模拟退火优化BP神经网络进行云相态分类

为利用中分辨率成像光谱仪(MODIS)多通道云图数据对云相态进行分类,提出了一种利用云图红外、可见光谱段数据,基于SABP神经网络的云相态分类算法.算法基于BP人工神经网络学习算法及模拟退火算法.在两算法结合的基础上,作出了有益于云相态分类的改进.基准网络模型为3层前馈BP网络,其中,隐层和输出层使用不同的激活函数.选取了5种光谱特征作为网络输入,输出两单元判定结果.为加速收敛速度,避免振荡,采用了一种自适应的变学习系数、惯性系数策略.个例分析表明,算法在中低纬度地区效果良好.

Smi ulated annea ling optmi ized BP-ANNmethod for c loud thermodynam ic phase retr ieva l

In o rder to re trieve c loud therm ody nam ic pha se usingM OD IS im age s, an ANN -ba sedm e thod ut ilizing M OD IS v isib le and in frared band s data and m ult i-spect ra l fea tu res w a s pu t fo rw ard. T h is m ethod com b ines the re la tive ly m ature BP art ificia l neu ra l ne tw o rk study a lg o rithm and sim u1a ted annea ling a lgorithm. S pecia l am endm ent s w e re m ade fo r be tte r cloud therm ody nam ic pha se re tr iev al perfo rm ance. A t rip le-lay er feed fo rw a rd BP-ANN base m ode lw h ich u sed div erse act iva t io n funct io ns w as se lected. F ive spec tral fea tures w e re selected as inpu t o f the ne tw o rkm ode l . the netw o rk ou tpu t laye r ha s tw o un its fo r speeding up and avo id ing v ibra tion e rro rs. A se lf-adap ted va ria t io nal study coe ff icient and iner tia coef f icien t st ra teg y w as adopted. C ase show presen ted the e ff iciency o f the m e thod in m id o r low lat itude reg io ns.