| A large number of crop straw are planted and output every year. While its efficiency is quite low, most of them are still being burned or discarded. It also causes huge waste of resources and produces garbage which contaminates the environment. Therefore, reasonable and effective utilization of straw resources is extremely important. Accurate detection of crop straw components content is an important prerequisite to improve the utilization rate of straw. At present, people still give priority to traditional chemical analysis method in the detection of lignin and cellulose. The method is obsolete and its detection speed is slow. Also, the labor cost is high as well as the damage to the samples is irreversible. Compared with conventional methods, near infrared spectroscopy is a method with fast detection and nondestructive characteristics. However, NIRS is influenced by a variety of factors. So it is right to choose different pretreatment methods to analyze in different areas and varieties. In order to establish a stable, re liable and accurate model, it is effective to eliminate the interference of background noise and specific physical factors. So we can extract the information related to chemical composition from the spectrogram.Therefore, soybean straw is used as the research object in the study. Combined with chemometrics, the NIRS method is used to model. The model completes the rapid detection of soybean straw main ingredient,namely cellulose and lignin content. At the same time, different pretreatment methods of spectral data are analyzed and demonstrated while modeling. The main research work and conclusions are as follows.First of all, samples of soybean straw with lignin and cellulose were put in normality analysis and multiple correlation analysis. On this account, selected soybean straw samples are reasonable and representative. They could be used for subsequent modeling and analysis.Secondly, different pretreatment methods were used in the straw samples, including eliminating abnormal samples, dividing the sample set and denoising processing of the original spectrum. K-S algorithm is adopted to divide calibration and validation sets of lignin and cellulose. The study takes advantage of Mahalanobis distance, 3-D and Hotelling T2 to distinguish and eliminate abnormal samples. Then, the precision of model of lignin and cellulose improve respectively.Moreover, the study focuses on different denoising method and explore the influence of different denoising method on model accuracy. Derivative, normalization, smoothi ng were used in the study. Also, baseline correction, MSC, SNV and OSC algorithm were carried on discussion. Results indicate that accuracy and predictive ability of the model is improved after using denoising method. When using baseline correction combined with multiple scattering correction, soybean straw lignin model achieves the best result. the validation set d ecision coefficient is 0.7620502. When using the second derivative method combined with smooth processing, denoising effect of cellulose is best, the validation set decision coefficient is 0.7703286. It is profound to study the influence of different denoising method on model accuracy in deep, for it supports theory of strengthening model precision and robustness. Also, it lays a solid foundation for the extension model of transitivity.Finally, using Partial Least Square and Support Vector Machine, lignin and cellulose model are established. And it is turns out that it is workable to predict the content of lignin and cellulose of soybean straw. Also, prediction ability of lignin and cellulose model is higher by using PLS regression.In conclusion, the study provides support for the modeling of other components of legumes. At the same time, the study gives a certain reference value about improving detecting precision of the model in the field of biomass in the future. |
PDF Full Text Request