摘要
针对遥感影像建筑物提取过程中出现的边缘模糊和提取精度低的问题,以LiDAR数据中的高度信息作为高光谱影像的补充,提出一种多模态遥感影像建筑物提取方法。该方法首先利用随机森林模型对高光谱遥感影像进行特征波段优选,以实现对高光谱影像的降维。其次,提出一种针对多模态遥感影像的建筑物提取网络EAFINet,该网络根据两种数据的特性对通道数进行设计,并引入金字塔切分注意力和密集空洞空间金字塔池化,以实现更高精度的建筑物提取。最后,利用连通域滤波对模型输出结果进行分类后处理,进一步优化提取结果。在Houston 2018数据集和MDAS数据集上的实验结果表明,该方法能够实现较高精度的建筑物提取,优于当前主流方法。
To address the issues of edge blurring and low extraction accuracy in building extraction from remote sensing images,this study proposes a multimodal remote sensing image building extraction method by using height information from LiDAR data as a supplement to hyperspectral images.Firstly,the method employs a random forest model to select optimal feature bands from hyperspectral remote sensing images,achieving dimensionality reduction.Secondly,a building extraction network for multimodal remote sensing images,named EAFINet,is proposed.This network is designed based on the characteristics of the two types of data and incorporates pyramid split attention and dense atrous spatial pyramid pooling to achieve higher precision in building extraction.Finally,connected domain filtering is used for post-classification processing of the model output to further optimize the extraction results.Experimental results on the Houston 2018 dataset and the MDAS dataset demonstrate that this method can achieve high-precision building extraction,outperforming current mainstream methods.
作者
曲云昊
王昶
QU Yunhao;WANG Chang(School of Civil Engineering,University of Science and Technology Liaoning,Anshan,Liaoning 114051,China)
出处
《遥感信息》
北大核心
2025年第2期139-148,共10页
Remote Sensing Information
基金
辽宁省教育厅面上项目(LJKMZ20220638)。
