Precision Responsibility Classification of Space Events (PReCiSE)

Web Sensing LLC seeks to address Space Force’s problem of identifying and classifying space events for space Domain Awareness (SDA) and responsible space action. Both DoD and private industry on orbit space assets are in a constantly contested, shared space environment that needs awareness in order to act responsibly and still carry out their objectives. We believe the critical part of solving this problem is the ability to take the existing space object detection and identification technology and radically advancing that to detect, identify, and classify not just objects, but events conducted by those objects so as to determine the level of threat and (ir)responsibility of that event. Our primary objective here is to develop a software system based in a combination of advanced computer vision and artificial intelligence/machine learning (AI/ML) techniques capable of rapidly and accurately identifying and classifying arbitrary space events. Currently, a lot of research, algorithms, and AI/ML models exist space object detection and classification within images and video[1-9], but there is not the same existing research and technology around the movement of those objects as events and applying classification methods to those as shown in Figure 1. This leaves a incredible knowledge gap in the space domain for one’s space assets. In Phase I, we will conduct a feasibility study using both convolutional neural networks (CNN) and broad learning system (BLS) styles of AI/ML model to assess the method with the most promise going forward in terms of extensibility and accuracy. We will conduct transfer learning re-training on each of these methodologies using the open-source data and pre-defined event classification criteria accumulated earlier. This will enable us to rapidly train, test, and select the best AI/ML learning method for space event classification. Once we have decided on a class of model, we will assess it’s structure and ability to be used for a final product, and then begin develop, test, and initial assessment of that model; as well as integrating this model into our larger system and interfaces. Our final system will be based in either the cloud or our self-hosted compute clusters, accessible via a web-based RESTful API with clearly defined and extensible interfaces. Our interfaces will be developed to be easily integrated with existing full-stack tools such as Confluence, AWS EX2, GitLab, Kubernetes, etc. We will seek to enhance and transition this product in a follow-on Phase II. We expect to collaborate with Space Force and other transition partners to utilize the entirety of the technology portfolio and capabilities to meet specific accuracy, integration, and computing at scale needs.