Sequential classification of active sonar returns from systematically distributed objects

Active sonar returns as seen on a range-bearing display often show objects distributed in range and bearing. Objects may also be distributed and depth but these show up as being distributed in range and bearing unless depth can be estimated with sufficient accuracy. Examples of distributed objects are man made underwater structures, underwater geological features, and schools of fish to name a few. Even for a point object multipath will cause the object to appear distributed in range. When these returns represent similar information, for example, if they are noisy versions of each other, they can be combined to enhanced classification. But the information can be combined in many ways. Four methods are considered to combine information. The first method is direct averaging of the input data vectors from the multiple objects. This assumes that the data are noisy versions of each other. The second method sums the loglikelihood ratios (LLR) of the multiple objects. This assumes that the LLR´s are noisy versions of each other. The third method forms the union of the loglikelihood ratios of the multiple objects. In this method only one of the objects need satisfy the classification criteria. In the fourth method which is the most stringent the intersection of the loglikelihood ratios of the multiple objects is formed. This method requires that all of the objects must satisfy the classification criteria. In the paper the four methods will be compared with the same data set