Optic Associated Team

 

OPTIC: Optimal Inference in Complex and Turbulent data.

 

 

The OptIC associated team focuses on nonlinear signal processing for universe sciences, with a strong emphasis on data fusion in earth observation and monitoring. The extension and development of a strong collaboration between the Inria's GEOSTAT team and the Indian Institute of Technology (IIT) Roorkee, department of electronics and computer engineering (Prof. D. Singh's group), complements the expertise in universe sciences and earth observation of partners (ONERA, CNRS) that are already involved in research actions with GEOSTAT.

Nonlinear physics puts strong evidence of the fundamental role played by multiscale hierarchies in complex and turbulent data: in these data, the information content is statistically localized in geometrical arrangements in the signal's domain, while such geometrical organization is not attainable by classical methods in linear signal processing. This is one of the major drawbacks in the analysis of complex and turbulent signals. The goal of this associated team is to show that inference of physical variables along the scales of complex and turbulent signals can be performed through optimal multiresolution analysis performed on nonlinear features and data extracted from the signals, resulting in novel and powerful approaches for data analysis and fusion between different acquisitions (in temporal/spatial/spectral resolutions). This program needs both strong expertise in the physical processes beyond the acquisitions and the application of nonlinear physics ideas on the behavior of the acquired physical phenomena.

Our teams focus on five different aspects of image and signal analysis for remote sensing:

  • Feature detection, together with the spatial, data and temporal scales at which these features are defined. Standard edge detection routines are not adapted for objects like oceanic currents which present evanescent boundaries. However, these usually present a definite pattern across multiple scales in both the spatial (e.g. between 50 and 100km) and data domains (e.g. sea surface temperature difference the order of a few °C). The GEOSTAT team is expert in multi-scale non-linear analysis. We are building on this experience to define innovative feature descriptors, in order to better analyse remotely sensed data.
  • Data Fusion, of multispectral, SAR, and signals acquired at different spatial resolutions (e.g. floaters, altimetry) and having different spectral characteristics (e.g. chlorophyll concentration CHL_a). The IIT team is expert in leveraging the physics behind each type of signal, in order to provide the most informative maps for a given objective (e.g. tracking underground coal mine fires). Our data fusion methodology is thus driven by applications, enhancing some targeted feature for classification, or a better monitoring of global changes of natural processes (e.g. oceanic currents, partial pressure pCO2 fluxes).
  • Super-resolution. The public availability of low-resolution MODIS data is cost-effective, but limited in precision. Some applications, such as monitoring drought, must operate on objects smaller than provided in the freely available data. One way to compensate this lack of resolution is to unmix the content of each given pixel. This can be done based on both a physical model, shared information between each data source, and accounting for the spatial structure around each pixel (e.g. edges). We rely on field measurements in order to provide the ground truth for a corpus of well-registered locations, which together encompass a wide variety of objects (e.g. urban, crops, etc). We then train our super-resolution algorithms, and quantitatively assess our super-resolved maps on how well they improve the performance of the final applications (e.g. classification of land usage). A similar methodology, using the location of tracked floaters, allowed us to validate super-resolved ocean dynamics from altimetry and sea surface temperature data.
  • Classification. We use all the standard indicators (e.g. Normalized Difference Vegetation Index) but, thanks to the above super-resolution, data fusion and feature detection approaches, we aim at producing better feature descriptors for the final applications. Our main projects concern land use monitoring, agricultural applications and ocean/atmosphere couplings.
  • Optimal data representations, by means of an adapted wavelet basis, specific to the data set under consideration. We use sparse representations based on dictionaries generated from optimal wavelets and more classical matching pursuit algorithms, in order to reconstruct complex signals. Our main application consists of the use of optimal wavelet decomposition for non-linear reconstruction of turbulent phase in adaptive optics.

 

Key participants (France)

 

 

Key participants (India)