The topic is available also in Hungarian.
The electronic nose technology mimics the way how humans detect odours. There are various application domains, as medtech (detecting diseases based on biomarkers), industry 4 (detecting gases of carcinogen nature), food safety (freshness of meat, microbiological contamination). The student will work on the research of deep learning techniques to process sensor data. The task can be labeling (do we smell a banana or a raspberry) or the estimation of the concentration (do we smell butadiene and in which concentration). In the former case, a feed-forward (FF) neural network is involved for classification. In the latter case, an FF is involved for regression.
The electronic nose is basically a MOx based sensor fusion device. The electric conducting capability (resistance / impedance) of a MOx sensor changes depending on the actual molecule that reacts with the surface of the sensor. Furthermore, such sensors are coated with a membrane, which restricts the set of molecules that can reach the surface of the sensor. The sensors available on the market differ on their physical characteristics, thus the conducting capability of each sensor changes differently based on the actual molecule to be detected. We work with a sensor fusion solution meaning that multiple different sensors are combined, which leads to a kind of fingerprinting approach.
MOx sensors can be involved in various application domains, e.g., pharma, food, cosmetics, petrochemistry, medtech. The role of deep learning is to match the sensor signals to the diagnosis. Its task is to find the interdependencies in the sensor data measured. The techniques to be applied are, e.g., sensor data preprocessing, deep feed forward network, recurrent neural network.
The task of the student is to work on deep learning algorithms that processes the raw sensor data in order to provide high accuracy classification, e.g., the diagnosis (Covid+/Covid-).