Sniffing out disease? Undergraduate student helps build a virtual nose with Arduino components

An undergraduate student from the School of Computing has published a paper while making a significant contribution to a project investigating the potential of teaching computers to smell.

While the capacity of computers to recognise speech and images has been well-explored in recent years, there is comparatively little research into teaching them to recognise and differentiate between smells.

However, a virtual nose has the potential to be extremely useful as, for example, a low-cost form of medical examination - as it is now known that certain diseases can be detected through smell.

Patrick Fox, a third-year student of Computer Science and Mathematics MSci, BSc, began investigating this topic in the second year of his course. He worked with researchers from the Miguel Hernández University of Elche in Spain, who were building an Arduino platform equipped with MQ sensors – a class of sensor often used to detect gases such as carbon monoxide, methane and ammonia.

The team wanted to create a device capable of distinguishing between different kinds of olive oil – effectively “smelling” the type and quality of each based on the evaporated chemicals detected. Olive oil was chosen as a useful substance to test the device with, being difficult to distinguish by sight and more readily available than pre-diagnosed patients.

Patrick’s role in the project was to develop the software used to analyse the data collected by the device, as well as performing the data processing and advising on the mathematics of computing smells. Drawing on his experience studying a second-year module in linear differential equations, he had the idea of introducing Fourier transforms, a mathematical tool that breaks waveforms into new representations characterised by sines and cosines.

Using Fourier transforms as an intermediate step before applying an algorithm to classify the sinusoidal data (no pun intended), the team from Miguel Hernández were able to devise a generic signature for each oil that gave the input for the classifier.

Ultimately, the device was capable of classifying the type of olive oils with an accuracy of 88% to 91%, and the quality of the oils with an accuracy of 61% to 77%.

As well as advancing the understanding of the practice of classification by evaporated chemicals, the project demonstrated that it is possible to achieve a high degree of accuracy on a low budget: the device cost only around €30 to build and could easily be replicated in remote parts of the world, where medical diagnosis can be costly and time-consuming.

Patrick co-authored the paper detailing the group’s research, “DFT based classification of olive oil type using a sinusoidally heated, low cost electronic nose”, which was published in the journal Computers and Electronics in Agriculture.

He will give a presentation and take questions about his paper at the upcoming Leeds Data Science Engagement and Employability Conference at Leeds University Student Union on May 8th.

Further reading

DFT based classification of olive oil type using a sinusoidally heated, low cost electronic nose, Computers and Electronics in Agriculture, Martin J.Oates, Patrick Fox, Lucia Sanchez-Rodriguez, Ángel A.Carbonell-Barrachina, Antonio Ruiz-Canales