Non-destructive detection of dry matter content based on Vis-NIR spectroscopy and wavelength selection

A rapid and non-destructive method for predicting dry matter (DM) content in leeks was developed using visible and near-infrared (Vis-NIR) spectroscopy, coupled with what we believe to be a novel wavelength selection algorithm. Reflectance spectra (397.7-1716.7 nm) were acquired from 288 leek samples collected from three production areas in Nantong, China, and DM content was determined by oven-drying. The full-spectrum partial least squares (PLS) model yielded moderate prediction accuracy, with RP2 of 0.7963 and RMSE_P of 1.14%. To improve performance, the iterative ranking-based variable elimination PLS (IRIVE-PLS) algorithm was proposed, which integrates multiple importance metrics to iteratively eliminate uninformative wavelengths. The algorithm autonomously identified the red-edge region (680-780 nm) as the most informative spectral feature, enriching its proportion from 9.5% in the full spectrum to 10.6% in the selected set. The IRIVE-PLS model achieved excellent prediction performance, yielding RP2 of 0.9683 and RMSE_P of 0.45%, significantly outperforming conventional wavelength selection methods. The proposed approach provides an accurate, interpretable, and non-destructive alternative for leek quality assessment, with strong potential for online sorting applications in the vegetable industry.