地理空间信息2024，Vol.22Issue(1)：29-32,4.DOI:10.3969/j.issn.1672-4623.2024.01.007
基于提示信息的分割大模型建筑物提取方法研究
Research on Building Extraction Method of Segmentation Model Based on Prompt Information
摘要
Abstract
To address the issue of poor building extraction performance on the test set by existing segmentation models,we proposed a precise building extraction method based on segmentation model prompt information.Firstly,we used the segmentation model to roughly extract the building areas from the original images.Then,we generated the buildings'target bounding boxes and key points from these areas to serve as prompt information for the segmentation model.Finally,we guided the precise segmentation of building targets by prompt information.To evalu-ate the proposed method,we conducted segmentation experiments on multiple remote sensing images.The results show that compared to existing segmentation methods,this method improves the precision and recall rates from 88.71%,58.38%to 91.49%,83.74%,respectively.关键词
目标分割/分割大模型/建筑物提取/深度学习Key words
object segmentation/segmentation model/building extraction/deep learning分类
天文与地球科学引用本文复制引用
向浩,祁信舒,吕现伟,梁思,沈佳洁..基于提示信息的分割大模型建筑物提取方法研究[J].地理空间信息,2024,22(1):29-32,4.基金项目
湖北省自然资源厅科技资助项目(ZRZY2023KJ08). （ZRZY2023KJ08）
