Application of machine learning and artificial intelligence techniques to improve autonomy in maritime surveillance radar systems

Brown, Angus (2023) Application of machine learning and artificial intelligence techniques to improve autonomy in maritime surveillance radar systems. PhD thesis, University of Glasgow.

Abstract

Executive Summary.

Current maritime radar surveillance missions are typically carried out using an airborne platform with one or more operators on board. The workload of human operators is a bottleneck in surveillance performance as they can only perform on a single platform discontinuously. Additionally, with progress being made towards the use of remotely operated UAVs for radar surveillance missions, an increase in radar operational autonomy is required to maximise the UAV’s surveillance potential.

Consequently, the focus of this research is to improve the autonomy of current maritime radar surveillance missions. By reducing the workload of the current radar operator, then surveillance missions can be performed for longer. This research breaks the autonomy of the radar surveillance mission into two aspects: the platform operator autonomy and the radar operator autonomy. However, the implemented autonomous methods must “complement rather than compete with one another".

In order to implement algorithms for the platform operator autonomy and radar operator autonomy, a maritime radar surveillance simulation and user interface is required. Consequently, this work outlines a real-time maritime surveillance radar simulation and graphical user interface which can be used to carry out missions in the same manner as an operator would with a real system.

For the platform operator autonomy aspect, there is a trade-off between maximising information obtained from the surveillance search area and minimising fuel consumption. The research presented here provides an approach for the optimisation of a UAV’s trajectory for maritime radar wide area persistent surveillance to simultaneously minimise fuel consumption, maximise mean probability of detection, and minimise mean revisit time. Quintic polynomials are used to generate UAV trajectories due to their ability to provide complete and complex solutions while requiring few inputs. A wide area search radar model is used within this article in conjunction with a discretised grid in order to determine the search area’s mean probability of detection and mean revisit time. The trajectory generation method is then used in conjunction with a multi-objective particle swarm optimisation algorithm to obtain a global optimum in terms of path, airspeed (and thus time), and altitude. The performance of the approach is then tested over two common maritime surveillance scenarios and compared to an industry recommended baseline.

In terms of the radar operator autonomy, imitation learning, as opposed to other forms of machine learning, are advantageous as they act in the same manner as the operator, thus reducing the deviation from the current operational standard and allowing for easier system qualification and human operator interaction. The developed radar simulation and interface is used to obtain operator decision data from a human operator. The operator data is then used with two imitation learning methods, namely Bayesian networks and inverse reinforcement learning, with the methods used in place of the operator with their performance compared and their suitability discussed.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: This work was supported by Leonardo MW Ltd [grant number SELEX/GU/2015/SOW01] and the ESPRC [grant number EP/N509176/1].
• Subjects: T Technology > T Technology (General)
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Funder's Name: Engineering and Physical Sciences Research Council (EPSRC)
• Supervisor's Name: Anderson, Dr. David
• Date of Award: 2023
• Depositing User: Theses Team
• Unique ID: glathesis:2023-83738
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 25 Jul 2023 09:52
• Last Modified: 25 Jul 2023 09:55
• Thesis DOI: 10.5525/gla.thesis.83738
• URI: https://theses.gla.ac.uk/id/eprint/83738

Related URLs

• Article DOI
• Article DOI