A SOLSA BRGM team took the opportunity to participate in the SiDoS workshop (Massive Sequential Data Similarity: definition, calculation and optimization).
SiDoS is the first workshop to focus on the optimization of distance computation on large volumes of sequence data. This theme is at the crossroads of HPC (High Performance Computing) and data analysis and mining. The aim of the workshop is to structure the French community dealing with massive sequences and their similarity computation.
SOLSA participation
Oral pesentation by Nathan Bodereau and Théophile Lohier from BRGM "Optimization of high-resolution hyperspectral data processing for the description of drill cores"
Abstract: Exploring ground resources has become challenging as the need for metalliferous raw materials or for characterisation of underground pollution has increased. Hyperspectral imaging of drill cores has been gaining popularity, especially in mining field, as it allows a fast and reliable estimation of mineral distribution. However, with the increasing resolution of hyperspectral images, the size of the datasets (> 1 Tb) to process is exploding, impairing the capacity of the mineral mapping algorithms to work in near real time. To ensure runtimes compliant with the operational requirements, we develop a framework allowing to reduce the computational cost of hyperspectral images processing. It embeds different algorithms for hyperspectral data pre-processing, spectral and spatial reduction as well as mineral mapping. These last two treatments heavily rely on similarity metrics. We demonstrate the capabilities of this framework by processing a 10-meter core sampled from the closed tin mining site of Abbaretz (France).