Hyperspectral Imaging (HSI) is a well-established technology able to capture images across hundreds of different wavelengths (or channels) in the electromagnetic spectrum. Therefore, while conventional imaging provides colour information in three channels (Red, Green, and Blue – RGB), HSI covers not only colour but also information from other regions invisible to the human eye, such as the Ultra Violet (UV) and Near InfraRed (NIR). Being able to simultaneously capture spatial and spectral data, HSI has a great potential for the automated inspection of assets in many industries. Additionally, the cost and size of HSI systems in recent years has been reduced, making this technology more accessible to industry.

However, introducing HSI into industry faces significant barriers mainly related to: (i) operating HSI systems require certain conditions that are unfeasible in many industrial contexts, (ii) industrial regulators require a clear/enhanced explainability before adopting new technologies such as HSI (and its related data processing), and (iii) interpreting the HSI data can be difficult due to its large and multi-dimensional nature.

The aim of the proposed PhD is to create novel AI-based techniques and tools to support overcoming the abovementioned barriers and facilitate the effective deployment and adoption of HSI technology in challenging industrial environments. These tools are to be embedded in the next generation of intelligent autonomous sensor technologies (Industry 4.0 and beyond) for condition monitoring and inspection of high-value assets and decision support in key sectors such as nuclear, energy, and manufacturing, but also expandable to many others.

The methodology planned for developing the proposed research relies on the replication of industrial (challenging) environments under lab conditions. For instance, data can be captured using a changing distance between sensor and sample under study. Then, the effect that this change has on the captured data can be investigated, modelled and reproduced. The same concept applies to other changes in the environment, for example, illumination conditions or the presence of partially transparent media between sensor and sample. The latter is a case of particular interest, as in a number of industries, it is often to evaluate assets via gloveboxes, so that data can only be captured from outside the box and, therefore, there is transparent material between sensor and sample (e.g., acrylic or polycarbonate panels) obstructing the vision of the sensors. AI can potentially be used to recover the obstructed vision.

Further information

An exciting opportunity to combine fundamental research on hyperspectral imaging and artificial intelligence to enable effective application in industrial environments. This is also a great opportunity for gaining expertise in operating high-spec hyperspectral systems covering UV, VNIR and SWIR ranges.