首页 | 本学科首页   官方微博 | 高级检索  
     

结合像元级与对象级的滨海湿地变化检测方法
引用本文:吴瑞娟,何秀凤,王静. 结合像元级与对象级的滨海湿地变化检测方法[J]. 地球信息科学学报, 2020, 22(10): 2078-2087. DOI: 10.12082/dqxxkx.2020.190417
作者姓名:吴瑞娟  何秀凤  王静
作者单位:1.内江师范学院地理与资源科学学院,内江 6411002.河海大学地球科学与工程学院,南京 2111003.武汉大学资源与环境科学学院,武汉 430072
基金项目:国家自然科学基金项目(41830110);国家自然科学基金项目(41871203);内江师范学院科研资助项目(2019YZ02)
摘    要:滨海湿地是动态且脆弱的生态系统,遥感变化检测技术为滨海湿地动态变化监测提供了有效手段。为解决像元级变化检测对配准误差敏感及其椒盐现象严重,对象级变化检测受分割参数影响较大且过程繁琐等问题,本文提出了显著图引导的结合像元级与对象级变化检测方法。首先,提取湿地亮度、归一化差异植被指数、归一化差异水体指数三个特征,得到特征差异影像;其次,利用最大对称环绕显著性检测算法生成显著图,采用结合模糊C均值和马尔可夫随机场方法对显著区域进行分割得到初始像元级变化检测结果;最后,在面向对象分割的基础上,通过构建对象的不确定性指数自适应选择训练样本,采用随机森林分类器进行分类得到最终变化检测结果。利用江苏盐城滨海湿地资源三号影像进行实验,结果表明,结合像元级与对象级方法的湿地变化检测总体精度为93.51%,与像元级、对象级方法相比,虚检率分别降低了29.04%和22.78%。

关 键 词:遥感变化检测  滨海湿地  视觉显著性检测  不确定性  模糊C均值  马尔可夫随机场  特征提取  资源三号卫星  
收稿时间:2019-08-01

Coastal Wetlands Change Detection Combining Pixel-based and Object-based Methods
WU Ruijuan,HE Xiufeng,WANG Jing. Coastal Wetlands Change Detection Combining Pixel-based and Object-based Methods[J]. Geo-information Science, 2020, 22(10): 2078-2087. DOI: 10.12082/dqxxkx.2020.190417
Authors:WU Ruijuan  HE Xiufeng  WANG Jing
Affiliation:1. School of Geography and Resource Science, Neijiang Normal University, Neijiang 641100, China2. School of Earth Sciences and Engineering, Hohai University, Nanjing 211100, China3. School of Resource and Environmental Sciences, Wuhan University, Wuhan 430072, China
Abstract:Coastal wetlands are dynamic and fragile ecosystems, and they have taken place obvious changes, which affected by siltation and erosion, coastal development and utilization, therefore it is of great practical significance to timely monitor coastal wetlands changes. Remote sensing change detection technology can obtain the changes occurred in different times by mathematical model analysis, so it provides an effective way to monitor the dynamic changes of coastal wetlands. From the perspective of analysis unit of remote sensing change detection technology, change detection methods can be divided into pixel-based change detection methods and object-based change detection methods. Pixel-based change detection methods are sensitive to image registration errors, and their salt-and-pepper phenomena are also serious, while object-based methods are affected by image segmentation parameters, and often complicated for users. In order to solve the problems above, saliency-guided change detection combining pixel-based and object-based methods is proposed, in which the scene characteristics of coastal wetlands are taken into account. Firstly, the brightness, Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) features are extracted, and the feature difference images are then obtained. Secondly, the Maximum Symmetric Surround (MSS) saliency detection algorithm is used to generate the saliency maps of feature difference images, and then the combination of Fuzzy C-means (FCM) with Markov Random Field (MRF) is used to extract the initial change detection result at the pixel level. Finally, multi-scale segmentation algorithm is utilized for object-oriented image segmentation, in which Rate of Change of Local Variance (ROC-LV) is used to estimate the optimal segmentation scales. The uncertainty index of segmentation objects is constructed to adaptively select training samples, and these training samples are used to train random forest classifier which is used to obtain the final change detection results. The experiments are carried out using Ziyuan-3 images in Yancheng coastal wetlands, Jiangsu Province, the results show that the proposed saliency-guided change detection combining pixel-based and object-based methods obtains the best change detection result when the segmentation scale and the uncertainty threshold are 55 and 0.7 respectively, the proposed method obtains the highest overall accuracy and accuracy ratio compared with traditional pixel-based, object-based, SG-PCAK, and SG-RCVA-RF methods, overall accuracy of our proposed method is 93.51%, which is higher than SG-PCAK method of 5.95%, false rate is reduced by 35.96% and accuracy ratio is improved by 29.24%, compared with SG-PCAK method. False rate is reduced by 29.04% and 22.78%, compared with the pixel-based method and object-based method respectively. Accuracy ratio of our proposed method is improved by 14.23%, compared with SG-RCVA-RF method. Therefore, the experimental results demonstrate the proposed change detection method improves the accuracy of monitoring coastal wetlands changes, compared with traditional change detection methods.
Keywords:remote sensing change detection  coastal wetland  visual saliency detection  uncertainty  fuzzy C mean  Markov random field  feature extraction  Ziyuan-3 satellite  
点击此处可从《地球信息科学学报》浏览原始摘要信息
点击此处可从《地球信息科学学报》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号