MBDA is expanding its research into Computational Optical Sensing and Imaging (COSI). One area of interest is Structured Imaging using, for example, Ladar (Laser Detecting and Ranging) to map a scene in real-time. The image of the scene is formed from thousands of pulses, with each pulse being directed to a particular point within the scene. Within this context, several topics are of particular interest, including the following examples:
• How best can Compressed Sensing techniques be applied to this problem to inform the sampling strategy in order to facilitate scene recovery from the lowest number of samples? • What are the optimal sampling strategies to achieve different imaging goals, such as target detection and tracking versus area surveillance? A SI system will need to operate in a variety of roles. • To what extent can data generated during the reconstruction of the previous frame or frames inform the reconstruction of the current frame? If, for example, an iteratively reweighted l1 minimization method was used to reconstruct a scene, could the weights from the previous frame be used to facilitate the reconstruction of the current frame? If so, under what conditions does this hold to be true? MBDA is also interested in augmenting current sensors using COSI techniques. One example would be to replace a high grade lens with a lower grade lens and correct the optical aberrations using deconvolution. Within this context, MBDA is particularly interested in the degree to which Artificial Intelligence (AI)-based techniques can complement or supplant more traditional techniques that rely upon a forward model of the image formation process. A potential limitation when using AI techniques is their reliance on large quantities of training data, however, it seems likely that AI techniques could be used to characterise a sensor and thereby form part of the image reconstruction process. In the example above, AI techniques could be used to characterise the spatially variant Point Spread Functions associated with a low grade lens, which would then be used in a traditional deconvolution process. The aim of this element of the research would be to outline: • The pros and cons of AI-based approaches versus traditional approaches, using metrics such as runtime performance and the resulting quality of the recovered image. • The minimum requirements for using an AI-based approach under different contexts, such as sensor characterisation versus full image recovery, e.g. number of training sets required.
The research will provide a balance between theoretical analysis, algorithm development and programming. MBDA will provide access to expertise, models and data as necessary.
The project is expected to contribute to the fields of structured imaging, compressed sensing and computational imaging.
Essential Criteria: Candidates will have a 1st or 2:1 in Maths, Physics or a related field such as Electronic Engineering. Candidates will have an aptitude and interest in computational imaging techniques, such as super-resolution and sparse image reconstruction. The candidate will need to be comfortable in a self-led research project, but able to collaborate with a team of scientists and engineers when required.
Desirable Criteria: A good understanding of signal processing techniques and theory will prove to be highly beneficial, as will experience with MATLAB, Python or other scientific programming languages.
Flexible Working: MBDA is committed to making work a comfortable and enjoyable experience. We are constantly improving our sites and facilities, creating a lively and open working environment that has earned us 11th place in the Sunday Times “Best Big Companies to Work For” ranking.
MBDA operate a flexi-time scheme, where employees can vary their start and end of work times around a set of core hours, in order to let you fit your work around your home life schedule. Further details are available on request.
EngD project not available to non UK/EU applicants.