Non-destructive Detection the Content of Acid Detergent Fiber in Corn Stalk Using NIRS

Abstract

Near-infrared spectroscopy (NIRS) with its nondestructive and high-efficiency advantages can be qualified for the quantitative detection. This study demonstrated that NIRS combined with random forest (RF) algorithm was applied as a rapid analytical method to predict the content of acid detergent fiber (ADF) in corn stalk. In order to select representative samples for modeling, Kennard-Stone (KS) method was proposed as a tool to partition samples. Then after optimized by various pretreatment methods, the performance of RF model was enhanced. Subsequently, the combination of correlation coefficient method (CCM) and linear discriminant analysis (LDA) performed on the spectra was used to reduce data redundancy and improve the accuracy of model. It turned out that the performance of RF calibration model was best when the data's dimension reduced from 1050 to 8. The determination coefficients (R2), root mean square error (RMSE), residual predictive deviation (RPD) of test set were 0.9923, 0.3759 and 11.3356, respectively. Finally, the overall results indicated that the proposed method provided a nondestructive and effective technical to predict ADF content in corn stalk.