Correlation Filter Synthesis Using Neural Networks.
Abstract
Excellent results were obtained using neural networks to synthesize filters for optical correlators, including filters for both cluttered backgrounds and target rotation angles not used in training. The most significant results employed new stretch and hammer neural networks which constitute an important and enduring advance because they train with guaranteed upper bounds on computational effort and generalize with guaranteed lower bounds on smoothness and stability. These results indicate good prospects for training neural networks to synthesize filters for a wide range of target distortions, and this approach has clear advantages compared to searching stored filters. Neural networks, Optical pattern recognition, Optimizing algorithms, Target recognition.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1993
- Accession Number
- ADA278124
Entities
People
- David L. Flannery
- Steven C. Gustafson
Organizations
- University of Dayton