一种基于PCA的面向对象多尺度分割优化算法

蒋晨琛; 霍宏涛; 冯琦

doi:10.13700/j.bh.1001-5965.2019.0398

一种基于PCA的面向对象多尺度分割优化算法

doi: 10.13700/j.bh.1001-5965.2019.0398

蒋晨琛¹,
霍宏涛^1, ,,
冯琦^{2, 3}

1.
中国人民公安大学信息网络安全学院, 北京 100038
2.
中国人民公安大学公安遥感应用工程技术研究中心, 北京 100038
3.
中国人民公安大学公安军民融合中心, 北京 100038

基金项目:

国家重点研发计划 2017YFC0822405

中国人民公安大学基本科研业务费 2019JKF330

详细信息

作者简介:
蒋晨琛女, 硕士研究生。主要研究方向:遥感数据挖掘、遥感技术与应用

霍宏涛男, 博士, 教授, 博士生导师。主要研究方向:模式识别、图像处理、公安遥感应用

通讯作者:
霍宏涛, E-mail: huohongtao@ppsuc.edu.cn

中图分类号: TP753
计量
- 文章访问数: 806
- HTML全文浏览量: 88
- PDF下载量: 242
- 被引次数: 0
出版历程
- 收稿日期: 2019-07-19
- 录用日期: 2019-10-11
- 网络出版日期: 2020-06-20

An object-oriented multi-scale segmentation optimization algorithm based on PCA

JIANG Chenchen¹,
HUO Hongtao^{1
, ,},
FENG Qi^{2, 3}

1.
Institute of Information Cyber Security, People's Public Security University of China, Beijing 100038, China
2.
Remote Sensing Center of Public Security, People's Public Security University of China, Beijing 100038, China
3.
Civil-military Integration Center for Public Security, People's Public Security University of China, Beijing 100038, China

Funds:

National Key R & D Program of China 2017YFC0822405

Basic Scientific Research Operating Expenses of People's Public Security University of China 2019JKF330

More Information

Corresponding author: HUO Hongtao, E-mail: huohongtao@ppsuc.edu.cn

摘要

摘要:
多尺度分割是图像面向对象分类的基础，针对不同区域特征最优分割尺度确定的主观性以及采用聚类算法时聚类中心确定的随机性，提出了一种联合降维与聚类算法的面向对象多尺度分割优化算法。该算法首先利用主成分分析法（PCA）降维排序后的结果产生初始聚类中心；然后采用K-means聚类和度量每一个像素点合并的概率，从而得到适应不同研究区域内不同尺度地物的分割结果。采用多个影像数据库，通过引入聚类评价指标（内部评价指标和外部评价指标）、分割评价指标（分割精度、过分割率和欠分割率）并结合现有的图像分割方法及原始的K-means算法、与PCA降维后的K-means聚类对比分析。研究结果表明：经过降维处理后进行的聚类算法稳定性更高；与传统的聚类算法相比，结合PCA降维更能自动识别最优分割尺度；降维技术和聚类算法联合之中，目视和定量评价指标表明经过降维预处理后的聚类能得到更高质量的分割结果。
- 主成分分析法(PCA) /
- 聚类 /
- 面向对象 /
- 多尺度 /
- 图像分割
Abstract:
Multi-scale segmentation is the basis of remote sensing images object-oriented classification. The paper proposes an object-oriented multi-scale segmentation optimization algorithm which combines dimension reduction technique with clustering algorithm aiming at the subjectivity of optimal segmentation scale determination of different regional features and the randomness of clustering center determined when using clustering algorithms. In this method, the initial clustering center is generated using the result of dimension reduction and sorting by Principal Component Analysis (PCA). Then the probability of merging each pixel is calculated by K-means clustering algorithm, so as to obtain the multi-scale segmentation results suitable for different scales in different research areas. This paper comparatively analyzes, in combination with the existing image segmentation methods and the original K-means algorithm, the K-means clustering segmentation after PCA dimension reduction, using multiple image databases, through a series of clustering evaluation indicators (internal and external evaluation indicators) and segmentation evaluation indicators (segmentation accuracy, over-segmentation rate and under-segmentation rate) to evaluate the result of different methods. The results show as follows: first, the method of the clustering algorithm after dimension reduction is more stable than the original clustering algorithm; second, compared with the traditional clustering algorithm, the PCA dimension reduction can identify the optimal segmentation scale more automatically; third, in the combination of dimension reduction technology and clustering algorithm, visual and quantitative evaluation indexes show that the clustering after dimension reduction preprocessing can get higher-quality segmentation results.
- Principal Component Analysis (PCA) /
- clustering /
- object-oriented /
- multi-scale /
- image segmentation

HTML全文

图 1 本文技术路线

Figure 1. Technical route of this paper

下载: 全尺寸图片幻灯片

图 2 图像分割PCA过程简图

Figure 2. A brief diagram of PCA process in image segmentation

下载: 全尺寸图片幻灯片

图 3 基于PCA的K-means算法聚类中心选择

Figure 3. Clustering center selection of K-means algorithm based on PCA

下载: 全尺寸图片幻灯片

图 4 K-means聚类K值的选择

Figure 4. Selecting parameter K in K-means clustering

下载: 全尺寸图片幻灯片

图 5 UC Merced Land Use Dataset场景影像

Figure 5. UC Merced Land Use Dataset scene images

下载: 全尺寸图片幻灯片

图 6 GF-2遥感影像

Figure 6. GF-2 remote sensing images

下载: 全尺寸图片幻灯片

图 7 遥感影像ROI提取流程

Figure 7. ROI extraction process of remote sensing image

下载: 全尺寸图片幻灯片

图 8 人工影像数据集

Figure 8. Artificial image data sets

下载: 全尺寸图片幻灯片

图 9 ESP分割及尺度选择

Figure 9. ESP segmentation and scale selection

下载: 全尺寸图片幻灯片

图 10 棋盘分割结果

Figure 10. Chessboard segmentation results

下载: 全尺寸图片幻灯片

图 11 四叉树分割结果

Figure 11. Quadtree based segmentation results

下载: 全尺寸图片幻灯片

图 12 对比度分割结果

Figure 12. Contrast split segmentation results

下载: 全尺寸图片幻灯片

图 13 多尺度分割结果

Figure 13. Multiresolution segmentation results

下载: 全尺寸图片幻灯片

图 14 光谱差异分割结果

Figure 14. Spectral difference segmentation results

下载: 全尺寸图片幻灯片

图 15 PCA降维结果

Figure 15. PCA dimension reduction results

下载: 全尺寸图片幻灯片

图 16 联合PCA-K-means分割结果

Figure 16. Segmentation results of combining PCA-K-means

下载: 全尺寸图片幻灯片

图 17 K-means随机聚类与PCA-K-means评价

Figure 17. Evaluation of K-means random clustering and PCA-K-means

下载: 全尺寸图片幻灯片

图 18 PCA-K-means外部指标评价

Figure 18. External index evaluation of PCA-K-means

下载: 全尺寸图片幻灯片

图 19 分割结果评价

Figure 19. Evaluation of segmentation results

下载: 全尺寸图片幻灯片

表 1 GF-2卫星有效载荷技术参数

Table 1. Technical parameters of GF-2 satellite payload

波段序号	波长范围/μm	波段名称	空间分辨率/m
Band 1	0.45~0.52	Blue	4
Band 2	0.52~0.59	Green	4
Band 3	0.63~0.69	Red	4
Band 4	0.77~0.89	NIR	4
PAN	0.45~0.90	Panchromatic	1
注：数据来源于中国资源卫星应用中心

下载: 导出CSV

表 2 GF-2卫星绝对辐射定标系数

Table 2. Absolute radiometric calibration coefficient of GF-2 satellite

波段序号	定标系数	GF-2 PMS 1	GF-2 PMS 2
PAN	Gain	0.1503	0.1679
PAN	Bias	0	0
Band 1	Gain	0.1193	0.1434
Band 1	Bias	0	0
Band 2	Gain	0.1530	0.1595
Band 2	Bias	0	0
Band 3	Gain	0.1424	0.1511
Band 3	Bias	0	0
Band 4	Gain	0.1569	0.1685
Band 4	Bias	0	0
注：Le=Gain·DN+Bias，Le为卫星载荷通道入瞳处等效辐射亮度, Gain和Bias分别为定标系数增益和偏移量，单位均为W/(m ²·sr·μm), DN(Digital Number)为卫星载荷观测值，无单位。

下载: 导出CSV

表 3 误差平方和

Table 3. Index of sum of the squared error

数据集	SSE		簇
数据集	PCA- K-means	K-means	簇
飞机场	3130471.24	3154435.88	7
棒球场	4087335.06	4163850.17	8
建筑	8363422.52	8385022.59	4
高尔夫球场	11631195.71	11631195.71	3
河流	39832127.26	39832127.26	2
跑道	7319143.02	7319143.02	4

下载: 导出CSV

参考文献(25)

[1]	邓书斌.ENVI遥感图像处理方法.[M].2版.北京:高等教育出版社, 2014. DENG S B.ENVI remote sensing image processing methods[M].2nd ed.Beijing:Higher Education Press, 2014(in Chinese).
[2]	JOZDANI S E, MOMENI M, JOHNSON B A, et al.A regression modelling approach for optimizing segmentation scale parameters to extract buildings of different sizes[J].International Journal of Remote Sensing, 2018, 39(3):684-703. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=10.1080/01431161.2017.1390273
[3]	HU Z W, ZHANG Q, ZOU Q, et al.Stepwise evolution analysis of the region-merging segmentation for scale parameterization[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2018, 11(7):2461-2472.
[4]	高仁强, 欧阳建, 陈亮雄, 等.面向对象的鹤地水库高分辨率影像分类研究[J/OL].测绘科学: 1-14(2018-11-19)[2019-07-02].http://kns.cnki.net/kcms/detail/11.4415.P.20181115.1629.002.html. GAO R Q, OU Y J, CHEN L X, et al.Research on high resolution image classification of Hedi reservoir based on object-oriented method[J/OL].Science of Surveying and Mapping: 1-14(2018-11-19)[2019-07-02].http://kns.cnki.net/kcms/detail/11.4415.P.20181115.1629.002.html(in Chinese).
[5]	HU L, CHAN K C C.Fuzzy clustering in a complex network based on content relevance and link structures[J].IEEE Transactions on Fuzzy Systems, 2016, 24(2):456-470. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=f6d62149890d0d2f239d8226d442e278
[6]	ZHANG H, SHI W Z, HAO M, et al.An adaptive spatially constrained fuzzy c-means algorithm for multispectral remotely sensed imagery clustering[J].International Journal of Remote Sensing, 2018, 39(8):2207-2237. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=10.1080/01431161.2017.1420934
[7]	FAN J C, WANG J.A two-phase fuzzy clustering algorithm based on neurodynamic optimization with its application for PolSAR image segmentation[J].IEEE Transactions on Fuzzy Systems, 2018, 26(1):72-83.
[8]	王慧贤, 靳惠佳, 王娇龙, 等.k均值聚类引导的遥感影像多尺度分割优化方法[J].测绘学报, 2015, 44(5):526-532. http://d.old.wanfangdata.com.cn/Periodical/chxb201505009 WANG H X, JIN H J, WANG J L, et al.Optimization approach for multi-scale segmentation of remotely sensed imagery under k-means clustering guidance[J].Acta Geodaetica et Cartographica Sinica, 2015, 44(5):526-532(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/chxb201505009
[9]	SIDDIQUI F U, ISA N A M.Enhanced moving K-means (EMKM) algorithm for image segmentation[J].IEEE Transactions on Consumer Electronics, 2011, 57(2):833-841. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=f5787723a74b4461296ad883f8ca04e2
[10]	姜文斌, 刘丽萍, 孙学宏.基于自适应权重法的K-means模型对遥感图像分割[J].计算机应用与软件, 2019, 36(5):231-234. http://d.old.wanfangdata.com.cn/Periodical/jsjyyyrj201905041 JIANG W B, LIU L P, SUN X H.K-means model based on adaptive weight method for remote sensing image segmentation[J].Computer Applications and Software, 2019, 36(5):231-234(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/jsjyyyrj201905041
[11]	CHEN S Y, SUN T L, YANG F Q, et al.An improved optimum-path forest clustering algorithm for remote sensing image segmentation[J].Computers & Geosciences, 2018, 112:38-46. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=a9ba1323df65bfec0d1433ef4a9dccc9
[12]	马彩虹, 戴芹, 刘士彬.一种融合PSO和Isodata的遥感图像分割新方法[J].武汉大学学报(信息科学版), 2012, 37(1):35-38. http://d.old.wanfangdata.com.cn/Periodical/whchkjdxxb201201010 MA C H, DAI Q, LIU S B.A new method of remote sensing image segmentation based on PSO and Isodata[J].Geomatics and Information Science of Wuhan University, 2012, 37(1):35-38(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/whchkjdxxb201201010
[13]	CHEN J Y, ZHENG H B, LIN X, et al.A novel image segmentation method based on fast density clustering algorithm[J].Engineering Applications of Artificial Intelligence, 2018, 73:92-110. http://d.old.wanfangdata.com.cn/OAPaper/oai_doaj-articles_9a8268c3fbd7093f2833dc9bd6e046b8
[14]	FAN J C, HAN M, WANG J.Single point iterative weighted fuzzy C-means clustering algorithm for remote sensing image segmentation[J].Pattern Recognition, 2009, 42(11):2527-2540. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=0bc51794892b8479cdd46a1aea36dd98
[15]	YANG Y, NEWSAM S.Bag-of-visual-words and spatial extensions for land-use classification[C]//ACM Press the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems(ACM GIS).New York: ACM, 2010: 270.
[16]	中国资源卫星应用中心.数据服务平台[EB/OL].(2015-09-02)[2019-07-02].http://www.cresda.com/. China Centre for Resources Satellite Data and Application.Data service platform[EB/OL].(2015-09-02)[2019-07-02].http://www.cresda.com/(in Chinese).
[17]	杉山将.图解机器学习[M].许永伟, 译.北京: 人民邮电大学出版社, 2015: 143. SUGIYAMA M.Graphic machine learning[M].XU Y W, translated.Beijing: People's University of Posts and Telecommunications Press, 2015: 143(in Chinese).
[18]	JIANG Q C, YAN X F.Parallel PCA-KPCA for nonlinear process monitoring[J].Control Engineering Practice, 2018, 80:17-25. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=04f42304818e42208ac070204eeea5d5
[19]	JAIN A K.Data clustering:50 years beyond K-means[J].Pattern Recognition Letters, 2010, 31(8):651-666. http://d.old.wanfangdata.com.cn/NSTLQK/NSTL_QKJJ0218023812/
[20]	HARTIGAN J A, WONG M A.Algorithm AS 136:A K-means clustering algorithm[J].Journal of the Royal Statistical Society, 1979, 28(1):100-108. http://d.old.wanfangdata.com.cn/OAPaper/oai_doaj-articles_d9ef8caab2204fcde05264964740e263
[21]	周本金, 陶以政, 纪斌, 等.最小化误差平方和k-means初始聚类中心优化方法[J].计算机工程与应用, 2018, 54(15):48-52. http://d.old.wanfangdata.com.cn/Periodical/jsjgcyyy201815009 ZHOU B J, TAO Y Z, JI B, et al.Optimizing k-means initial clustering centers by minimizing sum of squared error[J].Computer Engineering and Applications, 2018, 54(15):48-52(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/jsjgcyyy201815009
[22]	傅立伟, 武森.基于属性值集中度的分类数据聚类有效性内部评价指标[J].工程科学学报, 2019, 41(5):682-693. http://d.old.wanfangdata.com.cn/Periodical/bjkjdxxb201905015 FU L W, WU S.A new internal clustering validation index for categorical data based on concentration of attribute values[J].Chinese Journal of Engineering, 2019, 41(5):682-693(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/bjkjdxxb201905015
[23]	DAVIES D L, BOULDIN D W.A cluster separation measure[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 1979, 1(2):224-227. http://d.old.wanfangdata.com.cn/OAPaper/oai_arXiv.org_astro-ph%2f0306174
[24]	LI M, DE BEURS K M, STEIN A, et al.Incorporating open source data for Bayesian classification of urban land use from vhr stereo images[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2017, 10(11):4930-4943. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=54a578fe3d41046446e344a40267fa05
[25]	CHEN Y Y, MING D P, XU L, et al.An overview of quantitative experimental methods for segmentation evaluation of high spatial remote sensing images[J].Journal of Geo-information Science, 2017, 19(6):818-830. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dqxxkx201706011