首页 | 期刊介绍 | 编委会 | 道德声明 | 投稿指南 | 常用下载 | 过刊浏览 | In English
引用本文:朱玉玲,王建步,王安东,王锦锦,赵晓龙,任广波,胡亚斌,陈晓英,马毅.融合浅层特征的深度卷积神经网络互花米草遥感监测方法[J].海洋科学,2019,43(7):12-22.
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  下载PDF阅读器  关闭
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 1054次   下载 938 本文二维码信息
码上扫一扫!
分享到: 微信 更多
融合浅层特征的深度卷积神经网络互花米草遥感监测方法
朱玉玲1, 王建步1, 王安东2, 王锦锦3, 赵晓龙1, 任广波1, 胡亚斌1,4, 陈晓英1, 马毅1
1.自然资源部第一海洋研究所, 山东 青岛 266061;2.山东黄河三角洲国家级自然保护区管理局, 山东 东营 257091;3.珠海欧比特宇航科技股份有限公司, 广东 珠海 519000;4.大连海事大学, 辽宁 大连 116026
摘要:
基于2018年10月份黄河口入海两侧的LANDSAT8 OLI影像,提取植被指数和缨帽变换分量共9维光谱特征,构建融合浅层特征的8层深度卷积神经网络(deep convolutional neural network,DCNN)分类模型,开展互花米草(Spartina alterniflora Loisel)遥感监测的方法研究,并从不同的浅层特征来具体分析互花米草的监测结果。结果表明:(1)在分类方法上,DCNN模型的总体分类精度最高,达到90.33%,与支持向量机(support vector machine,SVM)、随机森林(random forest,RF)分类器相比,精度分别提高4.78%、2.7%,互花米草的生产者精度分别提高了2.56%、0.47%,说明在滨海湿地遥感影像分类中,DCNN有着更好的应用潜力;(2)融合浅层特征后,DCNN的总体分类精度和互花米草的识别精度分别提高了0.34%和3.25%,有效提高了对互花米草的监测能力。其中,融合归一化植被水分指数(NDⅡ)浅层特征的DCNN分类方法中,互花米草的识别精度提高最多,为2.56%,比值植被指数(RVI)次之,为2.32%。研究结果可为互花米草的监测与管理提供技术与数据支撑。
关键词:  深度卷积神经网络(deep convolutional neural network,DCNN)  浅层特征融合  湿地分类  互花米草(Spartina alterniflora Loisel)  黄河口
DOI:10.11759/hykx20190130001
分类号:P237
基金项目:国家自然科学基金项目(61601133,41706209);高分海岸带遥感监测与应用示范项目(41-Y30B12-9001-14/16)
Remote-sensed monitoring of Spartina alterniflora using deep convolutional neural network method with fusion of shallow features
ZHU Yu-ling1, WANG Jian-bu1, WANG An-dong2, WANG Jin-jin3, ZHAO Xiao-long1, REN Guang-bo1, HU Ya-bin1,4, CHEN Xiao-ying1, MA Yi1
1.First Institute of Oceanography, Ministry of Natural Resources, Qingdao 266061, China;2.Administration Bureau of the Yellow River Delta National Nature Reserve, Dongying 257091, China;3.Zhuhai Orbit Aerospace Science & Technology Co., Ltd., Zhuhai 519000, China;4.Dalian Maritime University, Dalian 116026, China
Abstract:
Although the deep convolutional neural network (DCNN) model has been applied to remote sensing image classification, the focus is generally on urban areas, and studies on multispectral image classification in coastal wetlands are relatively scarce. Based on the LANDSAT8 OLI imagery on both sides of the Yellow River estuary in October 2018, in this study, a total of nine-dimensional spectral features of vegetation index and the tasseled cap transformation components were extracted, and an eight-layer DCNN classification model with shallow features was constructed to identify Spartina alterniflora. The results of S. alterniflora monitoring were analyzed from different shallow features. The results indicated the following:(1) Compared with the support vector machine (SVM) and the random forest (RF) classifier, the DCNN model had the highest overall classification accuracy, reaching 90.33%; its accuracy was higher than those of the SVM and RF by 4.78% and 2.7%, respectively, and the producer's accuracy of S. alterniflora was higher by 2.56% and 0.47%, respectively. This shows that the DCNN has better application potential in coastal wetlands classification. (2) After the fusion of shallow features, the DCNN overall classification accuracy and its recognition accuracy of S. alterniflora were higher than those of SVM and RF by 0.34% and 3.25%, respectively, and the model effectively improved the monitoring ability of S. alterniflora. In addition, in the DCNN classification method that combines the shallow features of normalized vegetation water index (NDⅡ), the recognition accuracy of S. alterniflora increased the most, which was 2.56%, followed by the ratio of vegetation index (RVI), which was 2.32%.
Key words:  deep convolutional neural network (DCNN)  fusion of shallow feature  wetland classification  Spartina alterniflora  the Yellow River estuary
版权所有 《海洋科学》 Copyright©2008 All Rights Reserved
主管单位:中国科学院 主办单位:中国科学院海洋研究所
地址:青岛市市南区福山路32号  邮编:266071  电话:0532-82898755  E-mail:marinesciences@qdio.ac.cn
技术支持:北京勤云科技发展有限公司