Parallel Computing Applied to Geographic Information Systems

GIS-PAC - ITDC-135

Keywords: parallel processing, GIS, performance evaluation, optimisation, neural networks, pattern recognition

Start Date: 1 February 95 / Duration: 30 Months

Objectivies and Approach

The objectives of the project are two-fold: theoretical and practical. The theoretical aspect is to deepen and generalise the design of an efficient and performing algorithms parallelisation methodology. This methodology is based on algorithms task decomposition and inter-task precedence analysis. The second objective consists in applying this methodology on a practical case study, i.e., the design of a vectorisation and shape recognition tool in geographic information systems (GIS) by using neural networks (NN) concepts.

Progress and Results

Choice of an adequate NN architecture for pattern recognition of cartographic data and implementation on a TN 310 transputer based machine.
Parallelization of thinning algorithms on a TN 310.
Design of efficient parallel linear algebra algorithms and performance evaluation on a TN 310 MIMD machine.
Design of efficient parallel recursive matrix multiplication and inversion algorithms. Performance evaluation on a TN 310 MIMD machine.

Information Dissemination Activities and/or Exploitation

Until now, the dissemination activities include the supervising of the projects of three students at Faculté des Sciences de Tunis, one presentation during the Information Day on CEE-IT projects held in June 1996 at Tunis and one communication in FRI'96 (Forum de la Recherche en Informatique) held in July 1996 at Tunis.

Coordinator

Faculté des Sciences de Tunis (FST)
Belvédère 1060, TN

EU Partners

Telmat Multinode, F

Non-EU Partners

Faculté des Sciences de Tunis, TN
Institut Regional des Sciences Informatiques et
des Télécommunications (IRSIT), TN

CONTACT POINT

Dr. Zaher Mahjoub
Tel: +216 1 512 600
Fax: +216 1 885 190