Fusarium graminearum and F. culmorum are considered some of the most dangerous pathogens of plant diseases. They are also considerably dangerous to humans as they contaminate stored grain, causing a reduction in yield and deterioration in grain quality by producing mycotoxins. Detecting Fusarium fungi is possible using various diagnostic methods. In the manuscript, qPCR tests were used to determine the level of wheat grain spoilage by estimating the amount of DNA present. High-performance liquid chromatography was performed to determine the concentration of DON and ZEA mycotoxins produced by the fungi. GC-MS analysis was used to identify volatile organic components produced by two studied species of Fusarium. A custom-made, low-cost, electronic nose was used for measurements of three categories of samples, and Random Forests machine learning models were trained for classification between healthy and infected samples. A detection performance with recall in the range of 88-94%, precision in the range of 90-96%, and accuracy in the range of 85-93% was achieved for various models. Two methods of data collection during electronic nose measurements were tested and compared: sensor response to immersion in the odor and response to sensor temperature modulation. An improvement in the detection performance was achieved when the temperature modulation profile with short rectangular steps of heater voltage change was applied.
Read full abstract