Odor Sensing Techniques:

ABSTRACT

With advances in technology and the never-ending goal of making life simpler for humans, it is obvious that odor sensing could lead to a better tomorrow. This chapter addressed the multiple cases in which odor sensing could be used and applied specially when identifying individuals. Various research has been carried out in this field using multiple other methods to assist create this field of studies. Most of research has been specifically focused on a single industry or field of application of odor sensing techniques. The work focused on and developed a system using artificial neural network with odor sensing techniques and laid the foundation for a general-purpose system that can be used for authentication and identification of individuals.

The expression “biometrics” is originated from the Greek words “bio” which means “life” and “metrics” which means “to measure”. The huge advancements in the field of computing processing enabled the emergence of automated biometric systems over the last few decades. A considerable number of these new mechanized approaches is developed by utilizing the ideas that were primarily imagined hundreds, even decades ago (Nancy, 2012).

Human face is one of the oldest and most fundamental characteristics utilized in identifying people. Since the start of the civilization, people have utilized appearances (faces) to distinguish known (familiar) and obscure (unfamiliar) individuals. This straightforward errand turned out to be progressively more testing as populaces expanded and as increasingly helpful techniques for movement brought numerous new people into-one little communities. The idea of human-to-human acknowledgment is likewise observed in social dominating biometrics, for example, voice and gait acknowledgment. People utilize these qualities, to some degree unwittingly, to perceive known people on an everyday premise.

Additional human special features have likewise been utilized since the commencement of human advancement as increasingly formal approaches to identify individuals. Stephen Mayhew in his article “biometrics update” (Stephen, 2019) highlights on the history and explorations of Biometrics methodologies and usage throughout the history of mankind. The following paragraphs spotlights on some different methods used throughout the history to identify individuals reported by Stephen Mayhew.

Currently, biometrics denoted a defining moment in the worldwide acknowledgment. It is widely used across industries, including mobile, businesses, airports, borders and healthcare and it is publically accepted. The most consumer awareness of biometrics technology is the incorporation of advanced biometric techniques into consumer devices. There are many applications for the use of Biometric Technology, but the most common ones are as follows: logical access control; physical access control; time and attendance; law enforcement; surveillance.

Customarily, authentication is achieved by utilizing an individual's belonging, something he has, or an individual's information, something he knows. For example, by possessing a house key, individuals are assumed to have authorization to enter the house, because he/she possesses the house key. The authentication device, in this case, is the house lock. Biometrics is been presented as the next huge thing in verification and authentication, supplanting or enhancing the idea of “things that you know” such as PINs and passwords. These traditional methods are the not the preferred approaches, if we aim to make individuals identification as transparent as possible. Biometrics technologies enables a substitution approach to authenticate individuals, using distinguishing physical characteristics of the human body, such as fingerprints, face recognitions, iris recognition or characteristics actions of individual such as written signature.

But despite lots of advances in the realm of biometric authentication, it's clear that there's still plenty of room for improvement. One of recent techniques used to identify individuals is through body odor or what is known as individual odor signature.

Scent, which alludes to odors or smells, can be used as a signature to recognize certain issues or sources of attention. These incorporate air contamination, environmental pollution, malady diagnostics, food classification, food spoilage detection and as a tool for crime investigations. Consequently, in the course of recent years there has been a developing interest, particularly in the territories of safeguard and national security, in the likelihood of utilizing human smell signatures as biometric identifiers. For many decades, dogs have been utilized by law requirement work force to recognize or track people by their smell signatures. The huge accomplishments of these exceedingly prepared animals give some confirmation of essential proof that people can relate to an unmistakable odor signature. Yet, there is no instrument that can substitute for the nose of an all-around prepared dog.

Odor composed of volatile organic compounds (VOCs) and it is well known that there are hundreds of VOCs in human odor (Sichu, 2009). If sufficiently nearby, the human olfactory system can smell out the human body odor with an affectability of 1 Parts-Per-Million (ppm) to sub Parts-Per-Billion (ppb) level (Pearce, 1997). Along these lines we utilize the affectability of a human olfactory system as a kind of perspective for human odor recognition in air matrices. It should be called attention to that this affectability level could be considered as a base prerequisite for human smell identification in air matrices. This is because, in many cases of human identification, it is still important to utilize the more delicate dog nose to smell out the odors. Li reported that it is believed that the limit of detection (LOD) of the dog nose in detecting VOCs in air matrices can be as low as Parts-Per- Trillion (ppt) levels (Sichu, 2009). Biological noses have both high affectability and specificity for smell out odors, and they are utilized as a model for Electronic Nose (Enose) design and advancement.

On the other hand, electronic noses are being created as frameworks for the automated recognition and characterization of smells, vapours and gasses. The two principle parts of an electronic nose are the odor detection framework and the computerized pattern recognition framework. The odor detection framework can be an array of several different sensing elements (e.g., chemical sensors), where each element measures a different property of the sensed odor, or it can be a single sensing device (e.g., spectrometer) that produces an array of measurements for each odor, or it can be a combination of both chemical sensors and a spectrometer. By showing a wide range of smells to the sensor array, a database of signatures is developed. This database of marked odor signatures is utilized to train the pattern recognition system. The objective of this preparation procedure is to arrange and configure the recognition system to deliver extraordinary mappings of every odor so that an automated distinguishing proof can be executed.

The aim of this chapter is to build on top of our research work in the field of odor sensing and electronic nose (Karlik & ALbastaki, 2005), (Albastaki, 2009), (Albastaki & Almutawa, 2013) and (Albastaki & Albalooshi 2018) to describe a framework for the detection of human odor signature using computerized artificial neural network system. This computerized system enormously builds the chances for utilizing human odor signature for personal authentication. Active researches in biometric person authentication emphasize the urgent demand for inexpensive, simple, and fast early qualitative authentication process. The chapter discusses the development of a computerized system using OMX-GR sensor to recognize human odor signature.

The remaining of the chapter is organized as follows: Related work of the field is discussed followed by a highlight on different types of existing biometric technologies used for authentication purposes. Finally, data collection and analysis will be described with the major findings of this research study.

Recent research reviled that the presence and variation of multiplicity of odor known as volatile organic compounds (VOC) in the surrounding of human body can be used as a mark for uniqueness of everyone known as human body odor signature. Therefore, to build a sold research foundation for measuring human body odor, the following paragraph will summarize the main findings in this field through reviewing the literature with a special attention given to detecting and sensing human body odor technologies and theories.

Rajan et al. in their article “Chemical Fingerprinting of Human Body Odor: An Overview of Previous Studies” (Rajan et al., 2014). They reviewed several theories touching the human body creation of scent and its characteristics. The article identified the primary human body scent to be responsible for individual identification through classifying human body odor into three different components and named them as primary odor, secondary odor and tertiary odor. The study focused and highlighted the past research that have been conducted to measure and anatomize the combination of VOCs existence in human body odor. Different methods of sampling are used in this study to collect human body odor by using a gauze pad continuing scent materials collected by direct contact with the human cuticle or evidence articles, using Scent Transfer Unit (STU), or by swiping the surface containing the scent with the gauze pad. Separation technology is utilized in this study to conduct the required analysis to extract scent compounds using the gas chromatography coupled with mass spectrometer. Alternative studies utilized the concept of electronic nose to collect human scents. Body odor and its use in forensic investigation are elaborated in this article by comparing the utilization of canine in identifying individuals and contrasting it with identifying individuals through their body productions of scents. Theories of Scent Productions are also investigated in this article. It identifies two theories for the production of scent: the Skin Raft Theory and Logical Source of human scent. The Skin Raft Theory by Syrotuck 1972 (Stockham, Slavin & Kift, 2004) demonstrates how the bacterial on the cuticle and the physical liquids shed by the Integumentary system participates in the uniqueness of the raft and as a result the scent produced. Whereas the second theory of scent production claims that the logical source of human scent might be one of the different secretions normal to the human cuticle (Tebrich, 1993). As a conclusion this article shows how different methods are used to establish the basis for utilizing human scents as the individualization biometrics. As a result, this study reviles that all the analytical studies indicate that human scent analysis is a viable method that can be used to identify human.

Bhattacharyya and his team (Bhattacharyya et al., 2009) in their article “Biometric Authentication: A Review”, reviewed the biometric authentication techniques such as figure print, face recognitions and IRIS in general with a special focus on some future development in the field. This research article demonstrates how different biometrics characteristics of humans are utilized to identify individuals. The majority of information technology related systems needs a reliable personal features and recognition schemes to either assure or find out the identity of an individual seeking these systems services. The aim of such systems is to guarantee that only a lawful customer and not anyone else accesses the rendered services. Utilizing biometrics, the identity of an individual can be confirmed or established. In this technique, the position of biometrics in the present security sector was portrayed. This article also described different views on the usability of biometric authentication technologies, comparison of distinct methods and their benefits and disadvantages. This research work then illustrates the degree of security in different biometrics techniques adopted by the researchers. Different techniques are evaluated to show the degree of security when applying these types of biometrics authentications methods. Evaluation factors such as False Accept Rate (FAR) and False Match Rate (MAR), False Reject Rate (FRR) or False Non-Match Rate (FNMR), Relative Operating Characteristic (ROC) and Equal Error Rate (EER) are adopted in this research work. The article concludes that Biometric authentication is extremely reliable, as it is much more difficult to fake physical human features than security codes, passwords and hardware keys.

Zhanna (Zhanna, 2019) in his article “Biometric Person Authentication: Odor” a human identification issue is described through the authentication of the odor. This is the method of view that is still under growth. There are no commercial apps available yet on the market. While recognition of human odor is not yet accessible, nowadays odor recognition is commonly used. The author claims that technology needs become progressively advanced and less costly, biometric instruments are becoming more common as a type of identification. Vendors are already selling fingerprint recognition technology to producers of automated teller machines on computer keyboards or iris recognition. The article raises and important question: can we use the odor to recognize individuals? Medical researcher Lewis Thomas first suggested a connection in the mid-1970s between immunity and body smell. A collection of immune genes has already been related by scientists to a distinctive human body. As a result of previous research, the author claims that it is totally evident that distinct body odors are produced by individuals with distinct immune genes. Each person has a distinctive body odor that is about thirty distinct odorants in conjunction. Human body odor's primary aim is not only to identify these whole parts, but also to estimate their concentration.

(Inbavalli & Nandhini, 2014) in their article “Body Odor as a Biometric Authentication” argue that research on biometrics has noticeably increased. Biometric systems such as Fingerprint, retinal scan, face, voice, iris, signature and hand geometry are in use today, but they have several drawbacks. The authors claim that recent studies disclosed human odor's uniqueness. Compared to other biometric identifiers such as iris, fingerprints and face recognition, the body odor as a biometric identifier has the smallest error rate (15%) and this mainly because that human odor cannot be replicated. Body odor shows strong authentication over other biometric technologies that have lately emerged. This article is about developing a model scheme that authenticates individuals based on their body odor. In conclusion this article claims that odor detection for authentication in the biometric domain is a novel concept. This would lead in improved safety systems if implemented. The added benefit is that this strategy is contactless.

An interesting research paper in this field is presented by (Oyeleye et al., 2012) “An Exploratory Study of Odor Biometrics Modality for Human Recognition”. They claim that the global security issues that are presently recurring and alarming have resulted to the growth and use of biometric methods for access control and human recognition. Although several biometrics for human recognition and access control usage have been suggested, investigated and assessed; it becomes apparent that each biometric has its strengths and constraints as each best fits a specific identification security application. There is no biometric modality, therefore, that is ideal for all applications. This opens a wide gap in introducing and applying newly emerging biometric methods for individual recognition and identification. Biometrics frequently applied or studied in security systems, however, include fingerprinting, face, iris, speech, signature, and hand geometry. The article argues that several newly emerging biometric modalities, including Gait, Vein, DNA, Body Odor, Ear Pattern, Keystroke and Lip, are less studied, understood, researched, promising to deliver better output, acceptability and circumvention. Finally, the article concludes that body odor, which considerably displays powerful safety potential over other newly emerging modalities, may prove very efficient for precise personal identification, little is known about its basic characteristics and human recognition suitability.

As a result of the above related work review, people with differing immunity genes produce different body odors. Odor detection for authentication in the biometric domain is a novel concept. This would lead in improved security systems if implemented.

Many biometric methods are now accessible; some of them are only at the research stage (e.g. the analysis of odors). However, a large amount of techniques is already mature and accessible commercially. Approximately ten distinct biometric technologies and methods are now a days accessible on the market. In this section we attempt to briefly describe the most common biometrics methods available nowadays.

Fingerprints Biometrics Technology: One of the oldest of all biometric methods is the identification of fingerprints. Figure 1 illustrates fingerprint biometrics process. This method maps the individual's fingerprint pattern and then compares the ridges, furrows, inside the template.

The fingerprint provided to the device is first searched in the database at the coarse stage and finer comparisons are then made to obtain the outcome. Since the last century, their use in law enforcement is well the earliest known and effectively let a fingerprint equal to crime association.

Iris Biometrics Technology: Each iris is a unique, complex pattern structure. This can be a mixture of features called corona, crypts, filaments, freckles, pits, furrows, striations, and ring. The iris is the colored tissue ring that surrounds the eye's pupil. Even twins have distinct patterns of iris, and the left and right iris of individuals is also distinct. Figure 2 illustrates iris biometrics process. Low intensity light source scans the iris and retina and compares the picture to the patterns stored in the database model. This is one of biometry's fastest technique.

Facial Recognition Biometrics Technology: The most natural means of biometric identification is facial recognition. The technique of separating one person from another is nearly every human being's capacity. The recognition of the face has never been regarded as a science until recently. It is possible to use any camera with adequate resolution to acquire the face picture. Usually the facial recognition systems use only the data on a gray scale. Colors are used only to assist locate the face in the picture. Facial scanning includes scanning the whole face and checking with the template of critical points and regions in the face. Figure 3 illustrates iris biometrics process.

Hand and Finger Geometry Biometrics Technology: Hand geometry biometrics technique is based on the reality that the hand of almost every individual is shaped differently and that after a certain age the shape of the hand of a person does not alter. This technique utilizes information such as length, shape, finger distance, general hand size as well as the comparative angle between the fingers. In conjunction with the Fingerprint scanning method, modern devices use this method. Different techniques are used for hand measurement. Most frequently, these techniques are based on either mechanical or optical principles. Figure 4 depicts a hand and finger geometry biometrics system.

Your whole hand provides unique identifiers beyond a mere finger. Hand geometry readers use a simple concept of measuring and recording the length, width, thickness and surface area of an individual's hand while guided on a plate, according to the National Science and Technology Council. Some scanners can measure the distance between the fingerprints and the knuckles, the curvature of the digits and even the shadow cast.

Voice Biometrics Technology: Also referred to as speaker authentication. The speaker verification principle is to evaluate the user's voice in order to store a voiceprint that is later used to identify or verify and individual. Verification of speakers and recognition of speech are two distinct tasks. The purpose of speech recognition is to find out what principle has been said while the speaker's objective is to verify who said that. Figure 5 illustrates the process of voice biometrics technology. This technique can confirm an individual based on speech patterns. Everything can be attributed to people by combining pitch, velocity and talking style. Users enroll by offering their credential template with a voiceprint. Speakers will utter an agreed sentence or word to obtain access, which may be prevalent to all people registered or a password particular to each individual.

Other Biometric Techniques: In the previous paragraphs we discussed the most popular biometrics techniques available and widely used in different applications. There are many less popular biometric techniques such as:

This research work tackles relatively a new and active research field for individual biometric authentication technique based on body odor. The biometrics of the body's odor is based on the reality that almost every human odor is distinctive. Sensors those can obtain the smell from non-intrusive areas of the body such as the underarm and back of the hand capture the smell. Every human odor consists of chemicals known as volatiles. The proposed system extracts them and transforms them into a template. Using body odor detectors raises the problem of privacy as the body odor carries a significant quantity of delicate private data.

This chapter aims on developing an Odor sensing system using OMX-GR odometer sensor as a hardware and Artificial Neural Network as software for general odor sensing system and applying it to identify individuals (Figure 6).

Various Artificial Intelligence techniques have been developed and are used in different fields. For this research, we analyzed three most relevant techniques: Artificial Neural Networks (ANNs), Decision Trees and XGBoost (Natale et al., 2003) and Decision Trees and Random Forest.

For this research, the algorithm we choose is Artificial Neural Networks over Decision Trees with XGBoost and Decision Trees with Random Forest. The reasons behind this are as follows. First, in terms of outputs, ANN is more flexible and can be used for single or multiple outputs. In addition, ANN provides the “Back propagation” feature, which requires minimal supervision and requires only input data to train itself, making it much easier to train for large amounts of data and more flexible for different types of applications, enabling the same network to be used in different fields. Lastly, with more training, the ' weights ' of an ANN's nodes are automatically adapted whenever trained to deliver more accurate outcomes, this characteristic can lead to our application being simplified so that less tech-savvy individuals can use it, thus improving its usefulness. The following paragraphs describe ANN briefly.

Artificial Neural Networks: ANN consists of nodes in various layers. As shown in Figure 7 the layers are input layer, intermediate hidden layer(s) and output layer. The links between nodes of adjacent layers has weights. The goal of training is to assign correct weights for these edges. These weights determine the output vector for a given input vector (Wilson & Baietto, 2009).

At the beginning, all edge weights are assigned randomly. The ANN is performed for each input in the training dataset. The obtained output is compared to the already known intended output, and if an error occurs, the error will be propagated back to the previous layer. The weights are adjusted by the error. This process is repeated until the error in the output is equal to or below a default limit.

A trained ANN is resulted once the algorithm is finished. The ANN is then at a functional stage to work with fresh inputs. This ANN can then be further trained by making the findings more precise by offering more input information.

As shown in Figure 8, the scope of this study is to produce a odor sensing scheme capable of distinguishing between individual sample odors using the hardware OMX-GR sensor and Artificial Neural Network (ANN)-based software. Hardware work and software execution details are outlined in this chapter's upcoming parts.

An OMX-GR smell sensor as shown in the figure 9 was used for odor sensing. It detects odors using two semiconductor gas sensors. Odor strength and classification (ID) calculated using an initial Shinyei technology technique is presented on the screen via its LCD display. There is no correlation between the OMX-GR value stated and the human sense of odor intensity value.

The ID is an integer with a range of 0 to 90 and an integer with a range of 0 to 999 is also the power or the strength of the odor. The response rate to various compounds and liquids is distinct as follows:

Using the custom-made application, an RS-232 cable was used to connect and read the strength and odor ID of the sensor.

Data collection and its analysis will be addressed in this chapter after we defined our problem and covered the necessary background theory. The argument will begin with data collection, in which the problems encountered in data collection will be examined and resolved. After that the gathered data will be evaluated, processed and lastly the outcomes.

Data Collection and Problems: Five body odor samples from underarm of two individuals’ persons were gathered using the OMX-GR sensor. However, many problems were confronted in odor samples collection as the sensor was old and could not be substituted with a fresh one owing to the budget constrain of the research.

First, we discovered that the sensor did not have a normal response time to the body odors even when the odors were submitted to the sensor from a uniform range. This meant that the response time of odors could not be used as an input parameter for the neural network, which could have resulted in more precise outcomes. The response time of three odors is stated in the sensor handbook from which it can be concluded that the sensor initially had a normal response time. Nonetheless, as the sensor had grown old, its response time might have been impacted.

Next, when we attempted to read distinct odors in fast succession, the sensor gave entirely distinct measurements to the odor measurements earlier acquired. After research, we learned that distinct sensors have distinct retrieval times in relation to response time. Unlike reaction moment, however, the retrieval time was not stated in the handbook of the sensor. Therefore, to define the retrieval time, we began by enabling the sensor a retrieval time of 5 seconds and checking the readings consistency. The consistency of the readings did not improve, so we increased the recovery time further, this continued until we had increased the recovery time to 10 minutes and the consistency of the readings had not improved, as sometimes the sensor would not respond to an odor even after 10 minutes of recovery time. We therefore found that the sensor no longer had a normal reaction or retrieval time.

Then it also came to our understanding that environmental variables, i.e. temperature and humidity, influence sensor measurements. This has also been verified in numerous researches. To decrease the impact of environmental variables, tubes of variable dimensions were used to directly expose odors to the sensor and limit the impact of the environment on them.

Even after taking all the above measures to enhance the measurements, coherent measurements were not acquired. Therefore, the following technique of purification was used to achieve coherent measurements.

Data Analysis and Data Cleaning: In data analysis, the readings of the 5 body odors for two individuals will be analyzed. A total of 600 readings were taken for two volunteer’s body odors (300 samples for each) over 5 sessions of 60 readings each. Each reading has an ID and strength. Figure 11 and 12 shows the density for IDs readings and strengths readings respectively.

Figure 11 shows the ID density values of the odor readings. The ID values of the first-person odor range between 63.5 and 66.5. Figure 12 shows the strength density values of the odor readings. The strength values of the first person’s odor range between 135 and 165.

The basic design and architecture are described in figure 13. The design basically collects individual body odors by a chemical sensor array and fed to the artificial neural network system for odor identification and classification. Consequently, the classification will be retrieved for authentication purposes.

In order to process the information and to identify the odors using the neural network, the measurements were split into a proportion of 20, 70 and 30 for validation, training and testing according to the results of the Neural Networks Division. This implies that 20 percent of the complete readings (1500) were randomly chosen for validation set. Of the remaining 1200 readings, 70 percent were randomly chosen to train the network and the remaining 360 readings (30 percent) were used to test the qualified network.

Different architectures were used to train the neural network to assess which one would be the most suitable. Training was done until an error rate of 0.1000 was obtained. The details of trainings using different architectures are given in table 1, where the architecture of the neural network is represented as number of input layers: number of neurons in the hidden layer: number of neurons in the output layer.

Table 1. Training results of different architectures

Architecture	Training Time	Epochs Completed	Validation Rate
2:2:1	1h 14m 04s	3,740,445	0.03968
2:3:1	0h 03m 59s	178,999	0.03673
2:4:1	0h 02m 41s	68,0998	0.03809
2:5:1	0h 02m 02s	65,957	0.03098
2:6:1	0h 01m 41s	4,876	0.03752
2:7:1	0h 01m 59s	36,613	0.03801

After analyzing the results obtained in table 1 we can conclude that the best architecture for the information submitted to the neural network was 2:6:1. Training time reduced as the amount of neurons in the hidden layer increased until the amount of neurons in the hidden layer was 7. The number of epochs reduced as the number of neurons in the concealed layer increased. However, there was no important change in the validation rate, but the 2:6:1 architecture also obtained the highest validation rate. A simple 2:2:1 architecture was not suitable as there were overlapping data in both ID and Strength values, and therefore the neural network took a lot of time to train itself to attain a maximum error rate of 0.1000. It should be noted that, regardless of which architecture is chosen for training, if the required maximum error rate is achieved during training, the testing and querying accuracy of the neural network will not be significantly affected by the architecture of the network.

After training the network with the purified information, testing was conducted to evaluate the network's efficiency. The neural network properly categorized all 360 test samples, which meant that a 100 percent precision rate was achieved as shown in Figure 14 produced by the program.

With advances in technology and the never-ending goal of making life simpler for humans, it is obvious that odor sensing could lead to a better tomorrow. This chapter addressed the multiple cases in which odor sensing could be used and applied specially when identifying individuals. Various research has been carried out in this field using multiple other methods to assist create this field of studies. Most of research has been specifically focused on a single industry or field of application of odor sensing techniques. Our work focused on and developed a system using Artificial Neural Network with an odor sensing technique and laid the foundation for a general-purpose system which can be used for authentication and identification of individuals.

The odor meter used in this research was an OMX-GR sensor which is mainly a semiconductor-based gas sensor. Artificial Neural Network was used as software, together with the smell sensor as hardware, which was built on top of the Sinapse framework and the framework itself was heavily modified to fit the research needs. Due to the sensor being old, and its other various limitations. Many methods had to be tried for data collection. Adjustments had to be made to the distance between the sample and the detector while gathering measurements, its retrieval times, and other internal variables to guarantee accurate measurements. The data was also cleaned and went through a purification process.

Data was collected and purified before it was fed to the ANN to train on and then the Artificial Neural Network was tested for accuracy. This method has been repeated for all the samples collected. Backpropagation algorithm was used to train and adjust weights to guarantee accurate outcomes.

Findings: The scheme established in this study could distinguish between different kinds of odors with a precision of between 93-100 percent. It indicates that Artificial Neural Networks can be used very effectively with smell sensors to detect multiple odors owing to the big quantity of information engaged in the nature of the issue and ANN's ability to learn and train further as more information is supplied, thereby improving its accuracy.

Limitations: There were numerous constraints affecting the study, primarily those linked to the smell sensor. The smell sensor used in this study is an OMX-GR sensor that was old. Because of this, the technique of information collection had to be adjusted to the restriction of the sensor. Some of these constraints linked to connectivity problems as the sensor's Serial Port utilizes legacy drivers that are uncommon to discover. Therefore, in order to make the sensor compatible, required drivers had to be implemented within the scheme. In addition, due to the absence of comprehensive sensor documentation, numerous experiments and tweaks had to be made to the information collection techniques to guarantee that the most precise information was obtained. Some of the information collection problems linked to the sensor's recovery and response time

Some constraints linked to software were also the restriction of the system used. This resulted to the development of numerous new features to make the system perform better and quicker and make it much easier to use.

Future Implementations: In the future, research could be performed to test the system with multiple other sensors that are contemporary and accurate to get measurements of more odors with more parameters, i.e. response time, to have a high precision with a greater number of odors. In addition, the capacity to develop a cloud-based data hub where different research could contribute with their information and samples to train the ANN could also lead to further innovations in the field of odor sensing. As far as sophisticated software is concerned, it can be improved by combining the ANN with a Genetic algorithm to avoid being stuck in local minima while training.

Albastaki, Y. (2009). An Artificial Neural Networks-Based On-Line Monitoring Odor Sensing System . Journal of Computational Science , 878–882.

Albastaki, Y., & Albalooshi, F. (Eds.). (2018). Electronic Noses and Technologies and Advances in Machine Olfaction . IGI Global. doi:10.4018/978-1-5225-3862-2

Albastaki, Y., & Almutawa, K. (2013). ANN Based Approach to Integrate Smell Sense in Multimedia Systems, Technology Diffusion and Adoption: Global Complexity, Global Innovation: Global Complexity, Global Innovation . IGI Global.

Bhattacharyya, D., Ranjan, R., Alisherov, F. A., & Choi, M. (2009). Biometric Authentication: A Review. International Journal of u- and e- Service . Science and Technology , 2(3), 13–27.

Inbavalli, P., & Nandhini, G. (2014). Body Odor as a Biometric Authentication. International Journal of Computer Science and Information Technologies , 5(5), 6270–6274.

Karlik, B. & Albastaki, Y. (2005). Bad breath diagnosis system using OMX-GR sensor and Neural Network for telemedicine, Computer Medicine ‘2005, Scientific-practical Conference [eHealth]. ElectronicHealthcare , 23–25.

Nancy, Y. L. (2012). Bio-Privacy: Privacy Regulations and Challenge of Biometrics. Academic Press.

Natale, D. C., Macagnano, A., Martinelli, E., Paolesse, R., D’Arcangelo, G., Roscioni, C., & D’Amico, A. (2003). Lung cancer identification by the analysis of breath by means of an array of non-selective gas sensors. Biosensors & Bioelectronics , 18(10), 1209–1218. doi:10.1016/S0956-5663(03)00086-1

Oyeleye, C. A., Fagbola, T. M., Babatunde, R. S., & Adigun, A. A. (2012). An Exploratory Study of Odor Biometrics Modality For Human Recognition. International Journal of Engineering Research & Technology, 1(9).

Pearce, T. C. (1997). Computational parallels between the biological olfactory pathway and its analogue. The Electronic Nose ‘: Part II. Sensor-based machine olfaction . Bio Systems , 41(2), 69–90. doi:10.1016/S0303-2647(96)01660-7

Rajan, R., Fakhuruddin, N., Hassan, N., & Nasimul Islam, M. (2014). Chemical Fingerprinting of Human Body Odor: An Overview of Previous Studies. Malaysian Journal of Forensic Sciences , 4(1), 33–38.

Sichu, L. (2009). Overview of Odor Detection Instrumentation and the Potential for Human Odor Detection in Air Matrices. MITRE Nano systems Group, MITRE Innovation Program and U.S. Government Nano-enabled Technology Initiative, Project No. 07MSR216 and 15095320, Dept. E552.

Stephen, M. (2019). Biometrics Updates. Retrieved from https://www.biometricupdate.com

Stockham, R.A., Slavin, D.L., & Kift, W. (2004). Specialized Use of Human Scent in Criminal Investigations. Forensic Sciences Communications, 1-12.

Tebrich, S. (1993). Human Scent and Its Detection. Retrieved from: https://www.cia.gov/library/center-forthe- study-of-intelligence/kentcsi/ vol5no2/html/v05i2a04p_0001.htm

Wilson, A., & Baietto, M. (2009). Applications and Advances in Electronic-Nose technologies. Sensors (Basel) , 9(7), 5099–5148. doi:10.3390/s90705099

Zhanna, K. (2005). Biometric Person Authentication: Odor. Techylib. Retrieved from https://www.techylib.com/en/view/nauseatingcynical/biometric_person_authentication_odor_2

Chapter 4

Odor Sensing Techniques: