结合互信息和遗传算法特征选择的多层次面向对象影像分类

项目名称： 结合互信息和遗传算法特征选择的多层次面向对象影像分类

项目编号： No.41201427

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 地理学

项目作者： 周小成

作者单位： 福州大学

项目金额： 25万元

中文摘要： 面向对象分类方法是当前高分辨率遥感影像分类时最有效的方法。针对面向对象分类中空间关系特征未充分利用和对象特征选择问题研究不足的问题，本项目拟开展结合互信息和遗传算法特征选择的多层次面向对象分类方法研究。以高分辨率卫星影像城市目标的有效识别为目的，本研究采用多尺度面向对象分析策略，充分考虑分割对象空间关系特征的表达和提取；针对对象光谱、形状、纹理和关系特征形成的"特征灾难"问题，提出基于互信息（MI）和遗传算法（GA）结合进行对象特征子集有效选择，采用支持向量基（SVM）进行特征子集分类的方法；定量评价和解释针对城市目标有效分类识别的特征子集，特别是空间关系特征。项目研究一方面可以克服面向对象影像分类中特征选择的盲目性和低效率问题，另一方面可以挖掘对象关系特征在城市目标识别中的作用，对于提高面向对象影像分类的效率和精度，促进高分辨率卫星影像信息提取的自动化程度具有重要意义。

中文关键词： 特征选择；互信息；遗传算法；高分辨率遥感；面向对象影像分类

英文摘要： The high spatial resolution satellite remote sensing data is most applied to extract information in urbans. Targets in Urban such as buildings and road et al, covers about 80 percent area in urbans. It has been an urgent problem for extract these targets accurately and high efficiently from high spatial resolution satellite remote sensing images.Object-oriented classification is a new method for extracting targets from high resolution remote sensing images. Remote sensing images can be classification based on muti-scale feature objects by object-oriented classification compared to the traditional classification method which is pixel-based. The Object oriented classification has been the most promise method in extracting the information from the high spatial resolution remote senisng image.Utilization of Spatial relation feature and feature selection are always ignored during the object oriented classifciation,which affects the accuracy and efficiency of classification. A hierarchical object oriented classification method based on mutual information feature selection will be researched in the study. In order to extract the urban target effectively.Spatial relate feature will be also extacted beyond spectral feature,shape feature and texture feature.And mutual information and genetic algorithm(GA)will be used for

英文关键词： feature selection；Mutual information；Genetic algorithm；High resolution remote sensing image；Object oriented classification