Discriminant analysis and comparison of corn seed vigor based on multiband spectrum

Abstract

Seed vigor is an important factor determining the viability of seeds, which is related to many seed-internal components. The vitality of seeds can be identified by assessing the internal information of seeds. In this study, seeds of the sweet corn variety “TuxpenoSweet” were used. Visible and near-infrared spectral ranges were assessed to explore the viability of seeds under different processing methods, and variable selection algorithms were used to optimize models. Heat damaged and artificially aged seeds were compared with normal seeds, and two spectrometers with different band widths were used to collect the spectra of seeds. Seed discrimination models were established based on full-wavelength data and effective variables were selected through competitive adaptive reweighted sampling. The discrimination accuracy of models based on both spectrometers exceeded 95%, and the discrimination accuracy of the model based on effective variables was higher than or consistent with that of the model established with full wavelengths. The experimental results corroborate the feasibility of using spectral ranges between 500–1100 nm or 1000–1850 nm to discriminate seed vigor. The model based on the effective variables extracted by competitive adaptive reweighted sampling reduces the amount of calculated data and either optimizes or maintains the accuracy of the discrimination rate. Effective variables, extracted from the spectrometer with the spectral range of 500–1100 nm, effectively reduced the test cost and the amount of data while ensuring high seed vigor discrimination accuracy.