Simplified techniques for recognition of target signatures

Practical algorithms are developed for recognizing the characteristic signature of a statistically modelled target when given a data record from a sensor observing a fixed spatial resolution element. Any one of many background objects could be present instead of the target. Since the background statistics may be poorly known, the objective is to devise an easily implemented detection method which can yield near-optimum performance without requiring a precise background model. It is shown that the simple chi-square test can meet this objective, in the sense that it should perform nearly as well as the optimum likelihood ratio test if the composite background probability density function can be adequately approximated by a constant over the detection region. This characteristic is implied when the false alarm probability can be made low. It is pointed out that the Chi-square test cannot be used to treat continuous data records. For this case, a continuous autocorrelation test is suggested which should have comparable performance. Dynamic modelling of the target stochastic process is suggested as a means for simplifying the implementation.