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

This page provides information regarding ATR effort under a DARPA  / ONR contract. The group includes Raphi Coifman from Yale, Quyen Huynh from CSS, Nathan Intrator (P.I.) from Brown and Tel-Aviv University and Truong Nguyen from Boston University.

July 2001:    PowerPoint Slides (Updated July 23, 01) 
April 2001:    Supporting Material  Supporting slides
November 2000:    Waveform Design and Decomposition for Biosonar   Past Work  
Aug/Sep 2000:    Broad Band Sonar    Quad Chart Summary
April 2000:    Presentation  html version  Poster  Supporting Material: 1  2 
June 1999:    Executive Summary    Summary Slide     Progress Report    Publications


  • Overview: Network Architectures
  • Hybrid Supervised/Unsupervised Training
  • The integrated classification machine
  • Overview: Wavelet Feature Extraction (Quadratic Discrimination)
  • Methodology

  • Sonar Image Acquisition
  • Preprocessing and Detection
  • Image Wavelet denoising
  • Multiscale Matched Filter
  • Post-processing at a Multiscale level
  • Hybrid Supervised/Unsupervised Training
  • The integrated classification machine
  • Dolphin Sonar links

    POD Dolphin click sounds recording
    Bat biosonar demo
    Sonar frequencies
    1998 ONR Workshop
    General info and links   Navy Related


    Side Scan Sonar Mine Detection
    Back Scatter Sonar
    6.2, 6.3 Transitions

    Related sites

    June_98 Contractors meeting
    Fast Mathematical Algorithms
    Signal Processing workshop (Feb 9-11)
    CSS Panama City  Publications
    CIS Washington University
    ONR  ONR/University 

    Side Scan Sonar Mine Detection

    The current data base consists of a 60-image set from a side-scan sonar (SSS0) collected at the Naval Surface Warfare Center (NSWC). They are encoded as 8-bit gray scale images, 1024 range cells by 511 cross-range cells. The 60 images contain 33 targets; some contain more than one target while others contain no targets. Non-target objects which look as targets appear throughout the images. A typical mine-like target consists of a strong highlight on its left side and a long shadow down range on its right side. Unfortunately the presence of clutter can mask this structure.

    A recent report describes some effort in wavelet denoising of these images.
    To optain the side-scan sonar or the backscatter data please contact Dr. Quyen Huynh 850-234-4158, or Dr. Edward Linsenmeyer 850-234-4161.

    Some images can be viewd and down-loaded here.

    Back Scatter Sonar

    This application involves an active backscatter data set of mine-like objects. The data was collected at the Naval Surface Warfare Center (NSWC) by Gerald Dobeck. The task is to distinguish between man-made and non-man-made objects. There are six objects in the data: metal cylinder, cone-shaped plastic object, water-filled barrel, limestone rock, granite rock, and a water-logged wooden log. The data-set contained seven different frequency bands, however in this preliminary study, only one frequency band, an FM sweep between 20 to 60 kHz was used. The targets were suspended in a large water tank, while cylindrical objects were suspended horizontally. Measurements were collected in 5 degree increments on a rotating target around a vertical axis. Every second measurement was used for testing, thus the train and test data were interleaved and both included measurements at 10 degree increments.

    A preliminary report appeared in SPIE-97. It describes some effort in wavelet denoising of these images.

    6.2, 6.3 Transitions

  • Close-Range Sensors and Autonomous Detection-Classification for Underwater Bottom Vehicles (UBVs).
    (Funded by Dr. Tom Swean, ONR code 321).
  • Broadband, Broadbeam, Biomimetic Minehunting Principles.
    (Funded by Dr. Harold Hawkins, ONR code 321)
  • Transition of Computer Aided Detection and Computer Aided Classification (CAD/CAC) Algorithms to the AQS-14 Sonar System.
    (Funded by Dr. Randy Jacobson, ONR code 321TS).
  • Recent Relevant Publications

  • Talk at the Darpa meeting Jun 23, 1998, Washington
  • Quyen Q. Huynh, Leon N Cooper, Nathan Intrator and Harel Shouval.  Classification of Underwater Mammals using Feature Extraction Based on Time-Frequency Analysis and BCM Theory IEEE Transactions On Signal Processing 46(5):1202--1207, 1998
  • N. Intrator, Q. Q. Huynh, Y. S. Wong, B. H. K. Lee.  Wavelet Feature Extraction for Discrimination TasksProceedings of The 1997 Canadian Workshop on Information Theory, Toronto June 3-4, 1997
  • N. Intrator, Q. Q. Huynh and G. Dobeck.  Feature extraction from acoustic backscattered signals using wavelet dictionariesProceedings of SPIE97, Apr., 1997.
  • Q. Q. Huynh, N. Neretti, N. Intrator and G. Dobeck.  Image enhancement for pattern recognitionUpdated Technical Report,Dec. 1998. Earlier version appeared in Proceedings of SPIE98, Apr., 1998
  • Ying-Jui Chen and Truong Q. Nguyen.  Sea Mine Detection Based on Multiresolution Analysis and Noise Whitening. Technical Report, Feb. 1999. Matlab Code
  • N. Intrator, Q. Q. Huynh & G. Dobeck.  Feature extraction and fusion of wide-band backscattered signals. Proceedings of SPIE, vol. 3710, April, 1999
  • N. Neretti, N. Intrator, Q. Q. Huynh  Mine-like Target Recognition via Expert Fusion Submitted to ICPR, December, 1999