CLUSTER ANALYSIS OF CITRUS GENOTYPES USING NEAR-INFRARED SPECTROSCOPY

摘要

There are many genotypes and varieties in the citrus family. Currently, citrus classification systems have significant divergences in varieties of species, and subgenus classification as well. In this study, near-infrared spectroscopy technique was used to acquire spectral information on the surface of citrus fruits. Cluster analysis was consequently conducted to identify citrus genotypes. Results indicated that the combination of 9-point moving average smoothing and multiplicative scattering correction was optimal for preprocessing spectral data. In the spectral range of 1,180-1,220 nm, the cumulative reliability of the first two principal components were greater than 99.4%, and sweet oranges were clustered into an independent class. In 1,280-1,320 nm, systematic clustering performed better than principal component clustering, and all other sour oranges, except Goutoucheng, were clustered into a single clade. With dimensions reduction, the cumulative reliability of first five principle components in full band of 1,000-2,350 nm reached up to 99.1%. Using principal component cluster analysis, pomelo and loose-skin mandarin were clustered together; sweet and sour oranges were clearly separated. Pomelo being clustered with loose-skin mandarin, implies that they may have a hybrid origin; Jiaogan Mandarin, Daoxian yeju Mandarin, Goutoucheng sour oranges, and Zhuhongju sour tangerine were clustered with sweet orange, which implies old varieties may contain similar characteristic matters as sweet orange; Given that Jinlong lemon and Ranpour lime were clustered with sour orange, they were proved to originated from sour orange. The study indicates the great potential of spectral analysis for citrus genotype identification and classification.