Pattern Recognition

Current research projects:

  • Moving Object Recognition Using DVS and NeuCube
  • Novel Methods for Radio Astronomy
  • Environmental Event Prediction (e.g. risk of earthquakes)

Methods for computer vision and image processing of dynamic objects

This project will develop a new method and an engineering system for fast video and image scene data analysis, including neuromorphic computation with dynamic vision sensors (DVS, also called artificial retina) and neuromorphic chips, both produced at INI/ETH Zurich.

Project Team

  • Prof. N. Kasabov
  • A/Prof. W. Yan
  • Prof. R. Klette (The University of Auckland)
  • A/Prof. D. Bailey (Massey University)
  • Prof. G. Indiveri (University of Zurich and ETH Zurich, Switzerland)
  • W. Cui
  • F. Alvi

Big data in radio-astronomy and pulsar recognition

This project will continue to look into the problem of complex pattern recognition from large and fast radio- astronomy data, particularly for the identification of dispersed transient and pulsar signatures, using evolving neuromorphic systems. We propose to apply techniques of neuromorphic systems, most particularly the NeuCube (Radioastronomy variant) architecture for spectro-temporal data modelling based on SNN. First experiments show promising results with very low computational cost. This is especially true when paired with a dedicated neuromorphic hardware platform, such as the SpiNNaker. With further development, the NeuCube has the potential to significantly improve the state-of-the-art in dispersed transient and pulsar search, and become a core technology in projects like the SKA.

Project Team

  • Prof. N. Kasabov
  • Prof. S.Gulyaev
  • A/Prof. B. Pfahringer (The University of Waikato)
  • Dr. G. Dobbie (The University of Auckland)
  • Prof. B. Stappers (The University of Manchester)
  • N. Scott
  • M. Mahmood

Predicting risk of earthquakes in New Zealand: dream or reality?

Predicting risk of earthquakes is a very challenging task, but our hypothesis is that with the SNN NeuCube framework it is possible to achieve it with a high accuracy subject to sufficient spatiotemporal data is available on-line. Preliminary experiments on small data from GeoNet showed promising results, but this project will develop a new model that treats for the first time the problem in its entirety, using all available fast and large stream data in an on-line mode (from the GeoNet site).

Project Team

  • Prof. N. Kasabov
  • A/Prof. R. Pears
  • R. Hartono