Research On Remote Sensing Information Extraction By Statistical Methods

Information extraction is the key for application of remote sensing. How promptly and precisely information can be extracted from the image greatly influences utilization of remote sensing technology. Only by improvements and innovations of extraction methods can advantages of the technology be fully used. Meanwhile the way is paved for deeper application of remote sensing technology.Based on the predecessors’ study of object spectrum, a basic assumption is established that grayscales of the same object obey gaussian distribution and research on information extraction by statistical methods has been taken on. The research starts in low dimentional feature space, by taking use of its intuitionistic exhibition, and discipline for distribution of object’s grayscales is studied. In two dimentional spectral feature space the distribution of grayscales of the same object forms an oval which shows a feature of anisotropy. A new information extraction method is established based on spatial U-static mehod, which is originally used for analysis of spatial singularity, and introduced to remote sensing information extraction to acquire spatial anisotropy parameters. This method is used in Bayinshan for ferrous alteration information extraction, and its validity is proved by comparing with field survey data.After Summarizing and analyzing the problems encountered before, a more perfect data model which is applied in high dimentional feature space is studied in order to improve the precision of information extraction. In high-dimentional spectral feature space, gaussian mixture model is used to describe pixel data. In the paper, markov random field and simulated annealing are used to develop gaussian mixture model and the new model could use spatial correlation and steadily acquire global optimum solution in the information extraction process. According to extracting the information of Pine, the new method demonstrates a prominent improvement. This method is also used in Bayinshan for ferrous alteration information, and has a better performance than that of principal component analysis.In this paper, a remote sensing information extraction research using statistical method is carried on. The research begins in low dimentional feature space and gets improved in high dimentional feature space. For the research, two methods are formed, including spatial U-static method and gaussian mixture model based method, which is a beneficial exploration in efficient use of remote sensing data.