摘要
在分析遥感红外图像特点的基础上,提取了灰度共生矩阵的能量、惯性、熵等14个特征量用于红外图像纹理分析。以最小判别熵可分性判据作为准则,利用遗传算法搜索最优特征子集,实现了遥感红外图像的特征选择。为了验证此算法特征选择的有效性,设计了RBF网络分类器,对遥感红外图像进行分类识别,其结果证明,基于最小熵和遗传算法所得到的特征子集可以简化网络结构,减少训练时间,提高样本的识别概率。
Based on the characteristics of infrared remote sensing image, 14 features are extracted to analyze their texture attributes, such as energy, inertia and entropy of co-occurrence matrix-based features. For features selection from infrared remote sensing image, the genetic algorithm （GA） is used to search for the optimum feature subset based on minimum entropy separability criterion. For validation of features selection, radial basis function （RBF） networks are designed to classify the infrared remote sensing image. The result of classification proves that the feature subset selected could simplify the structure of classifier, reduce training time and enhance the recognition probability.
出处
《遥感信息》
CSCD
2005年第5期3-5,11,共4页
Remote Sensing Information
基金
国防预研项目(41101010506
41322020102)
关键词
特征选择
遗传算法
熵
识别
features selection
genetic algorithm （GA）
entropy
recognition
作者简介
陈修桥（1976～）,男,江苏新沂人,博士生,研究方向图像处理与目标识别.
