Automated Generation of Electronic Warfare Libraries Navy SBIR FY2013.1 Sol No.: Navy SBIR FY2013.1 Topic No.: N131-036 Topic Title: Automated Generation of Electronic Warfare Libraries Proposal No.: N131-036-0948 Firm: Lakota Technical Solutions, Inc. PO Box 2309 Columbia, Maryland 21045 Contact: William Farrell Phone: (410) 381-9780 Web Site: www.lakota-tsi.com Abstract: Metrological differences between tactical sensors result in a wide range of accuracy, completeness, and correctness of features used for object classification. Thus, it is typical practice for Subject Matter Experts (SMEs) to develop the required reference libraries so that they are tailored to each tactical sensor type and, in some cases, sensing environments. The process of developing this tailored library is often called "coloring" the reference library. The SMEs objective is to optimize the classification performance (w.r.t. some metric(s)) using a feature-based classifier. The proposed Hierarchical Emitter Library Optimization (HELO) technology mimics the coloring process in order to generate an Emitter Library that optimizes the classifier performance while avoiding the need for a labor-intensive manual process that requires SME knowledge. HELO employs a general Hierarchical Evolutionary Programming (H-EP) based upon the Genetic Programming (GP) optimization paradigm to achieve this optimization. This approach provides a computationally scalable process that rigorously quantifies the performance of the classification algorithms without knowledge of its algorithms. Using the performance assessment of the classifier, a large set of potential emitter libraries (population) is iteratively refined (evolved) until an optimal (or sufficiently good) emitter library is generated. The solution is hierarchical because the evolution of the population is achieved jointly in two stages: (1) evolution of the coloring functions to generate a parameter library for a particular parameter type and (2) evolution of the set of parameters used by the classifier. Benefits: This H-EP approach is beneficial because it generically applies to the optimization of any feature-based classifier that relies on a reference library for feature correlation. This is accomplished by employing a fitness function that assesses the performance of the classifier against candidate reference libraries without knowledge of the classifier algorithm(s). This "black box" approach means that the H-EP approach can be quickly adapted to a wide range of classification problem domains. In addition, the H-EP avoids the need to have SME operators develop the reference libraries and allows for repeatable/comparable optimization results across users with varying levels of expertise. Return