This work provides a new, to the best of our knowledge, approach to
constructing linear models for object detection in a scene.
Specifically, we use representative training data in order to estimate
the parameters describing a generalized wavelet model for the express
purpose of detecting the presence of maritime targets in a scene. The
parameter estimates are taken as those that maximize the probability
of detecting the targets for a fixed probability of false alarm. The
approach is then demonstrated on a database of short-wave infrared
imagery containing various watercraft. Results are then compared to
some of the more standard wavelet bases used in detection
applications.