CAJ | 학술논문

当前的茶叶分级研究主要基于纹理特征构造分类器,但易受采样过程中的光照、噪声影响.本文提出了结合经典的SIFT(Scale-invariant feature transform)特征描述子在自然光条件下的茶叶分级问题,并使用多类AdaBoost算法对样本进行分类.单幅图像的提取结果显示,SIFT特征对带瑕疵的图片仍具有很好的描述能力.在采集的90幅3级茶叶样本上的实验结果显示,纹理特征+SIFT特征取得了比单组特征更好的分类性能.
당전적다협분급연구주요기우문리특정구조분류기,단역수채양과정중적광조、조성영향.본문제출료결합경전적SIFT(Scale-invariant feature transform)특정묘술자재자연광조건하적다협분급문제,병사용다류AdaBoost산법대양본진행분류.단폭도상적제취결과현시,SIFT특정대대하자적도편잉구유흔호적묘술능력.재채집적90폭3급다협양본상적실험결과현시,문리특정+SIFT특정취득료비단조특정경호적분류성능.
Much research work focuses on the texture feature of the image in the tea grading, but it is prone to the illumination and the noise during the sampling. This paper proposes the combination of the texture and the SIFT features for the recognition of the tea grades under the condition of the natural light, where multi-class Ad?aBoost is used for the classification. The extracted images show that it is robust to the image noise. Moreover, the cross-fold results on 3 categories tea image dataset including 90 examples demonstrate that the texture +SIFT achieves a better performance than one group of the features alone.