NEURO-MTI : Neuromorphic Moving Target Identification

The guiding vision for the USAF 2030 Science and Technology Strategy is that the Air Force dominates time, space, and complexity in future conflict. Complex, Multimodal, Sensor Big Data has the added advantage of significantly reducing identification errors such as False Positives or Misidentified Events and False Negatives or Missed Events. However, conventional inference approaches currently utilize a sequential, non-adaptive, AI/ML processing flow consisting of Data Collection (days to months), followed by Curation (days to months) and Modeling/ Training (days to months). Instead, the Air Force views Neuromorphic Learning as a Game Changing Computing Approach that is adaptive and intelligent and will significantly reduce Mission Execution time through a cycle of Observation (seconds) followed by Analysis (seconds) and Learning (seconds). Ā SaraniaSat aims to enable the USAF/USSF 2030 vision through its proposed NEURO-MTI project aimed at the demonstration of a powerful, combined approach of implementing its previously demonstrated Deep Learning Algorithms for Moving Target Identification onboard the Intel Loihi 2 Neuromorphic Chip currently under development by AFRL/RI. In close collaboration with AFRL, SaraniaSat will adapt its Convolutional Neural Network algorithms to operate on the LAVA Open-Source Software Framework also developed by Intel Corporation in response to the lack of clear, productive programming models for true Neuromorphic Computing. LAVA is modular and extensible with ease of adoption by virtue of the fact that all libraries and features in LAVA are exposed through Python, with optimized libraries and underlying C/C++/Open CL where necessary to improve performance. Ā SaraniaSat will execute the proposed SBIR Phase I NEURO-MTI project in close collaboration with AFRL/RI by identifying potential USAF/USSF Customers and End Users and their applications for High-Performance Edge Computing. Potential applications include Moving Target Identification (MTI), Intelligence Surveillance and Reconnaissance (ISR), and Missile Early Warning particularly within Highly Contested Environments (HCEs). The overarching requirements that SaraniSatÆs proposed NEURO-MTI solution aims to deliver are: Robust, Actionable Information with Low Latency derived through Multimodal sensor fusion and rapid data processing using advanced AI/ML algorithms. SaraniaSat will build on its successful past work of demonstrating the identification of single flying aircraft (AMTI) within highly cluttered, 100 sq. km multispectral satellite scenes in under 45s. Additionally, the SBIR Phase I work will lay the design and foundation for a Phase II Proof-of Concept Demonstration of achieving AMTI using SaraniaSatÆs AI/ML algorithms operating on the Intel Loihi 2 Neuromorphic Chip system.