顾及边缘及权重的极限学习机高光谱图像分类

Hyperspectral image classification using extreme learning machine considering edge and weight

针对当前高光谱图像分类中缺少考虑地物分布边缘信息以及空间特征融合时对图像的适应权重问题,提出一种顾及边缘及权重融合的极限学习机高光谱图像分类方法。方法首先利用引导滤波提取顾及地物边缘信息的滤波特征,其次利用拓展形态学多属性剖面提取图像的结构信息,然后通过试错法实验确定面对不同图像时两种信息的融合权重,最后利用极限学习机完成分类。利用三组标准实验数据证明本文方法具有较好精度提升及运算时效性,在面向小训练样本时同样具有适用性。

Aiming at the lack of considering the edge information of ground object distribution and the adaptive weight of image in spatial feature fusion for hyperspectral image classification,a extreme learning machine hyperspectral image classification method considering edge and weight fusion was proposed.Firstly,guided filtering was applied to extract the filtering features which took into account the edge information of ground objects.Secondly,the structure information of images was extracted by extended morphological multi-attribute profile.Then,the fusion weights of two kinds of information were determined by trial and error experiment.Three groups of standard experimental data were explored to prove that the proposed method has good accuracy improvement and operation timeliness,and is also applicable to small training samples.