基于复合核的相关向量机高光谱图像分类

发布时间：2018-05-27 03:43

本文选题：高光谱图像分类 + 复合核　；参考：《湖北大学》2014年硕士论文

【摘要】：高光谱遥感是将成像技术和细分光谱技术结合在一起的多维信息获取技术,伴随着成像光谱技术的迅速发展,高光谱图像分类识别技术日渐完善,在各种应用领域得到了广泛运用和发展.传统的常用分类算法最大似然估计和神经网络在使用维数高、波段之间相关性大的高光谱数据时则会遭遇维数灾难现象.支持向量机是解决维数灾难问题的核方法之一,与支持向量机相对应的相关向量机,却只需要更少的相关向量和更快的测试时间,就可得到与支持向量机相接近的准确率,相关向量机用于高光谱图像分类,有着自身的优势和不足. 作为另一种核方法的相关向量机,很好的解决了维数灾难问题.基于相关向量机分类准确率不高的事实,考虑到高光谱数据本身具有空间信息等结构特点,为了弥补这一不足,本文以相关向量机为基本模型展开研究,将高光谱图像的空间信息与光谱信息融合在一起,并提出了一种新的基于复合核的相关向量机高光谱图像分类算法.本文主要研究工作如下： (1)将相关向量机理论应用到高光谱分类图像中,提出了改进型相关向量机分类算法,在真实的高光谱数据上进行模拟实验,并与传统的支持向量机算法进行较为详细的比较. (2)提出了基于复合核的相关向量机高光谱图像分类算法.该算法使用复合核,融合了空间特征与谱信息,既顾及到起分类主作用的谱特征,又利用了高光谱数据的空间差异性.考虑到权和核中谱信息和空间信息两者之间所占比值的不同,继而提出了广义的复合核,避免了平衡参数的调节,同时探讨了不同窗口大小对各类分类器的分类精度的影响,从而保证了实验的可靠性和高效性. (3)本文提出的基于复合核的办法能较好利用高光谱图像数据的空间信息,在真实的高光谱数据实验上也验证了所提出的算法的有效性.
[Abstract]:Hyperspectral remote sensing is a multi - dimensional information acquisition technology combining imaging technology with subdivision spectrum technology . With the rapid development of imaging spectrum technology , high spectral image classification and recognition technology has been widely used and developed . Support vector machine is one of the kernel methods to solve the problem of dimension disaster . Support vector machine is one of the kernel methods to solve the problem of dimension disaster .

Based on the fact that the classification accuracy of the correlation vector machine is not high , the spatial information and the spectral information of the hyperspectral image are fused together in consideration of the fact that the high spectral data itself has spatial information and so on , and a new high spectral image classification algorithm based on the composite core is proposed . The main research work is as follows :

( 1 ) The correlation vector machine theory is applied to hyperspectral classification image , and the improved correlation vector machine classification algorithm is proposed , and the simulation experiment is carried out on the real hyperspectral data , and compared with the traditional support vector machine algorithm .

( 2 ) The high spectral image classification algorithm based on the complex kernel is proposed . The algorithm uses the complex kernel , combines the space feature and the spectral information , takes into account the spectral characteristics of the main function of the classification , and uses the spatial difference of the hyperspectral data . Considering the difference between the spectral information and the spatial information in the kernel , the generalized composite core is proposed , the influence of different window sizes on the classification accuracy of various classifiers is discussed , and the reliability and efficiency of the experiment are ensured .

( 3 ) The method based on the complex kernel presented in this paper can better utilize the spatial information of hyperspectral image data , and verify the effectiveness of proposed algorithm in real high spectral data experiment .
【学位授予单位】：湖北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【相似文献】