【作者】 王晓飞； 何宏昌； 史翔宇； 阎秋静； 陈向南；

【Author】 WANG Xiao-fei;HE Hong-chang;SI Xiang-yu;Yan Qiu-jing;CHEN Xiang-nan;Key Laboratory of Electronic Enginerring, College of Heilongjiang Province, Heilongjiang University;

【机构】 黑龙江大学黑龙江省普通高等学校电子工程重点实验室；

【摘要】 为了实现对高光谱图像中的目标自动检测,提出了一种基于空间上下文单类分类器的目标检测算法。对该算法所采用的空间与光谱结合的特征、SVDD分类器原理、本文算法流程等进行研究。首先,分析了支持向量数据描述(SVDD,support vector data description)的单类分类原理。接着,结合高光谱图像特点,介绍了如何利用空间上下文信息和光谱特征作为SVDD分类器输入特征。然后,在分析比较空间光谱结合单类分类器性能的基础上,说明了采用该算法原理。最后,给出了该算法的具体实现方法。实验结果表明:该方法优于常规的直接利用光谱信息的CEM等算法,在AVIRIS成像的某国外海军基地数据上,检测飞机目标的精度达到了90%以上。基本满足目标检测的稳定可靠、低虚警率、高识别率等要求。更多还原

【Abstract】 Objetive: purpose: In order to implement the automatic target detection in hyperspectral image, this paper proposes a target detection algorithm based on spatial–contextual one classification. Features of combining space and spectrum used for the algorithm, principles of SVDD classifier and the algorithm process are studied. Method: First, the single class classification principle of support vector data description(SVDD)is analyzed. Second, Combining with the characteristics of hyperspectral image, this paper introduces how to use spatial and spectral features as the SVDD classifier input. Then, the principle of the algorithm is explained by comparing and analyzing the single class classifier performance combined space and spectrum. Last, The concrete realization method of the algorithm is given. Result: Experimental results show that, this method is superior to the conventional CEM algorithm, a foreign naval base’ data in AVIRIS imaging, the accuracy that detects the aircraft target is more than 90%. Conclusion: stable and reliable, low false alarm rate, high recognition rate of the target detection are met.更多还原