摘要
归一化是对样本数据进行分类处理的一个重要过程,论文提出了基于样本概率分布均衡技术的归一化算法,通过合并低概率区间和分离高概率区间的非线性方法,改变样本分布的不均衡性,使样本接近均匀分布。并将算法应用到显微镜细胞决策树与BP神经网络识别系统中,通过与最小—最大值归一化和零—均值归一化对比,可以得出,识别的速度和效果到得到了很大的提高。
Normalization is an important process in samples classifying. The paper advanced a normalization algorithm based on the technology of probability distribution equalization, and changed the im-balance of samples＇ distribution by uniting low-probability areas and splitting high-probability areas, in order to make the samples＇ distribution as even distribution. At last, the algorithm was used in the system of decision tree and BP neural network in microscope ceils recognition, and was compared with those using min-max normalization and zero-even normalization. It was proved that the algorithm improved the recognition accuracy and speed.
出处
《信号处理》
CSCD
北大核心
2009年第4期636-638,共3页
Journal of Signal Processing
关键词
归一化
概率分布
均衡
决策树
BP神经网络
normalization
probability density function
equalization
decision tree
BP neural network
