| 摘 要: 协同表示分类方法已经被越来越多地应用在高光谱图像分类中，但协同表示方法因重视稀疏性忽略局部性而不能充分地刻画高光谱图像特征，导致分类精度不高。针对这一问题，提出了空谱融合下局部判别嵌入核协同表示方法(LPKCRC)。首先，利用空谱特征学习模型对高光谱图像进行特征学习；其次，利用图嵌入矩阵提取数据局部几何结构和局部判别信息，并将其作为流行正则项引入CRC中，同时利用核的特性对高光谱数据进行核映射。实验结果证明，该算法在Indian Pines和Salinas两个高光谱数据集分类结果中都优于其他相应的算法，能够提高分类准确率。 |
| 关键词: 核协同表示；局部流形结构；空谱特征学习；高光谱图像；稀疏性 |
|
中图分类号: TP181
文献标识码: A
|
| 基金项目: 国家自然科学基金项目(61772250)；辽宁省自然科学基金(20170540574)；辽宁省教育厅科学研究项目(LJ2019014). |
|
| A Hyperspectral Image Classification Algorithm under Space Spectrum Fusion for Local Projections Embedding Kernel Collaborative Representation |
|
CAO Yichang, YAN Deqin, CHEN Lang, LIU Deshan
|
(College of Computer and Information Technology, Liaoning Normal University, Dalian 116081, China)
1352499417@qq.com; yandeqin@163.com; chenlangstudy@163.com; deshanliu@yeah.net
|
| Abstract: Collaborative representation classification methods have been increasingly applied to hyperspectral image classification. However, collaborative representation methods fail to describe characteristics of hyperspectral images as they emphasize sparsity and ignore locality, which leads to low classification accuracy. To solve this problem, this paper proposes a Locality Projections Kernel Collaborative Representation Classification (LPKCRC) method under space spectrum fusion. Firstly, learning model of space spectrum features is used to learn the feature of hyperspectral images; secondly, graph embedded matrix is used to extract local geometric structure and local discriminant information of the data, which is introduced into CRC (Collaborative Representation Classification) as a popular regular term. At the same time, kernel mapping is performed on hyperspectral data using the kernel characteristics. Experimental results prove that the proposed algorithm is superior to other corresponding algorithms in classification results of Indian Pines and Salinas hyperspectral data sets, and it can improve the classification accuracy. |
| Keywords: kernel collaborative representation; local manifold structure; space spectrum feature learning; hyperspectral image; sparsity |
