Comparative evaluation of stored-pattern classifiers for radar aircraft identification
Abstract
Classifier performance is evaluated in terms of misclassification probability, classification bias, computational economy, and the effect of changes in classifier parameters. These parameters include dimensionality, assumed value of the noise level, and the number of subclasses in each class. The problem of classifier design when more than one noisy pattern can be used to make a decision is investigated. This is the problem of efficient utilization of cumulative evidence to improve classifier performance. One of the procedures evaluated operates by taking a majority vote over multiple decisions of a given classifier. It is shown that the misclassification probability of a majority vote procedure increases monotonically with the classification-bias of the given classifier.
- Publication:
-
NASA STI/Recon Technical Report N
- Pub Date:
- 1976
- Bibcode:
- 1976STIN...7713277S
- Keywords:
-
- Aircraft Detection;
- Classifiers;
- Optical Transfer Function;
- Pattern Recognition;
- Radar Detection;
- Computer Aided Design;
- Decision Making;
- Noise Reduction;
- Communications and Radar