SAM在遥感影像通用语义分割中的应用
Application of SAM in General-Purpose Semantic Segmentation of Remote Sensing Images
李帅 1初启凤 1袁如金 1何月 2张开硕 1李佳明1
作者信息
- 1. 黑龙江地理信息工程院,黑龙江哈尔滨 150081
- 2. 自然资源部第三地形测量队,黑龙江哈尔滨 150025
- 折叠
摘要
Segment Anything Model(SAM)是Meta公司最新发布的一个处理图像分割任务的通用模型,与FCN、U-Net等只能处理特定类型图片的图像分割模型不同,SAM具有较强的泛化能力.本文简述了目前主要的图像分割模型及特点,提出了基于SAM的遥感影像通用语义分割方法,研发了软件平台,并详细阐述了各功能模块的设计思路.实践表明,SAM在遥感影像通用分割方面具有良好的应用效果,研究成果在自然资源调查监测、农作物种植结构提取等工作中得到广泛应用.
Abstract
Segment Anything Model (SAM) is a general model for processing image segmentation task newly released by Meta and it has generalization capabilities which is different from FCN,U-Net and other image segmentation models that can only process specific types of images. This paper briefly describes the main segmentation models and characteristics,proposes a general semantic segmenta-tion method for remote sensing images based on SAM,develops a software platform,and elaborates the design ideas of each functional module. Practice has shown that SAM has a good application effect on general-purpose image segmentation of remote sensing images. The research results have been widely used in the investigation and monitoring of natural resources and the extraction of crop planting structures.
关键词
SAM/遥感影像/语义分割/深度学习/自然资源Key words
SAM/remote sensing images/semantic segmentation/deep learning/natural resources引用本文复制引用
基金项目
黑龙江省测绘地理信息局2022年基础测绘科技与标准计划(一期)(202202)
出版年
2024
NETL
NSTL
万方数据