Discriminant Analysis of Bamboo Leaf Types with NIR Coupled with Characteristic Wavelengths
Chu Bing-quan,Zhao Yan-ru,He Yong
DOI: https://doi.org/10.3964/j.issn.1000-0593(2017)06-1718-05
2017-01-01
Spectroscopy and spectral analysis
Abstract:Bamboo leaves are rich in many kinds of biological active components such as flavonoid, phenolic acid and polysaccharide, which have demonstrated good effects of anti-oxidant, blood lipid regulation, anti-cancer, cardiovascular and cerebrovascular protection etc. However, the content of active constituents exhibits great differences among different bamboo species. The traditional identification of bamboo species is mainly based on the observation of bamboo leaf size, bamboo texture and branching height etc. which has the disadvantages of low efficiency and high error rate. Therefore, to distinguish the varieties of bamboo with a rapid and accurate method is an important task in the development and utilization of bamboo leaves. A near-infrared (900 similar to 1 700 nm) hyperspectral technique was used to identify 12 bamboo species from different regions of China. Principal component analysis (PCA) was applied to make the cluster analysis. PCA X-loading (XL) and Random Frog (RF) algorithm was chosen to extract spectral feature and 6 characteristic wavelengths (931, 945, 1 217, 1 318, 1 473 and 1 653 nm) and 12 characteristic wavelengths (1 052, 1 140, 1 163, 1 177, 1 180, 1 193, 1 230, 1 241, 1 477, 1 483, 1 629 and 1 649 nm) were selected respectively. Then, the spectra based on the selected wavelengths were set as the input values of Least Squares-Support Vector Machine (LS-SVM) model to perform the discriminant analysis. At last, the properties of the three LS-SVM models were evaluated with Receiver Operating Characteristic curve (ROC curve). Results showed that (1) in the three LS-SVM models, the recognition rate of full band, XL algorithm and RF algorithm were 99. 17%, 95. 83% and 95. 83% respectively, (2) the area under the curve (AUC) in ROC curve were all reach over 0. 98. In conclusion, the bamboo leaves from different regions could be identified by near-infrared hyperspectral technique combined with chemometrics methods, which provided a theoretical foundation for efficient utilization of bamboo leaves.