首页 > , Vol. , Issue () : -
近年来，低秩表示(Low-rank Representation, LRR)在高光谱遥感影像分类中的应用越来越广泛，如何利用LRR准确地对地物进行分类已成为高光谱遥感研究的一个挑战。针对以上问题，本文设计了基于弹性网络的低秩表示(Low-rank Repre-sentation based on Elastic Net, ENLRR)方法，并将该算法扩展到了核空间提出了基于弹性网络的核低秩表示方法(Kernel ENLRR, KENLRR)。低秩表示分类可以充分利用影像的全局信息，它的基本思想是利用尽可能少的训练样本的线性组合来表示整个测试影像，再通过表示系数和训练样本对目标影像进行恢复重建，并根据最小重构误差准则判断每个像素的类别。提出的ENLRR方法的基本思想是在LRR中引入弹性网络思想，利用系数矩阵的核范数和F-范数代替秩函数进行低秩优化求解。为了更好地解决非线性数据的分类问题，在ENLRR 方法中引入核函数，提出KENLRR方法，通过邻域滤波核函数将原数据映射到高维特征空间中，实现空谱联合分类，进一步提高分类精度。实验部分选用三组高光谱遥感数据，利用提出的算法与SVM、KNN、ELM、LRR、MFLRR、LSLRR和KLRR等对比算法进行地物分类。结果表明，提出的两种算法在高光谱遥感地物分类方面效果较好，而且具有良好的稳定性和适应性。与LRR算法相比，提出的算法在Washington DC数据集上的精度分别提高了4.55%和6.74%，在Purdue Campus数据集上的精度分别提高了14.22%和23.30%，在高分五号(Gaofen-5, GF-5)黄河口湿地数据集上的精度分别提高了8.45%和15.40%，而且结果也表明KENLRR算法具有最佳的分类表现。精确的分类结果为分析地物分布格局提供了技术支撑，也证明了在本文提出的两种算法在高光谱遥感影像分类上的优越性。
Objective: Recently, low-rank representation (LRR) has attracted much attention in hyperspectral re-mote sensing imagery classification. However, nuclear norm is used to replace rank function of the co-efficient matrix in much LRR researches. There is a certain deviation between rank and nuclear norm which affect the classification accuracy to a certain extent. In addition, the performance of LRR method will be limited when classifying nonlinear data. To solve the above problem and achieve higher accu-racy, LRR based on elastic net (ENLRR) and the extended kernel version of ENLRR (KENLRR) are proposed for hyperspectral imagery classification. Method: The basic LRR method can represent the whole test image by as few training samples as pos-sible, then reconstruct the target image according to the representation coefficient matrix and training samples, and calculate the class of each pixel by the minimum reconstruction error criterion. To better solve the low-rank optimization problem, ENLRR method is proposed. The main idea of ENLRR is to introduce elastic net into LRR model, which replaces the rank function with the combination of nuclear norm and Frobenius norm of the coefficient matrix. To better classify nonlinear data, modified KENLRR method is proposed by introducing kernel tricks in ENLRR algorithm, and the neighborhood filter kernel function is adopted to map the original data into a high-dimensional feature space, which can obtain spatial-spectral joint information for better classification. Results: In the experiments, three popular hyperspectral datasets are adopted, and the proposed meth-ods and SVM, KNN, ELM, LRR, MFLRR, LSLRR and KLRR comparison methods are used to conducted classification. From the experimental results, the proposed methods are effective in accurately distin-guish ground objects and have well stability and adaptability. Compared with LRR method, the overall classification accuracies of ENLRR and KENLRR are respectively improved by 4.55% and 6.74% in Washington DC dataset, 14.22% and 23.30% in Purdue Campus dataset, and 8.45% and 15.40% in Gaofen-5 (GF-5) Yellow River Delta dataset. And it is demonstrated that KENLRR method can provide the best performance for hyperspectral remote sensing imagery classification. Conclusion: It is a vital research direction to achieve higher accuracy in hyperspectral imagery classi-fication. The performance of ENLRR method by introducing elastic net into LRR is acceptable, and the result of ENLRR is more robust and efficient. Furthermore, the application of kernel tricks can improve the accuracy greatly. KENLRR method can better classify nonlinear data and acquire higher accuracy. The high-quality classification results provide technical support for analyzing the distribution pattern of ground objects. And the higher accuracies demonstrate the effect of introducing elastic net and ker-nel tricks in LRR method, and prove the superiority of proposed methods in hyperspectral remote sensing imagery classification.