基于UE-Net6的无人机遥感影像城市地表水提取方法  

Urban Surface Water Extraction from UAV Remote Sensing Image Based on UE-Net6

作　　者：卞艳[1] 宫雨生[1] 马国鹏 华前程 李源 BIAN Yan;GONG Yu-sheng;MA Guo-peng;HUA Qian-cheng;LI Yuan(School of Civil Engineering,University of Science and Technology Liaoning,Anshan 114051,China;The Bureau of Non-ferrous Geology of Liaoning Province 103 Limited Liability Branch Company,Dandong 118008,China;Liaoning Shengji Construction Foundation Engineering Co.,Ltd.,Anshan 114051,China)

机构地区：[1]辽宁科技大学土木工程学院,辽宁鞍山114051 [2]辽宁省有色地质103队有限责任公司,辽宁丹东118008 [3]辽宁盛基建设基础工程有限公司,辽宁鞍山114051

出　　处：《水电能源科学》2022年第5期26-29,155,共5页Water Resources and Power

基　　金：国家自然科学基金青年基金项目(41801294);武汉大学测绘遥感信息工程国家重点实验室珞珈一号特别开放研究基金(18T07)。

摘　　要：针对利用高分辨率、背景复杂、波段少的无人机遥感影像提取城市地表水的方法少且提取精度不高等问题,采用深度学习法,构建了不同深度的U-Net网络模型(5、6、7层)提取城市地表水,对比发现U-Net6模型效果最优;同时,为避免神经元失活和模型过拟合现象,采用ELU代替ReLU并引入Dropout正则化对U-Net6网络进行改进,进而提出了一种以ELU为激活函数、网络层数为6的基于无人机遥感影像的城市地表水自动提取方法—UE-Net6方法,从而实现了复杂背景下水体信息的精确提取。为验证所提方法的优越性,试验选取同样的训练集与测试集,分别对经典U-Net、SegNet、FCN及UE-Net6模型进行对比试验。结果表明,UE-Net6方法的水体提取精度明显优于其他模型的提取精度。In view of the use of high resolution, complicated background, less band of UAV RS image extraction methods of urban surface water and low extraction accuracy, the method of deep learning was used to build different depth of U-net network model(5, 6, and 7 layer) to extract urban surface water, and the U-net6 model had best effect. In order to avoid the neuron inactivation and model overfitting, the ELU was used to replace the ReLU, and Dropout regularization was introduced to improve U-net6 network. Thus, an automatic extraction method of urban surface water based on UAV RS image, UE-net6, which takes ELU as activation function and has 6 network layers, was proposed, so as to achieve accurate extraction of water body information under complex background. In order to verify the proposed method, the same training set and test set were selected to carry out comparative tests on the classic U-net, SegNet, FCN and UE-net6 models, respectively. The results show that the water extraction accuracy of UE-NET6 method is obviously better than that of the comparison model.

关 键 词：无人机遥感影像 水体提取 深度学习 激活函数 卷积神经网络 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] TV221.1[自动化与计算机技术—控制科学与工程]