/ ISR and Security

Acquiring and maintaining a situational awareness of a space require vigilant and robust systems – combining both local and remote sensing platforms. Furthermore, as new disruptive hardware and software challenge legacy systems there is a need to modernize those systems and implement new tools to streamline the workflows of users. Also, new data resources are needed to expand into new domains, environments or fuse various sensor platforms for better automation in low shot detection contexts. On the other hand, the exponential increase of satellite products posed a looming big data problem for analysts. These kind of Intelligence, Surveillance and Reconnaissance (ISR) applications in the security and defense domain typically involve computer vision algorithms and methods, but the evaluation of the data resources used to train and evaluate such tools are also important. There is a need to reduce the burden on providing site security and actionable intelligence from big data sources. Also, while fully automated kill chains are against current DOD policy, there is a need for tools to assist in the lining up of targets correctly. To this end, Lynntech has developed a number of ways to evaluate automated detection or classification systems against various challenges, posed by a particular dataset, environment, or adversarial entity. The ability to visualize how a tool “sees” a dataset is a particular need for transferring tools between domains. We also specialize in enhancing the realism of physics-based simulations of data in various domains to facilitate better “in real life” (IRL) performance of derivative tools. The overall goal is to reduce the data burden, and increase the robustness of systems by leveraging the best that legacy systems have to offer with newer unsupervised learning techniques.

Capabilities

Lynntech is intent on pursuing R&D in remote sensing and geo-spatial analytics, which involves measuring geo-referenced surfaces from a distance. Our R&D involves electro-optical/infrared (EO/IR) data analysis, radar imagery processing, machine learning techniques, and automated signal processing from airborne and spaceborne platforms. We have developed EO/IR technology for real-time airborne EO/IR video streams over the ocean. Spaceborne remote sensing products are now reaching a critical mass in terms of increased global spatial and temporal coverage and resolution of satellite observations. This leads to a big data problem in which automated tools are need to processed data streams into actionable geo-spatial intelligence. The application of machine learning techniques to solve these issues is a pressing need and rapidly growing. Specifically, we are developing powerful deep learning tools that can measure small-scale changes of the surface of the Earth, even from space. Improved information about surface deformations will lead to better prediction of and response to disasters, tracking manmade surface changes, infrastructure and industry planning, and natural resource management.

The increasing reliance on systems automated by some sort of AI that is usually machine learning based poses a security risk. On the one hand, many AI systems are used as “black boxes” where there is little explainability of performance failures. On the other hand, many machine learning-based tools have security vulnerabilities at either the training stage (data poisoning) or from adversarial example inputs (evasion attacks). Such manipulations result in misleading “confidence levels” of the tool’s output which can misdirect an AI system’s behavior.  For example, a self-driving car needs to be able to correctly identify and parse street signs 100% of the time in all weather conditions or in the presence of graffiti. Furthermore, disturbing biases in “go to” training sets leave some critical performance shortcomings, such as a reduction in performance in recognizing child pedestrians (which is exacerbated by newer datasets not allowing minors as content because of privacy protections). Lynntech has developed a number of ways to both manipulate AI system behavior, evaluate biases, and detect when a system is being challenged by its inputs. Such AI security evaluations will ensure the development and deployment of more robust and secure AI systems.

Maintaining as situational awareness of the security of a site or entry point requires constant vigilance for a small number of people tasked with security monitoring, in spite of long periods of inactivity or normal patterns of behavior. For example, repeated false alarms due to the presence of warm-blooded animals can dull the wits of security personnel in recognizing a human miscreant. Being able to provide assistance tools to either integrate a large number of sensors, or even sensors across domains is a pressing need to establish a solid site-security paradigm. Being able to track persons of interest across different modalities (using imaging and biometrics sensors) at all times of day is important for providing secure access to restricted zones and buildings. Another need is identifying and tracking persons within vehicles, or behind glass. Finally, the increased automation for the detection of anomalous behavior or events would greatly enhance the utilization of security monitoring infrastructure in real time.

Airborne search is one of the most operationally demanding missions facing aircrews today.  Detection requires constant, detailed scanning of a dynamic scene for a fleeting, subtle indication of the search target. This is especially difficult in maritime environments. The environment is physically taxing and missions typically extend over several hours. Current search methods rely on extraordinary human visual acuity and processing capability, unwavering focus, and luck. Revolutionary technologies that have the potential to drastically improve detection performance for small targets are needed to enhance airborne search. In order to help focus operators’ attention on regions of high interest, Lynntech has developed systems which analyze and look for thermal signatures in a small area from a geo-registered thermal map used to locate anomalies that represent high-probability targets. Our system’s real time processing reduces background noise thus increasing the signal to noise ratio (SNR) and, leading to more efficient target detection of faint thermal signatures. Using input from the embedded global positioning system / inertial navigation system (GPS/INS) and pixel-wise optical calibration of the an imager, the infrared signal from each imaged ground position is averaged in time, reducing the amplitude of surrounding background noise. Since our geo-registration is based on precise inertial navigation and calibration, rather than image-processing, no visual cues or registration points are necessary in the field of view. This means that, unlike standard solutions, no mosaicing, stitching, or image-feature based registration are used to align individual frames. Additionally, targets at or even below the single-frame resolution of the auxiliary imager can be detected through averaging.

Selected Project: Airborne Wide-Area-Search

Many real world problems cannot be easily solved using direct modeling and conventional imperative programming. Lynntech uses AI methods in combination with state-of-the-art electro-optical/infrared devices, embedded navigation systems, human-machine interfaces, and other resources to solve real-world problems such as maritime search and rescue, analysis of geospatial data to provide actionable intelligence, terrain monitoring, disaster recovery, computer vision for autonomous vehicles, assistive technologies for disabled users, data analytics, natural language processing, and mobile computing for Internet of Things (IoT) applications.

Operational Need – Airborne maritime search is one of the most operationally demanding missions facing aircrews today.  Detection requires constant, detailed scanning of a dynamic scene for a fleeting, subtle indication of the search target.  The environment is physically taxing and missions typically extend over several hours.  Current search methods rely on extraordinary human visual acuity and processing capability, unwavering focus, and luck.  Revolutionary technologies that have the potential to drastically improve detection performance for small maritime targets are needed to enhance airborne maritime search.

Lynntech Solution – WAIRI is an Aided Target Detection system (AiTD) for open-water and littoral airborne search, surveillance, and law enforcement.  Utilizing commercial-off-the-shelf hardware, a small wide Field of View thermal imaging turret detects probable targets on the water and cues the main turret to slew to the region of interest and track that position for a brief period at high magnification.  The operator views the main imager feed which displays live, narrow field-of-view, geospatially-fixed video of high-interest probability regions.  Minimal operator interaction is required as subsequent interest regions are automatically displayed along the search track.  A detected target can then be locked and tracked to aid the rescue surveillance mission.

Revolutionary Performance – WAIRI has demonstrated the potential to significantly outperform unaided IR imagers in detecting small maritime targets on fixed and rotary wing aircraft, day and night, over calm waters and sea state conditions up to level three (2 to 3 ft waves). Current maritime imaging operations present two key limitations: 1) deployed turrets have small fields of view because they have been designed for standoff inspection and thus have outstanding zoom-in and stabilization capabilities. 2) Detection is currently unaided: the operator directs the search and decides if a potential target is being seen. Over an extended open-ocean test spanning a 24-hour day, an unaided MWIR EO/IR system and WAIRI were compared on search effectiveness for small targets.  Using a Receiver Operating Characteristic measure, WAIRI was 98.4% effective in detecting targets compared to 57.2% for the unaided IR imager.