论文部分内容阅读
使用ASD Field3在武夷山实测的9种鲜茶叶数据,该数据经过预处理后,计算24种光谱指数,用于对9种茶叶的分类,用SVM-RFE对光谱反射率数据和光谱指数数据进行特征选择,最后用线性SVM和随机森林分类.检验SVM分类器的性能和SVM-RFE选择特征的适用性,其中SVM分类器在4个数据集中都达到了95%以上的分类精度.随机森林分类器在其中3个数据集达到90%以上的精度,一个70%的精度.研究表明SVM-RFE是一个稳定有效的特征选择算法,并且SVM的性能优于随机森林.
Using the data of 9 fresh tea leaves measured by ASD Field3 in Wuyi Mountain, 24 kinds of spectral indexes were calculated after pretreatment, which were used to classify 9 kinds of tea. The spectral reflectance data and spectral index data were characterized by SVM-RFE Finally, linear SVM and random forest classification are used to test the performance of SVM classifier and the applicability of SVM-RFE selection features, SVM classifier achieves more than 95% classification accuracy in all four datasets.SRF Among the three datasets, the accuracy of over 90% and the accuracy of 70% are achieved.Research shows that SVM-RFE is a stable and efficient feature selection algorithm, and the performance of SVM is better than that of random forest.