We designed and prototyped a real-time RF signal classification platform that works on down-converted RF front-ends. Wideband RF signals are captured, channelised and transformed into time–frequency representations, which are then classified by deep learning models to identify modulation types and signal classes in real time. The system is intended as a building block for spectrum monitoring, signal intelligence and automated RF environment awareness.

We developed a visual geo-localisation pipeline that estimates the position of a UAV by matching onboard camera images with georeferenced satellite or GIS imagery. The system extracts robust visual features from both views, performs large-scale image matching and returns the most likely geographic coordinates of the current frame. This approach enables localisation in scenarios where precise positioning must be derived directly from visual and map data.

18 months of project and clinical studies have been successfully completed. As a result of the project, it was confirmed that AFib detection can be performed with 88% accuracy, 93% precision and 88% sensitivity by analysing the remote PPG signals of the participants using only the camera.

In collaboration with Akdeniz University, the rigidity levels of Parkinson's patients are classified with 84.7% accuracy, 86.6% precision and 80.4% sensitivity.

It has been clinically proven that 93.6% accuracy, 94.1% precision, 93.7% sensitivity and 93.6% F1 value were achieved in the developed AFib diagnosis model.

In cooperation with Akdeniz University, Parkinson's disease is detected with 89.75% accuracy, 88.4% precision and 91.5% sensitivity using the human voice.

In the developed COVID-19 diagnosis model, 82.0% accuracy, 76.0% precision and 86.3% sensitivity values were achieved.

In cooperation with Akdeniz University, a social distance violation identification, warning and tracking system was developed from the camera.