[Link].

com/scientificreports

OPEN Trace explosive detection based

on fluorescence sensing and
similarity measures for time series
classification
Weize Shi, Yabin Wang, Piaotong Liu & Xin Li
Currently, almost all explosives involved in bombings are nitro compounds, especially
2,4,6-trinitrotoluene (TNT) is the most widely used. In order to detect and prevent potential explosive
threats in time, it is of great significance to detect trace TNT quickly and conveniently. We investigated
a fluorescence sensor and designed a trace explosive fluorescence detection system for detecting TNT
acetone solutions and common chemical reagents. Experiments were conducted on the detection
of TNT acetone solution with different concentrations, common chemical reagents, the influence
of different injection volumes and injection flow rates, and the influence of UV irradiation time. In
addition, the time series similarity measures, including the Pearson correlation coefficient, Spearman
correlation coefficient, Dynamic Time Warping (DTW) distance, and Derivative Dynamic Time Warping
(DDTW) distance, were used to classify the detection results. The results show that the limit of
detection (LOD) of the fluorescent sensor for TNT acetone solution is 0.03 ng/μL, and the response
time is less than 5 s. Moreover, the fluorescent sensor is specific, reversible and repeatable, and the
recovery response time is less than 1 min. In addition, the method of integrating the calculation of
Spearman correlation coefficient and DDTW distance can effectively classify the detection results.

Keywords Explosive detection, Trace detection, Fluorescent sensor, Time series similarity

Trinitrotoluene (TNT), also known as 2,4,6-trinitrotoluene, has the chemical formula C7H5N3O6. It typically
exists as a white or yellow crystalline solid, exhibiting high detonation power whereas maintaining stability
against impact and friction. The detonation of 1 g of TNT releases approximately 4000 J of energy. Due to its
widespread use in military and industrial applications, the rapid and sensitive detection of trace TNT is critical for
security and environmental monitoring. Existing trace TNT detection methods include infrared spectroscopy,
Raman spectroscopy, gas chromatography (GC), mass spectrometry (MS), ion mobility spectrometry (IMS),
electrochemical, colorimetry, and fluorescence detection.
Infrared spectroscopy provides fingerprint characteristics, but it requires high-quality samples and has
relatively low sensitivity. Kumar investigated long-range Fourier transform infrared (FTIR) spectroscopy for
explosive identification, measuring TNT, pentaerythritol tetranitrate (PETN), and ammonium nitrate (AN)
spectra at a 5 m distance using diffuse reflection1. He analyzed inorganic low explosives such as black powder
and ammonium nitrate-based salts using FTIR, verifying their feasibility through real-world case studies2.
Raman spectroscopy exhibits high sensitivity and fingerprint characteristics, but its scattering area is small and
can be influenced by factors such as optical system parameters. Zachhuber developed a 532 nm pulsed laser
Raman spectrometer for qualitative and quantitative analysis of PETN and AN, achieving detection at distances
of 5–9 m3. Östmark demonstrated the use of spectral imaging Raman spectroscopy for long-range detection
of explosives, successfully acquiring Raman spectral images of sulfur, ammonium nitrate, 2,4-dinitrotoluene
(DNT), and TNT from individual explosive particles within a 10 m range4. GC offers high separation efficiency
and sensitivity, but it requires high-quality samples, and its instruments are bulky, limiting portability. Cruse
utilized GC/vacuum ultraviolet spectroscopy to analyze explosives, successfully identifying seven explosive
compounds, including TNT5. Arce-Rubí employed GC–MS and GC-nitrogen phosphorus detection coupling to
identify eleven explosives, including TNT6. MS enables rapid detection and high accuracy, but it can only analyze
relatively pure samples and requires coupling with instruments capable of high separation. Hong developed a
hollow cathode discharge ion source coupled with mass spectrometry, which was successfully used for TNT

School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China. email:

wangyabin@[Link]; 3120235077@[Link]

Scientific Reports | (2025) 15:23943 | [Link] 1

[Link]/scientificreports/

and TNT-related explosives detection7. Burns detected five different classes of explosives, including TNT, by
ambient ionization MS8. IMS is widely applied for explosive detection, but it requires high-quality samples.
Kostarev studied the ion mobility spectra of TNT and hexogen (RDX) using laser ionization asymmetric IMS,
demonstrating that laser mechanisms offer higher selectivity9. Mullen employed drift-tube IMS combined
with secondary electrospray and corona discharge ionization to study the signal strength of TNT and DNT10.
Electrochemical detection exhibits high selectivity and sensitivity, but it depends on the concentration of
explosives, making it challenging to ensure consistent sensitivity. Sekhar used hydrocarbon and nitrogen
oxide potentiometric sensors for effective TNT detection11. Puttasakul et al. proposed a polyacrylamide gel
as an electrochemical gas sensor for TNT vapor detection12. Niu constructed Si-doped graphene nanosheets
and successfully applied them to detect nitroaromatic compounds13. Colorimetry provides fast detection and
is intuitive but it has low sensitivity and is greatly influenced by environmental conditions. Yardımcı developed
an ethylenediamine-magnetite nanoparticle sensor for TNT detection, whereas Thipwimonmas introduced a
polymer gel-based colorimetric sensor using RGB image analysis to detect TNT and related explosives14,15.
Fluorescence detection offers multiple advantages, including high sensitivity, good selectivity, fast response,
simplicity, portability, and low cost. However, certain fluorescent sensing materials may be affected by
environmental factors, and their preparation processes can be relatively complex. Swager incorporated the rigid
three-dimensional pentadecene into conjugated polymer chains, developing thin-film polymer sensors for TNT
detection16. Later studies introduced various polycyclic aromatic hydrocarbons, improving conjugation effects
and achieving fluorescence lifetimes in the nanosecond range17. Zheng synthesized linear conjugated polymers
and hyperbranched conjugated polymers utilizing Sonogashira coupling to detect TNT in a dichloromethane
solvent18. In addition to improvements in fluorescent materials, numerous studies have proposed more sensitive
and faster-response fluorescent probes19–26. Fiber optic fluorescence detection has high sensitivity, non-contact
detection, and strong environmental adaptability. However, its technological complexity, high detection costs,
and limited detection range significantly hinder its application. Yang developed a hollow-core photonic crystal
fiber sensor based on fluorescence quenching, achieving a detection limit of 0.340 ppb for TNT27. Lin designed
and fabricated a porphyrin-doped polymer fiber optic sensor, achieving a detection limit of 120 ppb and a
response time of 3 min (min) for DNT28. Cui utilized evanescent fields induced by “click” polymerization on
waveguide surfaces to prepare composite fiber optic tapered probes for sensing. They detected TNT vapor at a
concentration of 10 ppb, achieving a 30% quenching within 50 s and 65% quenching after 250 s29.
Recently, combining explosive detection techniques with advanced data processing methods has gained
increasing research interest. Banas integrated FTIR with machine learning algorithms to classify high-order and
low-order explosive residues, using multivariate statistical analysis for accurate TNT and PETN identification30.
Hwang applied Raman spectroscopy with principal component analysis (PCA) for rapid explosive detection,
preprocessing 700 Raman spectra from 14 explosives, achieving an average classification accuracy of 99.3%31.
However, existing literature has mainly focused on the improvement and development of fluorescence
materials, with little further analysis of the time series data obtained through explosive fluorescence detection in
combination with data processing methods.
In this study, a tube-type fluorescent sensor was designed. Through field experiments, TNT acetone
solution and common chemical reagents were tested, and the TNT concentration response and selectivity of
the fluorescent sensor were analyzed. The time series similarity measures, including the Pearson correlation
coefficient, Spearman correlation coefficient, Dynamic Time Warping (DTW) distance, and Derivative Dynamic
Time Warping (DDTW) distance, were used to classify the detection results, providing a method for judging
detection results.

Materials and methods

Fluorescent sensor
The fluorescent sensing material used in this study is LPCMP332. LPCMP3 was synthesized using a palladium-
catalyzed Buchwald-Hartwig cross-coupling reaction, as shown in Fig. 1a. Tetrakis(4-bromidephenyl)-ethylene,
1,3,5-tris(4-aminophenyl)benzene, Pd(dba)2, 2-dicyclohexylphosphino-2′,4′,6′-triisopropylbiphenyl (XPhos)
and sodium tert-butoxide (NaOtBu) were mixed, tetrahydrofuran was added, stirred and heated to 65 °C.
The solution was then cooled to room temperature and filtered. The mixed solution was then cooled to room
temperature, stirred and filtered. Subsequently, the final product was extracted by Soxhlet extraction with
methanol, tetrahydrofuran and chloroform and dried under vacuum at 50 °C. The ultraviolet–visible (UV–Vis)
spectra and emission spectra of LPCMP3 are shown in Fig. 1b. The maximum absorption peak was located at
400 nm and the maximum emission peak was located at 537 nm.
The interaction mechanism of LPCMP3 with TNT is photoinduced electron transfer (PET). When
nitroaromatics are added to the LPCMP3 solution and the fluorescent membrane, π–π stacking interactions
occur between the conjugated networks of LPCMP3 and nitroaromatics. Upon excitation, electrons are
transferred from the conduction band of LPCMP3 to the lowest unoccupied molecular orbital (LUMO) of the
nitroaromatics, leading to fluorescence burst. The LPCMP3 used in this study was purchased and provided by
the research team.
To apply the polymer to the practical detection of trace explosives, firstly, tetrahydrofuran (THF) was used
as the solvent, and 10 mg of the solid sample of fluorescent material was weighed and dissolved in 1 mL of THF,
and then left to stand for 30 min, protected from light, after being completely dissolved. Then, we configured the
fluorescent sensing material solution with a concentration of 0.5 mg/mL, and 20 μL of the solution was dropped
onto the surface of the quartz wafer using a micropipette, and the fluorescent film was obtained by spin-coating
with the TC-218 spin-coater at a rotational speed of 5000 rpm for 1 min.
Taking fluorescent thin films as the object of study, we further investigated the photophysical properties of
fluorescent sensing materials, including photostability and fluorescence quenching effects.

Scientific Reports | (2025) 15:23943 | [Link] 2

[Link]/scientificreports/

Fig. 1. Synthesis of fluorescent sensing materials and study of their photophysical properties. (a) Synthesis
of LPCMP3. (b) UV–Vis spectroscopy and emission spectrum of LPCMP3. (c–g) Emission spectrum of the
fluorescence intensity of fluorescent films F1–F5 changing with time under the irradiation of the excitation
light source. (h) Fluorescence intensity decay rate of fluorescent films F1–F5.

According to the different preparation processes, we categorized the fluorescent films into F1–F5. Among
them, the fluorescent film F1 was prepared according to the conventional process described above. The
fluorescent films F2 and F3 were corroded with 20% sulfuric acid solution for 10 min before spin coating. The
fluorescent films F4 and F5 were prepared by adding antioxidant 891 to the fluorescent sensing material solution
while etching the substrate, and were fully dissolved and placed away from light for 30 min. The fluorescent films
F1, F2 and F4 were obtained by placing the spin-coated fluorescent films in a dust-free environment and drying
them naturally for 30 min. The fluorescent films F3 and F5 were obtained by baking the spin-coated fluorescent
films in an oven at 60 °C for 15 min. The experimental results are shown in Fig. 1c–g.
In order to effectively evaluate the stability and service life of the fluorescent sensor, we calculated the
fluorescence intensity decay rate of the fluorescent film F1–F5 at different illumination times. The calculation
formula is:
( )
I
DR = 1 − × 100%(1)
I0

Scientific Reports | (2025) 15:23943 | [Link] 3

[Link]/scientificreports/

wherein, I0 is the emission peak value of the fluorescent sensing material in the initial state (0 min), and I is the
emission peak value of the fluorescent sensing material after continuous illumination for a certain period of
time.
It can be seen that after 1 min, 2 min, 3 min, 4 min, 5 min, 10 min, 30 min, and 60 min of illumination, the
fluorescence intensity decay rates of F1 were 7.69%, 11.65%, 14.92%, 17.67%, 21.07%, 27.49%, 36.76%, and
44.86%, respectively. The fluorescence intensity decay rates of F2 were 6.75%, 11.28%, 13.62%, 16.46%, 17.60%,
19.16%, 26.53%, and 33.34%, respectively. The fluorescence intensity decay rates of F3 were 4.90%, 7.03%, 8.31%,
10.86%, 11.74%, 15.27%, 21.38%, and 27.19%, respectively. The fluorescence intensity decay rates of F4 were
4.06%, 8.14%, 12.26%, 14.29%, 16.32%, 18.39%, 25.89%, and 31.54%, respectively. The fluorescence intensity
decay rates of F5 were 2.24%, 4.48%, 5.00%, 5.60%, 6.77%, 8.64%, 14.10%, and 21.13%, respectively. The results
are shown in Fig. 1h.
The results show that the fluorescent thin film F5, with a fluorescence intensity decay rate of 21.13% after
60 min of continuous illumination, still has high photostability and is able to meet the working life requirements
in practical applications. In addition, by pre-corrosion treatment of the substrate, micro-etching can be formed
on its surface to increase the specific surface area of ​​the fluorescent film and improves its light stability. Adding
antioxidants and high-temperature baking can improve the optical stability of fluorescent films, and high-
temperature baking has a greater improvement in the optical stability of fluorescent films.
At the same time, we tested the fluorescence emission spectrum of the newly prepared fluorescent film before
and after contact with TNT vapor. The experiments were performed using a 400 nm wavelength UV excitation
light source, and an Ocean Optics USB 2000 fiber-optic spectrometer, as shown in Fig. 2a. The spectra were
obtained by placing 1 mg of TNT powder into a 10 mL brown bottle, screwed the cap tightly and baked in an
oven at 60 °C for 5 min, and then the mouth of the bottle containing TNT powder was opened and exposed
to the fluorescent film. The emission spectrum of the fluorescent film before and after contacting with TNT
vapor were obtained by the spectrometer, as shown in Fig. 2b. It can be seen that the fluorescence intensity of
the fluorescent film decreased by 44.66% after 1 s of contact with TNT, whereas the fluorescence intensity of the
fluorescent film decreased by 87.52% after 5 s of contact.
The TNT acetone solution was added dropwise into the fluorescent sensing material solution and the
fluorescent film to obtain the changes of the fluorescence sensing solution before and after quenching.
Fluorescence quenching occurred in the fluorescent material solution and the fluorescent film, as shown in
Fig. 2c,d.
To improve the detection efficiency and the processing efficiency of the sensor, the substrate of the fluorescent
sensor was selected as a high-purity, high-temperature resistant quartz glass tube with an inner diameter of
0.6 mm. The inner side of the glass tube was rinsed sequentially with NaOH solution, pure water and anhydrous

Fig. 2. Fluorescence quenching effect based on LPCMP3 and design of fluorescent sensor. (a,b) Experimental
principle of the contact between fluorescent film and TNT vapor and emission spectrum. (c) The fluorescent
material solution. (d) The fluorescent film. (e) Tubular fluorescent sensor.

Scientific Reports | (2025) 15:23943 | [Link] 4

[Link]/scientificreports/

ethanol, and further cleaned with high-frequency ultrasonic waves for 5 min, and then left to dry before being
heated in a vacuum muffle furnace at 400 °C for 5 h. In addition, in order to increase the specific surface area of the
fluorescent sensors after coating, hydrofluoric acid (HF) and sulfuric acid (H2SO4) were mixed proportionally,
diluted, and then injected into the tubes slowly using an automatic injection pump was slowly injected into the
glass tube for 4 h continuously. The tubes were then rinsed with pure water and anhydrous ethanol, dried, and
placed in an oven at 120 °C for 24 h to obtain clean quartz glass tubes. Using a micropipette and a spin-coater,
the above mixed solution was injected into the glass tube and evenly spun-coated on the inside of the tube wall.
The prepared fluorescent sensor is shown in Fig. 2e.

System design
The trace explosive fluorescence detection system includes a core control circuit, an excitation light source, a
photodetector and a fluorescent sensor. Through the design of precision optical circuit and structure of the
design, the parts of the integration of the development of portable explosives detection device. The principle of
the trace explosive fluorescence detection system is shown in Fig. 3a.
The excitation light source operates at a wavelength of 400 nm, and the precision optical path employs a
reference light structure to maintain peak-to-peak stability within 0.05%. Peak-to-peak stability refers to the
difference between the maximum and minimum output power values, expressed as a percentage of the average
power, representing the fluctuation range of output power over time. The excitation light beam is divided into
two paths using a high-precision half-transmissive, half-reflective mirror. One path passes directly through the
mirror to the photodetector as reference light, whereas the other is reflected towards the sensor for detection.
By conducting real-time comparisons between the reference and detection light, the system effectively mitigates
zero-point drift caused by temperature fluctuations or prolonged detection periods. And the excitation light
source and the photodetector are placed at 90 degrees, as shown in Fig. 3b.
The core control circuit processes the weak optoelectronic signals collected by the photodetector within the
precision optical path. To ensure stable signal transmission, the STM32F4 series microcontroller was selected.
The photodetector features a 24-bit analog-to-digital conversion circuit with a gain exceeding 10,000, whereas
its current sensitivity is no greater than 1 pA. The portable explosive detection device, as shown in Fig. 3c, was
developed by integrating modules such as precision optical circuit, precision structure and control circuit.
A visual real-time detection platform is designed by LabVIEW software, as shown in Fig. 3d Various
indicators and parameters of the system can be monitored in real time, so that problems and abnormalities
can be found in time to avoid accidents. At the same time, real-time data and status information of the system
are obtained through real-time monitoring, which is conducive to real-time analysis and decision-making, and
effectively improves the work efficiency.

Fig. 3. Trace explosives fluorescence detection system design and experimental program. (a) The principle of
the trace explosive fluorescence detection system (Top view). (b) Mounting structure of excitation light source
and detector (Side view). (c) The portable explosive detection device. (d) Visual real-time detection platform.
(e) Experimental protocol for detection of TNT and chemical reagents.

Scientific Reports | (2025) 15:23943 | [Link] 5

[Link]/scientificreports/

Based on the trace explosives fluorescence detection system, we used experimental and data analysis
methods to investigate the basic working conditions, TNT concentration response, selectivity, and light decay.
The experimental protocol for detection of TNT and chemical reagents is shown in Fig. 3e.

Data processing
This section describes the data processing methods employed in this study. We used the trace explosive
fluorescence detection system to detect trace TNT and chemical reagents. The performance of the system was
tested based on the change amplitude of the detection results. The detection results were classified using the
time series similarity measures such as Pearson correlation coefficient, Spearman correlation coefficient, DTW
distance and DDTW distance33,34.
By integrating DTW distance, DDTW distance, and the Pearson and Spearman correlation coefficients, the
analytical capability for the time series similarity measures and trend variation detection was enhanced. The
detailed mathematical formulations are presented in Eqs. (2)–(5).
DT W − P earson = DT W × (1 − rp )(2)
DT W − Spearman = DT W × (1 − rs )(3)
DDT W − P earson = DDT W × (1 − rp )(4)
DDT W − Spearman = DDT W × (1 − rs )(5)

where rp is the Pearson correlation coefficient, rs is the Spearman correlation coefficient, DTW is the DTW
distance calculated using the FastDTW algorithm, and DDTW is the DDTW distance calculated using the
FastDDTW algorithm.
The Pearson correlation coefficient quantifies the degree of linear correlation between two datasets, X and Y,
and is defined as the covariance between the two variables divided by the product of their standard deviations:
cov(X, Y ) E[(X − µX )(Y − µY )]
ρX,Y = = (6)
σX σY σX σY

For a sample of size n, it is commonly expressed as:

∑
n
(Xi − X)(Yi − Y )
rp = √
i=1
√ (7)
∑
n ∑
n
(Xi − X)2 (Yi − Y )2
i=1 i=1

The Spearman correlation is the Pearson correlation applied to rank-transformed variables. For a sample size of
n, the original datasets X and Y are converted into rank variables R(X) and R(Y), and the Spearman correlation
coefficient is computed as follows:
cov(R(X), R(Y ))
rs = ρR(X),R(Y ) = (8)
σR(X) σR(Y )

where ρR(X),R(Y) represents the Pearson correlation coefficient calculated using the rank variable. cov(R(X),R(Y))
is the covariance of the rank variable. σR(X) and σR(Y) denote the standard deviations of the rank variables.
DTW distance and DDTW distance were computed using the FastDTW and FastDDTW algorithms,
respectively, significantly reducing the computational complexity compared to the original DTW and DDTW
methods. DTW distance and DDTW distance are determined by identifying the optimal alignment path between
two time series, minimizing their distance after warping. The difference between the two lies in the calculation
method used to establish the optimal alignment path.
Given time series X and Y of length m and length n, the warping matrix D is defined as:
 d(x , y ) d(x1 , y2 ) ··· d(x1 , yn ) 
1 1

 d(x2 , y1 ) d(x2 , y2 ) ··· d(x2 , yn ) 

D= .. .. .. .. (9)
. . . .
d(xm , y1 ) d(xm , y2 ) ··· d(xm , yn )

where d(xi,yj) represents the dissimilarity measure between xi and yj. Additionally, DTW distance employs
Euclidean distance for alignment, whereas DDTW distance is based on squared derivative differences.

Results and discussion

Basic working conditions
To evaluate the basic working conditions of the trace explosive fluorescence detection system, we conducted
experiments on the effects of injection volume and injection flow rate on the detection of TNT acetone solution.
Among them, the injection volume impact experiments were set up with sample volumes of 1 μL, 2 μL, 3 μL, 4
μL, and 5 μL, a flow rate of 150 mL/min, and a TNT acetone solution concentration of 1 ng/μL. The injection

Scientific Reports | (2025) 15:23943 | [Link] 6

[Link]/scientificreports/

flow rate impact experiments were set up with flow rates of 50 mL/min, 100 mL/min, 150 mL/min, 200 mL/min,
and 250 mL/min. The sample volume was 1 μL and the concentration of TNT acetone solution was 1 ng/μL. The
experimental results are shown in Fig. 4a,b.
The fluorescent sensor employs a fluorescent material exhibiting a fluorescence quenching effect;
consequently, the detection curve demonstrates a downward trend upon TNT exposure. Since the reversibility
of the fluorescence quenching effect, the curve subsequently exhibits an upward trend after reaching a certain
minimum, gradually returning to its initial state. To facilitate the consistency of experimental data processing,
we set the initial fluorescence intensity to 150,000 arbitrary units (a.u.).
Upon analysis, the detection curves for both injection volume and injection rate showed a decreasing and
then increasing trend, and the measured fluorescence intensity reduction values were 318,673, 274,094, 360,051,
392,852, 342,923, 442,580, 301,768, 415,978, 401,767, and 361,589, respectively, all reaching their maximum
changes within 5 s. The fluorescence intensity decay values were then divided by the initial value (150,000) to
obtain the fluorescence intensity decay rates, which were 21.24%, 18.27%, 24.00%, 26.19%, 22.86%, 29.51%,
20.12%, 27.73%, 26.78%, and 24.11%, respectively, as shown in Fig. 4c.

Fig. 4. The basic working conditions of the trace explosive fluorescence detection system. (a) Detection results
for different sample volumes (1 μL, 2 μL, 3 μL, 4 μL, and 5 μL). (b) Detection results for different sample flow
rates (50 mL/min, 100 mL/min, 150 mL/min, 200 mL/min, and 250 mL/min). (c) Fluorescence intensity decay
rates of the injection volume and injection rate experiments.

Scientific Reports | (2025) 15:23943 | [Link] 7

[Link]/scientificreports/

The variation in fluorescence quenching intensity across different sample volumes was relatively minor,
indicating that sample volume has a negligible effect on TNT detection. Among the flow rates tested, a flow rate
of 150 mL/min provided the most reliable and efficient detection performance.

TNT concentration response

To study the concentration response of TNT dissolved in acetone, we prepared TNT acetone solutions at
concentrations of 0.01 ng/μL, 0.03 ng/μL, 0.05 ng/μL, 0.1 ng/μL, 0.5 ng/μL, 1 ng/μL, 5 ng/μL, 10 ng/μL, 20 ng/μL,
40 ng/μL, and 60 ng/μL. A sample volume of 1 μL and a flow rate of 150 mL/min were used for detection. The
experimental results are shown in Fig. 5a.
Upon analysis, the detection curve initially exhibits a decrease followed by a subsequent recovery, with
fluorescence intensity reductions measured at 0, 10,326, 21,105, 44,772, 134,546, 415,978, 526,519, 711,856,
1,000,295, 1,297,865, and 1,482,775, all reaching their maximum changes within 5 s. The fluorescence intensity
decay value was then divided by the initial value (150,000) to obtain the fluorescence intensity decay rates, which
were 0.00%, 0.69%, 1.41%, 2.98%, 8.97%, 27.73%, 35.10%, 47.46%, 66.69%, 86.52%, and 98.85%, respectively, as
shown in Fig. 5b.
The fluorescence intensity reduction values increase with the increase of TNT acetone solution concentration,
indicating that the detection performance of the fluorescent sensor is positively correlated with the concentration
of TNT acetone solution.
However, for a concentration of 0.01 ng/μL, detection stability is only achievable under controlled laboratory
conditions. For practical applications, further enhancements in the fluorescent sensor and portable explosive
detection device are necessary. For TNT concentrations between 0.03 ng/μL and 0.5 ng/μL, setting a detection
threshold (e.g., when the fluorescence intensity decreases by more than 10,000 within 5 s) enables the detection
of TNT at concentrations of 0.03 ng/μL and above. At concentrations of 1 ng/μL and higher, the fluorescence
intensity decrease exceeds 400,000, ensuring effective detection. At concentrations of 40 ng/μL and 60 ng/μL, the
detection curve initially exhibits a rapid decline, followed by a gradual decrease around the one-second mark.
The results show that the detection limit of the fluorescent sensor is 0.03 ng/μL, and the response time is
less than 5 s. Furthermore, due to the reversible fluorescence quenching effect of the fluorescent material, the
detection curve gradually returns to a state near the initial intensity after a certain period of time, indicating that
the fluorescent sensor is reversible.

Selectivity
To investigate the selectivity of the fluorescent sensor, tests were conducted using various chemical reagents,
including acetone, pure water, dichloromethane, potassium permanganate, methanol, potassium chlorate
solution, sodium hydroxide solution, chloroform, petroleum ether, THF, potassium nitrate solution, ethanol,
acetonitrile, ethyl acetate, and grease, in addition to a blank control. A sample volume of 1 μL and a flow rate of
150 mL/min were used for detection. The experimental results are shown in Fig. 6a.

Fig. 5. The concentration response of TNT dissolved in acetone. (a) Detection results for TNT acetone
solutions at varying concentrations. (b) Fluorescence intensity decay rates of TNT acetone solutions at varying
concentrations.

Scientific Reports | (2025) 15:23943 | [Link] 8

[Link]/scientificreports/

Fig. 6. The selectivity of the fluorescent sensor. (a) Detection results for various chemical reagents,
including acetone, pure water, dichloromethane, potassium permanganate, methanol, potassium chlorate
solution, sodium hydroxide solution, chloroform, petroleum ether, THF, potassium nitrate solution,
ethanol, acetonitrile, ethyl acetate, and grease. (b) Fluorescence intensity decay rates of acetone, pure water,
dichloromethane, potassium permanganate, methanol, potassium chlorate solution, sodium hydroxide
solution, chloroform, petroleum ether, THF, potassium nitrate solution, ethanol, acetonitrile, ethyl acetate,
grease, and air.

Upon analysis, the fluorescence intensity reduction values of air, acetone, pure water, dichloromethane,
potassium permanganate, methanol, potassium chlorate solution, sodium hydroxide solution, chloroform,
petroleum ether, potassium nitrate solution, ethanol, acetonitrile, and ethyl acetate were 1823, 18,116, 4252,
18,220, 4751, 16,512, 8205, 8270, 3861, 5625, 212, 8141, 12,354, and 12,218, respectively. Additionally, the
fluorescence reduction values within 5 s were 1020, 913, 2453, 1092, 91, 1582, 673, 2936, 1338, 3981, 212, 931,
850, and 444, respectively. The fluorescence intensity decay values were then divided by the initial value (150,000)
to obtain the fluorescence intensity decay rates, which were 0.07%, 0.06%, 0.16%, 0.07%, 0.01%, 0.11%, 0.04%,
0.20%, 0.09%, 0.27%, 0.01%, 0.06%, 0.06%, and 0.03%, respectively.
For THF, the detection curve initially declined, followed by an increase, and then another decline, resulting
in a total reduction of 46,108. Additionally, the fluorescence reduction value was 7462 within 5 s, and the
fluorescence intensity decay rate was 0.50%.
And for grease, the detection curve exhibited an initial rise, followed by a decline, and then another increase,
with a maximum increase value of 84,322, reaching the maximum within 5 s, and the fluorescence intensity
decay rate was −5.62%. The results are shown in Fig. 6b.
However, the detection curve of grease was opposite to that of TNT acetone solution, which reduces the
possibility of false positives. The presence of grease in TNT acetone solutions may impact the detection process,
necessitating further investigation beyond the scope of this study.
Overall, the results confirm that common chemical reagents, including acetone, pure water, dichloromethane,
potassium permanganate, methanol, potassium chlorate solution, sodium hydroxide solution, chloroform,
petroleum ether, THF, potassium nitrate solution, ethanol, acetonitrile, ethyl acetate, grease, and air, do not
interfere with TNT detection, demonstrating the high selectivity of the proposed fluorescent sensor.

Light decay
To evaluate the light decay performance of the fluorescent sensor, different UV irradiation times were tested:
0 min, 10 min, 30 min, 60 min, 120 min, 180 min, 240 min, and 300 min. At each time point, the TNT acetone
solution detection experiment was first performed, then the empty detection experiment was performed. At each

Scientific Reports | (2025) 15:23943 | [Link] 9

[Link]/scientificreports/

time point, the TNT acetone solution detection experiment was performed, and the blank test was performed
1 min later. The experiments were performed with a sample volume of 1 μL, a flow rate of 150 mL/min, and a
TNT acetone solution concentration of 1 ng/μL. The experimental results are shown in Fig. 7a,b.
Upon analysis, for TNT acetone solution detection, the fluorescence response initially decreased and then
increased, with reduction values of 369,551, 271,268, 75,511, 87,035, 81,022, 49,987, 35,693, and 31,329, all
reaching their maximum change within 5 s. Even after 300 min of exposure, the fluorescence intensity reduction
measured remained above 30,000. The fluorescence intensity decay values were then divided by the initial value
(150,000) to obtain the fluorescence intensity decay rates, which were 24.64%, 18.08%, 5.03%, 5.80%, 5.40%,
3.33%, 2.38%, and 2.09%, respectively. And for blank tests, the fluorescent sensor exhibited light exposure-
induced signal variations of 97, 0, 0, 168, 123, 654, 1309, and 362, all remaining below 10,000. We obtained the
fluorescence intensity decay rates of 0.01%, 0.00%, 0.00%, 0.01%, 0.01%, 0.04%, 0.09%, and 0.02%, respectively.
The results are shown in Fig. 7c, indicating that the fluorescent sensor does not make false detections even after
300 min of continuous illumination.
The results show that the fluorescent sensor maintains stability over extended periods, allowing for continuous
and reliable detection of TNT acetone solutions at concentrations as low as 1 ng/μL for at least 300 min. Since the
TNT acetone solution detection was performed first and then the empty detection was performed 1 min later,

Fig. 7. The light decay performance of the fluorescent sensor. (a) Detection results for TNT acetone solution
under different UV illumination times. (b) Detection results for blank tests under different UV illumination
times. (c) Fluorescence intensity decay rates of TNT acetone solution detection and blank test.

Scientific Reports | (2025) 15:23943 | [Link] 10

[Link]/scientificreports/

there were no false positives in the detection, indicating that the fluorescent sensor is repeatable and the recovery
response time is less than 1 min.

Time series similarity measures

The results of the TNT acetone solution test and the chemical reagent test were flipped and summarized by the
y-axis, as shown in Fig. 8.
Building on the previous results, we conducted a similarity analysis between the detection data for TNT
acetone solutions (at concentrations of air, 0.01 ng/μL, 0.03 ng/μL, 0.05 ng/μL, 0.1 ng/μL, 0.5 ng/μL, 1 ng/μL,
5 ng/μL, 10 ng/μL, 20 ng/μL, 40 ng/μL, and 60 ng/μL) and common chemical reagents, including acetone, pure
water, dichloromethane, potassium permanganate, methanol, potassium chlorate solution, sodium hydroxide
solution, chloroform, petroleum ether, THF, potassium nitrate solution, ethanol, acetonitrile, ethyl acetate, and
grease.
To quantify similarity, we calculated the Pearson correlation coefficient, Spearman rank correlation coefficient,
DTW distance, and DDTW distance, as shown in Fig. 9. Additionally, we combined DTW and DDTW distances
with Pearson and Spearman correlation coefficients to derive DTW-Pearson, DTW-Spearman, DDTW-Pearson,
and DDTW-Spearman distances, as shown in Fig. 10.
The analysis reveals that Pearson and Spearman correlation coefficients can effectively distinguish TNT
detections from common chemical reagents but fail to differentiate blank tests from TNT signals. DTW distance
effectively identifies common chemical reagents but lacks sufficient resolution to differentiate various TNT
concentrations. DDTW distance performs well in detecting high-concentration TNT detection (≥ 1 ng/μL) but
struggles to differentiate low-concentration TNT detection (< 1 ng/μL) from background chemical reagents.
DTW-Pearson distance effectively identifies low-concentration TNT solutions. DTW-Spearman distance
enhances differentiation between TNT and common chemical reagents, allowing for accurate classification of
high and low TNT concentrations. DDTW-Pearson distance is well-suited for identifying low-concentration
TNT solutions. DDTW-Spearman distance demonstrates the best overall performance, distinguishing high-
concentration and low-concentration TNT detections from common chemical reagents.

Discussion
We studied a tube-type fluorescent sensor with fluorescence quenching effect and designed the trace explosive
fluorescence detection system. Through actual measurement experiments, the effects of different injection

Fig. 8. The results of TNT acetone solution detection and chemical reagent detection.

Scientific Reports | (2025) 15:23943 | [Link] 11

[Link]/scientificreports/

Fig. 9. Similarity analysis between TNT acetone solutions and common chemical reagents. (a) Pearson
correlation coefficient. (b) Spearman correlation coefficient. (c) DTW distance. (d) DDTW distance.

volumes and injection flow rates were studied, TNT acetone solutions and common chemical reagents of
different concentrations were detected, and the light decay law of the fluorescent sensor was studied by UV
irradiation time.
The results show that the limit of detection (LOD) of the fluorescent sensor for TNT acetone solution is
0.03 ng/μL, the detection range is 0.03–60 ng/μL, and the response time is less than 5 s. Moreover, the fluorescent
sensor is reversible and repeatable, and the recovery response time is less than 1 min. In addition, the fluorescent
sensor is specific, and there is no false detection in the detection of chemical reagents such as acetone, pure water,
dichloromethane, potassium permanganate, methanol, potassium chlorate solution, sodium hydroxide solution,
chloroform, petroleum ether, THF, potassium nitrate solution, ethanol, acetonitrile, ethyl acetate, and grease.
All the above experiments were repeated 5 times, with an accuracy rate of 100% and a false positive rate of 0%.
According to the analytical figures of other reported explosive detection technologies, including Raman
spectroscopy, MS, IMS, electrochemical, colorimetry, and fluorescence sensing, a comparative analysis was
conducted with this study in terms of selectivity, LOD, and response time, as shown in Table 1.
At the same time, the time series similarity measures, including the Pearson correlation coefficient, Spearman
correlation coefficient, DTW distance, and DDTW distance, were used to classify the detection results of TNT
acetone solution and common chemical reagents, providing a method for judging detection results.
Among all similarity calculation methods, DDTW-Spearman distance was the most effective, categorizing the
data into three distinct groups:

(a) Blank tests and low-concentration TNT detection (< 1 ng/μL).

(b) High-concentration TNT detection (≥ 1 ng/μL).

Scientific Reports | (2025) 15:23943 | [Link] 12

[Link]/scientificreports/

Fig. 10. Further similarity analysis between TNT acetone solutions and common chemical reagents. (a) DTW-
Pearson distance. (b) DTW-Spearman distance. (c) DDTW-Pearson distance. (d) DDTW-Spearman distance.

Detection methods Selectivity LOD Response time References

TNT 1.48 nM – 35
Raman spectroscopy
TNT, DNT, 2,4,6-trinitrophenol (picric acid) 0.01 mg/L (TNT) – 36

TNT, PETN, RDX, DNT, TATP, etc – – 37

MS
DNT 15 ng – 38

TNT, 2-amino-4,6-dinitrotoluene (2-ADNT), etc 1 ng – 39

IMS
TNT, hexamethylene triperoxide diamine (HMTD) 10 ng, 50 ng – 40

TNT 0.0003 pM – 41
Electrochemical
TNT 0.2 nM – 42

TNT, 2,4,6-trinitrophenol (TNP) 255 μM (TNT), 318.6 nM (TNP) <1 s 43

Colorimetry
TNT, RDX, etc 67.2 nmol/L (TNT), 50.3 nmol/L (RDX) – 44

TNT, DNT 0.34 ppb (TNT) 130 s, 160 s 27

DNT 120 ppb 3 min 28

Fluorescence sensing
TNT, TNP 324 nM (TNT), 21.1 nM (TNP) <1 s 43

TNT 0.03 ng/μL <5 s This work

Table 1. Comparative analysis of typical explosives detection methods.

Scientific Reports | (2025) 15:23943 | [Link] 13

[Link]/scientificreports/

(c) Common chemical reagent detection.

Conclusion
Fluorescence-based detection technology holds great promise and significant research value in the field of
explosive detection. This study developed a fluorescent sensor with fluorescence quenching effect on TNT and
designed a portable explosive detection device. Through actual measurement experiments, it was shown that the
fluorescent sensor has specific detection of trace TNT acetone solution. And the time series similarity measures
were used to classify TNT acetone solutions and chemical reagents of different concentrations, providing a
judgment basis for the subsequent research on the application of artificial intelligence algorithms.
Despite these promising results, certain limitations remain. whereas data processing methods have been
extensively applied in explosive detection technologies based on infrared and Raman spectroscopy, research
on fluorescence-based detection has primarily focused on material optimization rather than on advanced data
analysis techniques. Consequently, the potential of fluorescence detection for real-time data processing and
classification remains largely unexplored. To address these limitations, future research will combine fluorescence
detection with complementary technologies, such as infrared and Raman spectroscopy, develop multimodal
detection platforms, and establish intelligent detection systems based on big data and advanced machine learning
algorithms to enable real-time analysis and adaptive detection in dynamic environments.
This study introduced novel methodologies that enhance detection accuracy and reliability, laying a solid
foundation for further advancements in this field.

Data availability
The data that support the findings of this study are available from the corresponding author, Piaotong Liu, and
the co-corresponding author, Yabin Wang, upon reasonable request.

Received: 1 May 2025; Accepted: 23 June 2025

References
1. Kumar, M. et al. Stand-off detection of solid targets with diffuse reflection spectroscopy using a high-power mid-infrared
supercontinuum source.
2. He, N. et al. Identification of inorganic oxidizing salts in homemade explosives using Fourier transform infrared spectroscopy.
(2019).
3. Zachhuber, B., Ramer, G., Hobro, A., Chrysostom, E. T. H. & Lendl, B. Stand-off Raman spectroscopy: A powerful technique
for qualitative and quantitative analysis of inorganic and organic compounds including explosives. Anal. Bioanal. Chem. 400,
2439–2447 (2011).
4. Östmark, H., Nordberg, M. & Carlsson, T. E. Stand-off detection of explosives particles by multispectral imaging Raman
spectroscopy. Appl. Opt. 50, 5592 (2011).
5. Cruse, C. A. & Goodpaster, J. V. Optimization of gas chromatography/vacuum ultraviolet (GC/VUV) spectroscopy for explosive
compounds and application to post-blast debris. Forensic Chem. 26, 100362 (2021).
6. Arce-Rubí, S. et al. Identificación de explosivos Orgánicos en Indicios Post-Explosión mediante GC-MS y GC-NPD. TM ​h​t​t​p​s:​ ​/​/​d​
o​i​.​o​r​g/​ ​1​0​.​1​8​8​4​5​/t​ ​m​.​v​3​3​i​3​.​4​3​7​7​​​​ (2020).
7. Hong, H., Habib, A., Bi, L. & Wen, L. Gas phase ion-molecule reactions of nitroaromatic explosive compounds studied by hollow
cathode discharge ionization-mass spectrometry. Talanta 236, 122834 (2022).
8. Burns, D. et al. Ambient ionisation mass spectrometry for the trace detection of explosives using a portable mass spectrometer. Int.
J. Mass Spectrom. 471, 116735 (2022).
9. Kostarev, V. A., Kotkovskii, G. E., Chistyakov, A. A. & Akmalov, A. E. Detection of explosives in vapor phase by field asymmetric
ion mobility spectrometry with dopant-assisted laser ionization. Talanta 245, 123414 (2022).
10. Mullen, M. & Giordano, B. C. Combined secondary electrospray and corona discharge ionization (SECDI) for improved detection
of explosive vapors using drift tube ion mobility spectrometry. Talanta 209, 120544 (2020).
11. Sekhar, P. K. & Brosha, E. L. Trace detection of 2, 4, 6-trinitrotoluene using electrochemical gas sensor. IEEE Sens. J. 15, 1624–1629
(2015).
12. Puttasakul, T., Pintavirooj, C., Sangma, C. & Sukjee, W. Hydrogel based-electrochemical gas sensor for explosive material detection.
IEEE Sens. J. 19, 8556–8562 (2019).
13. Niu, F., Shao, Z.-W., Tao, L.-M. & Ding, Y. Si-doped graphene nanosheets as a metal-free catalyst for electrochemical detection of
nitroaromatic explosives. J. Colloid Interface Sci. 594, 848–856 (2021).
14. Yardımcı, B., Koç, Ö. K., Üzer, A., Hızal, J. & Apak, R. Ethylenediamine-bound magnetite nanoparticles as dual function
colorimetric sensor having charge transfer and nanozyme activity for TNT and tetryl detection. Microchim. Acta 188, 228 (2021).
15. Thipwimonmas, Y. et al. The development of digital image colorimetric quantitative analysis of multi-explosives using polymer gel
sensors. Sensors 21, 8041 (2021).
16. Yang, J.-S. & Swager, T. M. Porous shape persistent fluorescent polymer films: An approach to TNT sensory materials. J. Am. Chem.
Soc. 120, 5321–5322 (1998).
17. Yamaguchi, S. & Swager, T. M. Oxidative cyclization of bis(biaryl)acetylenes: Synthesis and photophysics of dibenzo[g,p ]chrysene-
based fluorescent polymers. J. Am. Chem. Soc. 123, 12087–12088 (2001).
18. Zheng, B., Li, Y., Tao, F., Cui, Y. & Li, T. Enhanced superquenching of the hyperbranched conjugated polymer for the detection of
nitroaromatic explosives. Sens. Actuators, B Chem. 241, 357–363 (2017).
19. Qiao, T., Kim, S., Lee, W. & Lee, H. An enhanced fluorescence detection of a nitroaromatic compound using bacteria embedded in
porous poly lactic- co -glycolic acid microbeads. Analyst 146, 4615–4621 (2021).
20. Zhang, Y. et al. Turn-on fluorescent probe with aggregation-induced emission characteristics for polyazoles. Mater. Chem. Front.
2, 1779–1783 (2018).
21. Yu, X. et al. Turn-on fluorescent detection of hydrogen peroxide and triacetone triperoxide via enhancing interfacial interactions
of a blended system. Anal. Chem. 91, 6967–6970 (2019).
22. Lee, J. Y. et al. Ratiometric turn-on fluorophore displacement ensembles for nitroaromatic explosives detection. J. Am. Chem. Soc.
142, 19579–19587 (2020).
23. Firuzabadi, F. D., Alavi, M. A., Zarekarizi, F., Tehrani, A. A. & Morsali, A. A pillared metal-organic framework with rich π-electron
linkers as a novel fluorescence probe for the highly selective and sensitive detection of nitroaromatics. Colloids Surf. A 622, 126631
(2021).

Scientific Reports | (2025) 15:23943 | [Link] 14

[Link]/scientificreports/

24. Zhang, C., Cheng, S., Zhuang, Q., Xie, A. & Dong, W. 18α-Glycyrrhetinic acid aggregation-induced emission probes for visual
fluorescence detection of explosive as well multi-functional applications. New J. Chem. 46, 1896–1904 (2022).
25. Zhang, W., Gao, B., Guo, X. & Dong, W. Polyfluorene based fluorescent sensor for sensitive and selective detection of picric acid.
Mater. Lett. 306, 130860 (2022).
26. Zhao, Y., Guan, Q. L., Bai, F. Y., Xing, Y. H. & Sun, L. X. An inorganic-organic hydrogen cluster: Fluorescence response of the high
efficient detection of Fe3+, OH− and nitro explosives. J. Mol. Struct. 1225, 129115 (2021).
27. Yang, J. et al. Fluorescence sensor for volatile trace explosives based on a hollow core photonic crystal fiber. Sens. Actuators, B
Chem. 306, 127585 (2020).
28. Lin, H. et al. Flexible porphyrin doped polymer optical fibers for rapid and remote detection of trace DNT vapor. Analyst 145,
5307–5313 (2020).
29. Cui, M., Zhuang, Y., Zou, G., Zhu, B. & Zhang, Q. Fabrication of composite optical fiber taper through “click” polymerization
initiated by evanescent field for sensing. Sens. Actuators, B Chem. 284, 243–249 (2019).
30. Banas, A. M., Banas, K. & Breese, M. B. H. Classification of the residues after high and low order explosions using machine learning
techniques on fourier transform infrared (FTIR) spectra. Molecules 28, 2233 (2023).
31. Hwang, J. et al. Fast and sensitive recognition of various explosive compounds using Raman spectroscopy and principal component
analysis. in (eds. Broach, J. T. & Isaacs, J. C.) 87090R (Baltimore, 2013). [Link]
32. Pan, L. et al. Luminescent and swellable conjugated microporous polymers for detecting nitroaromatic explosives and removing
harmful organic vapors. ACS Appl. Mater. Interfaces 11, 48352–48362 (2019).
33. Salvador, S. & Chan, P. FastDTW: Toward accurate dynamic time warping in linear time and space.
34. Keogh, E. J. & Pazzani, M. J. Derivative dynamic time warping. in Proceedings of the 2001 SIAM International Conference on Data
Mining 1–11 (Society for Industrial and Applied Mathematics, 2001). [Link]
35. Wang, Z. et al. Fabrication of flexible AuNPs@ polyimide heating chips for in situ explosives SERS sensing in nature samples.
Talanta 258, 124460 (2023).
36. Yang, Z.-W., Yu, Z., Zheng, J.-F., Wang, Y.-H. & Zhou, X.-S. Rapid detection of trace nitrobenzene in water via SERS using a
portable Raman spectrometer. Anal. Methods 16, 1531–1537 (2024).
37. Hong, H., Habib, A., Bi, L., Qais, D. S. & Wen, L. Hollow cathode discharge ionization mass spectrometry: Detection, quantification
and gas phase ion-molecule reactions of explosives and related compounds. Crit. Rev. Anal. Chem. 54, 148–174 (2024).
38. Ding, N. & Levis, R. J. Noncontact detection of absorbed dinitrotoluene using laser electrospray mass spectrometry. J. Am. Soc.
Mass Spectrom. jasms. [Link] (2025).
39. Zhang, X. et al. A study on the determination of the metabolites of 2,4,6-Trinitrotoluene using a dual-drift tube ion mobility
spectrometer. Talanta 281, 126857 (2025).
40. Zhou, Q. et al. Detection of nitro-based and peroxide-based explosives by fast polarity-switchable ion mobility spectrometer with
ion focusing in vicinity of faraday detector. Sci. Rep. 5, 10659 (2015).
41. Karazan, Z. M., Roushani, M., Hoseini, S. J. & Nasibipour, M. Designing an electrochemical aptasensor based on a microstructure
porous-covalent organic polymer for the determination of 2,4,6-trinitrotoluen in water and soil samples. Microchem. J. 204,
110975 (2024).
42. Chohra, M. et al. Electrochemical detection of 2,4,6-trinitrotoluene on L-cysteine-modified porous silicon electrode in dimethyl
sulfoxide solution. SILICON 16, 3111–3120 (2024).
43. Qiao, R. et al. Carbonized polymer dots for discrimination of 2,4,6-trinitrotoluene and 2,4,6-trinitrophenol. J. Hazard. Mater. 491,
137944 (2025).
44. Durmazel, S., Üzer, A. & Apak, R. “All-in-a-tube” detection of RDX and TNT: Old silver mirror reaction revived for nitro-explosive
quantification. Microchim. Acta 192, 344 (2025).

Acknowledgements
Throughout the writing of this dissertation, I have received a great deal of support and assistance. Firstly, I would
like to thank my supervisor, Yabin Wang, whose expertise was invaluable in formulating the research questions
and methodology. I would also like to thank my tutor, Yang Liu, for his valuable guidance throughout my studies.
Finally, I would particularly like to acknowledge the wonderful collaboration and patient support of my team-
mates, Piaotong Liu and Xin Li.

Author contributions
Conceptualization, X.L., Y.W.; investigation, W.S., P.L.; data curation, W.S., P.L.; methodology, Y.W., X.L.; formal
analysis, P.L.; writing original draft, W.S.; writing-review and editing, W.S., P.L., Y.W.; visualization, W.S., P.L.;
supervision, Y.W. All authors reviewed the manuscript.

Declarations

Competing interests
The authors declare no competing interests.

Additional information
Correspondence and requests for materials should be addressed to Y.W. or P.L.
Reprints and permissions information is available at [Link]/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Scientific Reports | (2025) 15:23943 | [Link] 15

[Link]/scientificreports/

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives

4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in
any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide
a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have
permission under this licence to share adapted material derived from this article or parts of it. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence
and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to
obtain permission directly from the copyright holder. To view a copy of this licence, visit ​h​t​t​p:​ ​/​/​c​r​e​a​t​i​ve​ ​c​o​m​m​o​
n​s​.​or​ ​g​/​l​i​c​e​n​s​es​ ​/​b​y​-​n​c​-​n​d​/​4​.​0​/​​​​.​​

© The Author(s) 2025

Scientific Reports | (2025) 15:23943 | [Link] 16