摘要
为了提高雷达影像提取水体的精度,本文以2023年Sentinel-1系列影像为数据源,在DeepLabV3+网络模型的基础上优化主干网络,并融合SE通道注意力机制,提出了一种改进的深度学习网络模型SEDeepLabV3+,针对改进的模型进行了消融试验,并以7月31日北京市昌平区水体提取为例,对该模型进行了验证。试验结果表明,使用改进后的SEDeepLabV3+方法提取水体时,平均交并比与像素准确率能够达到88.55%和93.49%,与DeepLabV3+、HRNet、U-Net相比,平均交并比分别提高了2.26%、2.31%和5.08%,平均像素准确率分别提高了0.76%、0.80%和3.07%,改进后的SEDeepLabV3+不仅具有更轻量级的网络结构,而且能够有效地提高水体提取精度和效率。
In order to improve the accuracy of water extraction from radar images,this paper uses Sentinel-1 series images from 2023 as the data source,optimizes the backbone network on the basis of the DeepLabV3+network model,integrates the SE channel attention mechanism,and proposes an improved deep learning network model SEDeepLabV3+.The ablation experiment is carried out for the improved model,and the model is verified by the water body extraction in Changping district of Beijing on July 31.The experimental results show that when the improved SEDeepLabV3+method is used to extract water body,the mean intersection over union and pixel accuracy can reach 88.55%and 93.49%.Compared with DeepLabV3+,HRNet and U-Net,the average intersection ratio is increased by 2.26%,2.31%and 5.08%,and the average pixel accuracy is increased by 0.76%,0.80%and 3.07%,respectively.The improved SEDeepLabV3+not only has a lighter network structure,but also can effectively improve the accuracy and efficiency of water extraction.
作者
赵兴旺
赵妍
刘超
刘春阳
ZHAO Xingwang;ZHAO Yan;LIU Chao;LIU Chunyang(School of Geomatics,Anhui University of Science and Technology,Huainan 232001,China;Key Laboratory of Aviation-Aerospace-Ground Cooperative Monitoring and Early Warning of Coal Mining-induced Disasters of Anhui Higher Education Institutes,Huainan 232001,China;Coal Industry Engineering Research Center of Mining Area Environmental and Disaster Cooperative Monitoring,Huainan 232001,China)
出处
《测绘通报》
北大核心
2025年第3期66-70,共5页
Bulletin of Surveying and Mapping
基金
安徽省自然科学基金(2208085MD101,2108085QD171)。
