Raman-based genetic purity identification of soybean seed production on line

Abstract

Raman spectroscopy can identify differences between a variety of seeds, but the difference between the hybrid seed, which is masked by significant signals such as internal components of the seed and external environment disturbance, and its parents is too small, and there is no prominent presentation showing variety differences in the spectral data but only slight changes in the local spectral bands. Therefore, we decomposed soybean spectra based on the empirical mode decomposition method and found that the Hilbert envelope spectrum of the decomposed intrinsic mode components presented local features that the raw spectrum could not detect. In this work, we tested the genetic and physical purity of eight varieties based on the combination method of empirical mode decomposition and Hilbert envelope spectrum transformation. The results demonstrated that the back propagation network identification model based on empirical mode decomposition and Hilbert envelope spectrum transformation had better identification accuracy, precision, recall, and F1 score (94.74, 95.83, 95.83, and 95.83) and the shortest iteration time (56.6 s) than traditional methods. This identification model realized fast and accurate identification of the genetic purity with its parents, as well as the physical purity, which meets the requirements of online production. Consequently, the Hilbert envelope spectrum transform of intrinsic mode components can not only highlight local features but also significantly compress spectral data at different time scales, which provides a promising tool for detecting minor local differences.