基于自适应矩阵的核联合稀疏表示高光谱图像分类  

Kernel Joint Sparse Representation Hyperspectral Image Classification Based on Adaptive Matrix

作　　者：陈善学[1] 夏馨 CHEN Shanxue;XIA Xin(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《遥感信息》2024年第2期19-27,共9页Remote Sensing Information

摘　　要：针对高光谱图像丰富的空间信息和光谱信息未充分利用的问题,提出了基于自适应矩阵的核联合稀疏表示高光谱图像分类的方法。在特征表示阶段,定义了自适应矩阵特征,通过结合自适应邻域块策略与非线性相关熵度量构成的特征来描述原始光谱像素,充分融合了形状可变的空间信息与非线性光谱信息。在分类阶段,考虑自适应矩阵和高光谱图像非线性,采用对数欧式核函数,构建了核联合稀疏表示模型,以获得重构误差。同时利用字典空间信息构建了矩阵相关性,引入平衡参数实现了稀疏重构误差与矩阵相关性的联合分类。在两个数据集上的实验结果表明,该算法充分利用了高光谱图像的空间信息、光谱信息,能够有效提高分类精度。Aiming at the problem of insufficient utilization of rich spatial information and spectral information in hyperspectral images,a kernel joint sparse representation method based on adaptive matrices is proposed for hyperspectral image classification.In the feature representation stage,adaptive matrix features are defined to describe the original spectral pixels by combining the feature of adaptive neighborhood block strategy with a nonlinear correntropy measurement,which fully integrates shape-variable spatial information and nonlinear spectral information.In the classification stage,considering the adaptive matrix and the nonlinearity of hyperspectral images,a kernel joint sparse representation model is constructed using a logarithmic Euclidean kernel function to obtain reconstruction errors.Meanwhile,matrix correlation is constructed using dictionary space information,and a balanced parameter is introduced to achieve joint classification of sparse reconstruction error and matrix correlation.Experimental results on two datasets demonstrate that the proposed algorithm fully utilizes the spatial and spectral information of hyperspectral images,and can effectively improve classification accuracy.

关 键 词：高光谱图像分类 核联合稀疏表示 自适应邻域块 自适应矩阵 矩阵相关性 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]