基于卷积神经网络的建筑物精细化提取

Refine Extraction of Buildings Based on the Convolutional Neural Network

现有图像分割方法往往受图像模糊和噪声的影响,提取的轮廓不准确。为了提取建筑物的精确轮廓,提出了一种基于卷积神经网络的集成方法,包括建筑物定位、形状判断、形状匹配等步骤。实验证明,无论是对DSM图像还是多光谱影像,该方法都能获得精确的建筑物轮廓。

The existing methods of image segmentation are often affected by image blurring and noise, so that the extracted contours are not accurate. In order to extract the accurate contours of buildings, this paper proposed an integrated approach based on the convolutional neural network, which included building locating, shape judgment and shape matching. The experimental result shows that the proposed method can obtain accurate building contours both for DSM and multispectral images.