摘要
遥感图像语义分割在农业、建筑物监测、城市规划等领域发挥着重要的作用,但传统的提取方法无法满足大规模生产且效率低下。针对该问题本文提出一种基于DeepLabV3+的遥感图像语义分割方法。首先,通过对原始数据变换生成多样化的训练数据集;然后,与FCN、U-Net、PSPNet三种语义分割方法比较。实验结果表明,本文方法在总体精度、准确率、交并比指标都达到最优,可实现遥感图像有效的提取,该研究可为遥感图像自动提取提供一定参考。
Semantic segmentation of remote sensing images plays an important role in fields such as agriculture,building monitoring,and urban planning,but traditional extraction methods fail to meet the needs of large-scale production and are inefficient.To address this issue,this article proposed a semantic segmentation method for remote sensing images based on DeepLabV3+.First,a variety of training data sets were generated by transforming the original data.Then,the proposed method was compared with three semantic segmentation methods,namely FCN,U-Net,and PSPNet.The experimental results show that the proposed method achieves the best overall precision,accuracy,and intersection over union and can effectively extract remote sensing images.This study can provide a reference for the automatic extraction of remote sensing images.
作者
梁静桦
梁杰文
LIANG Jinghua;LIANG Jiewen(Surveying and Mapping Insitute Lands and Resource Department of Guangdong Province,Guangzhou Guangdong 510700,China)
出处
《北京测绘》
2023年第12期1596-1600,共5页
Beijing Surveying and Mapping
基金
广东省科技计划(2021B1212100003)。
