摘要
目前对于卷烟牌号的鉴别多应用一些传统分类算法,这些传统算法用于归纳一个通用规则的训练样本数据较少,造成分类模型的准确度较低,且预测结果没有置信度衡量,在高风险领域的应用不足。针对传统分类算法的局限性,提出了基于转导推理的一致性预测算法。通过探索待测数据和样本序列之间的内在联系,运用Kolmogorov算法的随机性理论建立一种置信度机制,并应用随机性检测函数对置信度进行估算,这样可以很好地对烟叶和成品卷烟进行定性判别和分类。
At present,for the identification of cigarette brands,some traditional classification algorithms are used.These traditional algorithms are used to summarize the training samples with a general rule with less data,which results in a low accuracy of the classification model,and less application in high-risk fields because there is no confidence measure of the prediction.Aiming at the limitation of the traditional classification algorithm,a consistency prediction algorithm based on transduction inference is proposed.By exploring the intrinsic link between the data under test and sample sequence,Kolmogorov algorithm randomness theory was adopted to establish a confidence mechanism and random testing function was used to estimate the degree of confidence.In the light of this,it can perform qualitative identification and classification of tobacco and cigarette products effectively.
出处
《现代电子技术》
2012年第10期97-99,共3页
Modern Electronics Technique
关键词
近红外
一致性预测器
卷烟
牌号鉴别
near infrared
consistent predictor
cigarette
brand discrimination
