Spectral-spatial Joint Classification of Hyperspectral Image with Edge-preserving Filtering
-
摘要: 针对高光谱遥感影像分类过程中，高维数据引起的"维数灾难"以及空间邻域一致性信息没有得到充分利用的问题，提出一种基于边缘保持滤波（Edge-preserving filtering，EPF）的高光谱影像光谱-空间联合分类算法.该算法首先进行波段子集划分和主成分提取，构造新的低维特征集，在保存影像结构信息的前提下降低数据维度；其次利用支持向量机（Support vector machine，SVM）获得低维特征集的初始分类概率图；然后利用原始影像主成分对初始分类概率图进行边缘保持滤波，融合光谱信息和空间信息；最后根据滤波后分类概率图对应像素点值的大小确定每个像素的类别.在Indian Pines和Pavia University两组高光谱数据上进行仿真实验，相同实验条件下，本文算法都获得最高分类精度和最少的时间消耗.仿真结果表明本文算法在高光谱遥感影像分类任务中具有明显的优势.Abstract: To deal with the problem of "curse of dimensionality" caused by high dimension and the underutilization of spatial contexture information in classification of hyperspectral images, a new spectral-spatial joint classification method based on edge-preserving filtering is proposed. The proposed method consists of the following four steps. Firstly, the hyperspectral image is divided into several subsets of bands. By extracting the principal component of each subset, a new low-dimensional feature set is constructed. Secondly, the pre-classification result, which is obtained by support vector machines with the new feature set, is represented as multiple initial probabilistic maps. Then edge-preserving filtering is operated on each initial probabilistic map to merge the spectral and spatial information. Finally, the class of each pixel is determined by the maximum value of the corresponding filtered probabilistic maps. The proposed algorithm is examined by the Indian Pines and Pavia University hyperspectral datasets. On the same experimental conditions, the proposed method achieves the highest classification accuracy and the lowest time consumption, demonstrating obvious advantages in hyperspectral image classification.1) 本文责任编委 桑农
-
图 3 Indian Pines数据集波段子集$K$对分类精度的影响
Fig. 3 OA of the proposed methods with different numbers of band subsets $K$ in Indian Pines dataset
图 4 Indian Pines影像双边滤波器参数对分类精度的影响
Fig. 4 The influence of the parameters of Bilateral filter in Indian Pines image
图 5 Indian Pines影像引导滤波器参数对分类精度的影响
Fig. 5 The influence of the parameters of Guided filter in Indian Pines image
图 6 Indian Pines数据集不同方法分类效果图
Fig. 6 Classification maps for the Indian Pines dataset using different methods
图 7 数据集波段子集$K$对分类精度的影响
Fig. 7 OA of the proposed methods with different numbers of band subsets $K$ in Pavia University dataset
图 8 Pavia University数据集不同方法分类效果图
Fig. 8 Classification maps for the Pavia University dataset using different methods
表 1 Indian Pines数据集不同方法分类精度
Table 1 Classification accuracy for the Indian Pines dataset using different methods
类别 训练样本 测试样本 SVM (%) SVMCK (%) EPF-B-g (%) BFSVM-PC1 (%) BFSVM-PC3 (%) GFSVM-PC1 (%) GFSVM-PC3 (%) 1 8 38 73.68 75.00 90.91 ${\textbf{100.00}} $ ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ 2 143 1 285 77.37 89.31 93.17 ${\textbf{99.83}}$ 99.58 99.59 ${\textbf{99.83}}$ 3 83 747 78.13 88.81 ${\textbf{98.98}}$ 97.46 97.59 97.58 97.72 4 24 213 76.27 77.25 96.30 ${\textbf{98.98}}$ ${\textbf{98.98}}$ 98.51 98.53 5 48 435 92.30 94.85 99.02 98.82 98.82 ${\textbf{99.52}}$ 98.81 6 73 657 90.43 98.81 97.76 99.85 99.85 ${\textbf{100.00}}$ 99.85 7 8 20 88.88 93.75 ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ 8 48 430 97.48 98.18 ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ $ {\textbf{100.00}}$ ${\textbf{100.00}}$ 9 8 12 38.09 ${\textbf{100.00}} $ ${\textbf{100.00}}$ ${\textbf{100.00}}$ $ {\textbf{100.00}}$ $\textbf{100.00} $ ${\textbf{100.00}}$ 10 97 875 77.05 85.30 93.61 98.74 98.89 99.13 98.76 11 246 2 209 80.46 92.98 ${\textbf{95.38}}$ 94.47 94.13 94.01 93.73 12 59 534 79.34 87.16 91.58 96.32 96.15 ${\textbf{97.94}}$ $ {\textbf{97.94}}$ 13 21 184 87.80 97.91 99.46 $ {\textbf{100.00}}$ ${\textbf{100.00}}$ $ {\textbf{100.00}}$ ${\textbf{100.00}}$ 14 127 1 138 90.93 96.82 ${\textbf{97.76}}$ 97.14 97.15 95.90 97.57 15 39 347 74.89 73.10 94.74 96.65 96.35 96.20 ${\textbf{97.01}}$ 16 9 84 90.69 96.47 92.41 ${\textbf{98.67}}$ 98.51 98.48 98.59 OA (%) 82.87 91.28 95.84 97.42 97.33 97.29 ${\textbf{97.43}}$ AA (%) 80.86 90.36 96.32 98.56 98.50 98.55 ${\textbf{98.65}}$ $\kappa$ 系数 0.8038 0.9004 0.953 0.971 0.969 0.969 ${\textbf{0.971}}$ 时间(s) 159.42 210.44 159.47 34.00 33.25 ${\textbf{29.42}}$ 32.91 
下载: 导出CSV
表 2 Pavia University数据集不同方法分类精度
Table 2 Classification accuracy for the Pavia University dataset using different methods
类别 训练样本 测试样本 SVM (%) SVMCK (%) EPF-B-g (%) BFSVM-PC1 (%) BFSVM-PC3 (%) GFSVM-PC1 (%) GFSVM-PC3 (%) 1 265 6 366 92.91 96.97 ${\textbf{98.11}} $ 95.12 95.12 95.12 94.87 2 746 17 903 96.07 $ {\textbf{99.54}} $ 97.25 99.20 99.22 99.24 99.49 3 84 2 015 80.97 85.51 99.94 ${\textbf{100.00}}$ ${\textbf{100.00}}$ 99.83 ${\textbf{100.00}}$ 4 123 2 941 95.18 95.31 99.73 99.67 99.64 99.86 ${\textbf{100.00}}$ 5 54 1 291 98.09 99.85 ${\textbf{100.00}}$ ${\textbf{100.00}}$ $ {\textbf{100.00}}$ 99.92 99.92 6 201 4 828 88.72 96.21 ${\textbf{98.87}}$ 98.64 98.64 98.73 98.75 7 53 1 277 85.84 94.13 ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ 8 147 3 535 86.22 92.53 91.65 92.99 93.01 92.77 ${\textbf{93.61}}$ 9 38 909 ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ ${\textbf{100.00}}$ $\textbf{100.00}$ ${\textbf{100.00}}$ OA (%) 92.92 97.00 97.55 98.05 98.06 98.07 ${\textbf{98.23}}$ AA (%) 91.56 96.56 98.39 98.40 98.40 98.39 ${\textbf{98.52}}$ $\kappa$ 系数 0.9061 0.9602 0.967 0.974 0.974 0.974 ${\textbf{0.977}}$ 时间(s) 94.56 178.32 97.48 47.22 47.01 ${\textbf{28.98}}$ 47.37
下载: 导出CSV
-
[1] 潘宗序, 禹晶, 肖创柏, 孙卫东.基于光谱相似性的高光谱图像超分辨率算法.自动化学报, 2014, 40(12):2797-2807 http://www.aas.net.cn/CN/abstract/abstract18558.shtmlPan Zong-Xu, Yu Jing, Xiao Chuang-Bai, Sun Wei-Dong. Spectral similarity-based super resolution for hyperspectral images. Acta Automatica Sinica, 2014, 40 (12):2797-2807 http://www.aas.net.cn/CN/abstract/abstract18558.shtml [2] Du B, Zhang L P. Target detection based on a dynamic subspace. Pattern Recognition, 2014, 47 (1):344-358 doi: 10.1016/j.patcog.2013.07.005 [3] Plaza A, Benediktsson J A, Boardman J W, Brazile J, Bruzzone L, Camps-Valls G, Chanussot J, Fauvel M, Gamba P, Gualtieri A, Marconcini M, Tilton J C, Trianni G. Recent advances in techniques for hyperspectral image processing. Remote Sensing of Environment, 2009, 113 (S1):S110-S122 [4] Jia X P, Kuo B C, Crawford M M. Feature mining for hyperspectral image classification. Proceedings of the IEEE, 2013, 101 (3):676-697 doi: 10.1109/JPROC.2012.2229082 [5] Han M, Liu B. Ensemble of extreme learning machine for remote sensing image classification. Neurocomputing, 2015, 149:65-70 doi: 10.1016/j.neucom.2013.09.070 [6] Ramzi P, Samadzadegan F, Reinartz P. Classification of hyperspectral data using an AdaBoostSVM technique applied on band clusters. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014, 7 (6):2066-2079 doi: 10.1109/JSTARS.2013.2292901 [7] 倪鼎, 马洪兵.基于近邻协同的高光谱图像谱-空联合分类.自动化学报, 2015, 41 (2):273-284 http://www.aas.net.cn/CN/abstract/abstract18606.shtmlNi Ding, Ma Hong-Bing. Spectral-spatial classification of hyperspectral images based on neighborhood collaboration. Acta Automatica Sinica, 2015, 41 (2):273-284 http://www.aas.net.cn/CN/abstract/abstract18606.shtml [8] Zhang Z, Pasolli E, Crawford M M, Tilton J C. An active learning framework for hyperspectral image classification using hierarchical segmentation. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2016, 9 (2):640-654 doi: 10.1109/JSTARS.2015.2493887 [9] Fauvel M, Tarabalka Y, Benediktsson J A, Chanussot J, Tilton J C. Advances in spectral-spatial classification of hyperspectral images. Proceedings of the IEEE, 2013, 101 (3):652-675 doi: 10.1109/JPROC.2012.2197589 [10] Ji R R, Gao Y, Hong R C, Liu Q, Tao D C, Li X L. Spectral-spatial constraint hyperspectral image classification. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52 (3):1811-1824 doi: 10.1109/TGRS.2013.2255297 [11] Camps-Valls G, Gomez-Chova L, Munoz-Marí J, Vila-Francés J, Calpe-Maravilla J. Composite kernels for hyperspectral image classification. IEEE Geoscience and Remote Sensing Letters, 2006, 3 (1):93-97 doi: 10.1109/LGRS.2005.857031 [12] Benediktsson J A, Palmason J A, Sveinsson J R. Classification of hyperspectral data from urban areas based on extended morphological profiles. IEEE Transactions on Geoscience and Remote Sensing, 2005, 43 (3):480-491 doi: 10.1109/TGRS.2004.842478 [13] Tarabalka Y, Chanussot J, Benediktsson J A. Segmentation and classification of hyperspectral images using watershed transformation. Pattern Recognition, 2010, 43 (7):2367-2379 doi: 10.1016/j.patcog.2010.01.016 [14] Li X R, Pan J, He Y Q, Liu C S. Bilateral filtering inspired locality preserving projections for hyperspectral images. Neurocomputing, 2015, 164:300-306 doi: 10.1016/j.neucom.2015.01.021 [15] Kang X D, Li S T, Benediktsson J A. Spectral-spatial hyperspectral image classification with edge-preserving filtering. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52 (5):2666-2677 doi: 10.1109/TGRS.2013.2264508 -
计量
- 文章访问数: 2212
- HTML全文浏览量: 345
- PDF下载量: 715
- 被引次数: 0
