There are three means to decrease the number of attributes : delete irrelevant attributes to the task using fisher ' s discriminant ratio ; merge correlate attributes according to pearson ' s correlation coefficient ; reduce the dimensionality of the attribute vector by singular value decomposition 在属性选取时根据fisher函数把对分类影响小的属性删除，由pearson ' scorrelationcoefficient将相关联的属性合并，用singularvaluedecomposition减少属性向量空间的维度。
Firstly , based on the traditional ids , the method of improving ids is put forward by introducing the concept of user - interest , which adds user - interest to information entropy in the ids algorithm . therefore it makes the algorithm depend not only on train sets but also on field knowledge of studied attribute . in this way the improved algorithm increases the classification precision 首先，本文提出了在传统id3算法的基础上，引入用户兴趣度的概念来改进id3算法，在id3算法信息熵计算公式中，对信息量加权和增加用户兴趣度，使得算法不仅仅是依赖实例进行分类，而是在实例的基础上对具体属性加入先验知识调节其对分类的信息量，提高分类的准确性，并用实例验证了改进算法的有效性。
Thirdly , in the multi - spectral , there are some facts that different objects have the same spectrum and the same objects have different spectrum but they are in the region having different landform so that we can deduce the effect of the fact by segmenting the image according to the texture feature that reflect the landform into blocks in advance and then classifying each pixel according to its corresponding block spectrum information 第三,实际多光谱图像中存在着不少同物异谱和同谱异物现象,而它们往往处于不同的地形地貌之中,因此事先依据反映宏观地形地貌的纹理特征将图像分块,再有针对性地进行地物光谱分类是能够降低它对分类精度的影响。
First classifier chooses two classes whose matching distance between it and paper currency is bigger than others from all class . then in the second stage , we extract some new feature and improve the classifier to generate the last result . in the stage of defect detection for paper currency , we advances a homogeneity based algorithm for the detection of scratch and cracks appearing on paper currency , in which the homogeneity feature of the sensed paper currency image is first constructed to locate the pixels that probably been polluted , the image registration algorithm is subsequently used to overlay the sensed and reference paper currency image 在特征提取中,我们对基于方向块的特征提取方法进行了分析,在此基础上针对美元特点,对图像方向块的划分方式做了研究,并提出了基于几何距离的特征提取方法;在分类器设计中,我们采用了lvq网络对纸币进行学习与分类,并提出了一种具有两层结构的分类算法,第一层首先对输入的特征向量进行粗分类,选定与特征向量匹配距离最大的两类币种,进入第二层分类器;在第二层分类器中,我们通过研究进入该模块两类币种特征块的相关性,重新设计了特征向量,同时对分类器进行改进,最终实现对纸币的分类。
This type of strategy has two main shortcomings : 1 ) useful information for classification task contained in the matrix structure may be jeopardized in the vectorizing procedure ; 2 ) after vectorizing procedure computation complexity in classification task may increase substantially due to the vector pattern representation 这种方法存在着两个主要的缺点： 1 )矩阵模式中对分类有用的结构信息很可能会因为向量化的操作而遭到破坏； 2 )向量化的操作极大的增加了特征提取及随后识别的运算复杂度。
Secondly , the thesis puts forward that conditional probability of attribute to positive example can be used to compare the information which attribute provides so as to construct decision tree and to get classify rules . and a demonstration shows that the algorithm simplifies the decision tree ' s building process efficiently 其次，提出了利用属性对正例的影响度来比较属性对分类提供的信息量，进而选择分类属性构造决策树的条件概率决策树算法，同时实例计算说明该算法有效地简化了决策树的生成过程。
