|
Abstract:
New developments in the field of information technology in terms of processing algorithms and data mining methods enable more an opportunistic use of data banks of remote sensing images. Knowledge mining from satellite imagery and raster databases can be viewed as a case of spatial data mining. Data mining in image databases may have been seen as similar to image processing. However, in the case of data mining studies, very large amount of data are processed and analyzed, while traditional image processing usually concentrate on analysis of single or a few images. Until today, the analysis of images are mostly done manually or semi automatically. These conventional methods might prove useful for analysis of a small number of images, but might not be effective and efficient with a huge collection of remotely sensed imageries which might span terra bytes of data, such as those of Landsat archived and high-temporally resolved images such as MODIS. Moreover, the time taken for the analysis of remotely sensed imagery is also huge due to the nature of the data sets. Third, GIS operations and functions which are possible in the vector domain may not be feasible with the raster or image domain. This is mainly due to the nature of image database and the uncertainty involved within these databases. This problem gets aggravated with the change in spatial resolution. These limitations make the analysis of a series of imagery difficult. An alternative would be using data mining methods to extract the objects of significance and then to find their relevance from one set to another. So far, there is not any clear-cut definition for extraction and tracking of objects in space and time from remotely sensed imageries to follow such process.
In recent years, researchers at Center for Urban and Environmental Change, Indiana State University, have explored various methods of data mining to model, analyze, and discover knowledge about urban heat islands (UHIs) from MODIS, Landsat, and ASTER images. Our research focus on: 1) the development of data mining methods for extraction of UHIs, 2) tracking of UHIs in space and space-time, 3) methods for querying and analysis of the objects, 4) the automation of data mining process, and 5) application of the technique of association rule mining to quantitatively estimate the associations between the environmental, physical, and demographic variables with specific land use/cover types and UHIs. This talk intends to introduce our most recent research results.
|
