基于深度学习的高分遥感影像建筑物提取研究

Research on Building Extraction from High-resolution Remote Sensing Image Based on Deep Learning

传统的建筑物提取方法普遍存在过分依赖人工设计、自动化程度低、泛化能力弱等问题。随着深度学习算法在高分辨率卫星图像分类领域的应用,本文基于深度学习方法,以福建南安地区为研究区,以国产高分遥感影像为数据源,选取4块典型区域制作建筑物数据集,搭建U-Net和Mask R-CNN深度学习模型在自建的南安数据集上训练,从定量和定性的角度对比两种模型对建筑物提取的效果,最终选定精度更高的U-Net方法作为最终的提取算法;再对数据集中正负样本的比例进行调整,进一步提高了模型分割的精度,实现了基于深度学习方法的南安地区建筑物的识别和提取。

The extraction of buildings from high-resolution remote sensing images is of great significance to socio-economic construction,urban development planning,mapping,disaster assessment,and national defense.Traditional building extraction methods generally have problems such as excessive reliance on manual design,low degree of automation,and weak generalization ability.In recent years,deep learning algorithms have been applied in the field of high-resolution satellite image classification.Based on the deep learning method,this paper takes the Nan'an area of Fujian as the research area,uses domestic high-resolution remote sensing images as the data source,selects 4 typical regions to make a building data set,and builds the U-Net and Mask R-CNN deep learning models.The performance of the two models in building extraction was compared quantitatively and qualitatively,and the U-Net method with higher accuracy was selected as the final extraction algorithm.The proportion of positive and negative samples in the data set is adjusted to further improve the precision of model segmentation,and the recognition and extraction of buildings in Nan'an area based on deep learning method is realized.