Mixed-Variable Tuning with Particle Swarm Optimization

This work builds an implementation of particle swarm optimization that can be used to tune hyperparameters, such as those required to conFigure detection and classification algorithms for automatic target recognition. The proposed optimization method must include the capability to optimize over sets of parameters with diverse types. This is done by incorporating several previous developments from particle swarm optimization research into a single polymorphic and extensible framework. The method is tested by application to a set of standard test cases, and proof-of-concept for ATR algorithm parameter tuning is demonstrated for a detection algorithm on a set of dual-band synthetic aperture sonar data.