Pattern Recognition Device with Adaptive Preprocessing and Classification.

Abstract

A program to investigate the application of bionics techniques to the solution of aerial photography pattern recognition problems has been completed. The investigation included development of a unique circuit which will have application in pattern recognition and trainable system. Called an adaptive filter, this circuit is a band-pass filter which can be trained to adjust its center frequency in accordance with externally generated training signals. A bank of these filters was incorporated in the preprocessing stage of a breadboard pattern recognition system, and served to extract frequency information from an input video signal. Performance of this system was evaluated for the problem of terrain recognition from single-frame panchromatic aerial photography. The photography was converted to a suitable time-varying video signal by a flying-spot scanner. A training strategy enabled the system to refine its processing operations in order to improve classification performance with time. Successful performance of the system was demonstrated in a series of classification tests involving over 1900 terrain samples. Recognition accuracies ranged from 64.5 percent to 95 percent.

Document Details

Document Type
Technical Report
Publication Date
Sep 01, 1970
Accession Number
AD0714576

Entities

People

  • J. M. Idelsohn

Tags

DTIC Thesaurus Topics

  • Adaptive Filters
  • Aerial Photography
  • Classification
  • Filters
  • Flying Spot Scanners
  • Frequency
  • Pattern Recognition
  • Photographic Equipment
  • Photographic Materials
  • Photographic Recording Media
  • Photography
  • Preprocessing
  • Recognition
  • Training
  • Video
  • Video Signals

Fields of Study

  • Engineering

Readers

  • Computer Science/Computer Engineering/Data Science/Digital Signal Processing.
  • Computer Vision.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • Biotechnology