Artificial Intelligence & Machine Learning

The challenge from the client was to take image retrieval to the next level – improving the performance of traditional visual scene matching (VSM) algorithms at finding the best match to an image of a scene from a dataset of previously captured images.

The Plextek team decided to augment the traditional VSM algorithms with one of the latest ML techniques – Semantic segmentation. This was used to assign a label to each pixel in an image, based on which category it belonged to – tree, person, vehicle, etc. This enabled the VSM algorithm to prioritise those features which are likely to remain consistent over time (such as buildings) over those that are likely to change (e.g. cars, vegetation).

As traditional VSM algorithms were still performing the image retrieval, desirable properties such as explainability and consistency were retained – but with the added benefit of the improved performance offered by the ML semantic segmentation.

Rapidly deciding where to place transmitters and receivers for the best radio coverage in a complex urban environment is easier said than done. Yet time is of the essence in disaster or conflict zones, for example, where an effective communications network is vital.

That’s why the Plextek team turned to ML to create a cutting-edge coverage prediction tool. Once it is trained on numerous coverage maps generated using conventional tools, the ML model can accurately predict coverage for previously unseen environments in a fraction of a second, using only a standard PC or laptop.

Transmitter locations can then be optimised to achieve the desired coverage objectives for an effective communications network. The unprecedented speed and scalability of the technique is uniquely suited to urgent deployments or situations subject to rapid change.

Unauthorised drones flying near airports cause significant safety and security issues and extensive disruption to travellers and airlines. But differentiating between drones and birds can be a challenge for traditional radar, leading to a high rate of false alarm.

In response to this challenge, the Plextek team developed an automatic target recognition framework that enabled a wide range of different ML algorithms to be explored and optimised based on data from our PLX-U16 radar. The ML algorithms exploit the micro-Doppler effect caused by small, internal movements within an object – the rotation of rotor blades on a drone, for example, or wing movements relative to the body of a bird.

The result was the development of an ML classification technique that can accurately distinguish between birds and drones and even identify the type of drone.