可变形部件模型在高分辨率遥感影像建筑物检测中的应用

Building Detection from High Resolution Remote Sensing Imagery Based on a Deformable Part Model

摘要: 高分辨率遥感影像具有场景复杂、目标种类多样、同一目标呈现多种形态等特点，给建筑物检测带来困难。近年来，可变形部件模型（deformable part model，DPM）被广泛应用到模式识别领域，并且在自然场景的目标识别方面取得很好的效果。结合可变形部件模型，提出一种针对高分辨率遥感影像中建筑物的检测方法，将建筑物看作可变形部件的组合，通过训练得到其对应的参数模板，并采用滑动窗口的方式遍历待检测的影像，判断其中是否存在建筑物目标。通过对分辨率为0.5 m的高分辨率遥感影像的实验证明了方法的有效性。

Abstract: The characteristics of high resolution remote sensing imagery-complex scenes, diversity, various forms of one target, and so on-make automatic building detection difficult.. In recent years, the deformable part model (DPM) has become widely used in the field of pattern recognition, and effective for target recognition in natural scenes. In this paper, we propose a new method of building detection using high resolution remote sensing imagery, based on DPM. This method considers a building as a combination of deformable parts, obtains its template parameters by training, and traverses images with a sliding window to detect buildings. Experiments show the validity of the method.