DroneShield – Global Leader in ‘Micro-AI’ 


Detecting, tracking, and precisely disrupting signals in high-noise radio frequency environments is an extremely challenging technical problem that is growing in complexity. A single signal of interest can be buried in the background noise of communication signals like telephone, Wi-Fi, and radio, or even Electronic Warfare activities like jamming and GPS denial. To solve these problems, over the last six years, DroneShield has been developing deployed ‘Micro-AI’ architectures and solutions. The ‘Micro-AI’ approach is now paying dividends as the solutions can be deployed to extremely low SWaP (Size, Weight, and Power) devices and scale both horizontally and vertically, depending on the customers’ requirements.

‘Micro-AI’ was developed perpendicular to the now popular LLM (Large Language Model) and ultra-large models run by global technology companies, though the underlying mathematics is similar. Rather than being general-use AI which interacts directly with humans, ‘Micro-AI’ uses bespoke models designed for a specific purpose and usually serves as a building block in a larger program. It excels in a single type of problem that it is trained to address, and variations on that problem set. Due to these differences, ‘Micro-AI’ can be hosted entirely on small, portable devices and doesn’t require nearly as much power consumption, seeing applications across a range of small- and medium-scale consumer and industrial products.

The dataset used to train AI models very strongly affects their usefulness, which is even more true for the narrow focus of ‘Micro-AI’. DroneShield uses a rich data set collected from its test sites, supplemented with data from real situations encountered by clients around the world. The ‘Micro-AI’ is then deployed with a powerful combination of FPGA (Field-Programmable Gate Array) and industrial Nvidia GPU solutions to drive its AI systems for drone detection and mitigation. This dual-scale approach maximizes efficiency and performance, allowing for real-time analysis and rapid response capabilities.


DroneShield optimizes its radio frequency (RF) protocol detection and mitigation capabilities by harnessing FPGA technology to run AI directly at the edge – within the devices used for detection. This approach enables real-time analysis of RF data through parallelized AI algorithms on embedded circuits, while ensuring minimal power consumption. This makes it ideal for deployment in low SWaP military devices where energy efficiency and reliability are paramount. Utilizing multiple FPGAs with ‘Micro-AI’ concurrently at the edge enhances DroneShield's capacity to deliver responsive, low-latency security solutions tailored for diverse operational scenarios.

DroneShield supports these inbuilt models with further processing on Nvidia’s industrial-rated GPUs to achieve more complex analysis such as real-time object detection from video, precise motion tracking, and advanced Sensor Fusion. These GPUs are specifically designed to handle multiple, complex ‘Micro-AI’ algorithms efficiently, allowing rapid processing of video streams to identify and track small, fast-moving objects such as drones with high accuracy. By integrating Nvidia's powerful GPU compute solutions, DroneShield’s DroneOptID solution augments the human operator with higher degrees of visual processing and tracking.

DroneShield utilizes the full capability of Nvidia GPUs to deploy cutting-edge Sensor Fusion algorithms at the edge to significantly alleviate the cognitive burden on C2 (Command-and-Control) operators. The solution integrates data from many sensors of diverse types such as acoustic, thermal, radar, and other electromagnetic sensors, processed in real-time through a ‘Micro-AI’ architecture. This powerful solution is packaged up into a portable system to deploy at the edge, automating the integration of complex data streams, improving the accuracy of threat detection. This allows operators to focus on critical decision-making rather than manually analyzing data coming from all sensors – where they could easily miss critical information or take too long to identify a threat.

Previous
Previous

Decoding the New Jersey Drone Sightings – Insights and Leadership in C-UAS

Next
Next

Advancements in Machine Learning and AI for Counterdrone Detection and Response