A Feature Domain Space Transfer Method for Improving Identification of Maize Haploid Seed Based on Near-infrared Spectroscopy

Near-infrared Maize haploid identification technology enables the automatic identifying haploid seed efficiently. However, it is likely that the same identification task may degeneration another equipment when in large-scale applications. From the view of the noise influence in spectra, a method based on feature domain space transfer is proposed. The feature space of the current device is rectified by the feature domain space of the reference device. Through the transfer of the better feature space, the identification performance is able to be improved. The results of the proposed method show it improves the average identification accuracy referring to the results of without transfer, when the classifier is back propagation neural network (BPNN). It also performs better than the method of the spectral domain transfer of piecewise direct standardization (PDS).