CSS Brown Yale Boston
darpa
An Integrated Approach to Discrimination and Target Recognition
onr

Executive Summary

Our program integrates several key components of a pattern recognition system in an attempt to address the needs of challenging classification and detection problems. We intend to optimize concurrently key components such as signal representation and classification scheme so as to achieve optimal performance. In particular, we encorporate wavelet dictionaries and dimensionality reduction methods as well as sensor fusion at various levels in high dimensional real-world problems.

Goals

  • Construct a multi-level classification system that includes: signal preprocessing and enhancement using wavelets, wavelet dictionaries, partial differential equations and classical computer vision methods.
  • Introduce various feature extraction and detection methods, optimized for different sensors. These include robust linear and non-linear methods, different ways to find best basis representations and the design of optimal mother wavelets for discrimination.
  • Apply sensor fusion at various levels. The key relevant observation is that ensembles of experts improve their overall performance if the errors they make are independent.
  • Attempt to maximally exploit multi-sensor and multi-scale data as well as ensemble of independent experts.
  • Develop algorithms for fully automated detection and classification (DC) of sea mines in the difficult littoral region from acoustic and magnetic sensors. A special focus is on low target strength bottom mines that may be partially or fully buried.

    • Technical Contact

    Dr. Nathan Intrator
    Physics Department, Box 1843, Brown University
    Providence, RI 02912
    Phone   (401) 863-2187,   Sec.   (401) 863-2585
    Fax   (401) 863-3494
    Email   nin@cns.brown.edu
    Web Page   Brown   Mirror

    The Program Managers for this program are:

    Raphi Coifman     Quyen H. Huynh     Nathan Intrator     Truong Nguyen