Abstract—Density based Odor classification using EEG isan emerging issue nowadays.
As our environment becomes polluted with differentgases, so it is necessary to know that which gas is present in the air and inwhat density. Our work in this paper gives an elementary approach to solve thisproblem. We have used liquid stimuli with three different concentration levelsas Low (25% aroma and 75% water), medium (50% aroma and 50% water) and High (75%aroma and 25% water). General type-2 Fuzzy Classifier is used to classify thethree different density stimuli. An accuracy of 86% is obtained in thisexperiment. Thus, we can show that different density stimuli can be separablewith EEG signals.
The accuracy level can be further increased with other improvedclassifiers. Keywords- Liquor Density Based SmellClassification; General Type 2Fuzzy Set; Principal Component Analysis; EEG; I. IntroductionOdor classification plays an important role in understandingthe brain computation of object recognition 1, 2. Animals can use chemicalsignals to understand ecological information from the environment. Thosechemical signals frequently change in different concentration to conveydifferent messages. Here, in this paper, we try to recognize different odorousstimulus across different concentration level. This experiment can be used as apreliminary tool for diseases diagnosis in medical technology 4. Thisapproach can also be utilized in coal mines to observe the changes in odorsensing of coal mine workers during gas exposure 3.
Depending onpsychological dimensions of human odor, perception is a vital issue in olfactoryresearch. There has been extensive research in various disciplines for characterizedodorant quality and description using pattern recognition technique 5.Amongst all these studies, our work adds a new direction in the field ofolfactory classification.In literature 6, a recurrent neural networkmodel is designed to classify different aromatic stimuli and discriminate themusing EEG analysis.
The effects of each factor of human odor perceptualqualities are discussed in 7. Article 8 reveals that the odor qualityvaries with the concentration of the odorant stimulus. An electronic nose isdeveloped to classify odor and classify using neural network in 9-10.In the process of human smell perception, all thearomatic stimuli are sensed by the receptors which are located in the olfactoryepithelium 6. Odor molecules are then transferred through several hundredreceptors for perceiving a particular olfactory stimulus. According to thevarious concentration of aromatic stimulus, different kinds of EEG signals aregenerated from different brain regions.
Olfactory perception is highlyassociated with four parts of brain region namely prefrontal, frontal,temporal, and parietal. In this paper, we try to design an odorclassification depending on their molecular concentration. Here, we usedifferent types of smell stimuli like perfume, Dettol, Acetic Acid and Alcohol.Then the subjects are asked to perceive these smell stimuli in low, medium andhigh dilution according to different concentration levels. For classificationof odor perception we use here General Type-2 Fuzzy Set (GT2FS) InducedClassifier 11.
This proposed system is smart enough to classify the mixtureof different concentration level of different liquid aroma.The paper is divided into six sections. InSection II, we provide the basic overview of the proposed system. In SectionIII, we illustrate the details of classifier design.
Experimental details aregiven in Section IV. Classifier performance analysis is undertaken in SectionV. Conclusions are listed in Section VI. II. System OverviewThis section introduces an overview of theclassification based on odor concentration using EEG signal analysis.
The blockdiagram of the entire system is given in (Figure. 1). For recording EEGsignals, electrodes are placed on the scalp of human subjects. First, all thesmell stimuli are pre-processed and filtered to remove the eye-blinkingartifacts and other noise. Then the pre-processed signal is used to extract itsindependent features in the features extraction 12 block. We use PowerSpectral Density (PSD) 13 as the feature extraction technique.
However, allthe extracted features do not contain important information; hence we applyPrincipal Component Analysis (PCA) 14 to select most significant features as feature selection technique as well as toremove the unwanted features. The selected EEG features are applied to classifyodor concentrations using GT2FS 15 induced Classifier 16, 17.