| 基于KL-ISOMAP的高光谱图像彩色可视化 |
| |
| 引用本文： | 王立国,吴飞.基于KL-ISOMAP的高光谱图像彩色可视化[J].南京气象学院学报,2018,10(1):63-71. |
| |
| 作者姓名： | 王立国 吴飞 |
| |
| 作者单位： | 哈尔滨工程大学 信息与通信工程学院, 哈尔滨, 150001,哈尔滨工程大学 信息与通信工程学院, 哈尔滨, 150001 |
| |
| 基金项目： | 国家自然科学基金（61675051） |
| |
| 摘 要： | 标志点等距映射（L-ISOMAP）作为一种降维方法，在高光谱图像可视化中极具潜力.针对L-ISOMAP算法标志点代表性不足以及计算代价较高的问题，提出了基于K-medoids聚类算法的改进型L-ISOMAP算法（KL-ISOMAP），进而形成可视化方法.该可视化方法由以下几部分组成：1）通过改进型K-medoids算法选择标志点；2）根据相似度剔除相似像元；3）实现剩余像元的非线性降维；4）完成降维结果的可视化.实验结果表明，KL-ISOMAP算法一方面有效地提高了标志点的内在结构代表性，进而取得了更好的可视化效果；另一方面可通过对相似阈值的设置，加快算法的运算速度.由此形成的可视化方法有着较好的视觉效果、距离保持特性以及像元可分性.
|
| 关 键 词： | 高光谱数据 彩色可视化 标志点选择 L-ISOMAP算法 NK-medoids算法 |
| 收稿时间： | 2017/10/26 0:00:00 |
KL-ISOMAP-based color visualization for hyperspectral imagery |
| |
| WANG Liguo and WU Fei.KL-ISOMAP-based color visualization for hyperspectral imagery[J].Journal of Nanjing Institute of Meteorology,2018,10(1):63-71. |
| |
| Authors: | WANG Liguo and WU Fei |
| |
| Institution: | College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001 and College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001 |
| |
| Abstract: | Landmark-Isometric mapping (L-ISOMAP),as a dimensionality reduction method,has great potential in hyperspectral imagery visualization.There are two problems in L-ISOMAP algorithm,i.e.,the computational cost is high and the landmarks lack of representation for hyperspectral imagery.In this case,an improved L-ISOMAP algorithm,named KL-ISOMAP,is proposed based on K-medoids clustering algorithm.The KL-ISOMAP algorithm consists of the following steps:1)Selecting the landmarks by the improved K-medoids algorithm;2) Removing the similar pixels according to the similarity;3) Implementing the non-linear dimensionality reduction of the rest pixels;4) Implementing visualization on the reduced dataset.Experimental results show that KL-ISOMAP algorithm can improve the intrinsic structure representation of the landmarks and therefore improve the visualization performance.Furthermore,the algorithm can be speeded up by setting the similarity threshold.The visualization method is reasonable,feasible and of good visual effect,and has good performance in terms of feature distance and class separability preserving for hyperspectral imagery. |
| |
| Keywords: | hyperspectral dataset color visualization landmark selection L-ISOMAP NK-medoids |
|
| 点击此处可从《南京气象学院学报》浏览原始摘要信息 |
| 点击此处可从《南京气象学院学报》下载免费的PDF全文 |
|
