结合改进LBP和SRC的高光谱图像分类研究

Research on Hyperspectral Image Classification Combining Improved LBP and SRC

针对传统局部二值模型(local binary pattern,LBP)提取高光谱图像纹理特征信息量庞大的难题,提出一种基于对称旋转不变等价局部二值模型(symmetrical rotation invariant uniform LBP,SRIULBP)的高光谱图像特征提取方法,以缩减特征维度;针对稀疏表示分类(sparse representation classification,SRC)模型中稀疏字典冗余的缺陷,采用近邻思想,提出最近邻稀疏表示(nearest neighbor SRC,NNSRC)分类方法,实现高光谱图像的高效、高准确度分类。数据实验结合表明,SRIULBP能快速提取图像特征,提出的分类方法不仅在分类精度上优于其他稀疏表示分类算法,并且具有更强的时效性与泛化能力。

Aiming at the problem of the traditional local binary pattern(LBP)extracting the huge amount of hyperspectral image texture feature information,a hyperspectral image feature extraction method based on the symmetrical rotation invariant uniform LBP(SRIULBP)is proposed to reduce the feature dimension;aiming at the defect of sparse dictionary redundancy in the sparse representation classification(SRC)model,the nearest neighbor SRC(NNSRC)method is proposed by adopting the idea of nearest neighbors to achieve efficient and high accuracy classification of hyperspectral images.The combination of data experiments shows that SRIULBP can quickly extract image features.The proposed classification method is not only superior to other sparse representation classification algorithms in classification accuracy,but also has stronger timeliness and generalization capabilities.