顾及不确定性的多光谱遥感图像分类及其不确定性度量和表达

Uncertainty Measurement and Representation of Classification of Multi-Spectrum Remote Sensing Images Based on Uncertainty

针对传统遥感图像分类方法缺乏考虑空间自相关性、分类模糊性以及分类误差的空间表达的缺点,采用顾及空间数据的空间自相关性和分类模糊性的邻域EM算法进行多光谱遥感图像分类。构建了四个不确定性度量指标—模糊隶属度残差、相对模糊隶属度最大离差、模糊隶属度熵和相对模糊隶属度熵对其分类的不确定性进行度量和可视化表达,克服了采用误差矩阵和Kappa系数进行传统遥感图像分类精度评价缺乏空间信息分布的不足。

In order to overcome the deficiencies of traditional remote sensing image classification methods which spatial autocorrelation, classification fuzziness and spatial representation of classification error have not been considered, neighborhood EM algorithm considering spatial autocorrelation and classification fuzziness was adopted to classify the multi-spectrum remote sensing images. Four uncertainty measurement indexes--fuzzy membership residual, relative maximum fuzzy, membership deviation, fuzzy nenbership entropy and relative fuzzy membership entropy were founded to assessment the uncertainty of Neighborhood EM algorithm classification, which overcome the deficiencies of traditional uncertainty assessment methods of remote sensing images classification by error-matrix and Kappa coefficient