On Wednesday, October 10, 2018, Hoang Viet Tuan NGUYEN, PhD student in Information and Communication Sciences and Technologies (ICT) Computer Science at the Laboratory of Computer Science, Systems, Information and Knowledge Processing (LISTIC), will defend his thesis entitled " Taking data quality into account when extracting and selecting changes in time series of displacement fields in satellite imagery."
The defense will take place at 11 a.m. in lecture hall 052 at IAESavoie Mont Blanc, located at 4 chemin de Bellevue, Annecy-le-Vieux 74944 Annecy.
Thesis summary
This thesis deals with the discovery of knowledge from Time Series of Displacement Fields (STCD) obtained by satellite imagery. Such series now play a central role in the study and monitoring of natural phenomena such as earthquakes, volcanic eruptions, and glacier movement. These series are rich in both spatial and temporal information and can now be produced regularly at low cost thanks to space programs such as the European Copernicus program and its flagship Sentinel satellites. Our proposals are based on the extraction of Sequential Frequent Grouped (SFG) patterns. These patterns, originally defined for knowledge extraction from Satellite Image Time Series (STIS), have shown their potential in initial work aimed at analyzing a STCD. However, they do not allow the use of confidence indices intrinsic to STCDs, and the swap randomization method used to select the most promising patterns does not take into account their spatio-temporal complementarities, as each pattern is evaluated individually. Our contribution is therefore twofold. A first proposal aims to associate a reliability measure with each pattern using confidence indices. This measure makes it possible to select patterns supported by data that are sufficiently reliable on average. We propose a corresponding algorithm to perform extractions under reliability constraints. This algorithm is based in particular on an efficient search for the most reliable occurrences using dynamic programming and on pruning the search space using a partial push strategy, which makes it possible to consider substantial STCDs. This new method has been implemented on the basis of the existing SITS-P2miner prototype, developed within LISTIC and LIRIS to extract and classify SFG patterns. A second contribution aimed at selecting the most promising patterns is also presented. Based on an informational criterion, it takes into account both confidence indices and the way in which patterns complement each other spatially and temporally. To do this, confidence indices are interpreted as probabilities, and STCDs as probabilistic databases with only partial distributions. The informational gain associated with a pattern is then defined according to the ability of its occurrences to complement/refine the distributions characterizing the data. On this basis, a heuristic is proposed to select informative and complementary patterns. This method provides a set of patterns that are less redundant and therefore easier to interpret than those provided by swap randomization. It has been implemented in a dedicated prototype. Both proposals are evaluated quantitatively and qualitatively using a reference STCD covering Greenland's glaciers, constructed from Landsat optical data. Another STCD that we constructed from TerraSAR-X radar data covering the Mont Blanc massif is also used. In addition to being constructed from different remote sensing data and techniques, these series differ drastically in terms of confidence indices, with the series covering the Mont Blanc massif having very low confidence levels. For both STCDs, the proposed methods were implemented under standard conditions in terms of resource consumption (time, space), and the experts' knowledge of the areas studied was confirmed and supplemented.