基于超像素的协同分割变化检测方法
|
摘要
协同分割变化检测方法能够有效地克服椒盐现象,生成边界一致的多时相变化对象.但是对于大范围的实验区来说这种方法运算量大,耗时长.针对这一缺点,基于超像素的协同分割变化检测方法引入了超像素分割的思想,利用像素之间特征的相似性将像素分组,用少量的超像素代替大量的像素来表达图片特征.该方法将每个超像素块视为一个节点,减少最小割/最大流构建的网络流图中节点的数量,可以直接获得大范围变化检测的结果.以中国江西省南昌县高分一号影像以及Landsat TM影像为例进行试验,分割结果的总体精度在0.80～0.82之间,Kappa系数在0.65～0.61之间,计算时间从超过1 d提升至最长不超过4 h.试验表明,基于超像素的协同分割变化检测方法既能准确提取出变化图斑,又能极大地提升协同分割变化检测的运行速度.
The cosegmentation change detection method can effectively overcome the salt-and-pepper noise and generate a multi-temporal change object with consistent boundaries, but for a large range of experimental areas, the calculation amount is large and the loss time is too long. For the cosegmentation change detection algorithm is inefficient and the idea of superpixel segmentation is introduced. By using the similarity of features between pixels to group pixels, a small number of super pixels are used instead of a large number of pixels to express picture features. This method treats each superpixel block as a node, reducing the number of nodes in the network flow graph constructed by the min cut/max flow, and directly obtaining the result of the large-scale change detection. Taking the Gaofen-1 satellite image and the Landsat TM image as examples, the overall accuracy of the segmentation results is between 0.80 and 0.82, the Kappa coefficient is between 0.65 and 0.61, and the calculation time is shortened from one day to less 4 hours. Experiments show that this method based on superpixel cosegmentation change detection can not only accurately extract the change map, but also greatly improve the running speed of cosegmentation change detection.
The cosegmentation change detection method can effectively overcome the salt-and-pepper noise and generate a multi-temporal change object with consistent boundaries, but for a large range of experimental areas, the calculation amount is large and the loss time is too long. For the cosegmentation change detection algorithm is inefficient and the idea of superpixel segmentation is introduced. By using the similarity of features between pixels to group pixels, a small number of super pixels are used instead of a large number of pixels to express picture features. This method treats each superpixel block as a node, reducing the number of nodes in the network flow graph constructed by the min cut/max flow, and directly obtaining the result of the large-scale change detection. Taking the Gaofen-1 satellite image and the Landsat TM image as examples, the overall accuracy of the segmentation results is between 0.80 and 0.82, the Kappa coefficient is between 0.65 and 0.61, and the calculation time is shortened from one day to less 4 hours. Experiments show that this method based on superpixel cosegmentation change detection can not only accurately extract the change map, but also greatly improve the running speed of cosegmentation change detection.
引文
[1] 黄亮,左小清,於雪琴.遥感影像变化检测方法探讨[J].测绘科学,2013,38(4):203-206.HUANG Liang, ZUO Xiaoqing, YU Xueqin. Review on change detection methods of remote sensing images[J]. Science of Surveying and Mapping,2013,38(4): 203-206. (in Chinese)
[2] BAUDOUIN D, BOGAERT P, DEFOURNY P. Forest change detection by statistical object-based method[J]. Remote Sensing of Environment, 2006, 102(1-2):1-11.
[3] 袁敏,肖鹏峰,冯学智,等.基于协同分割的高分辨率遥感图像变化检测[J].南京大学学报(自然科学版), 2015(5):1039-1048.YUAN Min, XIAO Pengfeng, FENG Xuezhi, et al. Change detection from high-resolution remotely sensed images based on cosegmentation [J]. Journal of Nanjing University (Natural Science Edition), 2015(5):1039-1048. (in Chinese)
[4] 谢振雷.基于协同分割的遥感图像变化检测[D].北京建筑大学,2017.XIE Zhenlei. Change detection method based on cosegmentation for remotely sensed image[D]. Beijing University of Civil Engineering and Architecture,2017. (in Chinese)
[5] REN X, MALIK J. Learning a classification model for segmentation[J]. International Conference on Computer Vision, 2003(1):10-17.
[6] 栾庆祖,刘慧平,肖志强.遥感影像的正射校正方法比较[J].遥感技术与应用,2007(6):743-747+674.LUAN Qingzu, LIU Huiping, XIAO Zhiqiang. Comparison between algorithms of ortho-rectification for remote sensing images[J].Remote Sensing Technology and Application,2007(6):743-747,674. (in Chinese)
[7] 郑伟,曾志远.遥感图像大气校正方法综述[J].遥感信息,2004(4):66-70.ZHENG Wei, ZENG Zhiyuan. Review on methods of atmospheric correction for remote sensing images [J]. Remote Sensing Information,2004(4):66-70. (in Chinese)
[8] ACHANTA R, SHAJI A, SMITH K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2012, 34(11):2274-2282.
[9] 王春瑶, 陈俊周, 李炜. 超像素分割算法研究综述[J]. 计算机应用研究, 2014(1):6-12.WANG Chunyao, CHEN Junzhou, LI Wei. Review on superpixel segmentation algorithms [J]. Application Research of Computers, 2014(1): 6-12. (in Chinese)
[10] FORD L, FULKERSON D. Flows in networks[M]. Princeton: Princeton University Press,1962:208.
[11] 赵英时.遥感应用分析原理与方法[M].北京:科学出版社,2003:266-271.ZHAO Yingshi. Principles and methods of remote sensing application analysis [M].Beijing: Science Press,2003:266-271. (in Chinese)
[12] HARALICK R M, SHANMUGAM K, DINSTEIN I. Textural features for image classification[J]. Systems Man & Cybernetics IEEE Transactions on, 1973, 3(6):610-621.
[13] DINITS E A. Algorithm for solution of a problem of maximum flow in networks with power estimation[J]. Soviet Math Doklady, 1970, 11:754-757.
[14] 张安定.遥感原理与应用题解[M].北京:科学出版社,2016:105-107.ZHANG Anding. Remote sensing principle and application solution [M]. Beijing: Science Press,2016:105-107. (in Chinese)
[2] BAUDOUIN D, BOGAERT P, DEFOURNY P. Forest change detection by statistical object-based method[J]. Remote Sensing of Environment, 2006, 102(1-2):1-11.
[3] 袁敏,肖鹏峰,冯学智,等.基于协同分割的高分辨率遥感图像变化检测[J].南京大学学报(自然科学版), 2015(5):1039-1048.YUAN Min, XIAO Pengfeng, FENG Xuezhi, et al. Change detection from high-resolution remotely sensed images based on cosegmentation [J]. Journal of Nanjing University (Natural Science Edition), 2015(5):1039-1048. (in Chinese)
[4] 谢振雷.基于协同分割的遥感图像变化检测[D].北京建筑大学,2017.XIE Zhenlei. Change detection method based on cosegmentation for remotely sensed image[D]. Beijing University of Civil Engineering and Architecture,2017. (in Chinese)
[5] REN X, MALIK J. Learning a classification model for segmentation[J]. International Conference on Computer Vision, 2003(1):10-17.
[6] 栾庆祖,刘慧平,肖志强.遥感影像的正射校正方法比较[J].遥感技术与应用,2007(6):743-747+674.LUAN Qingzu, LIU Huiping, XIAO Zhiqiang. Comparison between algorithms of ortho-rectification for remote sensing images[J].Remote Sensing Technology and Application,2007(6):743-747,674. (in Chinese)
[7] 郑伟,曾志远.遥感图像大气校正方法综述[J].遥感信息,2004(4):66-70.ZHENG Wei, ZENG Zhiyuan. Review on methods of atmospheric correction for remote sensing images [J]. Remote Sensing Information,2004(4):66-70. (in Chinese)
[8] ACHANTA R, SHAJI A, SMITH K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2012, 34(11):2274-2282.
[9] 王春瑶, 陈俊周, 李炜. 超像素分割算法研究综述[J]. 计算机应用研究, 2014(1):6-12.WANG Chunyao, CHEN Junzhou, LI Wei. Review on superpixel segmentation algorithms [J]. Application Research of Computers, 2014(1): 6-12. (in Chinese)
[10] FORD L, FULKERSON D. Flows in networks[M]. Princeton: Princeton University Press,1962:208.
[11] 赵英时.遥感应用分析原理与方法[M].北京:科学出版社,2003:266-271.ZHAO Yingshi. Principles and methods of remote sensing application analysis [M].Beijing: Science Press,2003:266-271. (in Chinese)
[12] HARALICK R M, SHANMUGAM K, DINSTEIN I. Textural features for image classification[J]. Systems Man & Cybernetics IEEE Transactions on, 1973, 3(6):610-621.
[13] DINITS E A. Algorithm for solution of a problem of maximum flow in networks with power estimation[J]. Soviet Math Doklady, 1970, 11:754-757.
[14] 张安定.遥感原理与应用题解[M].北京:科学出版社,2016:105-107.ZHANG Anding. Remote sensing principle and application solution [M]. Beijing: Science Press,2016:105-107. (in Chinese)