NIR spectroscopy combined with chemometric algorithms has been widely used for seed authenticity detection. However, the study of seed genetic distance, an internal feature that affects the discriminative performance of classification models, has rarely been reported.
Therefore, maize seed samples of different genotypes were selected to investigate the effect of genetic distance on the authenticity of single seeds detected by NIR spectroscopy. Firstly, the Support vector machine (SVM) model was established using spectral information combined with a preprocessing algorithm, and then the DNA of the samples was extracted to study the correlation between genetic and relative spectral distances, the model identification performance, and finally to compare the similarities and differences between the results of genetic clustering and relative spectral clustering.
The results were as follows: the average accuracy of the models was 93.6% (inbred lines) and 93.7% (hybrids), respectively; Genetic distance and correlation spectral distance exhibited positive correlation significantly (inbred lines: r=0.177,
This study reveals the relationship between the genetic and relative spectral distances of seeds and the accuracy of the model, which provides theoretical basis for the establishment of the standardized system for detecting the authenticity of seeds by NIR spectroscopic techniques.