联合变化强度和相关系数最小错分概率变化检测阈值优化

针对最小错分概率的遥感影像变化检测方法依赖样本的问题,提出一种联合变化强度和相关系数最小错分概率的变化检测阈值优化方法。首先对两时相影像像元的变化强度和相关系数两特征的直方图和曲率进行数值分析,确定分割阈值的上下限范围;然后对阈值范围内的阈值进行迭代,得到总错分概率,选取最小错分概率对应的阈值为最佳阈值;最后对两幅影像进行叠加分析,实现变化区域的提取。实验结果表明:该方法是一种无需进行参数估计、非监督、可信赖的变化检测方法,且能达到与传统方法同等的变化检测精度。

Change Detection Threshold Optimization of Combining Change Intensity with Correlation Coefficient Based on the Minimum Error Probability

Since there is a problem in excessive dependence on prior knowledge of traditional change detection ap-proach of remote sensing image based on the minimum error probability this paper adopts a change detection threshold optimization method of combining change intensity with correlation coefficient based on the minimum er-ror probability. First of all the initial range of image segmentation threshold by histogram of change intensity image and correlation coefficient was determined on the premise of numerical analysis of the two characteristics. Then within the scope of the threshold the minimum error probability was calculated. In other words the minimum error probability was obtained by the iterative threshold and the corresponding threshold of it was the optimal threshold value. Finally the two images were executed overlay analysis to gain change results. The experiment results show that change detection precision of the method is the same as the traditional method and it s a reliable and unsuper-vised change detection with no need for parameter estimation.