摘要
基于大地测量型GNSS接收机获取的反射信号反演土壤湿度是GNSS领域的研究热点。为克服常规线性回归和BP神经网络算法等的缺陷,本文提出了一种基于深度置信网络的GNSS-IR土壤湿度反演方法。试验结果表明,基于该方法得到的决定系数、土壤湿度平均绝对误差和均方根误差分别为0.909 8、0.017、0.021,与线性回归和BP神经网络算法相比,与实测数据吻合度更高,可有效提高土壤湿度反演精度,证明了方法的有效性和可靠性。
Retrieving soil moisture based on reflected signals acquired by geodetic GNSS receivers is a research hotspot in the field of GNSS.In order to overcome the shortcomings of conventional linear regression and BP neural network algorithms,this paper proposes a GNSS-IR soil moisture retrieval method based on deep belief network.The results show that the coefficient of determination,the average absolute error and the root mean square error of soil moisture based on this method are 0.9098,0.017 and 0.021.Compared with the linear regression and BP neural network algorithm,they are more consistent with the measured data,and can effectively improve the accuracy of soil moisture inversion,which proves the validity and reliability of the method.
作者
陈堃
沈飞
曹新运
朱逸凡
CHEN Kun;SHEN Fei;CAO Xinyun;ZHU Yifan(School of Geography,Nanjing Normal University,Nanjing 210023,China;Key Laboratory of Virtual Geographic Environment(Nanjing Normal University),Ministry of Education,Nanjing 210023,China;Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application,Nanjing 210023,China)
出处
《测绘通报》
CSCD
北大核心
2020年第9期100-105,共6页
Bulletin of Surveying and Mapping
基金
国家自然科学基金(41904018)
江苏省自然科学基金(BK20190714)
武汉大学地球空间环境与大地测量教育部重点实验室开放基金(18-01-04)。
