摘要
针对BP神经网络遥感水深反演算法(简称传统BP算法)的缺点,提出了改进型BP神经网络遥感水深反演算法(简称改进型BP算法),其基本原理是在模型训练过程中反复运用粒子群算法对BP神经网络的权值和阈值进行优化以弥补传统BP算法的不足。试验表明:改进型BP算法的训练迭代收敛速度明显快于传统BP算法,浅水区的水深反演精度优于传统BP算法,且学习算法对初始权值和阈值不敏感。
BPANN algorithm is commonly used for estimating water depth from satellite imagery.In this paper,an improved BPANN algorithm is presented to overcome some disadvantages of BPANN algorithm.Its principle is that particle swarm optimization（PSO） is used to optimize the weights and thresholds of ANN in the process of training.The experiments show that improved BPANN algorithm has faster convergence speed and better generalization ability,it is not sensitive to initial weights and thresholds,and it can make more accurate results than BPANN algorithm.
出处
《测绘通报》
CSCD
北大核心
2017年第2期40-44,共5页
Bulletin of Surveying and Mapping
基金
上海市科委科研基金(14590502200)
关键词
遥感水深反演
传统BP算法
粒子群算法
改进型BP算法
权值和阈值优化
estimating water depth from satellite imagery
backpropagation-based artificial neural network algorithm（BPANN algorithm）
particle swarm optimization（PSO）
improved BPANN algorithm
optimization of initial weights and thresholds
