Identification of informative spectral ranges for predicting major chemical constituents in corn using NIR spectroscopy

• The corn NIR spectra was broken down under scope of overtones and combinations. • The sequential inspection of 129 sub-regions was done for each attribute of corn. • The more informative sub-regions and the corresponding bands were identified. • The interpretability was improved by the region-located important indexes. • The results are potentially useful for practical hardware or software applications. Many studies have been conducted using NIR spectroscopy to predict corn constituents; however, a systematic investigation of the spectral sub-regions under the scope of overtones and combinations has not been performed. In this study, the corn spectra were divided into second overtones (1100 - 1388 nm), first overtones (1390 - 1852 nm), and combinations (1852- 2498 nm). Then, using variable importance in projection and genetic algorithm, each region was inspected sequentially to identify the most informative sub-region for each attribute to improve interpretability. The identified spectral subsets were further tuned to select the most influential bands for each attribute. The sub-regions in combinations bands was most informative for predicting water (1908-2108 nm, 2 bands), oil (2176-2304 nm, 6 bands), and protein (2130-2190 nm, 3 bands), whereas the first overtones region was the best for predicting starch (1452-1770 nm, 5 bands). Results provided valuable information for potential hardware and software improvements.