论文部分内容阅读
高光谱图像具有光谱分辨率高、波段连续、数据量大、图谱合一等特点。然而较高的光谱分辨率会造成波段间相关性强,信息冗余多。所以如何从数百个高光谱波段中选出有利于识别或分类的波段组合成为了高光谱应用需要解决的问题。文章针对相邻波段间相关性较大的特点,提出一种改进的对波段相关矩阵进行全局搜索的子空间划分的波段选择方法。该方法克服了传统只利用相关向量对波段进行划分的缺陷,利用整个相关矩阵进行全局搜索划分,再在划分后的子空间内进行波段选择,从而降低了波段之间的相关性。文章最后使用上述方法对AVIRIS数据进行波段选择,并通过SVM方法对其进行地物分类,结果表明该方法较不进行子空间划分的波段选择方法有较高的分类精度。
Hyperspectral image has the characteristics of high spectral resolution, continuous waveband, large amount of data and one map. However, the higher spectral resolution will result in strong correlation between bands and more information redundancy. Therefore, how to select a combination of bands from hundreds of hyperspectral bands that are good for identification or classification has become a problem to be solved in hyperspectral applications. In this paper, aiming at the large correlation between adjacent bands, an improved band selection method for subspace partitioning is proposed to global search the band correlation matrix. This method overcomes the shortcomings of the traditional method that only divides the band by using the correlation vector, uses the entire correlation matrix to search for the global search, and then selects the band in the divided subspace to reduce the correlation between the bands. At the end of this paper, AVIRIS data is band-selected by the above method and classified by SVM. The results show that this method has higher classification accuracy than the band selection method without subspace partitioning.