基于多维专用特征空间的视觉模糊识别方法

A FUZZY VISUAL RECOGNITION METHOD BASED ON SPECIAL-USED MULTI-DIMENSION FEATURE SPACE

本大提出的视觉模糊模式识别算法建立在用特征源信息量分析和概率统计的方法为每个待识的模式构造的多维专用特征空间上,这种分而治之的特征筛选方法不仅减小了忽视有效特征的可能性,而且使特征分布的内在特性得以更精巧的利用.本算法在彼此交叉的专用特征空间中度量样品对各模式的隶属度,具有在通用特征空间无法达到的优点;同时,不同于一般依赖于语言变量和模糊规则的视觉模糊识别算法,其对先验知识的依赖性也大大地降低了.我们应用此算法生成了一个通用的工件自动识别系统,实验结果较为理想.

In this paper, we present a fuzzy visual recognition algorithm based on Special-used Multi-dimension Feature Space (SMFS) of each pattern instead of classical Common-used Multi-dimension Feature Space (CMFS), using information analysis of feature distribution and probability statistics. This is also a 'divide and conquers ' approach of feature selection and dimension reduction, and it not only shows less possibility of neglecting useful feature, but also make more use of the natural property of the feature distribution. Apart from general fuzzy visual recognition algorithm relying on language variable and fuzzy rule, this algorithm greatly reduces the dependency of priori information, even more, weighting the memberships of each pattern in several interweave SMFS shows advantages that can not be achieved in CMFS. We apply this method to the problem of Part Recognition on automated assembly lines, and develop a general purpose Part Recognition System. The result of the experiment is quite promising.