Feature Extractions and Recognition Methods for Concept Recognition and its Applications.

1)特征分类：判断所需的知识不是像真实的数字那样的数字分数，而是粗略的类别。在粗糙范畴的基础上引入了粒度的概念。我们将离散数据解释为分组问题。2）发现原型：判断所需的特征不是所有的知识和实验，而是一些抽象的数据。从这个角度出发，提出了一种在海量数据中发现典型原型的算法。原型不是一个点，而是一个有体积的物体。提出了一种原型更新算法，以适应随着时间的推移不断增加的数据。另一方面，我们定义了树结构的相似性。提出了一种利用相似度从多棵树中提取典型树的算法。3）基于个体情况的特征选择：提出了一种与分类器无关的特征选择方法。另一方面，从最优特征集依赖于类别和/或类别集的角度出发，提出了一种从随机选择的特征中提取最优特征的特征选择方法。

1) Classification of Features: The knowledge that is necessary for judgments are not numerical score like a real number but rough categories. A new concept "granularity" based on the rough categories was introduced. We interpreted discrete data as a grouping problem. Therefore an algorithm for discovering semi-optical answer for a grouping problem was proposed.2) Discovering Prototype: The features that is necessary for judgments are not all knowledge and experiments but some abstract data. From this point of view, an algorithm that discover typical prototype in enormous data was proposed. The prototype is not a point but a object has volume. A prototype update algorithm was also proposed to adapt increasing data with progress of time. On the other hand, we defined the similarity of tree structures. We proposed an algorithm that extracts some typical trees in many trees using the similarity.3) Feature Selection depend on Individual situations: A classifier-independent feature selection method was proposed. On the other hand, a feature selection method that extract optimum features in random selected features was also proposed from a point of view that optimum feature sets are depend on category and/or category set.