面向对象的多尺度加权联合稀疏表示的高空间分辨率遥感影像分类

Abstract

针对高空间分辨率遥感影像中的地物具有多尺度特性，以及各个尺度的对象特征对地物分类精度的影响具有较强的尺度效性，并结合面向对象影像分析方法和多尺度联合稀疏表示方法在高空间分辨率遥感影像分类中的各自优点，提出了一种面向对象的多尺度加权稀疏表示的高空间分辨率遥感影像分类算法。首先，采用多尺度分割算法获得多尺度分割结果并提取对象的多尺度特征；然后，根据影像对象的多尺度分割质量测度计算各尺度的对象权重，构建面向对象的多尺度加权联合稀疏表示模型；最后，采用2个国产GF-2高空间分辨率遥感数据集和1个高光谱-高空间分辨率航空遥感数据集（WashingtonD.C.数据）验证该算法的有效性。试验结果表明，与SVM、像素级稀疏表示、单尺度和多尺度对象级稀疏表示和深度学习等算法相比较，本文算法获得了较高的OA和Kappa分类精度，提高了各个尺度地物的分类精度，有效抑止了地物分类结果中的椒盐噪声现象，同时保持大尺度地物的区域性和小尺度地物的细节信息。

Alternate abstract:

In this paper, according to the multi-scale advantage for high spatial resolution remote sensing imagery and the influence difference among multi-scale objects for classification, the objected-oriented multi-scale weighted sparse representation classification algorithm is proposed by taking the advantages of object-based image analysis method and sparse representation classification algorithm. Firstly, the multi-scale segmentation results are obtained and the multi-scale features are extracted by the multi-scale segmentation algorithm; secondly, the object weights in each scale are computed according to multi-scale segmentation quality measure, and the objected-oriented multi-scale weighted sparse representation model is constructed; finally, the two domestic GF-2 high spatial resolution remote sensing images and one high-spatial and spectral resolution dataset (Washington D.C. data) were adopted to verify the proposed algorithm. The experiment results show that the proposed algorithm can obtain the highest classification accuracy with OA and Kappa,efficiently improve classification accuracy at each scale objects, reduce salt and pepper noise in the classification results, and respectively maintain the regional integrity in the large scale objects and the details in the small scale objects comparing with the traditional SVM, pixel sparse representation,single scale and multi-scale sparse representation and object-based deep learning methods.