基于邻域相似性的空谱联合稀疏表示的高光谱图像分类

Hyperspectral Image Classification Based on Neighborhood Similarity and Spatial Spectral Joint Sparse Representation

摘要: 传统的稀疏表示分类方法仅考虑图像数据的稀疏特性，并未利用邻域像元间的相似性与独特性，因此提出一种基于邻域相似性的空谱联合稀疏表示的分类方法来提高高光谱图像分类精度。该方法将像元间的稀疏特性和邻域信息结合起来，利用像元间的空间距离权重与光谱距离权重度量待测中心像元Y与邻域像元的相似性，即计算邻域权重，设定相似度阈值，选取与像元Y相似度高的像元从而得到最优邻域窗口，最后通过联合稀疏表示来确定像元Y的类别。实验结果表明，该方法能够有效提高分类精度，且在不同实验数据下具备良好的稳定性。

Abstract: The traditional sparse representation classification method only takes into account the sparsity of image data, and does not make use of the similarity and uniqueness between neighboring pixels. Therefore, a new method based on neighborhood similarity and spatial spectral joint sparse representation is proposed to improve the classification accuracy of hyperspectral image. The method combines pixel sparse features with neighborhood information, and uses the weight of spatial distance and spectral distance weight between pixels to measure the similarity between the center pixel Y and neighboring pixels, namely, calculate neighborhood weights, set similarity thresholds, select pixels with high similarity to pixel Y, and get the optimal neighborhood window. Finally, a class of pixel Y is determined by joint sparse representation. Experimental results show that this method can effectively improve classification accuracy and has good stability under different experimental data.