EEG is one of the most important methods of evaluating brain disorders and monitoring the electrical behavior of the brain. The EEG also has the major advantage of excellent temporal resolution, which enables it to study neural activity at a millisecond scale, and best approximates the neural timing. EEG allows for the analysis of the various rhythms generated by different cortical regions. The current produced by the electrical activity of neurons reaches the surface of the scalp. EEG offers a non-invasive method of recording the difference between the potentials that are generated by neural sources and annoying non-neural artifacts. As the signal induces important temporal and spatial variations, the electrode positions are determined using multiple channels settled by the international 10–20 standard. Recommendations for the use of EEG equipment in assessments are provided by the International Federation of Clinical Neurophysiology (Babiloni et al., 2020). The electrical signal is diffused from electrodes placed on the scalp to an external amplifier that intensifies the potentials. EEG signals are commonly detected between 20 and 150 μV in the 0.5–60 Hz band (Binnie et al., 1982). The signal is constantly sampled to provide a relevant temporal resolution to explore event-related potentials (ERP) and the EEG power spectrum. In research focused on frequency-based analyses (such as prefrontal lateralization of alpha or beta bands), a sampling rate of 128 Hz can be sufficient. When the objective is high time precision measurements (such as language-related high gamma activity), the EEG should ideally collect data at a high sampling rate (>500 Hz).

EEG analysis could be conducted in the time domain, frequency domain, or time-frequency domain. Extracting temporal features (e.g., amplitude, power, average periodicity, and synchronization) provides useful qualitative information for the classification. However, all of these temporal characteristics do not describe the signal in its integrity. During a seizure, for instance, the signal is not stable, and it is necessary to separate the frequency components to classify the seizure. Therefore, in addition to time-domain features, frequency domain exploration is also needed to detect and classify all types of seizures. A time-domain analysis provides better spatial information, thus poor frequency content information is required for EEG classification. The frequency-domain can provide time information when the function is windowed. The choice of window size is the biggest challenge in frequency analysis. Time-frequency analysis solves these two problems. Some EEG investigations consider that wavelet analysis is the best method for time-frequency analysis. Generally, authors apply a series of transformations e.g., Fourier transform (Radha et al., 2014), Short Time Fourier Transform (Görür et al., 2003), Wavelet transform (Fraiwan et al., 2012), Hilbert-Hung transform (Li et al., 2009), and Empirical Mode Decomposition (Hassan and Bhuiyan, 2016). Spectral analyses, based on Fourier Transform, are then commonly used to convert the time function into different frequencies and to calculate the amplitude in each frequency band. The frequency bands are universally classified as the following: slow and sleep wave delta (2–4 Hz), arousal wave theta (4–8 Hz), relaxation wave alpha (8–12 Hz), and active wave beta (13–32 Hz). Sensorimotor rhythm frequency bands (13–15 Hz) are related to the sensorimotor rhythm and entitled as low beta. Delta waves are commonly frontally located in adults and posteriorly in children. Theta waves are mainly recorded in frontal areas during low brain activities, sleep, or drowsiness and cognitive processing. Alpha waves are among the first rhythmic waves documented and are recorded during relaxed conditions at decreased attention levels and in a wakeful state. The alpha waves are located in the occipital area and can be induced by closing eyes. Beta waves are often recorded in frontal or central areas when the eyes are open and are related to consciousness, alertness, arousal, and motor behaviors (Barry and De Blasio, 2017). Cognitive processes such as attention, learning, and diverse types of memory occur during gamma frequencies (over 33 Hz). Unconventional classifications have also been analyzed in some studies (Caldwell et al., 2002; Gevins et al., 2003; Dahlstrom and Nahlinder, 2009; Holm et al., 2009; De Vico Fallani et al., 2012; Zhang et al., 2019a). It is worth mentioning that the frequency limits of specific waves are conventional, as there is no proper way of determining their exact values.

Many studies on oscillations in brain dynamics have indicated that during fatigue accumulation and sustained attention, increased EEG power is detected in theta frequency bands in the frontal, parietal, and central regions. Theta power increase has been detected during working memory load situations (Klimesch et al., 2007), visual tasks (Yamada, 1998), flight simulations (Smith-Jentsch et al., 2001; Dussault et al., 2004; Borghini et al., 2014), and air-traffic control simulations (Postma et al., 2005). A decrease in the alpha is known to occur during complex and cognitively demanding tasks. It has also been shown (Postma et al., 2005) that alpha and beta bands are different between the beginning and the end of a mental fatigue task. Increases in theta and decreases in alpha oscillations have also been associated with an increase in the accuracy of task performance (Klimesch et al., 1999). Interestingly, such an increase in EEG power in the theta band can be used to characterize a single task from a multi-task activity performed by pilots (Borghini et al., 2014). Varied EEG approach montages are given in complex higher-order cognitive operations.

Although great importance has been placed on classic unmovable EEG montages, they have disadvantages in that they are much bulkier and more time and effort consuming than the new generation of wearable EEG. The concept of wearable EEG improves upon the bulky and limited mobility of classic montages, using small devices that can record EEG outside of laboratory conditions. These miniaturized devices have the advantage of being able to detect EEG signals for short or long periods, for example when sleeping, which significantly improves the brain-computer interface (BCI) monitoring method. The importance of using the wireless EEG in BCI monitoring is to maximize wearability, enabling unconstrained mobility, usability, and reliability in operational environments.

As well as involving long preparation time and bulky design, classical scalp-mounted EEG are not suited to situations that involve environmental artifacts (e.g., aviation and space operation, or patients with a cochlear implant or hearing aid; Nogueira et al., 2019). To address these issues, novel EEG technologies use tiny electrodes that are placed externally around the ear (Debener et al., 2015; Bleichner and Debener, 2017) or involve in-ear electrodes (Looney et al., 2011). The cEEGrid electrode array (Debener et al., 2015) is a promising device with 10 electrodes printed onto a C-shaped flexible board, enabling it to fit around and measure EEG data behind the ear. It also uses a small amount of electrolyte electrode gel, which ensures low-impedance contact between the cEEGrid electrodes and the skin. The capacity is stable because the gel does not dry-up. The cEEGrid electrode array is then connected to a micro wireless amplifier and uses signal recordings from a cellphone. This accessible design means that it has potential applications in clinical settings, aeronautics, and other research areas.

With a reduced number of electrodes, the difference between cEEGrid performance and scalp EEG performance can be explained by the position of the reference electrode. The cEEGrid uses a local reference, with ground and recording electrodes allocated around the ear, giving out small signal amplitudes. However, for a scalp EEG with a smaller number of electrodes around the ear, the tip of the nose is used as a reference, and its position is farther from the recording electrode, and therefore gives higher amplitudes. In line with conventional EEG results, Debener et al. (2015) identified ERPs and alpha frequencies during an auditory oddball task with open eyes using the cEEGrid. Advanced explorations proved that the cEEGrid can detect neural signals to select the voice of a speaker with high precision, and can be even used as a BCI monitor for hearing aids (Mirkovic et al., 2016). Recently, the cEEGrid was successfully used in decoding selective attention in normal hearing and cochlear implant patients (Nogueira et al., 2019). Looney et al. (2011) presented the in-ear EEG device. It records to the same standards as the conventional scalp electrodes. However, the device relies on custom-made hearing aid earplugs, which take a wax impression of the ears (outer ear and external ear canal). The important benefits of in-ear EEG includes easy set-up; time of installation; accessibility for people with hair that is incompatible with EEG recordings; and durability because they are fixed in the ear canal. They are comfortable to wear, discreet, resembling earphones, earbuds, and earplugs, and facilitating everyday use. Contrary to scalp EEG, the in-ear EEG devices are easy to place without the presence of experienced assistants. They are held firmly in place and thus diminish motion artifacts. Since the electrodes are fixed on the earpiece surface, they offer a precise spatial positioning which decreases the inter-experiment variability. The feasibility of in-ear EEG for cognitive assessment has been studied in a few exploratory papers (Stochholm et al., 2016; Von Rosenberg et al., 2016; Zibrandtsen et al., 2016), which claim that in-ear EEG is a promising candidate for forthcoming explorations based on human monitoring technology (BCI, aviation, and space).

Despite advances in mobile EEG systems, around and in-ear EEG have some drawbacks. Compared to conventional scalp EEGs, these alternative systems have fewer electrodes and cover much-reduced regions. Therefore, it is recognized that its brain source analysis is less accurate than that of conventional scalp EEG. Note that physiological artifacts caused by the electrical activity of the skin are unavoidable, but may be relatively easy to deal with by temporal filtering and other post-processing procedures (Reis et al., 2014). The quality of the scalp EEG signal depends on the connection between the amplifier input and the skin surface. Wet electrodes based on conductive gel guarantee low impedance levels (<10 kΩ). Given that dry electrodes are placed on the skin without any gel application, the dry EEG system typically results in larger impedance than wet systems (Brown et al., 2010; Chen et al., 2014). To date, few studies have directly compared the data quality between these two systems. A recent study (Hinrichs et al., 2020) found that the resting state EEG power and ERP were comparable between the two systems. Di Flumeri et al. (2019) evaluated three different dry electrode types when compared with wet electrodes in terms of signal spectral features, mental state classification, and usability. The dry electrodes included a gold-coated single pin, multiple pins, and solid-gel electrodes. The results confirmed the high quality achieved by dry electrodes. They offered the same levels as wet electrodes with significantly reduced times of montage and increased the comfort of users. Although the signal quality is inevitably reduced, the dry electrodes are a reliable system for non-clinical and goal-oriented investigations, such as a comparison between two different mental states during real flight conditions.

Since the late 1950s, many studies have investigated changes in EEG rhythms during flight conditions (Carl et al., 1959) despite the noisy environment (e.g., vibration, wind, acoustic noise, physiological artifacts, and important pilot physical movements). The recorded EEG signals show specific features and changes in the power spectrum of the various frequency bands associated with flight performance (Carl et al., 1959). One EEG experiment (Callan et al., 2015) conducted under real flight conditions, showed that the neural signature of inattentional deafness (e.g., inadvertently missing an auditory stimulus) was revealed by a reduction in phase resetting in the alpha and theta band frequencies. Sauvet et al. (2014) have used a single EEG channel during real long-haul flights to detect low states of vigilance. Another EEG study involved a critical scenario in a flight simulator (Dehais et al., 2016), which detected the existence of an early and unconscious gating mechanism based on the association between the auditory alarm and the N100 and P300 amplitude. Di Stasi et al. (2016) analyzed the in-flight EEG activity of military helicopter pilots during real flights. They found that highly demanding procedures related to takeoff and landing were associated with higher power EEG frequency bands, whereas less demanding procedures were associated with lower EEG power over the same frequency bands. Several EEG investigations have focused on the mental workload of the operator during a flight task execution. For example, EEG revealed variations in alertness and could predict lower performance caused by increased mental workload during flight operations (Borghini et al., 2014). Theta changes were observed over the frontal brain areas when comparing the training improvements of novice pilots in flight simulation tasks (Borghini et al., 2014).

Workload perception was shown to be dependent on the level of experience, the abilities, or just the individual differences between pilots. The EEG was sensitive to workload level variance between novice and expert aircraft pilots during the execution of the identical task (Doppelmayr et al., 2008; Parasuraman and Jiang, 2012).

Recent studies exploring Air Traffic Controllers (Bernhardt et al., 2019) have computed an EEG-based workload index that could differentiate between task workload requirements exploring front-parietal brain function. Yet, EEG exploration has achieved traction in aviation and space operations, current studies face challenges related to the intrusive and bulky nature of the equipment (Caldwell et al., 2002), the discomfort of long preparation time, and dependence on gel electrodes (e.g., wet electrodes). In a recent study, Dehais et al. (2019) tested a dry EEG system on pilots during the low and high load traffic patterns associated with the passive auditory oddball. Analysis of ERPs and frequency characteristics has confirmed that dry EEG can be used to study cognition during highly ecological and noisy conditions. Recent developments in dry electrodes (Liao et al., 2012) allow them to reduce preparation time by eliminating the conductive gel or saline patch and preparing the skin to reduce the contact impedance. Some dry electrodes use a ring model with pins to touch the scalp (Hairston et al., 2014), while other dry electrodes use foam-based supplies covered in conductive textile materials. Other promising non-contact electrode techniques enable weak biopotentials to be amplified using a contact-free electrode-skin.

The EOG method recognizes the differences in potential changes induced by eye movements between two electrodes placed either horizontally or vertically around the eyes. Early eye movement studies revealed that the human eye is an electric dipole, comprising a positively charged cornea and a negatively charged retina (Anderson, 1937). When the eyeball moves in the direction of the electrode, the electrical potential increases and it decreases when the eye moves in the other direction. The voltage variation between the poles is known as corneoretinal potential and varies between 0.4 and 1 mV. The acquired potential varies based on the viewing angle, up to an angle of 30 degrees (Anderson, 1937). When the eyes do not move and are fixed, the potential does not vary. When the eye moves in the direction of the sensor, the potential greatly increases. Importantly, for people who are blind, or when people have their eyes are closed, the electrical changes remain. There are many configurations for electrode placement. Commonly, the horizontal electrodes are placed at the external borders of the eyes while the vertical ones are positioned above and below the eye (Singh and Singh, 2012). The horizontal EOG signal is the difference of the voltage between horizontal electrodes and the vertical EOG signal is the subtraction of the voltage between the vertical electrodes. Usually, the reference electrode is placed on the forehead (Barea et al., 2002) in the middle of the eyes (Yamagishi et al., 2006) or on the mastoids (Pettersson et al., 2013). Other atypical electrode placements have been extensively tested such as six electrodes positioned above and below both the eyes (Pettersson et al., 2013). Some typical and atypical electrode placements are presented in Figure 1.

Other studies have used a headband to place electrodes close to the skin and minimize noise, but without measuring the vertical eye movements (Chang, 2019). Adopting this technique, Yan et al. (2013) positioned five electrodes, one horizontal and one vertical around each eye in addition to the reference. Kanoh et al. (2015) placed electrodes near nasion and on both sides of the rhinion, which correspond to the bridge and nose pads of eyewear. Another remarkable placement was proposed by Manabe et al. (2013), who pioneered in-ear EOG signal detection by testing different material- electrode combinations that are suitable for daily use. They highlighted the relation between the in-ear electrical signal and eye movement but did not estimate the accuracy of the eye position. Favre-Felix et al. (2017) proposes a novel fitted ear EOG device that uses a single model taken from the participant's ear canals. Interestingly, they found a strong correlation between conventional EOG and ear EOG signals. Hence, ear EOG can be used to accurately estimate eye gaze in real-time situations, which is particularly relevant for aeronautics and the hearing aid industry. For example, a visually directed hearing aid could be faster and easier to manipulate than a hearing aid piloted by other conventional tools like a pointer or remote control (Hart et al., 2009).

In addition to these techniques, numerous tests have explored the placement options for mobile electrodes that measure changes in EOG voltage. JinsMeme (JINS Inc., Japan) implanted an EOG amplifier into eyeglasses. BioPac (CA, USA) connected a mobile EOG device to a stationary system (MP160) with wireless communication. Some other EOG systems have been connected to a wheelchair (Rajesh, 2014) or fixed into goggles that can be used in everyday environments (Bulling et al., 2009). One study (Acuna et al., 2014) has shown that a low-cost EOG system (<50 euros) can give good results for eye tracking. The performance achieved by this system is different depending on whether one considers vertical or horizontal movements; the monitoring of the latter being much more precise. Thus, an EOG system gives an accuracy of <3° on the horizontal plane (with movements ranging from −50° to +50°), and an accuracy of <4° on the vertical plane (movements ranging from −10° to +10°). For larger vertical movements (from −30° to +30°), the imprecision increases to 11°. Other systems could be used as an alternative way of measuring combined EEG and EOG including Muse (Toronto, Canada), Melomind (Paris, France), Open BCI (Brooklyn, NY, USA), and Emotiv system (San Francisco, USA). They present several advantages over conventional wired, bulky EEG devices in that they are economical, portable, and easy to use. In summary, the main reason for such diversity in electrode placement is due to the specificity of each experimental paradigm and whether other equipment is associated with EOG. Largely, these systems recognize eye movements with accuracy and further investigation of electrode placement and the quality of the EOG signal will be the subject of future research.

The main eye movements captured with EOG are fixations and saccades (Singh and Singh, 2012). Fixational movement is an attempt to capture the steady image on the retina. Depending on the quality of the processed information and the current cognitive load, fixation can last between 100 and 1,000 ms, with a mean range of 200–500 ms. The saccade movement changes the eye direction around the field of view and brings the object of interest into the foveal region. Saccades are characterized by a simultaneous and rapid change of the eye position between two fixation points. The duration of saccades is determined by the angular distance the eye travels during the movement. The saccades involve distinct patterns in the EOG signal. They are relatively easy to identify as the deflected amplitudes are above the common high-frequency noise level, and they are short (in duration). When the target is slowly moving, smooth pursuit movement allows us to maintain focus. Smooth pursuit eye movements are typically initiated with a delay of 100–180 ms relative to the onset of an unpredictable motion (Lisberger and Westbrook, 1985). It is difficult to distinguish these movements on EOG recordings because it can be confounded with linear signal drift. Other movements (such as vestibulo-ocular movements and optokinetic nystagmus) allow for the gaze to stabilize during head and body movement. Nystagmus is a rhythmic, often rapid, involuntary eye movement that takes place when the head moves rapidly and the eyes move in response to the moving fluid in the vestibular system (Kang and Shaikh, 2017). Physiological nystagmus occurs during the motions of the head (vestibular nystagmus), or in the presence of patterns in the visual field (optokinetic nystagmus). It is often accompanied by a feeling of disorientation or vertigo and can be a reliable indicator of vestibular pathology. Vergence movements rotate the eyes inwards and outwards, with slow 10°/s disconjugate movements. They allow the visual system to incorporate deep targets, permitting the perception of the world across three dimensions (Alvarez et al., 2005). Even if the eyeballs participate in the contraction of medial and lateral rectal muscles until the paired images are projected onto the fovea, they will activate shared and independent nerve regions. While saccadic and vergence eye movements rely on different muscles to restore the globes, they will activate both shared and distinct neural areas (Semmlow et al., 1998).

Numerous algorithms exist for detecting and modeling oculomotor parameters including eye blink, saccade direction, and fixation. Nolan et al. (2010) developed a fully automated statistical thresholding method that detects and classifies a portion of the signal as an eye blink if the potential of the portion exceeds the threshold. Niemenlehto (2009) analyzed EOG signals by exploring a detection method of a constant false alarm rate that detects saccades. Pettersson et al. (2013) detected eye blinks and saccades by identifying the threshold of the temporal derivative of the EOG signal. Behrens et al. (2010) explored the deviation of the eye-movement acceleration of the EOG signals. Toivanen et al. (2015) computed a real-time algorithm EOG signal that automatically detects blinks, saccades, and fixations and analyzes the temporal features of these reflexes. Although EOG signals have a good signal-to-noise ratio, due to their large amplitude and the relative ease with which they can detect saccades and blinks (Skotte et al., 2007), they are continuously contaminated by physiological artifacts such as electromyography (EMG). EMG artifacts are detected when the participant moves their facial muscles or body during EOG recording, such as jaw clenching, raising an eyebrow, or smiling. The forms and amplitudes of these artifacts depend on the types of movement, the position of the electrode, and sampling rates. High-frequency noise and smooth waveforms are beyond the potential range of eye movement-related signals and are usually differentiated. Median and low-pass filters are the most commonly used techniques to preserve these noises while retaining edge. The filters are based on split windows with varied cut-off frequencies and can be used for either online and/or offline analysis.

Besides EOG, other eye-tracking techniques include photo- and video-oculography, or scleral research lenses (Duchowski, 2017). These techniques allow the capture of eye movements with greater precision than EOG but involve complex set-ups (flight deck video-camera integration, for instance) and involve processing pipelines that demand a large amount of power (computer vision algorithms). Alternatively, EOG is a technique associated with simple electrodes that can be embedded in a headset, providing good insights into the mental state and brain activity of a participant with lower consumption of energy and simpler processing pipelines, which are useful in aeronautics.

The assessment of eye movements is particularly relevant for aeronautic (Peysakhovich et al., 2018) and neuro-ergonomic applications (Peysakhovich et al., 2019). Given the EOG's capacity to detect eye movements, it is an excellent candidate to be embedded into aeronautical systems. A number of important studies have related eye movement analysis to fatigue, mental workload, and cognitive performance in pilots. One of the most important causes of aviation accidents is provoked by human errors e.g., drowsiness or fatigue (Velazquez, 2018). Pilot fatigue and loss of control are considered by the Federal Aviation Administration as one of four common aviation hazards (Federal Aviation Administration, “Risk Management Handbook: U.S. Department of Transportation,” 2016). For example, a study described how two pilots from a commercial aircraft missed the target airport because both were sleeping (Borghini et al., 2014). Thus, EOG is considered to be an effective and predictive tool in detecting drowsiness markers such as a reduction in performance, and changes in the frequency of eye blinking (Oken et al., 2006). Interestingly, eye movement analysis can serve as a reference for the mental workload and state of pilots (Di Nocera et al., 2007). It has been shown that the workload reflected by eye fixation distribution varies according to the phases of the flight. The highest workload was noted during takeoff and landing, and the lowest during the cruise phase (Di Nocera et al., 2007). Brams et al. (2018) discuss how the gaze behavior of expert pilots differs from that of novices, suggesting relevant information about basic processes that explain the successful performance of expert pilots in flight. The authors explain that the expert pilots have expanded visual range that analyzes the global scene using the field of view next to the fovea, and that they then shift their attention to the pertinent area (Gegenfurtner et al., 2011). Experts have larger saccades that cover more areas and spend less time focusing on task-related regions. In addition to the important capacity to process information from multiple sources, this ability is also related to advanced cognitive performances. The gaze behavior of fighter pilots during flight conditions also varies according to altitude and speed (Svensson et al., 1997). An augmented workload can induce a lower percentage of eye fixation distribution outside and a higher percentage of fixation on the tactical display.

In addition to investigating the eye movement patterns underlying cognitive functions, alternative EOG (around the ear) may be a promising tool in real aeronautic environments. In numerous activities such as air traffic control or piloting an aircraft, the operators are equipped with peripherals (typically headsets). As the EOG requires only a few electrodes, it does not obstruct the visual field nor does it unnecessarily illuminate the eyes with infrared light. Therefore, this technique is convenient for head-mounted peripherals such as the audio headset used by pilots. Moreover, exploring EOG integration in control and communication peripherals may enhance human-system interaction and make psycho-physiological monitoring possible (e.g., based on blink rate or saccades). Such an approach could have numerous applications in aeronautics (fighters, helicopters, and unmanned aerial vehicle operation), naval systems, and control-command centers.