A MULTISTRATEGY LEARNING APPROACH FOR TARGET MODEL RECOGNITION, ACQUISITION, AND REFINEMENT

摘要

Target recognition systems are currently unable to modify their behavior automatically in environments where processing requirements change or novel situations are encountered. Most systems can not easily adapt to varying target appearances, considerable image noise, and target occlusion. More importantly, these systems are constrained by the selection of target models used for recognition; typically, the target model database is fixed and individual features within a target model remain static as well. The incorporation of machine learning technology into the target recognition process will allow the system to use situation context, to adapt in changing environments, and to improve the system's performance over time. This work describes an innovative approach which combines machine learning and target recognition into an integrated system. The system is called TRIPLE: Target Recognition Incorporating Positive Learning Expertise. It uses two machine learning techniques known as explanation-based learning and structured conceptual clustering, combined in a synergistic manner, which provide effective target model recognition, acquisition, and refinement capabilities. We provide an overview of the TRIPLE system and provide experimental results which illustrates the performance of the system.