Please wait a minute...
 
自然资源遥感  2021, Vol. 33 Issue (3): 63-71    DOI: 10.6046/zrzyyg.2020344
  技术方法 本期目录 | 过刊浏览 | 高级检索 |
基于多尺度超像素的高光谱图像分类研究
王华1,2(), 李卫卫2(), 李志刚2, 陈学业1, 孙乐3
1.自然资源部城市土地资源监测与仿真重点试验室,深圳 518034
2.郑州轻工业大学河南省食品安全数据智能重点实验室,郑州 450002
3.南京信息工程大学计算机与软件学院,南京 210044
Hyperspectral image classification based on multiscale superpixels
WANG Hua1,2(), LI Weiwei2(), LI Zhigang2, CHEN Xueye1, SUN Le3
1. Key Laboratory of Urban Land Resources Monitoring and Simulation, Ministry of Natural Resources, Shenzhen 518034, China
2. Henan Key Laboratory of Food Safety Data Intelligence, Zhengzhou University of Light Industry, Zhengzhou 450002, China
3. Nanjing University of Information Science and Technology, School of Computer and Software, Nanjing 210044, China
全文: PDF(4297 KB)   HTML  
输出: BibTeX | EndNote (RIS)      
摘要 

随着遥感技术的快速发展,高光谱遥感影像的分类方法研究受到普遍关注。现有高光谱遥感影像分类研究采用单一尺度下的超像素方法进行图像分割处理,无法确定最佳超像素个数,较易忽视图像细节信息,且单一核矩阵无法表征多特征信息导致分类精度降低。因此,本研究拟在多尺度下采用超像素分割方法对高光谱影像的第一主成分分量进行多尺度超像素分割处理,通过权值耦合多尺度空间光谱核与原始空间光谱核形成合成核来进行高光谱影像分类,并以Washington DC Mall高光谱影像为实验数据对本文方法进行测试与分析。实验结果显示,相较于对比方法,这一方法的有效分类精度最高提升6.93个百分点。结果证明该方法可以有效解决图像光谱无法自适应、光谱信息获取不全面的问题,能够显著提升高光谱影像分类精度。

服务
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章
王华
李卫卫
李志刚
陈学业
孙乐
关键词 RBF核函数多尺度超像素合成核SVM高光谱    
Abstract

With the rapid development of remote sensing technology, the research on the classification methods of hyperspectral remote sensing images has received widespread attention. However, existing studies on the classification of hyperspectral remote sensing images conduct image segmentation using a single-scale superpixel method. As a result, the optimal superpixel number cannot be determined, image details are liable to be omitted, and a single kernel matrix cannot characterize multiple feature information, thus leading to a decrease in the classification precision. Therefore, this study proposes to perform multiscale superpixel segmentation of the first principal component of hyperspectral images. Then it conducts hyperspectral image classification using the composite kernel obtained by coupling the multiscale spatial-spectral kernel with the original spatial-spectral kernel according to weights. Finally, it tests and analyzes the proposed method using the hyperspectral images of the National Mall in Washington, D.C. as experimental data. The test results show that the effective classification precision of this method is 6.93% higher than that of the compared methods. As proved by the results, this method can be used to effectively solve the problems such as the lack of self-adaption of image spectra and incomplete spectrum information acquired, thus significantly improving the classification accuracy of hyperspectral images.

Key wordsRBF kernel function    multiscale    superpixel    composite kernel SVM    hyperspectral image
收稿日期: 2020-11-02      出版日期: 2021-09-24
ZTFLH:  TP75  
基金资助:国家自然科学基金项目“领域知识驱动的土地利用空间优化配置与多情景模拟”(41771438);自然资源部城市土地资源监测与仿真重点实验室开放基金资助项目“融合多源数据的多粒度土地利用现状时空建模和系统研发”(KF-2019-04-038);数字制图与国土信息应用工程国家测绘地理信息局重点实验室开放研究基金资助项目“基于深度学习的城市地价评估样本点选择研究”(ZRZYBWD201911)
通讯作者: 李卫卫
作者简介: 王 华(1986-),男,博士,副教授,主要从事空间数据挖掘、空间决策支持技术研究。Email: whuwanghua@163.com
引用本文:   
王华, 李卫卫, 李志刚, 陈学业, 孙乐. 基于多尺度超像素的高光谱图像分类研究[J]. 自然资源遥感, 2021, 33(3): 63-71.
WANG Hua, LI Weiwei, LI Zhigang, CHEN Xueye, SUN Le. Hyperspectral image classification based on multiscale superpixels. Remote Sensing for Natural Resources, 2021, 33(3): 63-71.
链接本文:  
https://www.gtzyyg.com/CN/10.6046/zrzyyg.2020344      或      https://www.gtzyyg.com/CN/Y2021/V33/I3/63
Fig.1  模型流程图
Fig.2  超像素分割图像
Fig.3  多尺度滤波空间光谱核获取流程
Fig.4  单尺度超像素空间光谱核获取流程
Fig.5  原始空间光谱核获取流程
类别 训练集数量 测试集数量
住宅 500 3 365
公路 60 356
街道 50 89
草地 60 358
林地 60 345
水域 60 389
阴影 30 39
Tab.1  样本类别与样本集个数
Fig.6  权值ϑ对应的分类精度
类别 Ms-SSSK
/%
Ms-FSSK
/%
Ss-SSSK
/%
O-SSSK
/%
住宅 99.28 98.57 98.30 96.67
公路 98.32 94.36 92.68 87.55
街道 98.70 98.01 96.70 91.60
草地 97.35 95.25 92.12 89.16
林地 97.60 96.88 96.37 93.65
水域 98.65 96.31 98.71 96.87
阴影 98.16 96.98 96.60 87.73
总体精度/% 98.53 96.49 95.33 91.60
Kappa 0.960 1 0.948 9 0.921 6 0.903 1
Tab.2  分类结果对比
Fig.7  测试集分类结果
Fig.8  图像整体对比
Fig.9  图像局部对比
Fig.10  各模型相对误差结果图
[1] Goetz A F H. Three decades of hyperspectral remote sensing of the Earth:A personal view[J]. Remote Sensing of Environment, 2009, 113(s1):5-16.
[2] Uwe K, Andreas B, Udo S. Fusion trees for fast and accurate classification of hyperspectral data with ensembles of γ -divergence-based RBF networks[J]. Neural Computing and Applications, 2015, 26(2):253-262.
doi: 10.1007/s00521-014-1634-9
[3] 胥海威, 杨敏华, 韩瑞梅, 等. 用随机决策树群算法进行高光谱遥感影像分类[J]. 应用科学学报, 2011, 29(6):598-604.
Xu H W, Yang M H, Han R M, et al. Classification of hyperspectral remote sensing images using stochastic decision tree group algorithm[J]. Journal of Applied Sciences, 2011, 29(6):598-604.
[4] Guo Y H, Yin X J, Zhao X C, et al. Hyperspectral image classification with SVM and guided filter[J]. Eurasip Journal on Wireless Communications and Networking, 2019(1):1-9.
[5] 廖建尚, 王立国. 面向空间自相关信息的高光谱图像分类方法[J]. 农业机械学报, 2018, 49(6):215-224.
Liao J S, Wang L G. Hyperspectral image classification method for spatial autocorrelation information[J]. Transactions of the Chinese Society for Agricultural Machinery, 2018, 49(6):215-224.
[6] Kavita B, Vijaya M. Evaluation of deep learning CNN model for land use land cover classification and crop identification using hyperspectral remote sensing images[J]. Journal of the Indian Society of Remote Sensing, 2019, 47(11):1949-1958.
doi: 10.1007/s12524-019-01041-2
[7] 朱瑞飞, 马经宇, 李竺强, 等. 多层感知卷积神经网络的国产多光谱影像分类[J]. 光学学报, 2020, 40(15):1-21.
Zhu R F, Ma J Y, Li Z Q, et al. Classification of domestic multispectral images based on multilayer perceptual convolutional neural network[J]. Acta Optica Sinica, 2020, 40(15):1-21.
[8] Bera, Shrivastava. Analysis of various optimizers on deep convolutional neural network model in the application of hyperspectral remote sensing image classification[J]. International Journal of Remote Sensing, 2020, 41(7):2664-2683.
doi: 10.1080/01431161.2019.1694725
[9] Mohamad J, Mauro D M, Pierre C. Hyperspectral image classification based on mathematical morphology and tensor decomposit-ion[J]. Mathematical Morphology-Theory and Applications, 2020, 4(1):1-30.
doi: 10.1515/mathm-2020-0001
[10] Paulina, Koziol, Magda, et al. Comparison of spectral and spatial denoising techniques in the context of High Definition FT-IR imaging hyperspectral data[J]. Scientific Reports, 2018, 8(11):1444-1463.
doi: 10.1038/s41598-018-19829-6
[11] 杜培军, 夏俊士, 薛朝辉, 等. 高光谱遥感影像分类研究进展[J]. 遥感学报, 2020, 20(2):236-256.
Du P J, Xia J S, Xue Z H, et al. Review of hyperspectral remote sensing image classification[J]. Journal of Remote Sensing, 2020, 20(2):236-256.
[12] Wang P Y, Zhu H Q, Chen N. Umms:Efficient superpixel segmentation driven by a mixture of spatially constrained uniform distribution[J]. Ieice Transactions on Information and Systems, 2020, 103(1):181-185.
[13] Sanghyun P. Crab region extraction method from tidal flat images using superpixels[J]. Journal of Advanced Information Technology and Convergence, 2019, 9(2):29-39.
doi: 10.14801/JAITC.2019.9.2.29
[14] Vitaliy K, Grzegorz M. Persistence-based resolution-independent meshes of superpixels[J]. Pattern Recognition Letters, 2020, 131(3):300-306.
doi: 10.1016/j.patrec.2020.01.014
[15] Tang K W, Su Z X, Jiang W, et al. Superpixels for large dataset subspace clustering[J]. Neural Computing and Applications, 2019, 31(12):8727-8736.
doi: 10.1007/s00521-018-3914-2
[16] Song D M, Tan X, Wang B, et al. Integration of super-pixel segmentation and deep-learning methods for evaluating earthquake-damaged buildings using single-phase remote sensing imagery[J]. International Journal of Remote Sensing, 2020, 41(3):1040-1066.
doi: 10.1080/01431161.2019.1655175
[17] Amin B, Riaz M M, Ghafoor A. Automatic image matting of synthetic aperture radar target chips[J]. Radio Engineering, 2020, 29(1):228-234.
[18] Yuan X H, Guo S J, Li C W, et al. Near infrared star centroid detection by area analysis of multi-scale super pixel saliency fusion map[J]. Tsinghua Science and Technology, 2019, 24(3):291-300.
doi: 10.1109/TST.5971803
[19] 陈允杰, 马辰阳, 孙乐, 等. 基于边缘修正的高光谱图像超像素空谱核分类方法[J]. 电子学报, 2019, 47(1):73-81.
Chen Y J, Ma C Y, Sun L. Edge-modified superpixel based spectral-spatial kernel method for hyperspectral image classification[J]. Acta Electronica Sinica, 2019, 47(1):73-81.
[20] 刘忠林, 吴一全, 邹宇. 多尺度红外超像素图像模型的小目标检测[J]. 中国图象图形学报, 2019, 24(12):2159-2173.
Liu Z L, Wu Y Q, Zou Y. Multiscale infrared superpixel-image model for small-target detection[J]. Journal of Image and Graphics, 2019, 24(12):2159-2173.
[21] Kang X D, Duan P H, Li S T. Hyperspectral image visualization with edge-preserving filtering and principal component analysis[J]. Information Fusion, 2020, 57(5):130-143.
doi: 10.1016/j.inffus.2019.12.003
[22] Shi C, Pun C M. Multiscale superpixel-based hyperspectral image classification using recurrent neural networks with stacked autoencoders[J]. IEEE Transactions on Multimedia, 2020, 22(2):487-501.
doi: 10.1109/TMM.6046
[23] Sun L, Ma C, Chen Y, et al. Adjacent superpixel-based multiscale spatial-spectral kernel for hyperspectral classification[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 2019, 12(6):1909-1919.
[24] 刘纯, 洪亮, 陈杰, 等. 融合像素—多尺度区域特征的高分辨率遥感影像分类算法[J]. 遥感学报, 2015, 19(2):228-239.
Liu C, Hong L, Chen J, et al. Fusion of pixel-based and multi-scale region-based features for the classification of high-resolution remote sensing image[J]. Journal of Remote Sensing, 2015, 19(2):228-239.
[25] Ramzan M, Abid A, Khan H, et al. A review on state-of-the-art violence detection techniques[J]. IEEE Access, 2019, 7(9):107560-107575.
doi: 10.1109/Access.6287639
[26] Chang C C, Huang H T. Automatic tuning of the RBF kernel parameter for batch-mode active learning algorithms:A scalable framework[J]. IEEE Transactions on Cybernetics, 2019, 49(12):4460-4472.
doi: 10.1109/TCYB.6221036
[27] 尚坤, 李培军, 程涛. 基于合成核支持向量机的高光谱土地覆盖分类[J]. 北京大学学报:自然科学版, 2011, 47(1):109-114.
Shang K, Li P J, Cheng T. Land cover classification of hyperspectral data using composite kernel support vector machines[J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2011, 47(1):109-114.
[28] He J, Xu J L. Ultra-short-term wind speed forecasting based on support vector machine with combined kernel function and similar data[J]. Eurasip Journal on Wireless Communications and Networking, 2019(1):1-7.
[29] Akbari D, Moradizadeh M, Akbari M. Spectral-spatial classification of hyperspectral imagery using neural network algorithm and hierarchical segmentation[J]. The International Archives of the Photogrammetry,Remote Sensing and Spatial Information Sciences, 2019, XLII-2(W12):1-5.
doi: 10.5194/isprs-archives-XLII-2-1-2018
[1] 王嘉芃, 徐建国, 沈家晓, 张登荣. 德兴铜矿矿山重金属污染修复效果高光谱遥感评价[J]. 自然资源遥感, 2023, 35(3): 284-291.
[2] 郑宗生, 刘海霞, 王振华, 卢鹏, 沈绪坤, 唐鹏飞. 改进3D-CNN的高光谱图像地物分类方法[J]. 自然资源遥感, 2023, 35(2): 105-111.
[3] 张国建, 刘胜震, 孙英君, 俞凯杰, 刘丽娜. 基于弱监督鲁棒性自编码的高光谱异常检测[J]. 自然资源遥感, 2023, 35(2): 167-175.
[4] 苏腾飞. 基于边界信息的多尺度遥感影像分割质量非监督评价方法[J]. 自然资源遥感, 2023, 35(1): 35-40.
[5] 孔卓, 杨海涛, 郑逢杰, 李扬, 齐济, 朱沁雨, 杨忠霖. 高光谱遥感图像大气校正研究进展[J]. 自然资源遥感, 2022, 34(4): 1-10.
[6] 沈骏翱, 马梦婷, 宋致远, 柳汀洲, 张微. 基于深度学习语义分割模型的高分辨率遥感图像水体提取[J]. 自然资源遥感, 2022, 34(4): 129-135.
[7] 张鹏强, 高奎亮, 刘冰, 谭熊. 联合空谱信息的高光谱影像深度Transformer网络分类[J]. 自然资源遥感, 2022, 34(3): 27-32.
[8] 孙肖, 徐林林, 王晓阳, 田野, 王伟, 张中跃. 基于优化K-P-Means解混方法的高光谱图像矿物识别[J]. 自然资源遥感, 2022, 34(3): 43-49.
[9] 孙肖, 彭军还, 赵锋, 王晓阳, 吕洁, 张登峰. 基于空间统计学的高光谱遥感影像主成分选择方法[J]. 自然资源遥感, 2022, 34(2): 37-46.
[10] 晏红波, 韦晚秋, 卢献健, 杨志高, 黎振宝. 基于高光谱特征的土壤含水量遥感反演方法综述[J]. 自然资源遥感, 2022, 34(2): 1-9.
[11] 王茜, 任广利. 高光谱遥感异常信息在阿尔金索拉克地区铜金矿找矿工作中的应用[J]. 自然资源遥感, 2022, 34(1): 277-285.
[12] 曲海成, 王雅萱, 申磊. 多感受野特征与空谱注意力结合的高光谱图像超分辨率算法[J]. 自然资源遥感, 2022, 34(1): 43-52.
[13] 张大明, 张学勇, 李璐, 刘华勇. 一种超像素上Parzen窗密度估计的遥感图像分割方法[J]. 自然资源遥感, 2022, 34(1): 53-60.
[14] 吴琳琳, 李晓燕, 毛德华, 王宗明. 基于遥感和多源地理数据的城市土地利用分类[J]. 自然资源遥感, 2022, 34(1): 127-134.
[15] 温银堂, 王铁柱, 王书涛, 王贵川, 刘诗瑜, 崔凯. 基于多尺度分割的高分辨率遥感影像镶嵌线自动提取[J]. 自然资源遥感, 2021, 33(4): 64-71.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
京ICP备05055290号-2
版权所有 © 2015 《自然资源遥感》编辑部
地址:北京学院路31号中国国土资源航空物探遥感中心 邮编:100083
电话:010-62060291/62060292 E-mail:zrzyyg@163.com
本系统由北京玛格泰克科技发展有限公司设计开发