一种新的融合空间信息的半监督变化监测方法

A New Semi-supervised Change Detection Method of Fusing Spatial Information

基于改进的半监督FCM算法和马尔科夫随机场,提出了一种新的融合空间信息的半监督变化监测方法。首先将两幅遥感图像相减得到差值图像,并通过第4波段的差值给出了一种新的样本标记方法;然后,通过标记样本对差值图像利用半监督FCM算法进行聚类;最后,为了提高监测精度和去除聚类噪音点,利用像元点之间的空间邻接关系和马尔科夫随机场,通过更新后的隶属度矩阵得到了监测结果。为了验证本文方法的有效性,选取了两组TM遥感图像,监测了森林的变化。试验结果表明,改进的半监督FCM算法可以减少监测的漏检率,马尔科夫随机场方法可以很好地去除聚类过程中形成的噪声点,减少监测的虚检率。

Based on the improved semi-supervised fuzzy c-means （SSFCM） algorithm and Markov random field, this paper presents a new semi-supeivised change detection method considering spatial information. The difference image is firstly obtained by subtracting two original remote sensing images, and a novel labelling method is introduced by the values of band 4. Then difference image is clustered by SSFCM and labeled samples. To find good monitoring result and remove noise points, this work considers the spatial information of pixels and markov random field. The final monitoring result is obtained by updating the membership matrix. In order to show the validity of the proposed method, two TM remote sensing images are chosen to feed our method to detect the change of forest. The experimental results indicate that the improved semi-supervised FCM algorithm can reduce the missed alarm rate, and Markov random field method can remove the noise points in the process of clustering and reduce the false alarm rate.