基于最大联合概率的土地覆盖变化自动检测方法

Automatic Land-cover Change Detection Approach Based on Maximum Joint Probability

针对多时相遥感影像非监督变化检测采用不同方法获取的结果之间存在较大差异这一问题,提出了最大联合概率(MJP)决策融合方法。通过分析各种检测方法的检测概率与最终检测结果正确概率的关系,根据贝叶斯决策原理提出了MJP决策融合准则,并分析各种检测方法所获取结果的从属关系,将MJP决策融合应用于多个不同的检测结果,对多个不同结果进行了有效融合。选取武汉地区2个时相的TM影像为实验数据,对其土地利用变化进行研究,采用相关系数法获取差异影像,然后选取系列变化检测算法获取多种不同检测结果,最后采用本文所提出的融合方法对各种不同结果进行决策融合,获取最终变化检测结果。实验结果表明:基于MJP决策融合的变化检测算法与单一变化检测方法算法相比,具有更好的鲁棒性。这为遥感变化检测的决策融合提供了依据,是一种有效的利用遥感数据进行土地覆盖变化检测的方法。

The most common challenge of automatically unsupervised change detection in multitemporal, remotely sensing image is to find the best global threshold in difference images. It is obvious that different results will be obtained according to different threshold selection algorithms. In this paper, a decision fusion method based on maximum joint probability（MJP） is proposed in order to deal with this problem. By means of analyzing the relationship between the detection probability of each detection algorithm and that of final correct, several different results can be fused efficiently by MJP decision fusion algorithm. By taking multitemporal TM images at Wuhan, China, as experimental data, the different images are obtained by correlation coefficient. An array of change detection results are developed by a serial threshold selection algorithms. Final change detection results are gotten by fused results array. Experimental results show that, compared with each threshold selection algorithm, MJP fusion algorithm is more robust and satisfactory. The result confirms the validity of the proposed approach and the opproach is useful in land cover change detection.