Online Discriminant Model of Blood Spot Eggs Based on Spectroscopy

To explore the best method for online detection of blood spot eggs, fiber-optic spectrometer was used to collect the transmission spectra of eggs. Three different kinds of wavebands were extracted using competitive adaptive reweighed sampling (CARS), interval partial least squares (IPLS) and successive projections algorithm (SPA), respectively. The discrimination results of the models established with partial least square discriminant analysis (PLSDA) and support vector machine (SVM) showed that the CARS-screened wavebands had the best modeling results. Through stepwise Bayes discriminant analysis (SBDA) based on the spectral characteristic variables of CARS, five wavelength variables (509, 511, 526, 571 and 599 nm) were determined to be the characteristic variables for the detection of blood spot eggs finally. The detection accuracy of blood spot eggs and normal eggs by Bayes discriminant model using these five variables was 95%. These results show that stepwise Bayes discriminant model can effectively simplify the characteristic wavebands and improve the prediction accuracy, indicating that it can be applied to the real-time online detection of blood spot eggs. Practical ApplicationsThis study explore three different kinds of spectral characteristic variables were extracted using CARS, IPLS and SPA, respectively, only five characteristic variables are required in the stepwise Bayes discriminant model. This model can avoid the influence of excessive spectral variables on the accuracy of the model, so it is feasible to apply this model to real-time online detection of blood spot eggs.