摘要
现有的数字高程模型压缩方法大多从编码方式上进行优化,而很少利用其数据的自相关性。为此,提出了一种采用L-M训练算法的单隐层BP神经网络实现机载数字高程模型压缩的新方法,并给出了实现压缩的详细过程。论述了采用单隐层网络的理由,并根据机载要求的相对误差精度去选择最少的隐层节点数。通过选取ASTERGDEM30米分辨率的高精度数字高程模型进行了大量仿真,验证了所提方法的可行性和实用性。
The current compression of airborne Digital Elevation Model（DEM） is optimized mostly by coding method,and is seldom optimized by self-correlation of DEM.A new compression method of airborne DEM is presented,which is based on the Single Hidden Layer Back-Propagation（BP） neural network adopting Levenberg-Marquardt（LM） training algorithm,and the compression process is given in detail.The advantage of a single hidden layer network superior to the multi hidden layer network is discussed,and the least hidden nodes are selected to get the maximum compression ratio based on the relative error of the actual onboard accuracy required.The validity and feasibility of this method are verified by simulation.
出处
《航空工程进展》
2011年第3期339-343,共5页
Advances in Aeronautical Science and Engineering
基金
航空科学基金(08C53011)
关键词
数字高程模型压缩
BP神经网络
L-M算法
机载
digital elevation model compression
back-propagation neural network
Levenberg-Marquardt
airborne
