Determination of the optimal measurement sequence for a noisy pattern recognition system via the discrete maximum principle
✍ Scribed by M.H. Hammond Jr.; H.R. Weed
- Publisher
- Elsevier Science
- Year
- 1970
- Tongue
- English
- Weight
- 842 KB
- Volume
- 6
- Category
- Article
- ISSN
- 0005-1098
No coin nor oath required. For personal study only.
✦ Synopsis
An advanced procedure has been developed which will enable a decision maker to select the best sequence of making observations; the power of optimal control theory is utilized in still another field.
Sumnmry--The discrete Pontryagin Maximum Principle is applied to the problem of finding the best sequence of measurements of particular features in a pattern recognition system containing noise. The best sequence is defined as the one which minimizes a performance functional composed of the costs incurred in taking the measurements, the penalties from incorrect decisions, and a measure of the progress toward reaching the decision.
The approach places the problem in the format of a stochastic finite state system where each state corresponds to a set of quantized probabilities that the unknown pattern is in each of the possible pattern classes. State transitions are assumed to occur at discrete intervals and the corresponding transition equation is expressed as a four dimensional tensor equation developed from Bayes' Theorem. The discrete Maximum Principle is shown to be applicable after certain requirements are established concerning the changing region from which the control variable, i.e. the kind of measurement, is chosen. This changing region is found to be dependent upon the kinds of measurements previously used.
Compared to an exhaustive search technique, this new approach is shown to greatly reduce the number of basic calculations necessary to find the best sequence. *