Robust Maritime Target Recognition

Aretķ proposes a fully functional, robust maritime target recognition and classification capability for airborne EO/IR sensor systems. The architecture comprises 3 basic components: a segmentation process to transform the image chips into silhouettes suitable for template-based ATR, a hierarchical ATR classifier culminating with a matched filter, and the Classifier Database Manager (CDM) to maintain, refine and update the template database. The ATR foundation is the hierarchical classifier utilizing a scheme developed for periscope applications that proved successful for basic, gross and fine classification hierarchies with probability of correct classification success ranging from 90-99% depending on the number of pixels on target and the class hierarchy. When classified correctly, the range and Angle Off Bow accuracies exceed the specified requirements of 3% and 2░, respectively. AI/ML classifiers will be assessed for addition to the ATR framework to potentially form a hybrid classifier. CDM framework leverages deep learning methodologies developed in 2 ongoing Aretķ projects. Using target images, it maintains and refines the template database and incorporates agile procedures to add new target classes.