A Semi-Automated Earthquake Evacuation System Using Early

Warning Detection

TATSUKI FUKUDA,

Department of Data Science, Faculty of Data Science,

Shimonoseki City University,

2-1-1 Daigakucho, Shimonoseki City, Yamaguchi, 751-8510,

JAPAN

Abstract: - This paper presents the development and validation of a semi-automated evacuation route system

designed for rapid response during seismic events. Given Japan’s frequent earthquakes, ensuring timely

evacuation is a critical issue. The proposed system utilizes the Earthquake Early Warning (EEW) system to detect

seismic activity and offers residents the option to automatically open doors or windows, securing evacuation

routes before structural deformation occurs. A prototype of the system was developed using a Raspberry Pi

and tested with pre-recorded EEW signals. Results show that while the system reliably detected EEW and

transmitted notifications to residents, the time required for window opening reached 14 seconds, indicating

room for improvement in response time. Future work will focus on reducing this delay through local server

implementation and bypassing cloud-based systems. This system not only aids in disaster response but also has

potential for everyday applications, such as baby cry detection, ensuring its continued relevance in daily life.

Key-Words: - Earthquake, Evacuation system, Semi-automated, Earthquake Early Warning, Raspberry Pi.

Received: March 19, 2024. Revised: August 14, 2024. Accepted: September 11, 2024. Published: November 29, 2024.

1 Introduction

Earthquakes are among the most devastating natural

disasters in Japan, and due to their frequency

and magnitude, earthquake countermeasures are an

urgent issue that concerns society as a whole.

Geographically, Japan is located at the convergence

of four tectonic plates, making it a country with a very

high frequency of seismic activity. Approximately

18% of the world’s earthquakes with a magnitude of 6

or higher occur in Japan [1], a notably high proportion

compared to other countries, [2]. For instance, in

2023 alone, over 2,000 earthquakes were observed in

Japan, with 19 of them reaching a magnitude of 6 or

higher, [3].

Despite decades of global research, earthquake

prediction technology has yet to achieve full accuracy.

Current earthquake forecasts rely on factors such

as seismic activity history [4], crustal deformation

[5], and the behavior of underground faults [6].

Statistical methods and physical models are employed

to predict the timing, location, and magnitude of

earthquakes, but due to the complex nature of seismic

phenomena, perfect prediction remains technically

challenging, [7]. For example, research based on

plate tectonics theory suggests that earthquakes occur

when stress accumulated along plate boundaries is

released, but the timing and magnitude of such

releases are inherently uncertain, [8].

Additionally, some studies have reported

anomalous geomagnetic variations as potential

precursors to earthquakes, which has led to further

investigation into their predictive capabilities, [9],

[10]. Other phenomena, such as unusual animal

behavior [11], and changes in hot spring activity

[12], have long been regarded as possible earthquake

precursors, and some researchers are exploring the

scientific validity of these observations as predictive

tools.

In terms of post-earthquake response, Japan has

implemented an ”Earthquake Early Warning” system,

provided by the Japan Meteorological Agency, to

rapidly disseminate information after an earthquake

has been detected. This system detects the initial

seismic wave (P-wave) and issues an alert before the

more destructive secondary wave (S-wave) arrives,

giving people critical time to take shelter. Unlike

traditional prediction methods, this system focuses on

mitigating damage after the earthquake has occurred,

playing a crucial role in saving lives. However, both

current prediction technologies and the Earthquake

Early Warning system have limitations. Improving

prediction accuracy remains an important goal, and

the development of better preemptive measures for

large-scale earthquakes is necessary.

In earthquake-prone Japan, ensuring rapid

evacuation during seismic events is a critical issue,

and the implementation of systems to aid evacuation

efforts is essential for reducing casualties caused by

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

earthquake-related disasters.

In light of this, the goal of this study is to develop

an automatic evacuation door-opening system for use

during earthquakes. Currently, securing evacuation

routes during seismic events relies on manual actions,

such as opening doors or windows. However, when

an Earthquake Early Warning is received, individuals

should prioritize their safety, and many cases have

been reported where building deformation or damage

after the earthquake made it difficult to operate doors

and windows, [13].

To address this issue, this paper proposes a

semi-automated evacuation door-opening system.

Specifically, the system is designed to automatically

open evacuation routes in response to an Earthquake

Early Warning, allowing residents to evacuate quickly

and safely. Based on the results of the preliminary

survey described in Chapter 3, concerns were raised

regarding the reliability of a fully automated system

and its functionality during power outages, resulting

in a preference by most users for a semi-automated

approach. In response to these needs, the proposed

system supports residents in securing evacuation

routes promptly in the event of an earthquake while

also allowing for manual intervention. A prototype

of this system, utilizing a Raspberry Pi, has also been

developed and will be introduced in this paper.

If implemented, the proposed system is expected

to contribute to faster evacuation and improved safety,

while also playing a role in preventing secondary

disasters following earthquakes. Furthermore, even

in situations where voluntary evacuation is not

possible, the semi-automated system can facilitate

rescue efforts by allowing emergency responders

swift access to the building interior, thus enhancing

the efficiency of rescue operations.

2 Preliminary

2.1 Convolutional Neural Network (CNN)

Convolutional Neural Networks (CNNs), a type

of neural network, are widely used for pattern

recognition tasks, particularly in the domains of

image and audio data. CNNs are highly effective in

feature extraction through convolutional operations.

Having demonstrated superior performance in image

processing applications [14], CNNs have been widely

adopted in numerous studies and have significantly

contributed to advancements in processing not only

images but other data types as well.

The basic structure of CNNs consists of

convolutional layers, pooling layers, and fully

connected layers. The convolutional layer applies

filters (kernels) to the input data, performing

convolution operations to extract local features from

image or audio data. The convolution operation can

be expressed by the following equation:

ij =

M−1

m=0

N−1

n=0

x(i+m)(j+n)wk

mn +bk(1)

Here, yk

ij represents the output from filter k,xij is

the input data, wk

mn are the filter weights, and bkis

the bias. This operation allows CNNs to capture local

features from images or audio signals.

The pooling layer serves to reduce the size

of the feature map extracted by the convolutional

layer. Typically, max pooling is used, which selects

the maximum value within a specific region to

downsample the data, reducing its dimensions and

preventing overfitting. The output of the pooling

layer is defined as:

ij =max

(m,n)∈P

y(i+m)(j+n)(2)

Here, y0

ij is the output after pooling, and P

represents the pooling region.

In the fully connected layer, the output from

the pooling layer is flattened and connected to

every node in the next layer. This layer is

responsible for producing the final classification

results. For example, in tasks such as image or speech

recognition, a Softmax function is often applied at

the output layer to compute the probabilities for each

class.

One of the major advantages of CNNs is their

ability to share parameters and utilize a local receptive

field, which significantly improves computational

efficiency. By applying filters to large-scale data like

images, CNNs can efficiently learn local features,

resulting in a substantial reduction in the number of

parameters compared to fully connected networks.

This leads to faster learning and helps prevent

overfitting.

Another key feature of CNNs is their hierarchical

approach to learning. In the early layers, the

network captures low-level features such as edges and

corners, while deeper layers progressively learn more

abstract features, such as object shapes and patterns.

This hierarchical feature extraction is particularly

effective in tasks such as speech recognition and

image classification, [15].

2.2 Speech Recognition using CNN and

Spectrograms

In speech recognition technology, CNNs have found

broad application not only in image recognition but

also in processing audio data. A particularly effective

approach involves converting audio signals into

spectrograms and then inputting these spectrograms

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

into a CNN to efficiently learn the frequency

characteristics of the audio, [16].

However, since audio signals exist in the

time domain, they are not directly suitable for

processing by CNNs. To address this, Short-Time

Fourier Transform (STFT) is employed to convert

time-domain signals into the frequency domain,

producing a spectrogram. The formula for STFT is

expressed as follows:

X(t, f) = Z∞

−∞

x(τ)w(τ−t)e−j2πf τ dτ (3)

Here, x(τ)represents the audio signal, w(τ−t)

denotes the window function, trepresents time, and

findicates frequency. Through this transformation,

the frequency components of the audio signal

are visualized over time as a spectrogram. The

spectrogram is represented as a three-dimensional

dataset, where the horizontal axis shows time, the

vertical axis represents frequency, and the color

intensity reflects the energy strength. For instance,

a spectrogram created from a recording of an

Earthquake Early Warning is shown in Figure 1. By

feeding the generated spectrogram into a CNN, the

model can learn and infer from the data.

Figure 1: Spectrogram created from a recording of an

Earthquake Early Warning. The horizontal axis

represents time, the vertical axis represents frequency,

and the color intensity reflects energy strength.

CNNs extract features from spectrograms to

use in classification tasks. The convolutional

layers of CNNs typically respond to various

frequency components and temporal variations

in the spectrogram, allowing the model to learn

distinctive audio patterns. For example, the filter w

in the convolutional layer, which reacts to different

frequency components, is expressed as follows:

ij =

M−1

m=0

N−1

n=0

X(i+m)(j+n)wk

mn +bk(4)

Here, Xdenotes the spectrogram, wis the

convolutional filter, brepresents the bias term, and yk

represents the output feature map.

Additionally, in speech recognition,

Mel-Frequency Cepstral Coefficients (MFCC)

are widely used alongside spectrograms. MFCCs

capture the characteristics of audio signals based on

a nonlinear frequency scale, mimicking the human

auditory system. The process for generating MFCCs

follows these steps:

1. Segment the audio signal into frames

2. Apply Short-Time Fourier Transform (STFT) to

each frame

3. Convert the frequency axis to the mel scale

4. Extract the log-amplitude spectrum

5. Apply the Discrete Cosine Transform (DCT) to

compute MFCCs

MFCCs, which take the logarithm of frequency

components, consider the nonlinearity of human

hearing, contributing to improved accuracy in speech

recognition, [17].

Recently, hybrid models combining CNNs with

Recurrent Neural Networks (RNNs) have also

been applied to speech recognition tasks. While

CNNs extract localized temporal features from

spectrograms, RNNs capture long-term dependencies

in the audio data. This hybrid architecture has been

shown to significantly enhance speech recognition

accuracy compared to traditional models, [18].

In this study, since complex speech recognition

is not required, a simple CNN is used to recognize

Earthquake Early Warnings from spectrograms.

3 Proposed System

This section describes the system proposed in this

study.

3.1 Preliminary Survey via Questionnaire

A preliminary survey was conducted to gather

information relevant to the development of the

system. The survey was conducted online from

February 4, 2020, to February 11, 2020, with a total of

874 respondents, consisting of men and women across

Japan aged between their teens and seventies.

The survey included the following questions:

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

• What is the maximum seismic intensity you have

experienced, and have you ever encountered

difficulty opening doors or windows during an

earthquake?

• If an automatic evacuation door-opening system

were available, would you want to use it?

• Would you prefer a fully automatic or

semi-automatic system for opening evacuation

doors?

The results of the survey are summarized in Table

1, Figure 2, and Figure 3.

Table 1: Percentage of respondents who experienced difficulty

opening doors or windows, based on the maximum

seismic intensity they have experienced. The higher the

seismic intensity, the more likely respondents were to

report difficulty.

Maximum seismic Percentage reporting

intensity experienced difficulty [%]

3 or lower 4.7

4 4.0

5 lower 5.0

5 upper 10.8

6 lower 14.4

6 upper 29.6

7 30.3

Figure 2: Survey results regarding the willingness to use an

automatic evacuation door-opening system.

As shown in Table 1, the percentage of

respondents who experienced difficulty opening

doors or windows increases with the seismic intensity

they experienced. Therefore, the greater the seismic

intensity, the more essential it becomes to secure

evacuation routes by opening doors or windows

in advance. However, in practice, the stronger the

earthquake, the more difficult it becomes to move to

Figure 3: Survey results on preferences for fully automatic or

semi-automatic evacuation door-opening systems.

a safe location while simultaneously opening doors

or windows, as it becomes challenging to walk freely

once the shaking starts.

In Figure 2, approximately 54.2% of respondents

expressed a willingness to use an automatic

evacuation door-opening system, indicating that

securing evacuation routes is a significant concern

for many. Conversely, some respondents provided

reasons for not wanting to use the system, such as:

• Concern that doors or windows might open

automatically when no one is at home.

• The system would be ineffective during power

outages, rendering its installation pointless.

• A desire to maintain manual control over the

opening and closing of doors or windows.

The concern that the system might open doors or

windows when no one is home highlights the need for

the system to confirm whether residents are nearby.

As for concerns regarding power outages, the system

would simply revert to its non-installed state in the

event of a power failure, so this does not present

a significant problem. However, it is important to

avoid mechanisms that lock doors or windows during

a power outage, preventing manual operation. The

possibility of adopting backup power solutions should

also be considered.

For those who wish to control the opening and

closing of doors or windows manually, this issue can

be addressed by opting for a semi-automatic rather

than fully automatic system.

A fully automatic system automatically opens

designated doors or windows without human

intervention when an earthquake occurs. In

contrast, a semi-automatic system prompts human

decision-making during an earthquake, and doors

or windows are mechanically opened only if the

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

resident chooses to do so. According to Figure 3,

55.9% of respondents preferred a semi-automatic

system, which slightly outnumbered those who

favored a fully automatic system. This suggests that

a semi-automatic system better aligns with users’

needs.

Based on these results, it was determined that a

semi-automatic system is the preferred option for the

system being developed.

3.2 System Overview

The flow of the proposed system is illustrated in

Figure 4.

Figure 4: Overview of the proposed system. When an

earthquake occurs and the reception of an Earthquake

Early Warning is detected, a confirmation message is

sent to the resident’s smartphone, prompting whether

to open windows or doors. The system will then

execute the opening process based on the resident’s

input.

When an earthquake occurs, an Earthquake

Early Warning is broadcasted via smartphones

or television. This system constantly monitors

environmental sounds using a microphone. Upon

detecting the Earthquake Early Warning sound, the

system sends a confirmation message to the resident’s

smartphone, asking whether they wish to open doors

or windows. If the resident opts to open them, the

system initiates the process to open the specified

doors or windows.

3.3 System Configuration

A prototype of the system has been created, as shown

in Figure 5. The system is built around a Raspberry

Pi 3 Model B+, with the operating system running

Ubuntu Server. A USB microphone is connected to

monitor environmental sounds.

3.4 Speech Recognition

The environmental sound input from the microphone

is divided into 3-second intervals and converted into

frequency spectrogram images within the Raspberry

Pi. These images are then fed into a machine learning

Figure 5: The developed prototype. The system is based on a

Raspberry Pi running Ubuntu Server with an external

USB microphone connected.

model built using Microsoft’s Azure AI Custom

Vision service. The model was pre-trained using the

following audio sources:

• Earthquake Early Warning sound from

smartphones

• Earthquake Early Warning sound from television

broadcasts

• Baby crying sounds

The Earthquake Early Warning sounds from

smartphones and television were treated as separate

audio sources due to their differences. The reason for

including baby crying sounds in the training will be

explained later. When the Azure AI Custom Vision

model predicts the Earthquake Early Warning with

more than 70% confidence, the system determines

that an Earthquake Early Warning has been detected

and sends a confirmation message to the resident

about whether to open the doors or windows. Figure

6 shows a screenshot of the program in operation.

The reason for including baby crying sounds is

as follows: While this system is designed for use

during Earthquake Early Warnings, in reality, such

warnings occur infrequently, and thus, the system

may not often be put to use. There is a concern

that users might forget about the system’s existence,

which could reduce its effectiveness in emergencies.

To mitigate this risk, it is preferable that the system

can be used regularly for other purposes, thereby

maintaining its presence in users’ minds.

For example, in households with infants, the

system could be set to notify residents when a baby

is crying. This would make the system useful in

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

Figure 6: Screenshot of the program in operation. The

audio input from the microphone is classified into

four categories: Earthquake Early Warning from

smartphone (eq-phone), Earthquake Early Warning

from television (eq-tv), baby crying (baby), and other

sounds (Negative).

daily life and keep users aware of it. Even in

households without infants, the system could serve

as a general notification service, alerting residents to

various events and making the system more practical

for everyday use.

3.5 Notifications to Residents

In Japan, the SNS application LINE is widely used,

so to lower the barriers to system adoption, a LINE

Bot was implemented. When the system determines

that an Earthquake Early Warning has been detected

via audio recognition, the LINE beacon function is

activated. If residents who have registered their

devices in advance are within the range of the

beacon, the system responds to the beacon signal. A

cloud server, pre-configured with an HTTPS server,

receives the beacon response via Webhook, and

sends a confirmation message to the resident’s LINE

application asking whether to open the doors or

windows (Figure 7). The resident can choose to

tap either ”Open” or ”Don’t open” on the displayed

options to issue a command.

For the cloud server, Microsoft Azure was utilized.

After receiving the resident’s response, the Webhook

transmits the result to the cloud server. If the resident

chooses to open the doors or windows, the signal is

sent to the local Raspberry Pi to initiate the opening

process.

3.6 Opening Doors and Windows

If the resident opts to open the doors or windows, the

system proceeds with the opening process before the

Figure 7: Screen showing the LINE-based confirmation message

asking the resident whether to open the doors or

windows. The message reads ”An Earthquake Early

Warning has been issued. Would you like to open the

evacuation routes?” with options to ”Open” or ”Don’t

open.”

earthquake causes deformation to the door or window

frames. For the prototype, the SAWOACSS product

from Sanko Co., Ltd. was used. SAWOACSS is a

device that allows windows to be opened and closed

via smartphone, and its operation can be controlled

using the Javascript-based TuyAPI. When the system

receives the signal via LINE, TuyAPI is used to

control SAWOACSS to open the windows. Figure

8 shows the installed setup, with the arrow pointing

to the SAWOACSS device mounted at the top of the

window.

Figure 8: The SAWOACSS device from Sanko Co., Ltd.,

installed to automatically open and close windows.

The arrow points to the SAWOACSS, mounted at the

top of the window.

A key point to consider is that TuyAPI is an

API for controlling devices within a local network.

Therefore, both the Raspberry Pi running TuyAPI and

the SAWOACSS must be on the same network. When

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

the cloud server sends the open signal to the local

Raspberry Pi, TuyAPI communicates the command

to the SAWOACSS, triggering the window opening

process.

4 System Verification and Discussion

System verification was conducted using the

developed prototype system.

4.1 System Verification

In this test, pre-recorded Earthquake Early Warning

sounds were played. Two types of recordings were

prepared: one from a television broadcast and the

other from a smartphone. Each recording was played

fifteen times, for a total of thirty trials. The time

required for each process was measured.

When the pre-recorded Earthquake Early Warning

sound was played, the Raspberry Pi, running the

sound recognition program, successfully responded.

Subsequently, a message was delivered to the

resident’s smartphone, allowing them to select

whether to open the door or window. Upon selecting

the option to open, the window was successfully

opened.

On average, it took 4.9 seconds from the

Earthquake Early Warning sound to the arrival of

the confirmation message. Additionally, it took an

average of 2.7 seconds from receiving the message

to issuing the command to open the window. From

the time the command was issued, it took 3.4 seconds

for the window opening process to begin. Depending

on the size of the window, it takes approximately 5 to

7 seconds to fully open a large window that connects

to a garden, allowing human passage. However, for

opening to a minimum width sufficient for entry, i.e.,

about 600 mm [19], it took around 3 seconds.

4.2 Discussion

First, all thirty trials successfully detected the

Earthquake Early Warning. Additionally, the system

ran for several hours during the experiment without

any false detections.

Although it took 4.9 seconds for the message to

be delivered via LINE after the Earthquake Early

Warning was triggered, residents would likely need

more than 4.9 seconds to hear the warning and

move to a safe location. Therefore, considering that

this system operates semi-automatically, the response

speed is considered sufficient.

Furthermore, it took a total of 14.0 seconds

from the Earthquake Early Warning sound to the

point where the window was opened to a width of

approximately 600 mm.

Earthquakes generate primary waves (P-waves),

which cause initial minor shaking, followed by

secondary waves (S-waves), which cause the main

shaking and often lead to structural deformations. The

time between receiving an Earthquake Early Warning

and the arrival of the S-waves, known as the lead time,

typically ranges from a few seconds to several tens

of seconds, [20]. Thus, it is crucial to complete the

window-opening process in under 10 seconds.

To achieve this, it would be beneficial to establish

the TuyAPI connection as soon as the Earthquake

Early Warning is detected, allowing the system

to initiate the opening process immediately after

receiving the command. Moreover, implementing

a local server instead of relying on a cloud server

or adopting a system independent of LINE could

further reduce latency, improving the system’s overall

response time.

5 Conclusion

In this study, we proposed and developed a

semi-automated evacuation door-opening system

designed to automatically open evacuation routes

during earthquakes and evaluated its prototype.

The system is designed to detect Earthquake Early

Warnings (EEW) and prompt residents with a

confirmation to open evacuation routes, thereby

supporting rapid and safe evacuation. The results

of system verification demonstrated that the system

reliably detected EEW, and the processes leading

to the resident’s decision to open evacuation doors

or windows functioned as expected. However, it

was found that the time required to open doors

or windows reached 14 seconds, indicating that

further improvements are necessary to ensure that

evacuation routes are opened before the arrival of

the primary seismic waves. Future efforts will focus

on optimizing the TuyAPI connection, implementing

local servers, and adopting independent systems

other than LINE to reduce latency and improve

response times.

In addition, this system includes features useful

in daily life. For instance, it can detect a baby’s

cry, offering potential support for child-rearing. With

further development, the system might even assess

why a baby is crying, adding a deeper level of

assistance for caregivers. Furthermore, by detecting

sounds from pets, such as a dog barking, the system

could alert users to possible issues at home when they

are away. Regular use of the system in daily life

would increase familiarity, contributing to smoother

utilization in the event of an earthquake.

In terms of real-world impact, implementing this

system is expected to enable rapid evacuation during

earthquakes, especially in a country like Japan,

where seismic activity is frequent. The system

could potentially become a standard safety feature

in many homes. Moreover, by providing everyday

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

functionalities, the system extends its utility beyond

disaster scenarios, offering continuous value in daily

life.

Considering potential user feedback,

improvements based on user concerns are crucial.

Survey results revealed concerns about the reliability

of fully automated systems and functionality

during power outages. Therefore, future system

enhancements should focus on increasing reliability

and safety by incorporating user feedback. Solutions

such as backup power supplies and manual override

features during emergencies are considerations for

future development.

By advancing this system, not only will it

accelerate evacuation during disasters, but it will

also enhance residents’ sense of security, contributing

to both emergency preparedness and everyday

convenience. The system is expected to have broad

societal benefits, potentially becoming a standard

safety feature in many homes.

Declaration of Generative AI and AI-assisted

technologies in the writing process

During the preparation of this work the authors used

Grammarly for language editing. After using this

service, the authors reviewed and edited the content

as needed and take full responsibility for the content

of the publication.

References:

[1] Ministry of Land, Infrastructure, Transport

and Tourism, ”River Data Book 2021,” Water

Management and National Land Conservation

Bureau, 2021.

[2] USGS, ”Lists, Maps, and Statistics,” United

States Geological Survey, 2021.

[3] Japan Meteorological Agency, ”Seismic

Activity in Japan, 2023,” Monthly Report

on Seismology and Volcanology (Disaster

Prevention Edition), pp. 78-94, 2023.

[4] N. Chakraborty, P. Murti, and M. Bueno

Rodriguez, ”Recent advances in earthquake

seismology using machine learning,” Earth,

Planets and Space, vol. 74, no. 1, pp. 1-13, 2022,

doi: 10.1186/s40623-022-01567-x.

[5] J. Lu, Q. Meng, and X. Gu, ”Possibility of

earthquake prediction based on monitoring

crustal deformation anomalies and thermal

anomalies at the epicenter of earthquakes

with oblique thrust faulting,” Acta

Geophysica, vol. 58, pp. 989–1001, 2016,

doi: 10.1515/acgeo-2016-0040.

[6] M. V. Rodkin, ”Earthquake prediction: Old

expectations and new results,” Seismic

Instruments, vol. 58, no. 2, pp. 215–223,

2020, doi: 10.1134/S0747923920020068.

[7] M. Wyss, ”Why is earthquake prediction

research not progressing faster?,”

Tectonophysics, vol.338, no.3-4, pp.217-223,

2001.

[8] T. Okazaki, T. Ito, K. Hirahara, and N. Ueda,

”Physics-informed deep learning approach

for modeling crustal deformation,” Nature

Communications, vol. 13, no. 7092, 2022, doi:

10.1038/s41467-022-34922-1.

[9] G. Cordaro, E. Venegas-Aravena, and D.

Laroze, ”Long-term magnetic anomalies and

their possible relationship to the latest greater

Chilean earthquakes in the context of the

seismo-electromagnetic theory,” Nat. Hazards

Earth Syst. Sci., vol. 21, no. 6, pp. 1785–1806,

2021, doi: 10.5194/nhess-21-1785-2021.

[10] K. Hayakawa and M. Molchanov, ”Monitoring

of ULF magnetic field variations for short-term

earthquake prediction,” Nat. Hazards Earth

Syst. Sci., vol. 15, pp. 745-756, 2015, doi:

10.5194/nhess-15-745-2015.

[11] N. Ueda, T. Okazaki, K. Ito, and K. Hirahara,

”Unusual animal behavior as a possible

candidate of earthquake prediction,” Appl.

Sci., vol. 14, no. 10, 4317, 2024, doi:

10.3390/app14104317.

[12] G. Du, S. Su, X. Chang, H. Ren, Z. Huo, and X.

Zhang, ”Hydrogeochemical characteristics and

genesis of Hongshuilantang Hot Spring and its

water temperature anomalies during the Rushan

earthquake swarm in Eastern China,” Front.

Earth Sci., vol. 14, no. 7, pp. 65-76, 2020, doi:

10.3389/feart.2020.00347.

[13] K. Sato and D. Isobe, ”Motion Behavior

Analysis of Doors in RC Buildings

under Seismic Excitation,” Proceedings of

Computational Engineering Conference, vol.

22, 2017.

[14] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner,

”Gradient-based learning applied to document

recognition,” Proceedings of the IEEE, vol. 86,

no. 11, pp. 2278–2324, 1998.

[15] A. Krizhevsky, I. Sutskever, and G. E.

Hinton, ”ImageNet classification with deep

convolutional neural networks,” in Advances in

neural information processing systems, 2012,

pp. 1097–1105.

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024

[16] O. Abdel-Hamid, A. Mohamed, H. Jiang, and

G. Penn, ”Convolutional neural networks for

speech recognition,” IEEE/ACM Transactions

on Audio, Speech, and Language Processing,

vol. 22, no. 10, pp. 1533–1545, 2014.

[17] B. H. Rabiner and L. Juang, ”Fundamentals of

speech recognition,” Prentice Hall, 1993.

[18] A. Graves, A.-r. Mohamed, and G. Hinton,

”Speech recognition with deep recurrent

neural networks,” in Proceedings of the IEEE

international conference on acoustics, speech

and signal processing (ICASSP), 2013, pp.

6645–6649.

[19] FUJISAWA KOUGYO Co., Ltd, ”Layout and

Comfortable Space Size,” https://www.

tokio-one.jp/column/layout-size/,

accessed 30th Sep. 2024.

[20] Japan Meteorological Agency, ”Characteristics,

Limitations, and Precautions for Use of

Emergency Earthquake Early Warnings,”

https://www.data.jma.go.jp/eew/data/

nc/shikumi/tokusei.html, accessed 30th

Sep. 2024.

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

The author solely contributed in the present research,

at all stages from the formulation of the problem to

the final findings and solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflicts of Interest

The author is affiliated with Shimonoseki City

University, a public university corporation located

in Shimonoseki City, which is the subject of this

research.

Creative Commons Attribution License 4.0

(Attribution 4.0 International , CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2024.20.8

Tatsuki Fukuda

E-ISSN: 2224-3488

Volume 20, 2024