My scientific activity is related to remote sensing image processing.
In the graduation diploma I developed methods to simulate SAR intensity images in ground and radar geometry and to detect zones with irreversible distortions.
During the MSc period my first main contribution was the introduction, in 2006, of Data Mining techniques for Frequent Sequential Pattern extraction from Satellite Image Time Series (SITS). The pixel value evolution was the criteria for characterization, discrimination and identification of terrestrial entities.
My second important contribution is made during the PhD when the spatial aspect of SITS was re-considered and new extraction algorithms based on connectivity constraints were developed. The 8NN connectivity measures offer information not only about the objects forming tendency but also about the stratified structure of extracted patterns and can be used for shape recognition. The extraction process becomes more efficient by the active implementation (push) of antimonotone connectivity constraints.
The unsupervised approach makes equilibrium in exploiting the 3 pixel dimensions (radiometric, temporal and spatial) and follows the stages of a KDD process. It implies the preprocessing methods for re-quantization of pixel values and post-processing methods for spatial and temporal localization of extracted patterns and their spatial regularization and fusion. I contributed to the development, testing and documentation of SPATPAM software (www.efidir.fr) dedicated to sequential pattern extraction under constraints.
The extracted Grouped Frequent Sequential Patterns proved to be useful for a preliminary and synthetic characterization of SITS dynamics. This fact explains the wide range of applications from multispectral (visible and infrared) to radar data (amplitude, coherence, interferometric and polarimetric) obtaining information about meteorological behavior and monitoring of glacier displacement, crustal deformation and crop evolution.
In the graduation diploma I developed methods to simulate SAR intensity images in ground and radar geometry and to detect zones with irreversible distortions.
During the MSc period my first main contribution was the introduction, in 2006, of Data Mining techniques for Frequent Sequential Pattern extraction from Satellite Image Time Series (SITS). The pixel value evolution was the criteria for characterization, discrimination and identification of terrestrial entities.
My second important contribution is made during the PhD when the spatial aspect of SITS was re-considered and new extraction algorithms based on connectivity constraints were developed. The 8NN connectivity measures offer information not only about the objects forming tendency but also about the stratified structure of extracted patterns and can be used for shape recognition. The extraction process becomes more efficient by the active implementation (push) of antimonotone connectivity constraints.
The unsupervised approach makes equilibrium in exploiting the 3 pixel dimensions (radiometric, temporal and spatial) and follows the stages of a KDD process. It implies the preprocessing methods for re-quantization of pixel values and post-processing methods for spatial and temporal localization of extracted patterns and their spatial regularization and fusion. I contributed to the development, testing and documentation of SPATPAM software (www.efidir.fr) dedicated to sequential pattern extraction under constraints.
The extracted Grouped Frequent Sequential Patterns proved to be useful for a preliminary and synthetic characterization of SITS dynamics. This fact explains the wide range of applications from multispectral (visible and infrared) to radar data (amplitude, coherence, interferometric and polarimetric) obtaining information about meteorological behavior and monitoring of glacier displacement, crustal deformation and crop evolution.
View contact details
