Received October 22, 2019, accepted December 6, 2019, date of publication December 13, 2019,

date of current version December 31, 2019.

Digital Object Identifier 10.1109/ACCESS.2019.2959626

A New Assessment Method of the Pilot Stress

Using ECG Signals During Complex
Special Flight Operation
SHUYU SHAO , QIANXIANG ZHOU , AND ZHONGQI LIU
School of Biological Science and Medical Engineering, Beihang University, Beijing 100191, China
Beijing Advanced Innovation Centre for Biomedical Engineering, Beihang University, Beijing 100191, China
Corresponding author: Qianxiang Zhou (zqxg@buaa.edu.cn)
This work was supported in part by the National Key Research and Development Program of China under Grant 2016YFC0802807, and in
part by the Defense Science and Technology Innovation Special Zone Project under Grant 18-163-12-ZT-002-026-01.

ABSTRACT With the continuous improvement in the man-machine-designed fighter aircraft, the pilot
training intensity has increased. Complex special operation of the fighter aircraft is greatly affected by flight
conditions and human control reliability, which imposes a great burden on the pilot psychophysiologically.
To understand the intensity of the stress on the pilot, which is a crucial parameter, accurate evaluation
of the changes in the stress intensity of the pilot during the flight is very important. This study aimed to
analyze the stress intensity of pilots by collecting electrocardiogram (ECG) signals during the long flight
involving 11 pilot trainees. First, an improved soft-threshold denoising method for the ECG data and an
improved differential threshold method were used to process the first-order difference and the threshold
value to the filtered signal to determine the R wave. Then, the multiscale analysis method was used to fuse
the area of the heart rate curve of the pilots. The method of flight training and the division of the stress stage
were derived. Further, the functional relationship of the stress intensity with the frequency of training was
constructed. All these effectively improved the performance of the flight operation training and helped make
a reasonable training plan. This was of great practical significance to the measurement and evaluation of the
stress state of the individuals in dangerous jobs, such as special flight operations, high-altitude skydiving,
bungee jumping, etc.

INDEX TERMS Complex special operation, ECG, HR, multiscale, pilot, stress.
I. INTRODUCTION
Stress is the basic physiological impact on the body to
resist the adverse factors of the external environment [1].
Stress overload or excessive stress time leads to physiological
disorders, and an ‘‘adaptive syndrome’’ or ‘‘general adap-
tation syndrome’’ occurs. The physiological changes pro-
duced by stress are mainly related to the sympathetic–adrenal
medullary system [2]. When stimulation acts on the body,
the neuroendocrine secretion of cortisol is activated by the
hypothalamus-pituitary-adrenal gland axis and the sympa-
thetic nervous system [3]. A series of physiological effects
are produced by activating cellular signal transduction net-
works in target organ cells by hormones such as epinephrine,
norepinephrine, and glucocorticoids. These ensure the
body’s ability to respond to emergencies and adapt to the
The associate editor coordinating the review of this manuscript and
approving it for publication was Mohammad Zia Ur Rahman
environment [4]. Objective physiological indexes can be
reflected in saliva biochemical index [5], [6], serum hormone
index [7], and gastrointestinal hormone levels [8]. Stress eval-
uation of electrocardiogram (ECG) signal mainly includes
heart rate [average heart rate (HR), maximum heart rate, and
minimum heart rate] and heart rate variability (HRV), which
were used to evaluate the stress level [9]–[11].
Due to the complex combat environment of the fighter
aircraft, technical difficulty, high requirements, high risk, and
large amounts of information received and processed by the
pilot, these comprehensive factors cause great psychological
and physiological changes in the fighter pilots [12], putting
the pilots in a state of high stress. Repeated moderate stress
can induce the establishment of an adaptive mechanism of the
stressor. The stress response often determines the ability of
the body to adapt to the stressor. In particular, the pilots face
a lot of complex, dangerous situations during take-off and
. landing, and the requirements of the control skills of pilots are

185360 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/ VOLUME 7, 2019
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
quite demanding. With the increase in the psychological load
and workload of the pilots, the transient stress stimulation
is the most intense work content at this time. The pilots
experience corresponding physiological changes, such as ele-
vated blood pressure, faster heart rate, muscle contraction,
and an adverse state of stress reaction, which affect their flight
performance greatly, including flight safety [13], [14].
A number of studies on pilot stress include HR [15],
HRV [16], [17], cortisol [18], and saliva [5], [6]. Physiolog-
ical, biochemical, endocrine, metabolic, and immunological
indicators of the human body change under stress [12]. The
biochemical data are mostly acquired by intrusion acquisition
and post-task acquisition. Because the extraction is untimely,
the stress level cannot be reflected accurately, and it is gen-
erally not appropriate to apply physiological measurements
to the actual flight process due to the particularity of the
pilot. The previous studies on pilot stress mainly focused
on the comparison between pilots with different missions
and under different training stresses, and relevant studies
often focused on laboratory or simulation experiments. Pilot
stress in special flight operations refers to the physiological
and psychological changes in the pilots performing these
operations.
The degree of change in stress intensity has not been stud-
ied in detail so far. At the same time, the special flight opera-
tion is greatly affected by flight conditions and human control
reliability, and the change in stress intensity during special
flight operation is extremely obvious. However, neither the
monitoring of special flight stress intensity nor the change in
stress intensity during the long-term special flight operation
has been studied to date. The ECG signal is a relatively stable
and regularly changing index. Many factors (breathing, blood
pressure, body temperature regulation, and renin–angiotensin
system) of the body can affect the fluctuation in the heart
rate, eventually leading to dynamic changes in the heart rate
through the interaction of sympathetic nerve and parasympa-
thetic nerve to the heart, and adapt to the physiological needs
of the body. During the pilot’s special flight operation, a series
of complicated and delicate operations need to be completed
in a short period of time, during which the pilot’s instanta-
neous heart rate is greatly improved. Therefore, an accurate
evaluation of the changes in the stress levels of pilots is vital.
This study aimed to study the real-time stress level of
fighter pilots during special flight operations and the changes
in stress levels during long-term training in a complex flight
environment based on the pilot’s ECG data. Moreover, how
to identify the pressure of pilots through quantitative analysis,
improve the performance of effective flight training, develop
a reasonable flight training plan, and provide a new ECG eval-
uation method for relevant types of surgical stress assessment
were also explored.
II. EXPERIMENTAL SETTINGS
A. PARTICIPANTS
A total of 11 male fighter pilots were recruited with advanced
trainer experience and with no special flight training subjects.
VOLUME 7, 2019
Their age range was 24–27 years (mean = 25.2; SD = 1.43),
with a flying time of 300–500 h. These pilots were not
allowed to take any drug or drink a drop of alcohol or caffeine
to exclude other factors. They were not allowed to be engaged
in strenuous exercise and to get adequate sleep in 2 days
before each special flight operation. None of the pilots had
any history of cardiovascular disease.
B. ECG SIGNAL ACQUISITION EQUIPMENT
The belt-style multi-parameter physiological signal monitor-
ing system was used to collect the ECG data during the pilot’s
special flight operation. The system was a belt and a machine
box fixed on the belt for collecting, processing, storing, and
sending data. The ECG acquisition system within the belt
consisted of three ECG electrodes and the relevant circuits.
The three-electrode dual-lead ECG signal could be transmit-
ted to the computer through Bluetooth for real-time display
and recording, and could be simultaneously recorded into
the built-in memory card for post-processing. The sampling
frequency of the ECG signal was 750 Hz.
C. EXPERIMENTAL DESIGN AND PROCEDURE
The experimental mission was the course of training on
short, high-performance flight maneuvers. The time was set
between 9 am and 2 pm every day. Before the special flight
operation, the pilots first performed a level flight, and then
the special flight training began when the plane took off
and climbed up (the slope was about 60◦ ). After reaching
a certain height, it started doing a fast half-roll reversal
and finally took a return phase. During each flight, three
to five special flight operations were performed at random,
and each special flight operation lasted for 2min to 3 min.
At the end of each flight, coaches with flight training experi-
ence scored the completion of each special flight operation,
with a score of 2 to 3. The experimental process was as
follows:
1) Before each mission, the tester fixed the monitoring
device at the appropriate position within the pilot’s
clothing.
2) The testers pressed the (power) on/off power button,
and the monitoring mode of the monitoring device
began working. At the same time, the tester assisted
the pilots in wearing clothes and ECG data acquisition
belts, which were automatically connected to the ECG
belt.
3) The pilot then entered the cabin, and the flight training
began; the ECG belt automatically collected the ECG
data of the pilot during the controlling process, and the
monitoring device received the ECG data during the
special flight mission in real-time.
4) At the end of the mission, the testers recorded the start
time of the flight, saved the ECG data, and recovered
the device.
Fig.1 summarizes the entire flight training process and the
training steps.
185361

FIGURE 1. Flight training process diagram. Note: Level flight: The plane
flies horizontally at a constant speed in a straight line, which is called a
flat flight, and is the most basic state of flight. This process lasts about
20 min. Special flight process: This process involves multiple special flight
action stages, three to five times of special flight, and 2–3 min each time.
III. METHODS
A. DATA DESCRIPTION AND PREPROCESSING
The ECG signals obtained from 11 long-monitored flights
were analyzed comprehensively. The flight training times
of each pilot are shown in Table 1. Because of the loss of
ECG data during part of the flight training organization and
the performance of ECG monitoring equipment, which are
labeled in Table 1.
TABLE 1. Special flight times of pilots.
B. ECG DATA DENOISING AND R WAVE DETECTION
The initial sampling frequency of the ECG signal of a pilot
was 750 Hz. For the ease of analysis, the data were resampled
and converted into 360 Hz. Fig. 2 shows the original ECG
signal of a pilot, collected by an ECG acquisition device in
the level flight.
FIGURE 2. Raw ECG signals.
Common ECG noise mainly includes baseline drift fre-
quency, which is usually less than 1 Hz and 50 Hz power
frequency interference. The interference of EMG noise fre-
quency is generally in a relatively large frequency range
of 5 Hz-2 KHz. As the collected data were long-collected
flight data and the ECG signal was a weak electric signal
185362
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
during each flight, the interference of power frequency and
the noise of the measurement system (such as electrode polar-
ization noise, electrode contact noise, and internal noise of
amplifying circuit) might lead to baseline drift. The fixed
frequency of 50 Hz voltage and its harmonic was defined
as power frequency interference. The overlaps of power fre-
quency interference and the ECG frequency band affected
the analysis of the ECG signal. Denoising studies on ECG
signals included empirical mode decomposition translation
transfer algorithms and nonlinear filters [19]–[22]. Wavelet
transform is a time-scale method for signal analysis, which
has the characteristics of multiscale resolution analysis and
the ability to characterize local features of signals in both
time and frequency domains [23], [24]. The telescopic and
translational transformations were used to refine the ECG
signals at multiple different scales, and the local features
of the ECG signal in the time domain and the frequency
domain were simultaneously reflected on different frequency
bands, highlighting the different characteristics of the ECG
signals [25]. The denoising algorithm based on wavelet trans-
form (WT) and the median filter was used, frequencies below
1 Hz in ECG signal belong to low-frequency noise caused by
baseline drift and motion artifacts. In the process of correcting
the baseline drift in the signal, we use the wavelet transform
to decompose the ECG signal on a multi-scale. The obtained
low-frequency approximation signal fully approximates the
baseline drift noise characteristics of the signal. By setting the
low-frequency approximation signal to zero and reconstruct-
ing the decomposed high-frequency detail signal, the baseline
drift-free signal can be obtained. About power frequency
interference and EMG interference, through multi-level sta-
tionary wavelet transform of ECG signal polluted by noise,
the mean square deviation of noise in detail signal of sta-
tionary wavelet transform is estimated layer by layer. Closed
values of each layer are selected to process the detail signal
of stationary wavelet transform separately, and then inverse
wavelet transform is carried out to filter out noise.
The principles of wavelet transform and signal singularity
detections were as follows:
The wavelet function was defined as (1),
dθ (t)
ψ (t) = (1)
dt
where θ (t) stands for the smooth Gaussian function, and
the first order was derivable, ψj (t) as ψ (t), with scaling
transformation under scale j, was defined as (2),
 
1 t
ψj (t) = ψ (2)
a a
The wavelet transformation of the signal f (t) was
expressed as (3),
dθj (t)
 
Wj f (t) = f (t) ∗ j
dt
d
f ∗ θj (t)


=j (3)
dt
VOLUME 7, 2019
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
The convolution of signal f (t) and θ (t) was regarded as
the low-pass smoothing process of signal f (t) according to
scale j. An eight-layered scale was chosen in this study
to ensure that the frequency band range covered the noise
band range of the baseline drift. When wavelet denoising
is performed, the choice of threshold value directly affects
the denoising effect, method of determining the optimal
wavelet function, number of decomposition layers, threshold
value, and threshold function, which are the crucial param-
eters for the wavelet threshold denoising algorithm for ECG
signals [26].
The choice of the threshold was based on the basic
model (4),
S = f (t) + σ e(t)
where S stands for the noise and useful ECG signal,
f (t) stands for the true signal, e(t) stands for the Gaussian-
white noise or other high-frequency signals, and σ stands
for the noise standard deviation. In this study, the improved
soft-threshold denoising method was used to denoise the
data ECG signal [26]–[28]. The threshold λ value was
deduced as (5),
√ ing point was detected at 1. The starting point of the R wave
λ = σ 2LnN (5)
where σ stands for the noise standard deviation in the
This showed that the improved threshold allowed the
reconstructed signal to better retain the information of the
original ECG signal, effectively suppressing the amplitude
attenuation of the R wave.
After the original ECG signal was denoised and the R wave
was detected, the wavelet decomposition could keep time-
domain information while separating the signal frequency
information. As the R wave has the characteristics of a steep
waveform, large amplitude, and narrow width, its correct
positioning becomes the key to correctly identify other wave-
forms and extract characteristic parameters. The R wave
detection steps were as follows.
First, the data within a period (1: t) was selected to deter-
mine the starting point of R wave detection, and t = 2000ms,
(4) recorded as A. The maximum value in this period was found
and recorded as Amax . The data within a period (t + 1 : 4
was selected and recorded as B. The maximum value in this
period was found and recorded as Bmax . If Amax > 1.8×Bmax ,
the starting point was near the R wave, and the detection
starting point was postponed and then detected using 2 × t.
If Amax < 1.8 × Bmax , the starting point was in the stationary
part of the electrocardiographic signal, and the detection start-
detection of the electrocardiographic signal was determined
by the aforementioned method.
After determining the starting point of the R wave detection
first layer of wavelet decomposition coefficients σ = of the ECG signal, an improved differential threshold method
1 1 PN 

0.6745 N k=1 W j,k
 , where N stands for the signal length was used, and the first-order difference and the threshold
and Wj,k stands for the wavelet coefficient. value were applied to the filtered signal to determine the
Then, the threshold function was expressed as (6), R wave. The processed ECG signal f (t) was selected, and
( the ECG signal possessed a width of about 10 s from the
sgn Wj,k Wj,k  − λ , Wj,k  > λ

 
 
˜
Wj,k = (6) beginning. A first-order forward difference operation was
Wj,k  ≤ λ

0,
As the number of decomposition layers increased,
the intensity gradually decreased. When the wavelet coef-
ficient was greater than the fixed threshold, a constant
deviation ranging from W̃j,k to Wj,k occurred, resulting in
the denoised signal and the real large approximation error
between signals [29]. Different thresholds were chosen for
different wavelet coefficients. The new threshold function in
the present study was defined as (7),
λ = σj 2 logNj /ln(z + 2j)
p
(7)
where Nj stands for the length of the wavelet coefficient on
the jth scale, σj stands for the noise standard deviation on
the jth scale, and z stands for the constant. The threshold
function was W̃j,k = Wj,k , that is, an asymptote, which
overcame the defect of the constant deviation of the original
threshold function to guarantee the continuity of the wavelet
coefficient estimate W̃j,k at ±λ, and to avoid the oscillation
conducted for the ECG signal in each second to obtain the
difference in maximum value. Moreover, 10 differential max-
imum orderings were involved. After removing the maximum
and minimum values, the arithmetic mean of the remaining
difference values was determined with the threshold value
fixed as λ,
X 
8
λ= mi /8 (9)
i=1
The maximum and minimum values of the differential
maximum were removed, which could eliminate the exces-
sive differential value caused by the occasional spike inter-
ference to avoid the missed detection due to the excessive
threshold. Further, some too small differential values were
also removed to avoid the threshold being too small, giving
rise to false detection. After trial and comparison, we found
that such treatment reduced the probability of missed detec-
tion and false detection.
of the reconstructed ECG signal.   At the same time, in order to verify the validity of the
For a given
 constantz, the larger the Wj,k , the smaller the
 
R-wave detection method in this paper, we use this method
deviation W̃j,k − Wj,k , and and Pan-Tompkins method to detect the R-wave of 7 ECG
 
cases in MIT-BIH ECG database in the United States. The
lim|wj,k |→∞ W̃j,k = Wj,k 0 Wj,k  ≥ λ
 
(8) results are as Table 2. For most ECG data, this method

VOLUME 7, 2019 185363


TABLE 2. Detecting results of R wave in MIT-BIT database of different methods.
has achieved good detection results, and the accuracy is
over 99.89%.
After detecting the eight R waves employing the initial
detection threshold, the method of moving the average was
used to modify the detection threshold. The time window
of length was set as T0 (T0 = 20ms) and that of the slide
as T0 , taking the ECG signal segment of the time width T0
as the matching signal. Then, the maximum value in the
matched signal segment was compared with the amplitude
threshold. If it was smaller than the threshold, no R wave
was observed in the ECG data, and T0 = (T0 + RR)/2,
where RR is the previous RR interval width. The RR interval
varied in real-time due to sample differences and differences
in each heartbeat. Therefore, the time window also adjusted
the width in real-time. Each time, a new R wave was detected,
the detection threshold was modified as described earlier, and
then the next one was similarly derived. The R wave detection
effect diagram is shown in Fig.3.
FIGURE 3. Effect diagram of R wave detection.
After obtaining the R wave peak of the ECG signal, the rel-
ative continuous instantaneous heart rate could be gained
merely by calculating the time of RR interval because the
RR interval represented the period of the heartbeat. That is,
HR = 60/RR interval. The original heart rate curve of pilots
in special flight operations is shown in Fig.4.
As shown in Fig.4, the special flight operation of the pilot
consisted of a series of special actions. According to the
corresponding relationship between the training action and
the real-time heart rate curve, the heart rate curve was divided
into three stages: level flight process, special flight process,
185364
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
FIGURE 4. HR of three periods of the flight process of one pilot.
and level flight process. During the special flight process,
each heart rate curve represented a flight action, and each
flight action lasted for about 2 min, of which the ascending
and descending periods of the heart rate curve were for about
1 min each. The periods before and after the special flight
operation were defined as a steady-state, and the ECG signal
at this stage was relatively stable; thus, it could be analyzed
using HRV-related indexes. During the special flight opera-
tion, the heart rate of the pilot changed rapidly and greatly,
and the change law was that the heart rate increased first and
then dropped. When the heart rate increased, the sympathetic
nerves played the leading role. When the heart rate dropped,
the parasympathetic nerves played the leading role. This stage
was defined as the period of flight operation.
Pilots need to face sensory and emotional stressors in spe-
cial flight operations. Transient stress stimulation will be the
most intense task at this time. If pilots have an adverse state of
the stress response, it will not only affect flight performance
but also affect flight safety in serious cases. Especially in the
take-off and landing phase, pilots will face many complex and
dangerous situations. At this time, the pilots’ manipulation
skills are highly demanded, and the increase of pilots’ psy-
chological workload will trigger a series of stress reactions.
The ECG signal analysis is suitable for the stable state of the
human body. Based on this, the heart rate data is divided into
three stages according to the characteristics of the flight train-
ing process: data before flight action, data during flight
action, and data after flight action. Before and after the flight
action, the ECG signal was relatively stable at this stage.
Time-frequency characteristic parameters of HRV during the
VOLUME 7, 2019
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
TABLE 3. Comparisons of time-frequency-domain indicators of HRV for different special flight operations.
flight action will be analyzed, time-domain parameters of
HRV contain standard deviation of RR intervals(SDNN),
root-mean-square successive difference(RMSSD), standard
deviation of the averages of RR intervals(SDANN) [30], and
we use autoregressive spectral analysis with Burg estimation
spectral components
for HRV frequency characteristic parameters [31], spec-
tral of HRV include low frequency(LF:0.04-0.15Hz), high
frequency(HF:0.15-0.4Hz), and LF/HF.
IV. RESULTS
A. TIME-FREQUENCY ANALYSIS OF HRV
As is shown in table 3, the results of HRV time-domain
analysis show that: For all individuals, only RMSSD tends
to increase after the flight, that is to say, RMSSD was higher
after flight than before (p < 0.05). RMSSD is the variance of
time difference between adjacent cardiac cycles, representing
short-term variability. It is generally used to evaluate the
activity of the parasympathetic nerve in a short period of
time. It shows that the activity of the parasympathetic nerve
increases in a short period of time after flying.
The results of Burg algorithm show that the high-frequency
component HF of HRV tends to decrease during the flight
(p < 0.05). HF denotes the activity intensity of the
parasympathetic nerve and represents the weakening of
parasympathetic nerve control during flight. For the human
body, HF indicates mild parasympathetic nerve activity,
while LF represents the combined action of the sympathetic
nerve or parasympathetic nerve. Individual LF/HF index
tends to decrease after performing special flight maneuvers,
and individual HF tends to decrease after performing special
flight maneuvers. HF indicates the activity intensity of the
parasympathetic nerve, which indicates that the parasympa-
thetic nerve control function is weakened and the sympa-
thetic/parasympathetic balance is reduced during the period
of flight. HRV is suitable for the analysis and comparison of
ECG data before and after the special flight, and quantitative
analysis, but not suitable for the description during special
flight training.
B. STRESS INTENSITY CALCULATION AND
MULTISCALE ANALYSIS
The change in the heart rate of pilots was intuitively seen
during the special flight operation and was governed by the
change law. During the complex special operation, the instan-
taneous heart rate of the pilot fluctuated greatly, which
VOLUME 7, 2019
reached 150 bpm or more, and returned to normal limits
after the flight. The instantaneous heart rate and instantaneous
heart rate changes during special flight training could be
observed, and the two indicators could be used as the basis
for real-time monitoring of pilot operations. As is shown in
Fig.4, the pilot’s heart rate during the special flight operation
was patterned by going up and then down.
To facilitate the calculation and accurate measurement of
the heart rate curve area during this period, nonlinear total
least-squares fitting was used [32], [33], and the form of the
nonlinear least-squares problem was defined as follows:
Cost function:
1 Xm
F (x) = (f (x) − yi )2 (10)
2 i=1 i
where fi (x) stands for the model function and yi stands for
the actual value.
For each iteration, the range in the differences between the
minimum value and the initial value of the current iteration
should not exceed the set search area. Taylor expansion of
the cost function at a point yielded the following results:
F x ∗ = F(x) + J (x)1x


(11)
where F(x) stands for the J (x) of function F, that is, the first-
order derivative of function F, and x ∗ = x + 1x. Moreover,
an appropriate 1x minimized kF (x ∗ )k2 .
Assuming that x was a local minimum, even if it was small
enough, a point x + 1x could not be found; therefore, its
corresponding function value was less than the function value
at x. At the same time, the point at which all the first-order
derivatives of the functions were equal to zero was taken as
the stationary point. The local minimum was the stationary
point of the function, but the stationary point might also be the
local maximum or the saddle point of the function. Therefore,
to distinguish whether a point was a local minimum, a local
maximum, or a saddle point, the second-order information
of the Taylor expansion was preserved, and the correlation
properties of the H (Hessian) matrix in the expansion were
determined as (12),
2F(x)T J (x) + 2J (x)T J (x)1x = 0
H 1x = −J (x)T F(x), H = J (x)T J (x) (12)
The H matrix here was an approximation of the Hessian
matrix, and the calculation of the Hessian matrix by the J (x)
was avoided. The result is shown in Fig.5.
185365

FIGURE 5. HR signals of pilots in special flight operations.
After the heart rate curve was fitted, when the number
of points at which the heart rate continuously increased
exceeded 100 bmp, the first point was defined as L1. When
the number of points where the heart rate continuously
dropped from a certain point exceeded 100 bmp, the point was
defined as L2; when the heart rate changed to a point within
±20 points after falling to a certain point, the first point was
defined as H. First, the range of L1–H–L2 was determined for
each pilot during each special flight training period, and the
data were integrated in terms of the ascending and descending
periods and the area of the heart rate curve during this period
was calculated.
The area of the heart rate curve of a single-flight opera-
tion was the stress response of the body to the flight task.
The area sequence of the heart rate curves of pilots in each
flight also contained a lot of internal change information.
The special flight operation training was a deterministic time
series. To a certain extent, the correlation of times before
and after the operation was measured using a multiscale
method [34], [35]. At the same time, the influence of an
abnormal fluctuation trend of the overall change reduced.
Therefore, a multiscale analysis method was used to predict
the number and range of training. For the original time series,
x = {x1 , . . . , xi , . . . , xn }, the coarse-grained time series was
constructed as (13),
(γ ) 1 Xjγ
yj = xi ,
γ i=(j−1)γ +1
1 ≤ j ≤ N /γ (13)
where γ stands for the scale factor when γ = 1 and the time
series is the original time series x.
According to the single-flight heart rate area sequence of
individual pilots, the stress intensity of group pilots changed
with special flight operation times. The intensity of the group
stress presented a downward trend over time, the fitting
results are shown in Fig.6.
Fig.6 shows that the instructor–pilot proficiency ratings
accompanied the ECG-based ratings, differences among nor-
mally distributed values between instructor–pilot proficiency
ratings and heart rate areas were analyzed by one-way
ANOVA, followed by a Turkey multiple comparison posttest
(p < 0.05). The result showed that as the number of
operation times increased, the heart rate area curve showed
185366
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
FIGURE 6. Instructor–pilot proficiency ratings accompanied the
ECG-based ratings.
a downward trend. The score of instructor–pilot gradually
increased, and the two tended to be stable.
V. DISCUSSION
Stress is the body’s adaptation to adverse factors in the
environment. Due to the complex environment of the fighter
aircraft, many comprehensive factors result in great psycho-
logical and physiological changes on the part of the pilots.
All these cause a state of high stress in the pilots. Monitor-
ing the real-time flight training and physiological function
status of a pilot is also very important. Objective physio-
logical indexes, such as saliva biochemical index [5]–[7]
and gastrointestinal hormone level [8], can be used for stress
assessment, but these objective indicators were obtained by
extracting hormones from the secretion components of oral
saliva, urine, and so on, or turning to invasive acquisition.
Therefore, real-time extraction and monitoring were difficult.
The complexity of the EEG equipment could be generally
used in a laboratory environment.
In this study, the ECG data of the long-term flight operation
of the 11 pilots were analyzed. An improved soft-threshold
denoising method for the ECG data and an improved differen-
tial threshold method were used to apply the first-order differ-
ence and the threshold value to the filtered signal to determine
the R wave. The study of ECG signals was mainly the analysis
of HR or HRV. HR was modulated by the combined effects of
the sympathetic and parasympathetic nervous systems on the
sinus node on a beat–beat basis. Therefore, analyzing changes
in the heart rate over time provided information about auto-
nomic function [36]. The analysis of HRV was considered
to be a noninvasive method for predicting cardiac autonomic
function more accurately [37], [38]. However, the research
on HR or HRV was mainly carried out through the compar-
ison before and after tasks or comparison of different time
periods [38]–[40].
Because of the adjustment of sympathetic/parasympathetic
effects and other physiological and psychological responses
in the special flight operation, the pilot’s heart rate changes
dramatically in a transient manner during the flight operation
training. Therefore, using HRV relevant methods to analyze
the stress intensity during special flight operation is not suit-
able. Given the changes in heart rate during special flight
VOLUME 7, 2019
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
training, a new analytical method, the area of heart rate curve,
was proposed in this study. Because of the correlation of
times before and after the operation was measured using a
multiscale method, Therefore, a multiscale analysis method
was used to predict the number and range of training. The
stress intensity was analyzed by the shape, area and heart rate
extremum of the heart rate curve. The heart rate curve was
fitted during the special flight training period, and the key
points such as the extreme value of the curve, the starting
position of the rise and the ending position of the fall were
determined. At the same time, the influence of other environ-
mental factors on the heart rate was reduced. The area cor-
relation of heart rate curve of the ECG signal during special
flight operation was calculated innovatively to quantitatively
evaluate the intensity of single-flight stress (with the increase
of special flight operation times, heart rate areas showed an
exponentially decreasing trend and the correlation coefficient
R2 = 0.998).
The results obtained by analyzing the area of heart rate
curve in the special flight operation time series for each
pilot showed that the pilot stress intensity changed with the
number of flight training sequences. The flight stress intensity
decreased exponentially with the change in the flight training
times, from a high-stress degree to a lower stress degree,
before finally stabilizing. Then, the statistical range of the
multiscale shift values of the pilot group on the same flight
training vehicle was taken as the range of the change in the
stress stability, and on this basis, the evaluation criteria for
transient stress were formed. At the same time, the stress
stage of special training can be divided into three stages:
Adjustment stage—this stage is the beginning of special
training. At this stage, the flight operation ability of the pilots
is usually not strong. The pilots’ proficiency is not good
enough, and they often lack confidence. The individual stress
intensity is also comparatively high, and the range of stress
intensity fluctuation is very large.
Adaptation stage—with the increase in the number of
special training, the operational proficiency and operational
ability of the pilots gradually increase, and at the same time,
the self-confidence is also enhanced. The stress level shows a
rapid downward trend, and the pilot is in the stress adjustment
stage.
Stable stage—after many times of training, operation pro-
ficiency and operation ability is very strong. The performance
of pilots show great confidence and the factors that cause
stress tend to be stable with the decrease of stress intensity
of pilots, and the fluctuation range is narrowed, and the stress
stability degree increases. Completely new homeostasis will
finally be achieved.
The major aim of this study was to substitute the pilot
ECG signal for ordinary subjects to avoid the influence of
physical and psychological differences in the subjects on
ECG tests and to eliminate laboratory simulation effects to
promote the generalization of the findings for this study.
However, the shortcomings lay in the insufficient availability
of subjective scores for each flight operation or the scores for
VOLUME 7, 2019
flight operations, or the combination of other subjective eval-
uation with physiological evaluation. In the next step, more
ECG data of the pilots should be collected and special flight
operation real-time monitoring system should be designed
according to the relevant algorithms of this study, which
might automatically identify special flight time, calculate the
number of special flight operations, and determine the change
in the stress intensity of the pilot during the flight.
VI. CONCLUSION
The central finding of the study was that the pilot stress was
high on the first exposure to demanding maneuvers, and it was
lessened and stabilized after repeated training on these same
maneuvers. Repeated moderate stress led to the establishment
of an adaptive mechanism of the pilot under stress. Although
substantial evidence suggests that this happens even on the
examination of the basic heart rate, it is not suitable for
quantitative analysis. The present study proposed a new quan-
titative analysis assessment method for the stress intensity
of the pilot based on the ECG signal and applied it during
complex special flight operation. The multiscale ECG signal
analysis is a dynamic measure that can reveal the adjustment
mechanism of the body in a special flight operation envi-
ronment and the adjustment mechanism of different aircraft
to deal with such an environment. The pilot stress intensity
varied with the number of flight training sequences: the flight
stress intensity decreased exponentially with the change in
the flight training times, from a higher degree of stress to a
lower degree of stress, till the final stabilization. This study
was novel in qualitatively proposing the concepts of stress
stage and stress stability. These concepts are applicable to
personnel selection; online evaluation of training effects; and
designing of training tasks, special flight operations, high-
altitude skydiving, bungee jumping, and other dangerous
assignments.
REFERENCES
[1] W. B. Cannon, The Wisdom of the Body, 2nd ed. 1939.
[2] L. Schwabe, L. Haddad, and H. Schachinger, ‘‘HPA axis activation by a
socially evaluated cold-pressor test,’’ Psychoneuroendocrinology, vol. 33,
no. 6, pp. 890–895, 2008.
[3] C. Wieslaw Jerzy, L. Jerzy, S. Malgorzata, and W. Bartosz, ‘‘Zinc in drug-
naive patients with short-illness-duration first episode major depressive
disorder: Impact on psychopathological features,’’ Neuro Endocrinol. Lett.,
vol. 35, no. 8, pp. 741–745, 2015.
[4] I. A. Aziz Ibrahim, Y. Kamisah, M. I. Nafeeza, and M. F. N. Azlina,
‘‘The effects of palm vitamin E on stress hormone levels and gastric lesions
in stress-induced rats,’’ Arch. Med. Sci. AMS, vol. 8, no. 1, pp. 22–29, 2012.
[5] P. Foley and C. Kirschbaum, ‘‘Human hypothalamus-pituitary-adrenal axis
responses to acute psychosocial stress in laboratory settings,’’ Neurosci.
Biobehav. Rev., vol. 35, no. 1, pp. 91–96, 2010.
[6] E. Kajantie and K. Räikkönen, ‘‘Early life predictors of the physiological
stress response later in life,’’ Neurosci. Biobehav. Rev., vol. 35, no. 1,
pp. 23–32, 2010.
[7] K. E. Rehm, L. Xiang, O. U. Elci, M. Griswold, and G. D. Marshall, Jr.,
‘‘Variability in laboratory immune parameters is associated with stress
hormone receptor polymorphisms,’’ Neuroimmunomodulation, vol. 19,
no. 4, pp. 220–228, 2012.
[8] B. H. Jonsson and P. M. Hellström, ‘‘Motilin-and neuropeptide Y-like
immunoreactivity in a psychophysiological stress experiment on patients
with functional dyspepsia,’’ Integrative Physiolog. Behav. Sci. Off.
J. Pavlovian Soc., vol. 35, no. 4, p. 256, 2000.
185367

[9] E. J. Hermans, M. J. Henckens, M. Joëls, and G. Fernández, ‘‘Dynamic
adaptation of large-scale brain networks in response to acute stressors,’’
Trends Neurosci., vol. 37, no. 6, pp. 304–314, 2014.
[10] M. Hasegawa, A. Hayano, A. Kawaguchi, and R. Yamanaka, ‘‘Assessment
of autonomic nervous system function in nursing students using an auto-
nomic reflex orthostatic test by heart rate spectral analysis,’’ Biomed. Rep.,
vol. 3, no. 6, p. 831, 2015.
[11] C. Grad, ‘‘Heart rate variability and heart rate recovery as prognostic
factors,’’ Clujul Med., vol. 88, no. 3, pp. 304–309, 2015.
[12] P. Froom, Y. Caine, I. Shochat, and J. Ribak, ‘‘Heat stress and helicopter
pilot errors,’’ J. Occupat. Med., vol. 35, no. 7, pp. 720–724, 1993.
[13] E.-M. Elmenhorst, M. Vejvoda, H. Maass, J. Wenzel, G. Plath, E. Schubert,
and M. Basner, ‘‘Pilot workload during approaches: Comparison of sim-
ulated standard and noise-abatement profiles,’’ Aviation, Space, Environ.
Med., vol. 80, no. 4, pp. 364–370, 2009.
[14] C. K. McClernon, M. E. McCauley, P. E. O’Connor, and J. S. Warm,
‘‘Stress training improves performance during a stressful flight,’’ Hum.
Factors, J. Hum. Factors Ergonom. Soc., vol. 53, no. 3, pp. 207–218, 2011.
[15] N. V. Dojčinovski and E. Missoni, ‘‘Physiological parameters and trace
elements before and after aerobatic flight,’’ in Proc. 53 Int. Congr. Aviation
Space Med./Eric, Donaldson, Mazurek, Krzysztof, 2005.
[16] M. S. Mumenthaler, N. L. Benowitz, J. L. Taylor, L. Friedman, A. Noda,
and J. A. Yesavage, ‘‘Nicotine deprivation and pilot performance dur-
ing simulated flight,’’ Aviation Space Environ. Med., vol. 81, no. 7,
pp. 660–664, 2010.
[17] F. Sauvet, J. C. Jouanin, C. Langrume, B. P. Van, Y. Papelier, and
C. Dussault, ‘‘Heart rate variability in novice pilots during and after a
multi-leg cross-country flight,’’ Aviation Space Environ. Med., vol. 80,
no. 10, p. 862, 2009.
[18] Z. Obmiński, M. Wojtkowiak, R. Stupnicki, L. Golec, and A. C. Hackney,
‘‘Effect of acceleration stress on salivary cortisol and plasma cortisol and
testosterone levels in cadet pilots,’’ J. Physiol. Pharmacol., Off. J. Polish
Physiol. Soc., vol. 48, no. 2, p. 193, 1997.
[19] M. Blanco-Velasco, B. Weng, and K. E. Barner, ‘‘ECG signal denoising
and baseline wander correction based on the empirical mode decomposi-
tion,’’ Comput. Biol. Med., vol. 38, no. 1, pp. 1–13, 2008.
[20] S. Poornachandra, ‘‘Wavelet-based denoising using subband dependent
threshold for ECG signals,’’ Digit. Signal Process., vol. 18, no. 1,
pp. 49–55, 2008.
[21] S. Pal and M. Mitra, ‘‘Empirical mode decomposition based ECG enhance-
ment and QRS detection,’’ Comput. Biol. Med., vol. 42, no. 1, pp. 83–92,
2012.
[22] S. Cuomo, G. D. Pietro, R. Farina, A. Galletti, and G. Sannino, ‘‘A novel
O(n) numerical scheme for ECG signal denoising,’’ Procedia Comput. Sci.,
vol. 51, no. 1, pp. 775–784, 2015.
[23] Z. Wang, F. Wan, M. W. Chi, and L. Zhang, ‘‘Adaptive Fourier decompo-
sition based ECG denoising,’’ Comput. Biol. Med., vol. 77, pp. 195–205,
Oct. 2016.
[24] S. A. Qureshi, I. Masood, M. Hashmi, S. Hanninen, M. Sarwar, and
A. Jameel, ‘‘Noise reduction of electrocardiographic signals using wavelet
transforms,’’ Electron. Electr. Eng., vol. 20, no. 3, pp. 29–32, 2014.
[25] U. R. Acharya, H. Fujita, M. Adam, O. S. Lih, V. K. Sudarshan, T. J. Hong,
J. E. W. Koh, Y. Hagiwara, C. K. Chua, C. K. Poo, and T. R. San, ‘‘Auto-
mated characterization and classification of coronary artery disease and
myocardial infarction by decomposition of ECG signals: A comparative
study,’’ Inf. Sci., vol. 377, pp. 17–29, Jan. 2017.
[26] Y. H. Peng, ‘‘De-noising by modified soft-thresholding,’’ in Proc. IEEE
Asia–Pacific Conf. Circuits Syst. Electron. Commun. Syst., Dec. 2002,
pp. 760–762.
[27] P. Ghorbanian, A. Ghaffari, A. Jalali, and C. Nataraj, ‘‘Heart arrhythmia
detection using continuous wavelet transform and principal component
analysis with neural network classifier,’’ in Proc. Comput. Cardiol., 2010,
pp. 669–672.
[28] D. L. Donoho, ‘‘De-noising by soft-thresholding,’’ IEEE Trans. Inf. The-
ory, vol. 41, no. 3, pp. 613–627, May 1995.
[29] Q. Zhang, L. Wang, P. Gao, and Z. Liu, ‘‘An innovative wavelet threshold
denoising method for environmental drift of fiber optic gyro,’’ Math.
Problems Eng., vol. 2016, no. 3, pp. 1–8, 2016.
[30] Z. Piotrowski and M. Szypulska, ‘‘Classification of falling asleep
states using HRV analysis,’’ Biocybern. Biomed. Eng., vol. 37, no. 2,
pp. 290–301, 2017.
[31] O. Faust, R. U. Acharya, A. R. Allen, and C. M. Lin, ‘‘Analysis of EEG sig-
nals during epileptic and alcoholic states using AR modeling techniques,’’
IRBM, vol. 29, no. 1, pp. 44–52, 2008.
185368
S. Shao et al.: New Assessment Method of the Pilot Stress Using ECG Signals
[32] C. Cartis, N. Gould, and P. L. Toint, ‘‘Approximation and optimization,’’
in Springer Optimization and Its Applications, vol. 145, C. Demetriou and
P. M. Pardalos, Eds., 1st ed. Cham, Switzerland: Springer, 2019, pp. 5–26.
[33] F. Vogt, ‘‘A self-guided search for good local minima of the sum-of-
squared-error in nonlinear least squares regression,’’ J. Chemometrics,
vol. 29, no. 2, pp. 71–79, 2015.
[34] B. R. Bakshi, ‘‘Multiscale analysis and modeling using wavelets,’’
J. Chemometrics, vol. 13, no. 3, pp. 415–434, 1999.
[35] G. Kutyniok and D. Labate, Shearlets: Multiscale Analysis for Multivariate
Data. Basel, Switzerland: Birkhüser, 2012.
[36] A. D. Droitcour, O. Boric-Lubecke, V. M. Lubecke, and J. Lin, ‘‘Range
correlation effect on ISM band I/Q CMOS radar for non-contact vital
signs sensing,’’ in IEEE MTT-S Int. Microw. Symp. Dig., vol. 3, 2003,
pp. 1945–1948.
[37] F. Lombardi, ‘‘Clinical implications of present physiological understand-
ing of HRV components,’’ Cardiac Electrophysiol. Rev., vol. 6, no. 3,
pp. 245–249, 2002.
[38] J. Kaczmarek, S. Chawla, C. Marchica, M. Dwaihy, L. Grundy, and
G. M. Sant’Anna, ‘‘Heart rate variability and extubation readiness in
extremely preterm infants,’’ Neonatology, vol. 104, no. 1, pp. 42–48, 2013.
[39] C. Heinze, U. Trutschel, T. Schnupp, D. Sommer, A. Schenka, J. Krajewski,
and M. Golz, ‘‘Operator fatigue estimation using heart rate measures,’’
in World Congress on Medical Physics and Biomedical Engineering,
September 7–12, 2009, Munich, Germany. Berlin, Germany: Springer,
2009, pp. 930–933.
[40] E. C.-P. Chua, W.-Q. Tan, S.-C. Yeo, P. Lau, I. Lee, I. H. Mien,
K. Puvanendran, and J. J. Gooley, ‘‘Heart rate variability can be used to
estimate sleepiness-related decrements in psychomotor vigilance during
total sleep deprivation,’’ Sleep, vol. 35, no. 3, pp. 325–334, 2012.
SHUYU SHAO is currently pursuing the Ph.D.
degree in biomedical engineering with the School
of Biological Science and Medical Engineering,
Beihang University, Beijing, China. His research
interests include human-computer interaction and
ergonomics.
QIANXIANG ZHOU received the Ph.D. degree in
systems engineering. His research interests are in
aerospace human engineering, image processing
and analysis, and physiological signal processing.
He was a Professor, doctoral supervisor, and has
graduated from the China University of Mining
and Technology, in 1996. He is currently a Profes-
sor in biological science and medical engineering
with Beihang University. His main research area
for nearly two decades has been image processing
and signal analysis.
ZHONGQI LIU received the B.S., M.S., and Ph.D.
degrees in man machine and environment engi-
neering from Beihang University. He is currently
a Lecturer in biological science and medical engi-
neering with Beihang University. His research
interests include computer vision and human-
computer interaction.
VOLUME 7, 2019