摘要
针对人工判读作业模式难以满足地表覆盖变化检测现势性要求的问题,该文围绕地理国情监测技术设计要求和实际生产特点,提出了一种基于地理国情先验知识的多特征模糊融合变化检测方法。利用地表覆盖像斑边界,根据地理国情监测不同地类的最小面积指标进行多尺度分割,获取同质的对象级像斑;构建像斑的光谱、纹理、形状等差值特征集;基于模糊集理论进行特征的自适应加权融合得到初步变化检测结果,根据技术规则剔除不符合要求的像斑,获得最终变化区域。结果表明,该方法能够根据不同地类变化有效地集成不同特征组合表征变化信息,可充分利用地理国情监测前期成果,具有较高的准确率。
Aiming at the problem that the manual operation pattern was difficult to meet the current requirements of land cover change detection,this paper proposed a multi-feature fuzzy fusion change detection method based on prior knowledge,which focusing on monitoring technology requirements and actual production characteristics of national geographic condition monitoring.At first,multi-scale segmentation was applied to obtain homogenous image object,according to the boundary of land cover image object and the minimum area index of different land type object.Then the subtraction feature sets of image spectrum,texture and shape and so on were constructed.After,based on the fuzzy set theory,the preliminary results were obtained by adaptive weighted features fusion.At last,removing change detection results objects that didn’t meet the technical rules to acquire the final results.The experimental results showed that the proposed method could effectively fuse different features to express change information according to different land cover types changes.This method could fully utilize the previous results of geographical conditions monitoring and had high change detection accuracy.
作者
薛昱晟
汪小钦
张因果
XUE Yusheng;WANG Xiaoqin;ZHANG Yinguo(Key Laboratory of Spatial Data Mining and Information Sharing of Ministry of Education,Fuzhou University,Fuzhou 350116,China;National&Local Joint Engineering Research Center of Satellite-Spatial Information Technology,Fuzhou 350116,China;Fujian Surveying and Mapping Institute,Fuzhou 350002,China)
出处
《测绘科学》
CSCD
北大核心
2019年第12期60-66,共7页
Science of Surveying and Mapping
基金
中央引导地方发展专项(2017L3012)
福建省测绘地理信息局科技专项资金资助项目(2017J01)
关键词
地理国情监测
地表覆盖
变化检测
模糊融合
多尺度分割
geographical condition monitoring
land cover
change detection
fuzzy fusion
multiscale segmentation