A novel strategy of near-infrared spectroscopy dimensionality reduction for discrimination of grades, varieties and origins of green tea

Abstract Supervised orthogonal locality preserving projection (SOLPP), a supervised manifold learning dimensionality reduction method, was employed to reduce near-infrared (NIR) feature dimension of green tea, and the newly low-dimensional variables were applied to identify green tea with different grades, varieties and origins. First, the continuum removal was applied to highlight the subtle difference of spectral characteristics between two groups from grades, varieties and origins of different green tea. Then, the capability of SOLPP for mining special effective features was assessed in comparison with the unsupervised methods such as principal component analysis (PCA), multidimensional scaling (MDS), and locality preserving projections (LPP). Support vector machine (SVM) was used to build discrimination models based on variables represented by PCA, MDS, LPP and SOLPP. Results indicated the data processed by SOLPP displayed better separability between two classes on the premise of preserving prior level information and local manifold structure in each dataset. The combination of both SOLPP and SVM could identify grades, varieties and origins of green tea with improving performance. The accuracy and Kappa values achieved 100% and 1.00 in all datasets. Satisfactory results indicated that SOLPP as a valuable NIR spectral feature mining and dimensionality reduction method can evaluate quality of agricultural products.