| Popis: |
In order to analyze the quality similarity of tobacco leaves in different tobacco planting areas, 21 quality indices of flue-cured tobacco were collected from different production areas, extracted principal components by factor analysis, classified by 3 clustering analysis methods, and the classification results were compared and statistically tested. The result indicated that: (1) The quality of tobacco leaves was different, and high degree of information overlap was detected among different indices of tobacco appearance, chemical and sensory quality; (2) The cumulative variance contribution rate of 5 extracted principal component factors was 85.445%, and the eigenvalues were 7.761, 4.758, 2.472, 1.674 and 1.278, respectively; (3) The results of 3 cluster analysis methods were not the same. The results of the weighted principal component distance cluster and the weighted principal component cluster were similar, which were different from the general principal component cluster; (4) The results of statistical test showed that the weighted principal component distance cluster method had the largest F-test value (5.900), the smallest sum of squares within the group (8.164), and the largest sum of squares between groups (19.267). And the weighting results for different principal component factors were more reasonable which also had more objective classification results. The cluster results of weighted principal component distance and weighted principal component were better than that of general principal component. And the cluster results of weighted principal component distance were more interpretable which had better the statistical test results. |
