摘要
针对数学拟合法在进行全球定位系统(GPS)水准拟合时,因受自身模型限制,导致GPS水准拟合精度不高的问题,该文提出了一种基于EGM2008模型和深度学习的GPS水准拟合法。首先使用深度学习中的分段线性整流函数(ReLU)作为神经元激活函数加快网络的收敛速度,然后利用自适应矩估计函数(Adam)作为优化函数加速获取最优解,并采用正则化丢弃法(Dropout)增强深度学习网络的泛化能力。通过实测数据计算表明:该文方法相比常用的多项式拟合法,丘陵地区外符合精度提高了约65%,达到1.7 cm;高差变化较大的山地外符合精度提高了约90%,达到1.2 cm。
Aiming at the problem that the accuracy of global positioning system(GPS) leveling is not high due to the limitation of its own model when the mathematical fitting method performs GPS leveling fitting, a novel GPS level fitting method was proposed based on EGM2008 model and deep learning technology. The rectified linear units(ReLU) as a neuron activation function in deep learning was used to accelerate network convergence, the adaptive moment estimation(Adam) function as an optimization function was used to accelerate the optimal solution, and the regularized Dropout method was used to improve the network ubiquity. According to the field data, the experiment results showed that, compared to the polynomial method, the external coincidence accuracy of hilly area was improved by about 65% to 1.7 cm, and the external coincidence accuracy was improved by about 90% to 1.2 cm in the large height-difference mountain area.
作者
董洲洋
徐卫明
庄昊
孟浩
DONG Zhouyang;XU Weiming;ZHUANG Hao;MENG Hao(Department of Military Oceanography&Hydrography&Cartography,Dalian Naval Academy,Dalian,Liaoning 116000,China;Troops 32023,Dalian,Liaoning 116000,China)
出处
《测绘科学》
CSCD
北大核心
2021年第4期57-62,共6页
Science of Surveying and Mapping
基金
国家自然科学基金项目(61071006)。
