Evaluation of a change detection method based on joint spatial-spectral information

The performances of a new change detection methodology based on the analysis of pre and post change scattergrams (2D spectral domains) has been evaluated, through a comparative study with two other methodologies: a pre-classification and a post-classification method. The pre-classification methodology uses an unsupervised method based on the so-called "difference image" technique by change vector analysis (CVA) of pairs of pixel values at both times. The post-classification methodology is inspired on an unsupervised method (AutoChange) also, using clustering in two phases for change detection and identification. Multiternporal data sets of three multispectral images acquired by the Landsat 7 Thematic Mapper (ETM+) sensor, corresponding to the geographical area of Madrid countryside have been used for the comparative study. In this paper, it has been proven that the proposed method presents some outstanding features as compared with other change detection methods available in the literature. The developed method can be applied to determine very easily and accurately training areas associated to degraded zones for a further classification process.