Discovery of change from the Earth observation satellite data by spatio-temporal data mining and a large scale visualization

数据挖掘是一种有效的数据挖掘方法，它可以有效地挖掘大规模数据库，并支持从数据库中发现知识，近年来，数据挖掘的目标已经从简单的表格数据转向更复杂的数据，如图结构。其中，时空数据挖掘是最有前途的领域之一，因为时空数据大量存在于科学、工程数据和真实的世界数据中。我们使用了高知大学储存的10年以上的气象卫星图像和22年以上的全球植被指数数据模型，这些数据是从美国气象局创建的。S. NOAA系列卫星（GIMMS）。经过三年的研究，我们利用真实的数据实现和检验了时空数据挖掘的一些重要过程：（1）半自动目标提取和跟踪及其在气象卫星图像中的云的应用，（2）归一化植被指数的统计建模，（3）时空相关性分析;（4）获取的模式在三维时空中的可视化和交互式可视化数据挖掘。

The approach of data mining is effective to explore of a large scale database and to support discovery of the knowledge from it. Recently, the targets of data mining have been changed from a simple tabular data to more complex data such as graph structure. Among them, mining of the spatio-temporal data is one of the most promising fields, because spatio-temporal data occurs very often in the scientific, engineering data and the real world data in a large scale.In this study, the framework of spatio-temporal data mining was examined by using the Earth observation satellite data as the test bed. We used the weather satellite images stored at Kochi University for more than 10 years and the Global model of vegetation indices data for more than 22 years, which are created from U. S. satellite NOAA series satellites (GIMMS) . Both data are well calibrated and maintained.As a result of three-year studies, some important process of the spatio-temporal data mining were implemented and examined by using the real data: (1) semi-automatic object extraction and tracking and its application to clouds in the weather satellite imagery, (2) statistical modeling of vegetation index for NDVI, (3) spatio-temporal correlation analysis and (4) visualization of obtained pattern in the 3D spatio temporal space and the interactive visual data mining.