多源遥感影像土地覆盖分类结果一致性评价与集成应用

Consistency Evaluation and Integration Application of Land Cover Classification Results from Multi-source Remotely Sensed Images

针对多源遥感影像土地覆盖分类结果一致性与分类精度改进的要求,对两组中等空间分辨率的光学影像进行土地覆盖分类,以支持向量机分类结果为基础,采用Kappa统计量、双错误测量、Q统计量、相同错误率从不同角度评价了不同分类结果的一致性。实验表明,多源遥感数据分类结果总体上常规一致性程度较好,二值先验一致性程度尚可,错误一致性程度较小;不同土地覆盖类别的一致性程度并不相同,有的类别甚至出现不一致现象。提出组合法和替换法两种策略以综合数据优点、实现多传感器数据集成应用,能够有效提高分类精度。

In order to evaluate the consistency of land cover classification results derived from multi-source remotely sensed images,two groups of medium spatial reesolution optical data are respectively classified by Maximum Likelihood Classification （MLC）, Support Vector Machines （SVM）and Decision Tree （DT）. The results of SVM classifier are used for consistency eval- uation owing to its higher accuracy than other classifiers. Kappa statistic,Double-fault, Q statistic, Same-fault are used to evaluate classification consistency of multiple source data,and the experiments show the general consistency of multi-source data classification results is good, the binary prior consistency is fine and consistency of errors is poor. The degrees of consistency are different according to different land cover classes,and some land cover classes may show inconsistency. Two strategies named as combination and replacement are experimented, which integrate the merits of different data and enhance the fusion of multisource data,so the classification accuracy is improved effectively.