An optimization strategy for detection of fertile pigeon egg based on NIR spectroscopy analysis

During the incubation of pigeon eggs, unfertilized eggs not only cause a waste of hatching resources but may also cause contamination from exploder eggs for other eggs. The fertilization status of pigeon eggs can be determined using the visible near-infrared (Vis-NIR) spectroscopy technique, allowing for the removal of unfertilized eggs. We have constructed a strategy for the individual and combined use of mean centralization, multivariate scatter correction, and standardized spectral preprocessing methods. The effect of the individual and combined preprocessing methods on the performance of the classification model under this strategy is explored using the mobile window partial least squares discriminant analysis (MWPLS-DA) model combined with grid search technology to find the optimal discriminant model. Experimental results have shown that the model effort of the MWPLS-DA model combined with the combined preprocessing method is generally better than the model with the individual preprocessing method. The model based on the combination of standardized and multi-scattering correction preprocessing methods (STD-MC) achieved higher classification accuracy than the model based on other preprocessing methods. The classification accuracy (ACC) of the model performed on the test set is 98.60%, which meets the requirements for fertilization identification of pigeon eggs.