基于深度置信网络与数学形态学融合的遥感影像建筑物变化检测

当前人工调查土地资源利用情况具有较高的人力成本且劳动强度大,对其实现自动变化检测具有较高的理论和应用价值。将深度置信网络(deep belief network,DBN)应用于高分辨率遥感影像的建筑物变化检测,但DBN在变化检测时存在由误判现象造成的建筑物完整度欠缺、空间存在噪声等问题,提出DBN与数学形态学融合模型对高分辨率遥感影像建筑物进行变化检测。在遥感影像预处理基础上,标记少量明显的变化与未变化样本,利用搜索窗口从标记的区域获取大量带有标签的样本训练融合模型分类器对建筑物进行变化检测,检测方法准确率为94.76%,召回率为87.63%,F_1为91.06%。实验结果表明,该方法可以为建筑物的变化检测提供有效依据。

Building Change Detection Based on Deep Belief Networks and Mathematical Morphology Fusion

In view of the current man-made survey of land resource utilization, the task has high labor cost and high labor intensity, and it has high theoretical and practical value for automatic change detection. In this paper, the Deep Belief Networks (DBN) is combined with the high-resolution remote sensing image to detect the change of the building. In order to solve the problem of lack of completeness of the building and noise in the space caused by the misjudgment of the DBN in the change detection, The DBN and mathematical morphology fusion model is used to detect building changes in high-resolution remote sensing images. On the basis of remote sensing image preprocessing, a small number of obvious changes and unchanging samples are marked, and a large number of tagged samples are trained from the marked area to train the fusion model classifier to detect changes in the building. The accuracy of the detection method is 94.76%, the recall rate is 87.63%, and the F1 value is 91.06%. The experimental results show that the method can provide an effective basis for building change detection.