基于KL散度的面向对象遥感变化检测

Research on object-oriented remote sensing change detection method based on KL divergence

遥感影像的变化检测从基于像素到面向对象,从阈值分割到相似性度量已有众多的研究成果;但在对面向对象遥感图像变化检测中,存在分割参数的选择、变化阈值的确定、对象变化程度的表达等问题。为此,提出一种基于相似度测度的面向对象遥感影像变化检测方法,并打破了以往仅以有/无变化的检测结果所呈现的表现形式。首先计算了图像对象分割的最优参数,得到了2个时相的图斑对象,并进行了空间叠加处理;然后利用KL相似度计算方法计算了图斑对象的相似度系数,利用直方图统计了该系数的自然聚类特征;再运用不同的自然聚类特征值,分级得到了图斑对象的变化程度;最后,分析了不同参数分割结果、不同分级方法对图像变化程度检测的影响,同时通过对比有/无变化的检测结果,验证了本研究所提方法的科学性和有效性。

The change detection of remote sensing image has many research results from face-to-face to object-oriented operation and from the threshold to the similarity measurement; nevertheless, there are many problems such as the selection of the segmentation parameters, the determination of the change of the object and the degree of the change of the object. In view of such a situation, this paper proposes a new method based on similarity measurement to detect the change. This method has broken the performance form which has been used to detect the change of the results. Firstly, the optimal parameters of image object segmentation are calculated, and then the image patches are obtained. After that, the similarity coefficients are calculated by KL similarity calculation method, and the natural clustering features of the coefficients are calculated. The results show that the changes of the national economic development, disaster prevention and land use management decision-making are obvious, which shows the scientific nature and effectiveness of this method.