Discriminant analysis of maize haploid seeds using near-infrared hyperspectral imaging integrated with multivariate methods

Abstract

Doubled haploid (DH) technology is one of the most practical and innovative methods used to effectively accelerate the process of maize breeding. Quickly and precisely separating haploid seeds from maize kernels with a large number of diploids is crucial to improve the efficiency of maize DH breeding. This study explored the applicability of near-infrared (NIR) hyperspectral imaging (HSI) to identify the maize haploid and diploid of seeds. The effects of maize seed marker methods (colour and oil content markers) and embryo orientations (embryo and non-embryo) on the identification model were considered. Three variable selection methods, namely, sub-window permutation analysis (SPA), competitive adaptive reweighted sampling (CARS) and uninformative variables elimination (UVE), were used to obtain optimal wavelengths related to the difference index of maize haploid and diploid seeds. While the full wavelength was replaced by the optimal wavelengths, the identification accuracies for the maize haploid and diploid seeds of SPA-PLSDA, CARS-PLSDA and UVE-PLSDA were 90%, 90% and 89.38%, respectively. The 20 most optimal wavelengths were determined following a comprehensive analysis of the three selection methods and the PLSDA model was built. The model achieved an accuracy of 90.31%, exceeding that of the PLSDA model with the full 216 wavelengths (89.06% accuracy). The results demonstrate the suitability of the model for two common colour and oil content markers, and both embryo and non-embryo maize seeds. This approach significantly contributes to the efficient and broad practicability of maize haploid screening solutions. • NIR-HSI imaging technology was applied for identification of haploid and diploid in maize seeds. • Three variable selection methods (i.e. SPA, CARS, UVE) were used for determination of optimum wavelengths. • The classification model was built by PLSDA algorithm, which considered different marker methods (color marker and oil content marker) and embryo orientations (embryo and non-embryo) of maize seed. • Applicability of the developed analytical method for classifying haploid and diploid of the maize seeds was demonstrated.