Rapid Automatic Image Categorization (RAIC; rake), software developed for commercial use, analyzes large unstructured multidimensional datasets such as earth-observation (EO) imagery, Synthetic Aperture Radar (SAR), and other multi-spectral datasets in minutes and then enables ad hoc labeling, searching, and anomaly detection in seconds. We have built a user interface (UI) on top of RAIC, which provides users with an intuitive interface to their dataset enhanced with RAIC. With the UI, users can select objects of interest, quickly locate other occurrences, and then automatically train an AI model for high-accuracy object detection from within the domain (e.g., EO or SAR) in which the object was identified. This technology provides the means to rapidly gather and annotate 100s to 1000s of data points in seconds, train an AI model (typical training times can vary from minutes to hours), and run that model across an entire dataset. Preliminary results suggest that this entire procedure can be reduced to less than 15 minutes on EO data of approximately 100GB in size. Preliminary results also show RAIC producing detection models that rival state of the art in terms of performance; however, RAIC does not require the otherwise typical man-months of human annotation and labeling. RAIC excels in domains where labeled data is extremely sparse and annotation tasks would otherwise be too time intensive or infeasible. Our RAIC technology has been designed around cloud solution architecture, which provides scalability into petabyte sized datasets. Additionally, were able to easily deploy our application in a low-latency manner across the US and worldwide to include off-cloud offerings that can run behind firewalls, aka secure environments. RAIC technology has dual use applicability for the 16th Air Force 363d ISR Wing and 480th ISR Wing groups who need faster ways to monitor change, extract key insights, and enable automation through AI models with minimal human machine interaction. Our proposal seeks to demonstrate the value and performance of our RAIC technology for enabling rapid object labeling, AI modeling, and object detection tasks across imagery types while minimizing extensive labeling effort. In our previous proof of technology exercise, our cloud-based RAIC evaluated 3-inch resolution EO imagery of a large US metropolitan area (400 sq. km.) and enabled a non-expert labeler to annotate objects of interest, build and run a detection AI from only a handful of selected objects in under an hour. Combining our initial results with the proposed SBIR work, we anticipate we can improve performance and showcase the time and labor benefit to Air Force ISR workflows of this dual user technology. Our team has experience in Fortune 100 companies across a wide area of the technology spectrum including EO imaging, SAR, FMV, lidar, multispectral imagery, 3D modeling, machine-learning engineering, AI processing, and cloud solution architecture deployments.
