Development and Demonstration of an Artificial Immune Algorithm for Mangrove Mapping Using Landsat TM
Development and Demonstration of an Artificial Immune Algorithm for Mangrove Mapping Using Landsat TM
复制标题
DOI：
10.1109/lgrs.2012.2221675
复制
发表时间：
2013-07
影响因子：
4.8
通讯作者：
Yanmin Luo;Minghong Liao;Jing Yan;Caiyun Zhang;S. Shang
中科院分区：
文献类型：
--
作者：
Yanmin Luo;Minghong Liao;Jing Yan;Caiyun Zhang;S. Shang
红树林是沿海生态系统的宝贵贡献者;在全球变化的背景下，了解红树林生态系统的动态十分重要。为了获得这方面的知识，遥感是一个不可或缺的手段，但它也带来了挑战，因为用传统的分类方法将红树林与其他土地覆盖类型区分开来的准确性有时不能令人满意。本文提出了一种改进的人工免疫算法（AIA），其中抗体代表候选解，抗原由适应度函数表示。多类协同进化与克隆选择的概念相结合，以确保并行计算的最佳聚类中心的每一个土地覆盖类型。采用面向聚类中心的抗体十进制编码方法，将类内方差和类间差异共同作为适应度函数。设计了基于抗体溶解度的选择算子和非均匀变异算子。应用此改进的AIA在中国东南部的张江河口的Landsat Thematic Mapper多光谱遥感图像，我们发现，AIA大大提高了分类精度比传统方法，表现出90%的整体精度（kappa系数= 0.88），并能够识别红树林以及（佣金10%，遗漏22%）。
Mangroves are valuable contributors to coastal ecosystems; knowledge of the dynamics of mangrove ecosystems is important in the context of global change. To obtain this knowledge, remote sensing is an indispensable means, yet it poses challenges since the accuracy is sometimes unsatisfactory in distinguishing mangroves from other land cover types with traditional classification methods. In this letter, we proposed a modified artificial immune algorithm (AIA), in which the antibodies represent the candidate solutions and the antigens are expressed by the fitness function. Multiclass coevolution was combined with the concept of clonal selection to ensure computation of an optimal clustering center in parallel for each land cover type. A cluster-center-oriented decimal encoding method for antibodies was adopted, and the inner class variance and the between-class difference together were used to formulate the fitness function. Furthermore, a design of the antibody solubility-based selection operator and nonuniform mutation operator was undertaken. Applying this modified AIA to a Landsat Thematic Mapper multispectral remote sensing imagery in the Zhangjiang estuary in southeastern China, we found that the AIA substantially improved classification accuracy over traditional methods, showing an overall accuracy of 90% (kappa coefficient = 0.88) and was capable to discern mangrove well (commission of 10% and omission of 22%).