Fusarium is a genus of fungi found throughout the world. It includes many pathogenic species that produce toxins of agricultural importance. These fungi are also found in buildings and the toxins they spread can be harmful to humans. Distinguishing Fusarium species can be important for selecting effective preventive measures against their spread. A low-cost electronic nose applying six commercially available TGS-series gas sensors from Figaro Inc. was used in our research. Different modes of operation of the electronic nose were applied and compared, namely, gas adsorption and desorption, as well as modulation of the sensor's heating voltage. Classification models using the random forest technique were applied to differentiate between measured sample categories of four species: F. avenaceum, F. culmorum, F. greaminarum, and F. oxysporum. In our research, it was found that the mode of operation with modulation of the heating voltage had the advantage of collecting data from which features can be extracted, leading to the training of machine learning classification models with better performance compared to cases where the sensor's response to the change in composition of the measured gas was exploited. The optimization of the data collection time was investigated and led to the conclusion that the response of the sensor at the beginning of the heating voltage modulation provides the most useful information. For sensor operation in the mode of gas desorption/absorption (i.e., modulation of the gas composition), the optimal time of data collection was found to be longer.
Keywords: Fusarium; application of e-nose; gas sensor; odor differentiation; pathogen detection.