期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2010
卷号：XXXVIII - Part 8
页码：719-724
出版社：Copernicus Publications
摘要：This paper describes the development of a novel time series modelling and spectrum anomaly-detection method, which takes into consideration wide-area seasonal changes. By taking advantage of both the high temporal resolution and the wide swath mode of multi-temporal satellite data, such as NOAA/AVHRR, MODIS, and SP OT/vegetation, it is possible to perform high-frequency monitoring of wide-area, land cover changes. However, since the multi-temporal satellite data are influenced by clouds and system noise, in many cases, they must be processed in order to accurately represent the actual surface conditions. We engineered a discrete time-series model using a self-organizing map (SOM) and a hidden Markov Model (HMM) to reduce the influence of clouds in order to improve the accuracy of the products. The spectral information of the pixels was first converted to nominal scale values, and the influence of clouds was eliminated through a time-series modelling using HMM. Since the anomaly-detection method requires a clustering of nominal vectors, dedicated software based on SOM algorithm was also developed. The data for anomaly detection is not dependent on the information of neighbouring pixels, and it is possible to detect an anomaly even if there is only one pixel
