Deep Rational 3D Geospatial Analytics for Generative AI

Under this SBIR Phase I project, we propose a joint deep neural network (DNN) and large language model (LLM) based framework for performing contextual reasoning for semantic scene understanding. The framework uses the power of supervised and unsupervised discriminative DNNs to discover hidden features from individual data streams with the ability of LLM models to perform reasoning in a noisy, multi-variate, correlated, and constrained setting while incorporating a corpus of contextual cues, common sense reasoning and domain knowledge. The system will utilize Novateur’s existing reasoning engine that has been developed and enhanced under multiple DoD-funded projects that combines the output of multiple DNNs operating on individual data streams by using ontologies to perform contextual reasoning and inference. The Phase I effort will focus on developing discriminative AI models for remote sensing, modules for querying LLM and refining the semantic outputs as well as test and evaluation on publicly available data. The proposed effort will build upon the experience and expertise of Novateur team in the areas of context-based reasoning, deep learning, computer vision, exploitation of remote sensing imagery, LLM-based context modeling, and semantic scene understanding.