摘要
针对现有协同神经网络参数优化方法的不足,提出了一种基于免疫克隆算法的参数优化方法.与在平衡注意参数条件下的算法和在不平衡注意参数条件下基于遗传算法和模拟退火算法的优化方法相比,新方法具有全局兼局部寻优能力,不易陷入局部极值,并且迭代步长是自适应调整的.对纹理图像与遥感图像的分类识别结果表明:新方法不仅具有更快的收敛速度而且具有更优的分类识别性能,同时验证了注意参数及所有参数对各原型模式之间竞争态势的影响,从而达到更佳的分类识别效果.
Due to the shortages of optimization algorithms available in synergetic neural network （ SNN ）, an algorithm of parameters optimization on immunity elonal algorithm （ICA） was proposed here. Compared with the algorithm under balanced attention parameters and that under unbalanced attention parameters on genetic algorithm （GA） and simulated annealing algorithm （SA）, the new algorithm has the global and local searching ability and is not easy to get into local optima. And the iterative step is adjusted adaptively. Experiments on textural images and remote sensing images show that the proposed algorithm has not only faster convergent speed but also better classification performance. Simultaneously, the effect of attention parameters and all parameters on the competition of prototype patterns is verified and then better recognition result can be obtained.
出处
《红外与毫米波学报》
SCIE
EI
CAS
CSCD
北大核心
2007年第1期38-42,共5页
Journal of Infrared and Millimeter Waves
基金
国家"863"计划(2002AA135080)
国家自然科学基金(60133010)
"十.五"国防预研资助项目(11307050103)
关键词
协同神经网络
注意参数
免疫克隆算法
图像分类
synergetie rieural network
attention parameter
immunity elonal algorithm
image classification
