Editor’s Note: Dr. Jeffrey D. Clark is chief scientist of intelligent systems and machine learning research at Riverside Research and an OPIR artificial intelligence and machine learning subject matter expert. Riverside Research is actively involved with USGIF’s Machine Learning & Artificial Intelligence Working Group. Guest posts are intended to foster discussion, and do not represent the official position of USGIF or trajectory magazine.

GEOINT analysts have stated their desire for innovation in the process of generating geospatial intelligence. For example, the National System for Geospatial Intelligence Strategic Concept of Operations (CONOPS) for 2015 provides guidance for using intelligent machine learning (ML) algorithms to aid the GEOINT analyst. The CONOPS lists “tool-assisted information generation” and “fully automated information generation” as key elements of GEOINT analytic information generation. It also talks about “augmenting analytic capabilities through artificial intelligence (AI) and knowledge processes (cognitive/rule-based inferencing, link analysis, pattern identification).”

Along the same lines, the National Air and Space Intelligence Center’s (NASIC) strategic plan for 2023 states, “We will pursue new ideas, innovative processes, and groundbreaking research to create and deliver intelligence more effectively and efficiently.”

The team of scientists, engineers, and analysts that works with overhead persistent infrared (OPIR) data in NASIC’s Geospatial Persistent Infrared Analysis Squadron heeded that directive and began incorporating ML/AI algorithms into the tool set they use to produce Intelligence Community mission-essential products. Incorporating those algorithms provided more accurate detects and tracks, produced faster and more efficient processes, and allowed for some processes to be automated.

There are many examples of ML and neural network algorithms that have been and are being applied at NASIC: principal component analysis, linear discriminant analysis, support vector machines, self-organizing maps, and artificial neural networks (ANNs). These algorithms have been incorporated to help with change detection, target detection, feature selection, object clustering, tracking, background suppression, false alarm rejection, classification, and other tasks. For example, the NASIC team successfully incorporated an ANN that suppressed false alarms by 31 percent and increased the positive predictive value by more than 15 percent.

Incorporating any of these algorithms into the analytic process also helps to solve a related issue: the ever-increasing big data problem, namely that there is too much information to accurately sort in a timely manner. The amount of data analysts must analyze is continually increasing. In order to catch pertinent information that is of interest to the Intelligence Community, intelligent automation must be in place, and ML/AI provides those tools. The automated processes also have to be fast and accurate to provide the warfighter with the right information at the right time. The ML/AI tools NASIC is leveraging free up analysts’ time, allowing them to focus on producing an accurate and detailed report for the IC mission. Another advantage of the ML/AI algorithms is their ability to “see” events that can be missed by the human eye. In essence, these algorithms perform triage as the first step in alleviating the big data problem.

The ML/AI algorithms that support the OPIR mission at NASIC are still in the research and development phase and were developed in conjunction with advisory and assistance services support from Riverside Research, which has also prototyped several advanced neural networks. Those prototypes proved the neural networks’ utility in working with OPIR data. Riverside Research helped guide and oversee NASIC’s incorporation of these algorithms and techniques through direct contract support and provides expertise to the broader GEOINT community through its Open Innovation Center AI/ML lab.

The NASIC team is now prototyping several advanced neural network/deep learning methods (such as convolutional neural networks and evolutionary algorithms) to demonstrate their ability to advance the OPIR capabilities and mission.

The insertion of AI techniques into GEOINT data processes and products produces more and better intelligence from OPIR data. In the future, we can look forward to highly intelligent machines further advancing our capabilities by fusing multiple intelligence products to produce a synergistic, highly intelligent product for the GEOINT Community.

Related


Posted by Dr. Jeffrey D. Clark