Continuous attribute discretization and application in Chinese wine classification using BP neural network

In this paper we devote to study some continuous attribute discretization algorithms, and based on them we build a Chinese wines classification system using BP neural network. According to the discretization method based on cluster analysis, we first discretize attributes by using fuzzy c-means cluster analysis guiding by the level of consistency of decision table, and then merge neighboring intervals using rough reduction as a suitable post-processing. Micrographs of Chinese wines show floccule, stick and granule of variant shape and size. Different wines have variant micro-structure and micrographs, we study the classification of Chinese wines based on the micrographs. In this work, ten Chinese wines are determined or classified in the system, such as Wuliangye, Luzhoulaojiao, Xushui, Jiannanchun, Maotai and et.al. For each wine, we collect micrographs of deferent resolution(300 nm × 300 nm, 1.5 ¿m × 1.5 ¿m and 5 ¿m × 5 ¿m). These images are sub-divided into 16 sub-image, then we compute total of 26 features for each sub-image. As the features are continuous attribute, we discretize them using method proposed in this paper. The discrete features are used as inputs of BP neural network, and the labels corresponding to the ten Chinese wines will be the target. The classification results for Chinese wines show the efficiency and advantage of the method proposed in the paper.