基于最小外接矩形的遥感影像建筑物轮廓优化方法

深度学习提取高分辨率遥感影像中的建筑物信息容易受到物体周围的阴影、植被等噪声干扰而使结果存在边界锯齿化、建筑物整体不规整等问题。本文提出了利用符合建筑物边界轮廓的最小外接矩形最大限度地拟合建筑物轮廓的思路。首先利用深度学习和建筑物验证处理得到的建筑物信息,对建筑物边界利用垂距法进行多边形的拟合;然后对多边形的最小外接矩形进行筛选,选取最合适的最小外接矩形边线段作为新的边界轮廓,以提高提取的精度。对多幅遥感影像进行了实验,结果表明,本文所提出的方法提高了深度学习提取的建筑物边界轮廓准确性,能更逼近真实建筑物的边界轮廓。

Remote Sensing Image Building Contour Optimization Method Based on Minimum External Rectangle

In deep learning,building information extracted from high-resolution remote sensing images is easily disturbed by shadows around objects,vegetation and other noises,resulting in jagged boundaries and irregular buildings.This paper puts forward the idea of fitting the building contour to the maximum extent by using the minimum external rectangle that conforms to the building boundary contour.First,deep learning and building verification applied to process the building information.The vertical distance method is explored to fit polygons of building boundaries.Then,the minimum peripheral rectangle of the polygon is screened,and the most appropriate edge line of the minimum peripheral rectangle is selected as the new boundary contour to improve the extraction accuracy.Experiments on remote sensing images show that the proposed method improved the accuracy and final accuracy of building boundary contour extracted by deep learning,and could more closely approximate the boundary contour of buildings in remote sensing images.