深度学习的BP神经网络在GNSS水准拟合中的应用  

Application of BP Neural Network with Deep Learning in GNSS Level Fitting

作　　者：董洲洋 徐卫明[1] 庄昊 邱广闻 DONG Zhouyang;XU Weiming;ZHUANG Hao;QIU Guangwen(Department of Military Oceanography and Hydrography,Dalian Naval Academy,Dalian 116000,China;32023 Troops,Dalian 116000,China)

机构地区：[1]海军大连舰艇学院军事海洋与测绘系,辽宁大连116000 [2]32023部队,辽宁大连116000

出　　处：《海洋测绘》2019年第5期26-29,共4页Hydrographic Surveying and Charting

基　　金：国家自然科学基金(61071006)

摘　　要：针对误差逆向传播 BP ( back propagation)神经网络在 GNSS 水准拟合中存在梯度消失、陷于局部最小点的问题,通过使用深度学习中的分段线性整流函数Relu( rectified linear units)作为神经元激活函数,自适应矩估计Adam ( adaptive moment estimation)算法作为网络优化函数,提出了一种基于深度学习的 BP 神经网络模型。研究结果表明:改进后的 BP 神经网络内外符合精度分别提高近 50%和 25%,可达 0. 9 cm 和 2. 4 cm,为 GNSS 水准拟合提供了新的思路。Aiming at the problem of gradient disappearance and local minimum in GNSS leveling fitting of BP neural network with error back propagation,a BP algorithm neural network model based on deep learning is proposed by using piecewise linear rectification function Relu as neuron activation function and adaptive moment estimation Adam algorithm as network optimization function. The results show that the internal and external coincidence accuracy of the improved BP neural network is improved by nearly 50% and 25%,respectively, reaching 0.9 cm and 2.4 cm,which provides a new idea for GNSS level fitting.

关 键 词：深度学习 GNSS 水准拟合 BP 神经网络 Relu 函数 ADAM 算法 

分 类 号：P228.4[天文地球—大地测量学与测量工程]