Detect and Identify Automated Labeler for Radar (DIALR)

Detect and Identify Automated Labeler for Radar (DIALR)  Prime Solutions Group, Inc. (PSG) proposes Detect and Identify Automated Labeler for Radar (DIALR) for Research and Development (R&D) and demonstration of an algorithm framework for automated labeling of target classes in radar imagery. This provides a baseline for automatically labeling multiple target classes in Synthetic Aperture Radar (SAR) imagery. Our proposal includes R&D in semi-supervised learning (SSL) for object classification and a design framework that seamlessly ties into NGA-Research’s (NGA-R) Research Data Environment (RDE) system.     A challenge for NGA and the Intelligence Community (IC) is the human-in-the-loop timeline associated with exploitation, analytics, and data curation, especially when managing the huge amounts of collected imagery from current systems. The volumes of collected Synthetic Aperture Radar (SAR) imagery represents an opportunity for developing technology to improve workflows and utilize data content in more efficient ways. Advancements in machine learning and deep learning (ML/DL) technology offers a platform to streamline the modeling value chain, particularly in data labeling.    The overarching objective of this Phase I work is to research, demonstrate, and characterize performance of an automated labeling algorithm that can label multiple classes of targets without a human in-the-loop. A critical challenge is associated with multiple targets in the imagery as it impacts automated labeling. Our research will focus on novel semi-supervised learning (SSL) algorithms that address this challenge. The algorithms are selected to maximize labeling accuracy and minimize bias to minimize the effect of false negative data. PSG has significant experience and multiple publications in this area which offers the advantage of a quick ramp-up and a greater probability of success than would otherwise be the case.    NGA-R is developing the RDE for improved pipeline management of labeled SAR data. This provides a web-based application to enable greater data utility and offer improved user workflow efficiency for the modeling mission. As the RDE developer, PSG is working with NGA-R and other Recon performers to develop the Extensible Markup Language (XML) structure for single and multi-object metadata for labeled SAR data. The XML is the principal interface between the labeling function and RDE. It carries the critical support information for labeled object data so that it is queryable and discoverable by many parameters including label taxonomy. Thus, users of RDE can leverage this valuable metadata as they create training, testing, and validation data for their models. PSG’s DIALR program also addresses the interface between labeling and RDE to offer a seamless functionality in NGA’s modeling value chain.