摘要
为了进一步提高支持向量机方法在超光谱图像分类中的性能,提出一种自适应加权核方法。该方法的关键是每个波段自适应权值的计算,考虑到超谱数据信息依波段分布不均匀及每个波段图像所含信息不同的特性,采用相邻波段图像间的相关系数及波段图像的归一化标准差之和作为该波段数据的权值,并给出了算法的具体实现步骤。实验结果表明:自适应加权核方法明显优于支持向量机方法,平均精度和总体精度分别提高了2.07%和2.28%,且对支持向量数目也有一定约减。
In order to improve the ability of the support vector machines for hyperspectral image classification, a kernel method based on adaptive weight coefficient was proposed. The key of the method was how to calculate the adaptive weight in each waveband. Taking into account both the non-uniform information distribution and the information difference in each band image, the sum of the correlation coefficients between adjacent bands and the normalized standard deviation of the band image were used as the weight of the band considered, q^e steps of algorithm were given in detail. The experimental results show that the proposed method is superior to the support vector machines, with the average accuracy and the overall accuracy in hyperspectral image classification increased by 2.07% and 2.28% respectively, and the numbers of the support vector reduced to some extent.
出处
《红外与激光工程》
EI
CSCD
北大核心
2011年第12期2535-2539,共5页
Infrared and Laser Engineering
基金
国家自然科学基金(60975009)
关键词
超光谱图像分类
核方法
自适应
相关系数
hyperspectral image classification
kernel method
adaptive
correlation coefficeints
