利用改进Fisher分类器进行遥感图像变化检测

Change detection in multitemporal remote sensing images based on local mean dynamic Fisher discriminant analysis

将多时相遥感图像变化检测问题看成一个分类问题,利用改进的动态Fisher分类器通过二维联合直方图检测变化区域.考虑图像邻域关系,提出基于局部均值的动态Fisher分类器,在引入图像空间关系的同时,根据当前检测结果动态调整训练参数,解决了由于初始训练数据选取不同而造成的不稳定性.该算法不需要假设分布模型,不受差异算子的影响,且将原有的像素级检测提升为上下文相关检测.实验结果表明,该算法提高了检测精度,检测结果稳定。

This paper proposes a novel change detection technique, which treats the detection problem as a classifier problem and uses the improved dynamic Fisher classifier to identify the changes in the joint intensity histogram. By considering the relationship between the pixel and its neighborhood, local mean dynamic Fisher discriminant analysis （LMDFDA） is proposed to introduce the neighborhood＇ s information. Meanwhile, the parameters of the classifier are adjusted according to the current detection result, which avoids the influences of initial conditions. The proposed method is distribution free, context-sensitive and not affected by comparison operators. Experiments show that the proposed algorithm is effective and feasible for real multi-temporal remote sensing images.