基于免疫网络的遥感影像分类算法

Classification of remote sensing image based on immune network

基于独特型免疫网络原理,提出了一种新型的分区记忆模式人工独特型网络模型,并利用其对卫星遥感数据进行了分类。该模型在结构上将免疫网络的记忆抗体划分为特异记忆抗体区和自由记忆抗体区。前者的主要功能是记忆各类别抗原的特异特征,后者为前者提供各种类型的抗体源。记忆抗体间按照亚动力学原理进行调节,实现免疫网络的寻优过程。基于上述分区,它在初次免疫响应过程中实现网络的搭建和训练,在二次免疫响应过程中实现信息提取。最后利用该模型对ETM数据进行地物分类,并与传统分类方法进行对比。结果表明:该模型的总分类精度和Kappa系数分别是92.6%和0.91,优于传统分类方法。

Based on idiotypic immune network theory,a Regional-memory-pattern Artificial Idiotypic Network（RAIN） is proposed to classify multi-spectral remote sensing image.The immune memory antibodies of RAIN model are divided into two regions： specific memory antibody and free memory antibody.Specific memory antibody has several specific subregions sensitive to specific antigens,while free memory antibody region supplied kinds of specific memory antibodies for the former region.The adjustment and optimization of RAIN are realized according to antibody metdynamics.The initialization and training of RAIN are realized in the primary immune response process,and information extraction is executed in the second immune response process.At last, RAIN is used for the classification of ETM data.Accuracy and Kappa coefficient of our method are 92.6% and 0.91 respectively, while that of traditional Parallelepiped,Maximum Likelihood and Minimum Distance are 81.8%,82.2%,71.8%,and 0.78,0.78,0.65. The results show that RAIN is superior to three traditional classification algorithms.