You are here

Threat Detection Using Artificial Intelligence and Machine Learning



OBJECTIVE: Adapt and apply multi-int sensing and machine learning to identify, understand and help mitigate threats to Air Force installations. 

DESCRIPTION: Machine learning techniques have evolved and developed rapidly in the last few years because of the availability of low cost massive computing and large scale labeled data sets. Pattern recognition and other forms of information extraction from video, imagery, text/data streams, or large collections of meta-data from diverse sources are increasingly practical and effective. Processors and algorithms operate at a speed that is increasingly compatible with real-time activities including control and operation of autonomous vehicles, instantaneous facial recognition, and natural language processing. The Air Force deploys and manages forward operating bases, aircraft assets at expeditionary airfields, and other various fixed or temporary supporting ground installations and facilities. Each of these faces threats to its operation, ranging from personnel, manned or unmanned vehicle intrusion, kinetic, electromagnetic and cyber disruption and corruption. These threats evolve on multiple timescales - sometimes quite rapidly “ and can increasingly incorporate elements in multiple domains. Machine learning has the potential to map and understand the installation or operation and then to characterize, monitor and highlight dynamic threats, intrusions and interference within the environment that indicate anomalous behavior and that might pose a threat. Such systems could make use of any available/existing data and ingest new sources of data, including sensors, from within the installation, its physical exterior, local electromagnetic sources or exchanges over public networks. Signals collection and mapping, video analytics such as facial recognition and gait analysis, airspace awareness or physical change detection of the surrounding environment based on vehicle-mounted sensors could be used. A key challenge is the availability of representative data and associated labels or truth identifiers, so that for such systems can be adequately trained. These data could be gathered and the systems trained in-situ, or by using synthetic generation or simulation. For the purposes of this solicitation, the focus will be on forward operating bases in areas with adjacent or nearby urban and semi-urban environments, with sparse road and other infrastructure, and complex mixes of allied and adversarial groups, as well as threat detection during transitional periods of base operation. The detection, characterization and early identification of threats to base personnel and property, including high value assets such as aircraft, is the primary objective of the system to be developed. The government will make available a dataset which will include multiple bands of electro-optical data, RF GMTI, and acoustic data. Training data does not have to include the government provided data set, regardless of data source it should be clearly identified in the proposal. 

PHASE I: Determine appropriate machine learning techniques for implementation of threat detection at forward operating bases embedded in civilian areas. Establish the relevance of these machine learning approaches based on their previous or ongoing application to other similar challenges or clear potential to support threat detection, indications and warning, and predictive avoidance options. Identify existing data sources that could be used to support threat detection, and possible new datasets that could augment existing sources to uncover connectivity or indications of patterns and information. Examine the feasibility of learning methods to characterize and identify threatening behavior or precursors, including the availability of training data or truth sets. Provide a plan for development and demonstration of these concepts, including the development of sensors, data collections, and necessary training data sources. 

PHASE II: Develop and demonstrate the concept and application identified in the Phase I project, including deploying and/or connecting the network of sensors, sources, or databases, training the system to detect threat patterns and generate actionable indications and warnings for operators. Evaluate the effectiveness in terms of probability of detection and false alarm rate for the threats, and reporting on probability of correct classification and probability of detection with a standard confusion matrix. For the phase II additional government assessment could be accomplished so a proof-of-concept software deliverable should be made that can be tested by the government in order to validate future investment. In order to make a product that the government can use, DISA and DoD guidance should be followed in terms of cybersecurity and thus operating systems should have the STIG ( applied with all ports and protocols documented in the final report. 

PHASE III: The contractor can pursue markets and applications in which detection of anomalies or dangers is required, and where multiple data sources are present in dynamic and uncertain environments. Applications include autonomous vehicles, infrastructure protection, security and management of large public events. 


1.Mitchell, Robert, and Ing-Ray Chen. "A survey of intrusion detection techniques for cyber-physical systems." ACM Computing Surveys (CSUR) 46.4 (2014): 55.

2.Pathan, Shafiqua T., et al. "A survey paper on a novel approach for image classification based on low level image processing algorithm from real time video." International Journal of Scientific and Technology Research 3.2 (2014).

3.Hogenboom, Frederik, et al. "A Survey of event extraction methods from text for decision support systems." Decision Support Systems 85 (2016): 12-22.


KEYWORDS: Artificial Intelligence, Machine Learning, Deep Learning, Autonomy, Autonomous Systems, Neural Networks, Facial Recognition, Anomaly Detection, Intelligent Surveillance, Threat Indications And Warning 

US Flag An Official Website of the United States Government