Continuous Attribute Discretization and Application in Chinese Wine Classification Using BP Neural Network
In this paper we devote to study some continuous at-tribute discretization algorithms,and based on them webuild a Chinese wines classification system using BP neuralnetwork.According to the discretization method based oncluster analysis,we first discretize attributes by using fuzzyc-means cluster analysis guiding by the level of consistencyof decision table,and then merge neighboring intervals us-ing rough reduction as a suitable post-processing.Micro-graphs of Chinese wines show floccule,stick and granuleof variant shape and size.Different wines have variant mi-crostructure and micrographs,we study the classification ofChinese wines based on the micrographs.In this work,tenChinese wines are determined or classified in the system,such as Wuliangye,Luzhoulaofiao,Xushui,Jiannanchun,Maotai and et.al.For each wine,we collect micrographsof deferent resolution(3OOnm × 300nm,1.5μm × 1.5μmand 5μm × 5μm).These images are sub-divided into 16sub-image,then we compute total of 26 features for eachsub-image.As the features are continuous attribute,we dis-cretize them using method proposed in this paper.The dis-crete features are used as inputs of BP neural network,andthe labels corresponding to the ten Chinese wines will bethe target.The classification results for Chinese wines showthe efficiency and advantage of the method proposed in thepaper
Xingbo Sun Xiuhua Tang Yueyun Lei
Sichuan University of Science & Engineering Dept.of Electronic Engineering Zigong,Sichuan 643000,P.R Sichuan University of Science & Engineering Dept.of Material & Chemical Engineering Zigong,Sichuan 6
国际会议
厦门
英文
896-900
2008-11-17（万方平台首次上网日期，不代表论文的发表时间）