Selective Information Discovery and Efficient Transmission (SIDET)

The volume and frequency of information resulting from the continued proliferation of Navy platforms and sensor systems inundate human operators and make it difficult for commanders to obtain complete Recognized Maritime Picture (RMP), maintain actionable Maritime Situational Awareness (MSA), or make fully informed time-dominant decisions. Programs Management Analytics & Technologies, Inc. (PMAT) proposes the Selective Information Discovery and Efficient Transmission (SIDET) solution supporting big data analytics techniques to identify, extract, compress, and decompress critical contents from sensor imagery products, to reduce data link bandwidth requirements. SIDET will combine interest model and leading Artificial Intelligence and Machine Learning (AI/ML) frameworks including conditional Generative Adversarial Networks (cGAN) to minimize the volume of excess Imagery and Full Motion Video (FMV) data sent through the bottleneck. SIDET will also incorporate PMAT’s high-volume data ingest information management, and processing pipeline infrastructure - Extensible Collaboration and Analytics PlatformTM (X-CAPTM) - which is currently part of NAVAIR’s MINOTAUR Family of Services (MFoS). X-CAP supports autonomous extraction of critical image content from key sensors on large volumes of image data while simultaneously minimizing image size requirements for the data link that will transmit images from aircraft to off board recipients.  In Phase I, PMAT, Inc., in conjunction with CUBRC, Inc., will develop a proof-of-concept for SIDET’s core functions noted below:  (1) Compressing image data using encoder ML techniques to feed through the low-volume data link bottleneck while cropping specific entities of identified mission relevance to be preserved exactly as they are.  (2) Decompressing image data with decoder ML techniques after it passes through the bottleneck, preserving visual fidelity of key entities with no loss of information while recreating the rest of the image as best as possible. The novelty of our approach is the combination of a Naval interest model, which includes inputs from multiple sensors applied to the task of determining rules for what entities of interest are to be cropped and preserved, with a cGAN architecture that will efficiently add context to the cropped image through a process called Inpainting. This data model will allow SIDET to identify and send only the most critical areas of imagery in high detail and fill in the remaining parts of the imagery using the best and most efficient approaches modern AI/ML have to offer. Productionizing SIDET could significantly minimize the amount of information parsed through the data bottleneck without losing critical actionable intelligence.