Unsupervised remote sensing image classification using an artificial immune network
Unsupervised remote sensing image classification using an artificial immune network
复制标题
DOI：
10.1080/01431161.2010.502155
复制
发表时间：
2011-09
影响因子：
3.4
通讯作者：
Yanfei Zhong;Liangpei Zhang;Wei Gong
中科院分区：
文献类型：
--
作者：
Yanfei Zhong;Liangpei Zhang;Wei Gong
本文将基于人工免疫网络的计算智能方法--人工免疫网络模型应用于遥感图像处理，以提高其智能性。aiNet已被用于聚类、优化和数据分析。然而，由于aiNet算法本身的复杂性和遥感图像数据量大的特点，aiNet算法在遥感图像分类中的应用受到了很大的限制。提出了一种基于aiNet的无监督人工免疫网络遥感图像分类方法（RSUAIN）。该方法能够自适应地获取自定义的克隆率、变异率等参数，利用免疫算子和生物学特性，如克隆、变异和记忆算子，对记忆免疫网络进行进化，并利用遥感图像进行遥感图像聚类。对不同类型的图像进行了三个实验，以评估该算法的性能，并将其与其他传统的无监督分类算法，如k-means，ISODATA（Iterative Self-Organizing Data Anaysis Techniques Algorithm）和模糊k-means进行了比较。RSUAIN在三个实验中的表现优于传统算法，因此可能为无监督遥感图像分类提供了一种有效的选择。
In this article, the artificial immune network (aiNet) model, a computational intelligent approach based on artificial immune networks (AINs), is applied to remote sensing image processing to improve its intelligence. aiNet has been utilized for clustering, optimization, and data analysis. Nevertheless, due to the inherent complexity of the aiNet algorithm and the large volume of data in remote sensing imagery, the application of aiNet to remote sensing image classification has been rather limited. This article presents an unsupervised artificial immune network for remote sensing image classification (RSUAIN) based on aiNet. The proposed method can adaptively obtain some user-defined parameters, such as clone rate and mutation rate, and evolve the memorial immune network by immune operators and biological properties, such as clone, mutation and memory operators, using the remote sensing image for the task of remote sensing image clustering. Three experiments with different types of images were performed to evaluate the performance of the proposed algorithm and to compare it with other traditional unsupervised classification algorithms, for example, k-means, ISODATA (Iterative Self-organizing Data Anaysis Techniques Algorithm) and fuzzy k-means. RSUAIN was observed to outperform the traditional algorithms in the three experiments and hence potentially provides an effective option for unsupervised remote sensing image classification.