采用独立阈值的遥感影像变化检测方法  

A Change Detection Method of Remote Sensing Images with Independent Threshold

作　　者：贾永红[1,2] 谢志伟[1] 张谦[3] 杨刚[4] 

机构地区：[1]武汉大学遥感信息工程学院,武汉430079 [2]武汉大学测绘遥感信息工程国家重点实验室,武汉430079 [3]华中科技大学自动化学院,武汉430074 [4]武汉大学资源与环境科学学院,武汉430079

出　　处：《西安交通大学学报》2015年第12期12-18,共7页Journal of Xi'an Jiaotong University

基　　金：国家科技支撑计划资助项目(2011BAB01B05)

摘　　要：针对在多时相遥感影像变化检测中常规阈值确定方法无法获取小比例变化量区域准确变化阈值,并导致变化检测失败的问题,提出了采用独立阈值的遥感影像变化检测方法。通过多时相遥感影像多尺度分割获取像斑,采用变化向量分析法计算像斑差异度;从像斑差异度中自适应选择满足期望最大化算法和贝叶斯最小误差率理论获取准确阈值条件的训练样本;将训练样本导入独立阈值法确定变化阈值,利用变化阈值对像斑差异度进行二值分割获得影像变化的检测结果。实验结果表明,采用独立阈值的遥感影像变化检测方法能够获得更准确的变化阈值,在城郊变化检测中平均漏检率较全局阈值法和局部阈值法降低了9.6%和17.24%,在城区变化检测中平均正确率较全局阈值法和局部阈值法提高了51.27%和35.42%。A change detection method of remote sensing images based on an independent threshold is proposed to solve the problems that the accurate change threshold could not be worked out by either the general global or the local threshold methods,and the failure of change detection happens in multi-temporal remote sensing images change detection if the prior probability of the class of changed pixels in the detection region is low.The multi-scale image segmentation is used to get image objects from the multi-temporal remote sensing images,and differences of image objects are calculated from each image object based on the change vector analysis.Then,training samples that meet the expectation maximization algorithm and Bayesian rule with minimum error rate are correctly selected from the difference of image objects using the adaptive sample selection method.The change threshold is finally obtained from the training samples by the independent threshold method,and the change detection result is derived.Experimental results show that the proposed method gains more accurate change threshold.Comparisons with the global thresholdmethod and the local threshold method show that the independent threshold method reduces the average miss rate by 9.6% and 17.24%,respectively,in the suburbs change detection,and improves the accuracy rate by 51.27%and 35.42%,respectively,in the urban change detection.

关 键 词：变化检测 小比例变化量区域 像斑 样本选择 期望最大化算法 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]