基于空谱特征的核极端学习机高光谱遥感图像分类算法

Classification Algorithm of Hyperspectral Remote Sensing Image Based on Spatial Spectrum Feature of Kernel Extreme Learning Machine

摘要:

近年来,局部二值模式(Local Binary Patterns,LBP)由于其在空间特征提取方面具有显著的优势被应用于高光谱遥感图像分类中,该算法在空间特征提取上虽减少类内方差,却忽视了用于区分不同地物类别的光谱特征。为避免在图像分类过程中提取单一特征导致特征提取不充分、分类效果不理想的问题,通过将空间特征和光谱特征进行矢量堆叠得到新的空谱特征向量。再将新的空谱特征向量引入到核极端学习机中,提出一种基于空谱特征的核极端学习机高光谱遥感图像分类算法(Space Spectrum feature Kernel Extreme Learning Machine,SS-KELM)。为验证所提算法的有效性,将使用两个高光谱图像数据集进行实验。实验结果表明所提SS-KELM算法的分类性能优于目前较为常见的传统分类算法。

Abstract:

In recent years, local binary patterns (LBP) has been applied to hyperspectral remote sensing image classification because of its significant advantages in spatial feature extraction. Although the algorithm reduces the intra class variance in spatial feature extraction, it ignores the spectral features used to distinguish different ground objects. In order to avoid the problem of insufficient feature extraction and unsatisfactory classification effect caused by single feature extraction in the process of image classification, a new space spectrum feature vector is obtained by stacking spatial and spectral features. Then a new space spectrum feature vector is introduced into the kernel extreme learning machine, and a space spectrum feature kernel extreme learning machine (SS-KELM) is proposed. In order to verify the effectiveness of the proposed algorithm, two hyperspectral image data sets will be used for experiments. The experimental results show that the performance of the proposed SS-KELM algorithm is better than the traditional classification algorithm.