GOTCHA II – Graph-based Object-oriented Transformers using Contextual Hierarchical Attention

In the domain of object detection, current approaches are constrained by fragmented sensor data, limited contextual understanding, and rigid detection methodologies. Commercial imaging satellites and overhead aerial platforms generate millions of images daily, capturing vast amounts of visual data that overwhelm manual exploitation capabilities. Automatic target recognition (ATR) models are employed to detect and classify static and moving targets, yet they face significant challenges, particularly with small targets of interest like vehicles, ships, and small manufactured structures that occupy only a few pixels per frame, rendering them nearly unidentifiable by both automated systems and human operators. Aptima Inc. and Patch Plus Consulting propose to continue developing Graph-based Object-oriented Transformers using Contextual Heterogeneous Attention (GOTCHA). GOTCHA will (1) generate space-time-modality-patch (STMP) graphs, analyzing node relationships to inform detection decisions (2) be enhanced with zero-shot and few-shot detection capabilities using language models, and (3) will employ data augmentation techniques that will improve robustness. Performance will be evaluated through human-in-the-loop experiments and comparative analysis against existing single- and multi-modality detectors. Finally, a framework for modular integration with existing object detection and classification systems will be developed, accompanied by documentation and demonstrations.