Spectral-spatial classification for hyperspectral imagery： a novel combination method based on affinity scoring  

作　　者：Zhao CHEN Bin WANG 

机构地区：[1]State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing 100875, China [2]Key Laboratory for Information Science of Electromagnetic Waves (Ministry of Education), Fudan University, Shanghai 200433, China [3]Research Center of Smart Networks and Systems, School of Information Science and Technology, Fudan University, Shanghai 200433, China

出　　处：《Science China(Information Sciences)》2016年第10期177-189,共13页中国科学（信息科学）（英文版）

基　　金：supported in part by National Natural Science Foundation of China (Grant No. 61572133);Research Fund for State Key Laboratory of Earth Surface Processes and Resource Ecology (Grant No. 2015-KF-01)

摘　　要：Recently, a general framework for spectral-spatial classification has caught the attention of the hyperspectral imagery （HSI） society. It consists of three parts： classification, segmentation and combination of the former results to make a refined labeled map. Seeing the potentials of the last part, we derive a novel combination rule based on affinity scoring （CRAS）. The core of the system is affinity score （AS）, which is derived from fuzzy logic. Every AS measures the degree, i.e., the affinity, by which a pixel belongs to a class. The score is essentially decided by three factors： local spatial consistency, spectral similarity, and prior knowledge. The method is compatible with basic classification and segmentation tools, thus saving the trouble of designing complex techniques for the other parts in the framework. Experimental results show that CRAS excels several basic techniques as well as various state-of-the-art methods in the area of spectral-spatial classification.Recently, a general framework for spectral-spatial classification has caught the attention of the hyperspectral imagery （HSI） society. It consists of three parts： classification, segmentation and combination of the former results to make a refined labeled map. Seeing the potentials of the last part, we derive a novel combination rule based on affinity scoring （CRAS）. The core of the system is affinity score （AS）, which is derived from fuzzy logic. Every AS measures the degree, i.e., the affinity, by which a pixel belongs to a class. The score is essentially decided by three factors： local spatial consistency, spectral similarity, and prior knowledge. The method is compatible with basic classification and segmentation tools, thus saving the trouble of designing complex techniques for the other parts in the framework. Experimental results show that CRAS excels several basic techniques as well as various state-of-the-art methods in the area of spectral-spatial classification.

关 键 词：hyperspectral imagery spectral-spatial classification affinity score local spatial consistency fuzzy superpixel 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置] TP311.131[自动化与计算机技术—控制科学与工程]