克隆选择算法在遥感影像分类中的应用

Clonal Selection Algorithm for Classification of Remote Sensing Imagery

针对遥感影像分类中的局部极值、鲁棒性等问题,提出基于克隆选择算法的遥感影像监督分类方法。所提方法将遥感影像各波段亮度值定义为抗原的属性,通过计算遥感像元与抗体的亲和力识别其类别,并采用实数制方式对抗体进行变异,在保证亲和力上升的同时,也保证了系统的多样性。该算法应用于广州市遥感影像数据分类的实验结果显示:在抗体的进化过程中,抗体的亲和力和识别能力不断提高,最终的分类精度达到92.9%。与最大似然法相比,克隆选择算法的分类精度更高。

A novel supervised classifier for remote sensing data by employing clonal selection algorithm （CSA） is presented for solving the problems, e.g. local optimum and robustness, in remote sensing imagery classification. In the classifier, band brightness is defined as antigen＇s attribute, and the image is classified into class with the maximum affinity by calculating the affinity between remote sensing pixel and antibody. Antibody＇s real-encoding mutation integrates affinity＇s ascending and system＇s variety. The concerned experiment shows that antibody＇s affinity increases along with its evolution, and CSA classification＇s overall accuracy is 92.9%. Comparing with conventional Maximum Likelihood, CSA can get better precision.