Classification of dynamic evolutions from satellitar image time series based on similarity measures

Abstract

With a continuous increase in the number of Earth Observation satellites, leading to the development of satellitar image time series (SITS), the number of algorithms for land cover analysis and monitoring has greatly expanded. This paper offers a new perspective in dynamic classification for SITS. Four similarity measures (correlation coefficient, Kullback-Leibler (KL) divergence, conditional information, normalized compression distance (NCD)) based on image pairs from the data are employed, resulting in a series of maps describing different types of changes observed in the original series. The proposed algorithm performs a classification of the newly developed time series using a Latent Dirichlet Allocation model (LDA). This statistical method was originally used for text classification, thus requiring a word, document, corpus analogy with the elements inside the image. The experimental results were computed using 11 Landsat images over the city of Bucharest and surrounding areas.

The SITSMining Framework - A Data Mining Approach for Satellite Image Time Series

“…Therefore, a time series is obtained for each pixel, such that each data point corresponds to the pixel value in one image of the SITS. We can thus grasp the information related to the behavior of each area in the images along time, and analyze it using data mining techniques, such as clustering (Kyrgyzov et al, 2007) and classification (Vaduva et al, 2011). The analysis of SITS using data mining is useful in agriculture, for example, for crops monitoring along seasons (Julea et al, 2011).…”