| In combination with the chemometrical methods,the infrared spectroscopy(IRS) analysis technology was used for study on discrimination of cocoon sexuality,in which a predicting accuracy was used as the evaluation index of calibration model.In order to provide early theoretical basis and concrete detection method for the practicality of cocoon sexuality discrimination using IRS analysis technology,several essential issues on IRS analysis technology in building the IR forecast modeling were investigated in terms of the spectral preprocessing methods,the derivative-smoothing window numbers choice,the building parameters choice,the comparison of modeling effect among different instruments.The main contents and conclusions are as follows:5 different varieties of cocoon samples were collected for the study,and the infrared spectra of samples were detected by different types of spectrometers,namely WQF-400N FT-NIR, NIR256-2.2T2 and TENSOR27 FT-IR which were from different manufacturers.All of the infrared spectra of cocoons detected by TENSOR27 FT-IR spectrometer after the second derivative,S-G smoothing,normalizing and suitable forecasting sample set choice,were used to build forecasting model,using PLS+BP regression analysis.The predicting accuracy of cocoons from Suzhen×Chunguang race was 92.9%,and the predicting accuracy of cocoons from Chinese race was 94.3%.With the PLS cluster analysis and the PCA cluster analysis of the near-infrared spectrum of cocoons from group 1 detected by WQF-400N FT-NIR and NIR256-2.2T2 spectrometers,the results show that PLS cluster analysis result was better than PCA cluster analysis result.PLS+BP calibration model was established with the comparison of predicting accuracy under different number of derivative-smoothing window,PLS principal components and BP network hidden-layer units.The choice of model parameters was very important to prediction ability of model.The prediction ability of model could be improved effectively,after choosing the best model parameters combination.The best WQF-400N FT-NIR type forecast model results were: the predicting accuracy of cocoons from Chinese and Japanese race was 96.0%and 100%, respectively.The best NIR256-2.2T2 type forecast model results were:the predicting accuracy of cocoons from Chinese,Haoyue×Jingsong,Jingsong×Haoyue and Suzhen×Chunguang race were 92.0%,90.9%,84.0%and 94.3%,respectively.The calibration model results under different instruments were compared.The results showed that the predicting accuracy of WQF-400N FT-NIR model was better than that of NIR256-2.2T2 model,of which the causes were the inconsistency of detection conditions and methods between the two NIR spectrometers,and so on.The spectrum data of cocoons from Haoyue×Jingsong,Jingsong×Haoyue and Suzhen×Chunguang races detected by NIR256-2.2T2 spectrometer were complicated to be a large number ofdataset,which contains 325 samples.PLS+BP calibration model was established, with the predicting accuracy of 86.7%.The result showed that it was feasible to identify the sexuality of varieties of hybrid cocoons by using IRS analysis technology.To sum up,it was concluded that the predicting accuracy of the discrimination of male and female cocoon was high by using the PLS + BP model,no matter which from large number of samples,small number of samples or varieties of hybrid samples detected by IR spectrometer.So it was feasible to identify the cocoon sexuality by using IRS analysis technology.The results provided a new way to realizing nondestructive testing about discrimination of cocoon sexuality by using IRS analysis technology. |
