结合L0平滑和超像素的高空间遥感影像非监督分类

杨泽楠; 黄亮; 王枭轩; 方留杨; 宋晶

doi:DOI:10.13442/j.gnss.1008-9268.2019.04.005

结合L0平滑和超像素的高空间遥感影像非监督分类

doi: DOI:10.13442/j.gnss.1008-9268.2019.04.005

1.
昆明理工大学国土资源工程学院,云南昆明 650093
2.
云南省高校高原山区空间信息测绘技术应用工程研究中心,云南昆明 650093
3.
云南省交通规划设计研究院有限公司,云南昆明 650093

详细信息

作者简介:
黄亮 (1985-),男,博士，主要研究方向:遥感影像变化检测.

通信作者:
黄亮 E-mail:kmhuangliang@163.com.

计量
- 文章访问数: 367
- HTML全文浏览量: 53
- PDF下载量: 82
- 被引次数: 0
出版历程
- 刊出日期: 2019-08-15

Unsupervised classification of high spatial remote sensing image combining L0 smoothing and superpixel

1.
Faculty of Land Resources Engineering, Kunming University of Science and Technology, Kunming 650093, China
2.
Surveying and Mapping Geo Informatics Technology Research Center on Plateau Mountains of Yunnan Higher Education, Kunming 650093, China
3.
Broad Vision Engineering Consultants, Kunming 650093, China

摘要

摘要: 针对基于像元的非监督分类方法对高空间遥感影像分类时易形成“椒盐”噪声和产生大量错分、漏分的问题,提出了一种结合L0平滑和超像素的非监督分类方法．首先采用L0算法对高空间遥感影像进行平滑操作,减少大量图像噪声及冗余信息;然后采用简单的线性迭代聚类（SLIC）超像素方法处理平滑后图像,进一步抑制椒盐现象的同时降低处理复杂度,得到初始聚类图;最后采用K-means非监督分类方法得到最终分类结果图．为验证本文提出的方法,选取3景高空间遥感影像作为实验数据．试验结果表明,采用提出的方法能准确对地物分类,且总体精度分别达到了72.46％、77.55％和78.44％,Kappa系数分别达到0.788、0.779和0.779．提出方法能有效解决分类中存在的“椒盐”现象,可提高分类精度,对高空间遥感影像分类具有一定的参考价值．
- 高空间遥感影像 /
- L0平滑 /
- 超像素 /
- 非监督分类 /
- 分类精度
Abstract: Aiming at the problem that the unsupervised classification method is easy to form “salt and pepper” noise and generate many errors and missed points in the classification of high spatial remote sensing images, an unsupervised classification method of high spatial remote sensing image based on L0 smoothing and superpixel is proposed. Firstly, a L0 algorithm is instituted to smooth the high space remote sensing image and reduce a range of image noises and redundant information. Then, a superpixel method of SLIC (Simple Linear Iterative Clustering) is used for that further inhibiting the salt and pepper phenomenon while reducing the processing complexity, and the initial clustering map is obtained. Finally, the K-means unsupervised classification method is established to receive the final classification result image. Furthermore, three high spatial remote sensing images are selected as experimental data to verify the method proposed in this paper. The experimental results show that the proposed method can achieve accurate classification of features, and the overall accuracy is 72.46％, 77.55％ and 78.44％ respectively. The Kappa coefficients are 0.788, 0.779 and 0.779 respectively. The proposed method can effectively solve the phenomenon of “salt and pepper” in the classification, improve the classification accuracy and have certain reference value for high spatial remote sensing image classification.
- high spatial remote sensing image /
- L0 smoothing /
- superpixel /
- unsupervised classification /
- classification accuracy

HTML全文

参考文献(20)

[1]	黄昕,张良培,李平湘.基于多尺度特征融合和支持向量机的高分辨率遥感影像分类［J］.遥感学报,2007, 11(1): 48-54.
[2]	元晨.高空间分辨率遥感影像分类研究［D］.西安:长安大学, 2016.
[3]	余雄,张著洪.基于颜色特征的自适应图像分类算法及其应用［J］.贵州大学学报(自然科学版),2017,34(1): 62-65.
[4]	刘娜娜,李景文,李宁.基于图论分割的多光谱图像非监督分类方法［J］.北京航空航天大学学报, 2009, 35(5):544-546,554.
[5]	HUANG J Z,NG M K,RONG H Q,et al. Automated variable weighting in K-means type clustering［J］.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(5):657-668.DOI: 10.1109/TPAMI.2005.95.
[6]	杨玉梅.基于信息熵改进的Kmeans动态聚类算法［J］.重庆邮电大学学报(自然科学版), 2016, 28(2): 254-259.
[7]	LI X,LIU H F.Greedy optimization for K-means-based consensus clustering［J］.Tsinghua Science and Technology, 2018,23(2):184-194. DOI: 10.26599/TST.2018.9010063.
[8]	熊霖,唐万梅.基于K-means++的多分类器选择分类研究［J］.重庆师范大学学报(自然科学版), 2018,35(6):88-96.
[9]	RENATO C A,MIRKIN B. Minkowski metric feature weighting and anomalous cluster initializing in Kmeans clustering［J］.Pattern Recognition,2012,45(3):1061-1075. DOI: 10.1016/j.patcog.2011.08.012.
[10]	杨杰,郎丰铠,李德仁.一种利用Cloude-Pottier分解和极化白化滤波的全极化SAR图像分类算法［J］. 武汉大学学报(信息科学版),2011, 36(1):104-107.
[11]	刘峰,王颖.基于多通道Gabor滤波器的纹理图像非监督分类［J］.遥感信息, 2009(5): 19-22.
[12]	ARSALAN G,ALI M.An unsupervised feature extraction method based on band correlation clustering for hyperspectral imageclassification using limited training samples［J］. Remote Sensing Letters, 2018. DOI: 10.1080/2150704X.2018.1500723.
[13]	朱腾,余洁,李小娟,等.基于超像素与Span-Pauli分解的SAR影像分类［J］.华中科技大学学报(自然科学版), 2015, 43(7):77-81.
[14]	CHEN Y,BRUZZONE L, ZHAO H S, et al.Superpixel-based unsupervised band selection for classification of hyperspectral images［J］. IEEE Transactions on Geoscience and Remote Sensing,2018：1-16. DOI：10.1109/TGRS.2008.2849443.
[15]	甘霞,朱福喜,冯浩.基于熵正则L0梯度最小化模型的图像平滑方法［J］.电视技术,2018, 42(6):17-23.
[16]	宋晶.高空间分辨率遥感影像语义分割方法研究［D］.昆明:昆明理工大学, 2018.
[17]	WU X,LIU X H,CHEN Y F,et al.A graph based superpixel generation algorithm［J］.Applied Intelligence,2018，48 (11):4485-4496. DOI: 10.1007/s10489-018-1223-1.
[18]	NGVYEN T.Optimal ground control points for geometric correction using genetic algorithm with global accuracy［J］. European Journal of Remote Sensing, 2015,48(1):101-120. DOI: 10.5721/EuJRSS20154807.
[19]	ULVCAN-ALTUNTAS K, ILHAN F. Enhancing biodegradability of textile wastewater by ozonation processes: optimization with response surface methodology［J］.Ozone: Science and Engineering, 2018, 40(6):465-472. DOI: 10.1080/01919512.2018.1474339.
[20]	杨泽楠,黄亮,王枭轩.结合DEM的面向对象高分三号SAR影像高原山区水系提取［J］.昆明理工大学学报(自然科学版), 2019, 44(1):39-46.