First International Workshop on “Temporal Analysis of Satellite Images”, EARSEL, Mykonos, Greece, May 2012

I will present an oral presentation at 1st International Workshop on Temporal Analysis of satellite images, EARSeL.

My presentation will show results of a new automatic method which permit to select dataset of images (date selection), classify the best datasets and fuse these classification results to improve quality.A lot of tests are presented to validate the process especially for different sensors like Spot, Formosat or Landsat.

You can find the abstract : Hyper temporal dataset selection of multisource remote sensing images for classification improvement.

Coming soon : Sentinel-2 Preparatory Symposium (ESA), Frascati, Italia, April 2012

“The European Space Agency is organising the first “Sentinel-2 Preparatory Symposium”, which will be hosted in ESA-ESRIN, Frascati, Italy, from 23 to 27 April 2012.”

In context of Land use and Cover change (LUCC), we will present a poster on “Estimation of carbon capture and storage from LUCC. We compute land use change maps in the southwest of Toulouse (France), from satellite classifications during 2002 to 2010. This study was conducted in two parts: (a) changes in land use between 2002 and 2009; (b) crop rotations year by year. Tests are done on Formosat-2 images and in progress on Spot2-4-5 images.

The methodology consists in (1) classification to obtain land cover maps for each year (2); regrouping the main land cover theme in the classification result (Woodland, Water, Built up area / mineral surfaces, Grassland, Crops; (3) regularization (smoothing) the classified images by using segmentation; (4) contour erosion: to compare classified images, it is necessary to eliminate mixed pixels (pixels on border of two classes) which correspond to false changes and imply change errors in statistical results; (5) comparison of classified images from change matrices: change-classes are created; (6) Crop rotation: comparison of N classifications in N years and obtaining change matrices and change image; (7) the soil carbon stocks are estimated using a mathematical model (INRA, 2002) from statistics obtained by the land change map.
Land use change maps for the main classes (forest, croplands and pastures) are used to estimate Soil Organic Carbon (SOC) changes by applying a simple model based on regional observations (Soussana and Arrouays 2004). Crop rotation maps will be employed to refine this approach. Regional yield statistics will be computed to estimate Net Primary Production (NPP) and crop carbon balance by using a similar methodology as proposed by West et al. (2010).

Poster presentation to SPIE Defense+Sensing 2012

Poster presentation at SPIE conference in “Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery”. The paper is named “New feature for evaluation of supervised classification” and will be presented at Baltimore, Maryland, USA the 24th of April 2012.

The paper will describe all the quality indices which could be used to evaluate classification quality with pros and cons of every method and propose a new one to evaluate new characteristics of classification images.

QT

I’m using QT platform for my c++ programs developpement, it’s very powerful, multi-platform, good debuger and a memory analyser. Since now, i do not have  enought time to try ihm part of the platform, but it seems great (kde is based on qt). A huge documentation is available, very simple to use and i hope next upgrades keep the same way. For information, i use it on Ubuntu 11.04 (64bits) and windows Seven (64bits), no problems !

Hello world!

Welcome every body on my new blog !

“Ceci” sera mon journal de bord relatant mes travaux actuels et futurs, toutes suggestions étant la bienvenue cela va sans dire. J’essayerai au maximum de tenir cette page à jour, d’y inclure le plus de liens possibles et variés.