ABSTRACT
A total of 1,176 grain samples representing 10 different single- and double-mutant genotypic classes of specialty starch corn were used for developing various classification models based on near-infrared transmittance spectra. The genotypes used included amylose-extender (ae), dull (du), sugary-2 (su2), waxy (wx), ae wx, ae du, ae su2, du wx and du su2. Two-class classification models (only two genotypes compared) were developed using partial least squares analysis (PLS) while three-way and multiclass models were examined using principal component analysis (PCA). The effectiveness of the calibrations was evaluated by examining the percentage of unknown grain samples incorrectly classified. In general, two-class models performed better than multiclass models. However, they did not show improvement when discriminating among genotypes with overlapping amylose contents such ae du vs. ae and ae su2 vs. ae. Three-way models including double-mutants and their corresponding single-mutant counterparts had misclassification percentages typically <5% using 14 PCA factors but again, with the exception of models including genotypes with overlapping amylose contents such as ae du vs. ae vs. du. The best multiclass model using all 10 genotypic classes simultaneously revealed only two classes (ae su2 and du) with misclassification rates >10% based on 16 PCA factors. This study demonstrates that, depending on the material to be considered, near-infrared transmittance spectroscopy could be useful when segregation of specialty starch hybrids grain from other grain types is necessary.
