II. EXPERIMENTAL SYSTEM
The important problem in the efficient performance of an electronic nose used for recognition of the forgery of coffee
brands is its sensitivity to small changes of odor. The change of concentration of odor at small amount of the mediocre coffee brand added to the high-quality one represents quite an important measurement problem, particularly if we associate the measurement with the interference effects occurring due to the changes in humidity or temperature of the environment. The conventional construction of the electronic nose is not able to make such recognition with a sufficiently high accuracy.
In this paper, we propose to apply the so-called differential nose system introduced in [17]. The two main components of
this system are the sensing subsystem and the automated pattern recognition subsystem, as shown in Fig. 1. The sensing system is composed of two chemo sensor arrays composed of the same types of sensors [17], working in parallel. One of them is the so called “measurement array,” and the second is the “reference array.” Both act at the same environmental conditions. The signals measured by both arrays are subtracted from each other and produce the differential signal, which is subjected to further processing. The general scheme of the solution, presenting both subsystems, is presented in Fig. 1. The streams of air are delivered from the odor acquisition place to two chambers containing identical gas sensor arrays. Inside the measurement channel, there is a vapor of the coffee coming from the sample of coffee beans placed in it. The reference channel is supplied only from the outside air and is free from the coffee smell.
Both arrays provide simultaneous information of the actually measured odors (as with a traditional e-nose), but only their
differential signals are stored and processed in further stages. Owing to this, we avoid the procedure of baseline calibration since we process only the differential signal. Observe that only differential signals are processed, which means that the effects of the drift of the Metal-Oxide-Semiconductor (MOS) sensors and a long time sensor exposure are greatly reduced. Additionally, such organized measuring system enables to make all measurements in online mode since we avoid the lengthy procedure of the baseline estimation.