Multi-Objective Machine Learning for Concealed Weapon Detection

Concealed Weapon Detection (CWD) is an important area of research in the defence and security community. This is a result of a number of high-profile terrorist attacks which have resulted in loss of life and damage to public infrastructure. The threat faced by the security forces is diversifying and current technology is struggling to meet new requirements. The technologies currently in use at airports include metal detection portals, millimetre wave imaging systems, x-ray scanners and ion mobility spectrometers. Current approaches to solving this problem attempt to use computer algorithms to determine whether a scan represents a “threat” or “not a threat”. However, this has two significant short-comings: (1) The design and tuning of such algorithms is not always optimal; and (2) These systems are not able to detect multiple categories of threat at once, i.e. determining whether something is a “gun” or “knife” or “explosive device”, instead of just “threat” or “not threat”. Solving these challenges will allow a system to search for multiple threat types simultaneously (and thus quicker) and be able to determine what type of threat it is more accurately. This will allow the security forces to react to specific threats in a more controlled manner as they will know the type of threat presented. For example, detecting an improvised explosive device has much greater implications for how to handle the threat than detecting a knife.

This project will employ the state-of-the-art in machine learning algorithms, to produce concealed weapon detection algorithms that can detect multiple categories of threat according to decision maker preferences. Examples of machine learning techniques that may be employed include multi-objective optimisation, genetic optimisation techniques, deep learning (deep convolutional neural networks), keypoint detection, manifold embedding, graph-based community detection, image segmentation methods.

Objectives
1. To produce a critical review of the field of neuro-evolution and concealed weapon detection, with an emphasis on detecting multiple categories of threat.
2. To develop and benchmark a state-of-the-art many-objective neuro-evolution algorithm for solving real-world concealed weapon detection problems.
3. To develop a novel algorithm for incorporating decision maker preferences into the neuro- evolution process.
4. To evaluate the effectiveness of the novel preference driven many-objective neuro-evolution algorithm on synthetic test suites and real-world concealed weapon detection problems, using statistically rigorous analysis.
5. To evaluate the deployment of a multi-objective concealed weapon detection classifier to detect explosives, firearms, and benign objects. The datasets will be supplied by the company.

How to apply:

Applications are made via our website using the Apply Online button below. If you have an enquiry about this project please contact us via the Email NOW button below, however your application will only be processed once you have submitted an application form as opposed to emailing your CV to us.

The PhD Studentships are open to UK, EU and International students. Candidates for a PhD Studentship should demonstrate outstanding qualities and be motivated to complete a PhD in 3 years and must demonstrate:

• outstanding academic potential as measured by either a 1st class honours degree or a Master’s degree with distinction or equivalent Grade Point Average (GPA)
• an IELTS (Academic) score of 6.5 minimum (with a minimum 6.0 in each component) for candidates for whom English is not their first language and this must be evidenced at point of application.
In addition to satisfying minimum entry criteria, BU will look closely at the qualities, skills and background of each candidate and what they can bring to their chosen research project in order to ensure successful completion.

Additional Eligibility Criteria:
ESSENTIAL: Computer programming experience, preferably in Python and/or MATLAB, with a desire to implement algorithms using Python.
DESIRABLE: Machine learning engineering experience, preferably using popular libraries such as Keras, TensorFlow, PyTorch, and others.