基于形态运算的二值网格域描述单类分类方法  

Binary Grid Domain Description Based on Morphology for One-class Classification

作　　者：高峰[1] 曲建岭[1] 郭超然[1] 孙文柱[1] 

机构地区：[1]海军航空工程学院青岛校区,山东青岛266041

出　　处：《计算机与现代化》2015年第3期57-61,共5页Computer and Modernization

摘　　要：针对样本数不平衡的分类问题,提出一种基于形态运算的二值网格单类分类方法。该方法首先将样本分布空间划分成等尺寸网格,而后根据训练样本分布将网格分为目标网格和背景网格。包含样本的网格称为目标网格,不包含样本的网格称为背景网格。最后对目标网格进行形态学闭运算和开运算形成训练样本的域描述。在人工数据集和真实数据集上将该分类方法与其他典型分类方法进行了对比实验。结果表明,该方法分类精度较高、训练速度较快,是一种有效的单类分类方法。A one-class classifier of Binary Grid Domain Description（ BGDD） based on morphology is proposed for solving unbalanced samples classification problems. In this method,the sample space is first divided into grids. Then,an approximate domain description can be obtained by putting samples into these grids. These grids are divided into object grids and background grids,the grid which contains at least one sample is defined as the object grid,while the grid without any sample is defined as the background grid. Next,morphological closing and opening operations are applied to object grids to obtain the domain description of the training samples. Experiments based on both artificial and real-world datasets were done and comparative experimental results were present. Experimental results show that the BGDD classifier is an effective classification method for high classification accuracy and fast training speed.

关 键 词：形态运算 网格 单类分类 域描述 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]