面向对象分类特征优化选取方法及其应用

与传统基于像元的分类方法比较，面向对象的分类方法可利用的地物信息更加丰富，然而如何从众多信息中筛选出能够有效提取不同地物的分类特征，从而提高分类效率与精度，是使用面向对象方法分类时急需解决的问题。SEaTH算法（分离阈值法）是一种有效的自动选取分类特征并计算阈值的方法，但其只考虑了类间距离，容易存在信息的冗余，从而对分类精度造成一定影响。本文在SEaTH算法的基础上，综合考虑了特征间的相关性、类间距离以及类内距离，对SEaTH算法进行了优化，并将改进前后的两种方法运用到广东省肇庆市TM影像及环境一号卫星影像土地覆盖分类中进行对比分析。实验结果表明，改进后的方法筛选出的特征在提取地物上更为有效，尤其使耕地的分类精度提高了12.26%，使分类总体精度由80%提高到了85.26%。改进后的方法对不易获取多时相影像的地区的土地覆盖分类具有重要意义。

Feature selection and its application in object-oriented classification

More information extracted from the remote sensing images can be used by object-oriented classification than the traditional pixel-based classification. However, how to select effective information without redundancy quickly to improve the efficiency and accuracy of object-oriented classification is a key problem currently. A method called SEaTH (SEparability and THresholds) can select features and compute thresholds automatically but it just depends on Jeffries-Matudita distance to separate classes, which may cause information- redundancy and then impact on classification result. In this paper, a new method based on SEaTH is proposed. Then the method which takes account of the Jeffries-Matudita distance, the inter-class distance and the correlation of different features is used in the classification of multi-resolution remote sensing images of Zhaoqing. The result shows that the method presented in this paper can select more effective information than original SEaTH algorithm especially on extracting farmland which accuracy is improved by 12.26% and the total accuracy is also improved form 80% to 85.26%. It is important to the classification when multi-temporal data are difficult to be obtained.