中文摘要：

不平衡数据分类是机器学习的研究热点之一.传统的机器学习分类算法通常假定用于训练的数据集是平衡的,不能直接应用于不平衡数据分类.利用朴素贝叶斯和决策树对数据不平衡的敏感度不同,提出一种基于投票的不平衡数据分类集成算法.基分类器选择NB和C4.5,通过投票平均方法进行分类决策;并选择公开的不平衡数据集进行实验验证.实验结果表明,该算法能有效提高不平衡数据的分类性能,特别是对正类（少数类）的误报率较低,具有良好的鲁棒性.

英文摘要：

In allusion to the problem of SMD LED visual positioning, an algorithm based on rectangle fitting is proposed. By exploiting the geometric feature that most SMD LED lamp beads are rectangular, firstly the global threshold segmentation is used to process the captured image according to its background grey scale. Then the con- tour coordinates of the target component are divided and equal interval sampled into four groups by vertexes. Finally the rectangle fitting algorithm bas6d on the orthogonality of adjacent sides of rectangle is used to position the target component. The simulation indicates that the algorithm has both high accuracy and strong robust, which has met the requirement of high-accuracy chip mounter.