改进DeeplabV3+模型的河流水体提取  

River Extraction Based on Improved DeeplabV3+Model

作　　者：张晗涛 胡荣明[1] 姜友谊[1] 胡亚轩[2] ZHANG Hantao;HU Rongming;JIANG Youyi;HU Yaxuan(College of Geomatics,Xi’an University of Science and Technology,Xi’an 710000,China;The Second Monitoring and Application Center,CEA,Xi’an 710054,China)

机构地区：[1]西安科技大学测绘科学与技术学院,西安710000 [2]中国地震局第二监测中心,西安710054

出　　处：《遥感信息》2023年第3期146-152,共7页Remote Sensing Information

基　　金：国家自然科学基金项目(42171394、41972315)。

摘　　要：为了探究深度学习DeeplabV3+模型在河流水体提取的潜力,分别构建了ResNet-50、ResNet-101、ResNet-152、Xception共4种不同骨架网络的DeeplabV3+模型,开展河流水体提取研究。通过河流水体提取结果对比分析,确定了最优骨架网络模型为ResNet-50,在此基础上提出了改进的DeeplabV3+模型,并与最邻近分类法、随机森林分类法、支持向量机分类法、原始DeeplabV3+模型法等分类方法的分类结果进行比较。结果表明:改进的DeeplabV3+网络模型能有效提取河流水体目标,增强小面积河流水体识别能力,减少河流水体漏分现象,提高河流水体提取效果。改进后的DeeplabV3+网络模型在高分辨率遥感影像河流水体提取方面具有可行性,为后续该领域的进一步研究应用提供了参考。In order to explore the potential of deep learning DeeplabV3+model in river water extraction,DeeplabV3+models with different skeleton networks in ResNet-50,ResNet-101,ResNet-152,and Xception are constructed to carry out river water extraction research.Through the comparative analysis of the results of river water extraction,it can be seen that the optimal skeleton network model is ResNet-50.On this basis,an improved DeeplabV3+model is proposed and compared with other classification methods,such as nearest neighbor classification,random forest classification,support vector machine classification,original DeeplabV3+model and so on.The results show that the improved DeeplabV3+network model can effectively extract river water objects,enhance the recognition ability of small area river water bodies,reduce the phenomenon of river water leakage,and improve the effect of river water extraction.It can be seen that the improved DeeplabV3+network model is feasible in river water extraction from high-resolution remote sensing images,which provides a reference for further research and application in this field.

关 键 词：深度学习 高分辨率遥感影像 河流水体提取 DeeplabV3+ 卷积神经网络 

分 类 号：TP237[自动化与计算机技术—检测技术与自动化装置]