Peach variety identification using near-infrared diffuse reflectance spectroscopy

Abstract

More than 1000 peach varieties with significant differences in qualities are cultivated in China. Distinguishing peach varieties is not only needed by peach sellers, but also demanded by consumers. To offer information on identifying peach varieties, near-infrared (NIR) diffuse reflectance spectra between 833 and 2500nm were collected for four peach varieties, 100 samples for each variety. Kennard–Stone algorithm method was used to divide all samples into calibration set (320 peaches) and prediction set (80 peaches). Eight principal components (PCs), 1067 and 10 characteristic wavelengths were extracted by principal component analysis (PCA), uninformative variable elimination based on partial least squares (UVE-PLS) and successive projections algorithm (SPA) from full spectra (FS) with 2074 initial wavelengths, respectively. Least squares support vector machine (LSSVM) and extreme learning machine (ELM) were used to establish peach varieties identification models using the FS, selected PCs and characteristic wavelengths as input variables. Experimental results showed that all models based on PCA reached 100% accuracy for identifying the four peach varieties. The accuracy of LSSVM models based on UVE-PLS also reached 100%. This study indicated that peach varieties could be distinguished successfully by using NIR spectroscopy.