STATISTICAL DECISION THEORY OF UNDERWATER TARGET CLASSIFICATION - 'SCORE'.
Rept. no. 1 (Final),
QUEENS COLL FLUSHING N Y DEPT OF PHYSICS
Pagination or Media Count:
Basic information theory and decision theory is reviewed. It is then applied to the classification situation by assuming a constant rectangular probability density distribution of the rejection hypothesis. This permits the incorrect and correct acceptance probabilities and the incorrect and correct rejection probabilities to be expressed in terms of a decision parameter which depends on the threshold, the probability density level of the rejection hypothesis and on the variance of the acceptance hypothesis. Sets of calculated curves of error probabilities as functions of the available data types are plotted against the decision parameter. The maximum value, average value, and expectation value of the likelihood ratio are obtained and plotted as functions of the decision parameter. The information increase is obtained similarly. These functions are then extended to non-constant rectangular probability density distributions of the rejection hypothesis. The above theory is illustrated by a series of simulated submarine classification examples in which the effect of deteriorating data - missing types of data - is shown. These illustrative examples show that the correct decision is maintained in spite of the deterioration of data.
- Acoustic Detection and Detectors