基于深度学习的遥感影像建筑物提取方法研究

Research on Building Extraction Method from Remote Sensing Images Based on Deep Learning

随着遥感技术的不断发展,遥感影像数据中包含的地物信息越来越丰富,因此被广泛应用于土地信息变化、灾害检测、环境评估等各个领域。建筑物作为主要的地表地物之一,同时是人类活动的主要场所,与人口和经济密切相关,被广泛应用于城市规划、智慧城市建设等各个方面。然而传统的建筑物提取方法耗时耗力,因此使用以卷积神经网络为基础的深度学习方法成为遥感图像处理的主流方法。本文在参考了国内外建筑物提取方法研究的文献后,采用Unet++的网络模型进行建筑物提取,并根据IoU、kappa系数以及F1这三个精度指标,评价其提取结果。

With the continuous development of remote sensing technology, the ground object information contained in remote sensing image data is more and more abundant, so it is widely used in various fields such as land information change, disaster detection, and environmental assessment. As one of the main surface features and the main place for human activities, buildings are closely related to population and economy, and are widely used in urban planning, smart city construction and other aspects. However, traditional building extraction methods are time-consuming and labor-intensive, so the use of deep learning methods based on convolutional neural networks has become the mainstream method for remote sensing image processing. In this paper, after referring to the literature on building extraction methods at home and abroad, the network model of Unet++ is used to extract buildings, and the extraction results are evaluated according to the three accuracy indicators of IoU, kappa coefficient and F1.