hyperspectral imaging combined with regularized regression thus offers a workable route to estimating soluble sugar, protein, moisture, and lignin (cellulose was not estimable in the 400–1000 nm window) from a single acquisition in D. brandisii shoots, with trait-specific ROI placement. Models were trained on a single sampling batch that included six shoots of other bamboo species, and require independent D. brandisii-only validation across sites, seasons, and instruments.