摘要
归一化植被指数(NDVI)是地表植被信息的重要表征,针对传统GNSS干涉遥感(GNSS-IR)技术NDVI反演中数据利用不充分的问题,本文融合多轨多频GNSS卫星数据,采用随机森林(RF)、BP神经网络(Back Propagation Neural Network)、深度置信神经网络(DBN-DNN)3种机器学习算法,分别建立了基于单频/双频、单星/多星数据的GNSS-IR模式NDVI反演模型,并对其普适性进行了验证.结果表明:机器学习算法能够充分利用多轨多频GNSS数据,有效提高NDVI反演精度,其中DBN-DNN算法的建模精度最高,相比于BP网络和随机森林,其平均相关系数分别提高了2.57%,20.17%,平均均方根误差分别降低了1.97%,8.40%;L_2频多星数据最适用于NDVI反演,相比于双频、L_1频、单星数据,其平均相关系数分别提高了12.31%,19.63%,26.07%,平均均方根误差分别下降了6.33%,9.33%,12.50%.
Normalized difference vegetation index(NDVI) is an important representation of surface vegetation information. To solve the problem of insufficient use of data in NDVI inversion of traditional GNSS interferometric reflectometry(GNSS-IR), multi-orbit and multi-frequency GNSS satellite data were fused, and three machine learning algorithms, random forest(RF), back propagation(BP) neural network, and deep belief network deep neural network(DBN-DNN), were used to build NDVI inversion models of GNSS-IR mode based on single frequency/dual-frequency and single satellite/multi-satellite data, and the universality were verified. The results show that the machine learning algorithm can make full use of multi-orbit and multi-frequency GNSS data and effectively improve the NDVI inversion accuracy. Among them, the modeling accuracy of DBN-DNN algorithm is the highest. Compared with BP network and random forest, its average correlation coefficient is increased by 2.57% and 20.17% respectively, and the average root mean square error(RMSE) is reduced by 1.97% and 8.40% respectively;L_(2) frequency multi-satellite data is most suitable for NDVI inversion. Compared with dual-frequency, L_1 frequency and single satellite, its average correlation coefficient is increased by 12.31%, 19.63% and 26.07% respectively, and the average RMSE is decreased by 6.33%, 9.33% and 12.50% respectively.
作者
刘一
郑南山
丁锐
张克非
鞠海龙
LIU Yi;ZHENG Nanshan;DING Rui;ZHANG Kefei;JU Hailong(Key Laboratory of Land Environment and Disaster Monitoring,MNR,China University of Mining and Technology,Xuzhou,Jiangsu 221116,China;School of Environment and Spatial Informatics,China University of Mining and Technology,Xuzhou,Jiangsu221l16,China)
出处
《中国矿业大学学报》
EI
CAS
CSCD
北大核心
2023年第5期1014-1021,共8页
Journal of China University of Mining & Technology
基金
国家自然科学基金项目(41974039)
国家自然科学基金联合重点项目(U22A20569)
自然资源部国土环境与灾害监测重点实验室开放基金(LEDM2021B11)
教育部高等学校学科创新引智计划项目(2020-10969)
江苏高校优势学科建设项目(140119001)。
