摘要
针对现有遥感影像变化检测方法常存在的检测结果破碎、虚检较多、对数据匹配要求高等问题。提出了一种融合像素级和对象级的遥感图像变化检测方法。利用光谱和纹理信息构建单高斯模型,在多尺度上进行像素级变化检测。然后,以像素级检测结果为种子区域,同时在变化前后影像上区域生长,融合生长结果提取变化对象。最后,依据检测需求对变化对象进行特征分类并滤除虚警。实验结果表明,该方法降低了虚检,保持了变化区域的结构完整性,在变化前后图像分辨率存在一定差别时仍有较高的检测精度。
The traditional change detection methods of multi-temporal remote sensing images are al ways trouble at broken detected results, high false alarm rate and being sensitive to the quality of remote data. In this paper, a new method which combined pixel-level and object-level change detection was proposed. The spectral information and text information are merged by a single Gaussian model constructed to set the change threshold for each pixel. Based on the model, pixel-level change detection are carried out at multi-scales. The results of pixel-level change detection are used as seed areas for region growing at multitemporal images separately. Change objects are extracted from the grow results. Finally, the change objects are classified and false alarms are rejected according to detection requirements. Experiment results demonstrate that the proposed method can effectively reduce the false alarms, maintain the structure of changed region and keep the accuracy even the multi-temporal images have different spatial resolutions.
出处
《测绘科学》
CSCD
北大核心
2017年第5期106-112,共7页
Science of Surveying and Mapping
基金
复旦大学电磁波信息科学教育部重点实验室开放基金项目(EMW201506)
关键词
变化检测
单高斯模型
多尺度
变化对象
区域生长
change detection
single Gaussian model
multi-scale
change object
region grow