# A New Association Approach for Multi-Sensor Air Traffic Surveillance Data Based on Deep Neural Networks

## TL;DR

This paper introduces a new machine learning-based method for associating multi-sensor air traffic data, improving performance for various aircraft types.

## Contribution

The novel M-SIOTA algorithm uses deep neural networks for data association without requiring aircraft dynamics models.

## Key findings

- The M-SIOTA algorithm improves data association performance across different flight scenarios.
- The approach works for all aircraft classes and sensor configurations without needing dynamic models.
- Validation on diverse datasets shows consistent performance improvements.

## Abstract

Air Traffic Services play a crucial role in the safety, security, and efficiency of air transportation. The International Civil Aviation Organization (ICAO) performance-based surveillance concept requires monitoring the actual performance of the surveillance systems underpinning these services. This assessment is usually based on the analysis of data gathered during the normal operation of the surveillance systems, also known as opportunity traffic. Processing opportunity traffic requires data association to identify and assign the sensor detections to a flight. Current techniques for association require expert knowledge of the flight dynamics of the target aircraft and have issues with high-manoeuvrability targets like military aircraft and Unmanned Aircraft (UA). This paper addresses the data association problem through the use of the Multi-Sensor Intelligent Data Association (M-SIOTA) algorithm based on Deep Neural Networks (DNNs). This is an innovative perspective on the data association of multi-sensor surveillance through the lens of machine learning. This approach enables data processing without assuming any dynamics model, so it is applicable to any aircraft class or airspace structure. The proposed algorithm is trained and validated using several surveillance datasets corresponding to various phases of flight and surveillance sensor mixes. Results show improvements in association performance in the different scenarios.