Multispectral Visual Incremental Knowledge Assimilation System (MUVIKAS)

Automated Target Recognition (ATR) technologies offer potential for automated detection and recognition of targets of interest in imagery data with enhanced accuracy. Most existing ATR algorithms are trained on fixed datasets and cannot change during deployment. As a consequence, these ATR systems are likely to have degraded performance when deployed at unseen environments that are not covered by the training data. Whenever new data or new classes of targets need to be added, the baseline approach is to retrain the ATR system from scratch. Such offline retraining usually demands significant amount of computations which causes operational downtime, and is not suitable for online during-mission analysis. Current systems also lack an efficient framework to interact with operator for online learning and adapting to new environments.In this project, UtopiaCompression Corp. (UC) proposes to build a novel MUlti-spectral Visual Incremental Knowledge Assimilation System (MUVIKAS) as a software program that facilitates in-situ target detection and classification in multi-spectral imagery with human in the loop. The system self-adapts to varying scene backgrounds and to inputs from human operator, based on a novel incremental ATR algorithm that perpetually assimilates new and relevant information into the existing knowledge database in an incremental fashion.