基于多态蚁群算法的高光谱遥感影像最优波段选择  

Optimized Band Selection of Hyperspectral Remote Sensing Image based on Polymorphic Ant Colony Algorithm

作　　者：丁小辉[1,2] 李华朋[1] 张树清[1] 

机构地区：[1]中国科学院东北地理与农业生态研究所,吉林长春130102 [2]中国科学院大学,北京100049

出　　处：《遥感技术与应用》2016年第2期275-284,共10页Remote Sensing Technology and Application

基　　金：国家重大专项(21-Y30B05-9001-13/15-2);国家自然科学基金项目(41271196;41301465)

摘　　要：由于传统蚁群算法搜索空间大,算法时间复杂度高等,导致基于传统蚁群算法的高光谱数据波段选择算法(ACA-BS)耗时长,算法效率低下,且易陷入局部最优。而多态蚁群算法能大大缩小算法的搜索空间,降低算法时间复杂度。因此,研究设计了基于多态蚁群算法的高光谱数据波段选择算法(PACA-BS)。从算法运行时间、波段子集的类别可分性及信息量、总体分类精度等方面对算法进行对比分析。用于实验的数据为Hyperion和AVIRIS高光谱影像。实验结果表明:PACA-BS的运行时间较ACA-BS大大减少;对Hyperion影像进行降维时,基于PACA-BS的运行时间约为ACA-BS的一半。两种算法获得的波段子集的类别可分性大小较为接近,但PACA-BS获得的波段子集的信息量和总体分类精度优于ACA-BS。研究表明PACA-BS是一种效率较高的高光谱波段选择算法。Due to the large searching space,the time complexity of traditional Ant Colony Algorithm （ACA） is very high.Thus, the ACA-based band selection algorithms require a long time to run,and always suffer from local opti- ma.In comparison,the Polymorphic Ant Colony Algorithm （PACA） can significantly decrease the search- ing space and thus the time complexity.In considerations of this, this paper designed a PACA-based band selection algorithm （PACA-BS） for hyperspectral remote sensing imagery.Performance evaluation of algo- rithms was focused on the following aspects：computing time, separability of band sets, information amount and overall accuracy. Herein, Hyperion and AVRIS imagery were employed as data source. The results showed that the computing time of PACA-BS was markedly lower than ACA BS.Furthermore, band sets derived by both algorithms possess similar separability, however, the band sets of PACA-BS have larger in- formation amount,and thus generate a higher overall classification accuracy.The PACA-BS is thus proved to be a promising and optimized method for band selection of hyperspectral image.

关 键 词：高光谱 降维 波段选择 蚁群算法 多态蚁群算法 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]