摘要
为获取遥感影像混合像元中各组分的空间分布状况,提出一种新的遥感影像超分辨率制图方法,用于继混合像元分解之后的亚像元定位。将元胞自动机理论移植到不同空间尺度的演化上,建立基于神经网络的多尺度元胞自动机模型(ANN-CA),并利用该模型提取北京市海淀区城镇用地超分辨率信息。结果表明,该方法能有效表达图像像元之间的空间自相关性。
Mixed pixel is a familiar problem in remotely sensed image analysis and classification. The developing soft classification technology provides no indication of the spatial distribution of each class composition in the IFOV given by a pixel. In this paper, the theory of cellular automata （CA） is transplanted to the evolvement of spatial scale. On the basis of multi-scale neural-network-based CA model, a new technology of super-resolution mapping from remote sensing images is proposed, in order to give the appropriate spatial location of each class within the mixed plxel with the agreement of the different land cover fractions which are extracted from a soft classification.The method is tested on super-resolution mapping of urban area in Haidian District, Beijing City at different spatial scales, which indicates convenience and efficiency of the method for expressing the spatial correlation between pixels.
出处
《地理与地理信息科学》
CSCD
北大核心
2007年第3期42-46,共5页
Geography and Geo-Information Science
基金
教育部"新世纪优秀人才支持计划"
关键词
混合像元
亚像元
超分辨率制图
神经网络
元胞自动机
mixed pixel
sub-pixel
super-resolution mapping
neural networks
cellular automata
