Threat Detection Using Artificial Intelligence and Machine Learning

Applying machine learning based multi-sensor analytics to detect, identify, and associate threats to forward deployed installations requires a multidisciplinary approach. In addition, uniquely combining machine learning with multi-INT sensing significantly increases the capability to detect and understand threats from physical intrusion. However, traditional security and surveillance solutions typically process and analyze sensor outputs independently without intelligently combining information between multiple sensor types - leaving the capabilities of these systems limited or predictable. Therefore, we propose an effort to research and develop improved multi-sensor analytics methods based on machine learning, specifically for forward operating bases in areas with adjacent or nearby urban environments. More specifically, we propose combining a DLRT for object (e.g., video, person, etc.) detection and tracking with the RNN-LSTM method described in the last section for combined video, signal and acoustic detection, tracking, and notification (i.e., automatic sentences or tag sequences) of threats (both objects and behavior). We will establish the relevance of the proposed machine learning approaches based on our previous experience and ongoing efforts with similar applications. In addition, we will identify relevant data sources, both new and from our existing catalog of projects, that support threat detection and prediction.