Skip to main content
NCBI home page
ACS Omega logoLink to ACS Omega
. 2025 Feb 21;10(8):8052–8057. doi: 10.1021/acsomega.4c09130

Research on a Concentration Detection Method for Turbid Media Based on Multipath Combination

Zhe Zhao a,c, Yuyan Yang a,c, Xinming Dong d, Huiquan Wang b,c, Ruina Zhu b,c, Guang Han b,c, Jinhai Wang b,c,*
PMCID: PMC11886670  PMID: 40060885

Abstract

graphic file with name ao4c09130_0007.jpg

In order to improve the accuracy of concentration detection in turbid media, this study proposes a solution concentration detection method based on a variable-angle wedge-shaped sample cell. Each angle of the wedge-shaped sample cell corresponds to a specific optical path combination. By measuring at multiple wedge angles, we can expand a variety of optical path combinations, providing more information about the turbid media. In this study, transmission light intensity distribution characteristics of phantom solutions were collected at wedge angles ranging from 10 to 45°, with a 5° interval, and a multipath combination detection model was constructed. By analyzing the distribution characteristics of the transmission light intensity along the gradient direction, multipath combination information was fused, and a calibration model was constructed using partial least-squares regression. The results show that, compared to the detection method using a fixed-angle to construct multiple optical paths, the multipath combination method effectively improves the accuracy of concentration detection in turbid media, with a prediction set correlation coefficient (Rp) reaching 0.995. Therefore, this study proposes a new method to enhance the accuracy of turbid media detection by utilizing scattering characteristics through the construction of a multipath combination model by varying the wedge-shaped sample cell angles.

1. Introduction

Turbid media are widely present in fields such as food, biomedicine, and chemical engineering, and refer to homogeneous media that exhibit both absorption and scattering effects.1−3 Rapid and high-precision analysis of their concentration can significantly promote the development of these fields, making the chemical analysis of turbid media a research hotspot.4−6 Current methods for concentration detection in turbid media mainly include culturing methods, dye reagent kits, and vectorial structured light techniques.7−9 However, these methods are limited by experimental conditions, often suffering from long detection times, complex detection processes, and narrow applicability, making it difficult to achieve rapid, nondestructive, and high-precision detection of turbid media components. This has hindered the advancement of chemical analysis in turbid media.

Scattering effects, traditionally considered as interference factors in detection methods, significantly reduce the signal-to-noise ratio (SNR) of optical detection, thereby affecting its accuracy. Additionally, due to the presence of scattering effects, the transmission of light within the medium differs greatly from pure absorption conditions in terms of attenuation, transmission path, and the distribution of light intensity on the transmission surface.10 The impact of scattering effects on the Beer–Lambert law makes it difficult for conventional optical detection methods to achieve high-precision concentration measurements in turbid media. Therefore, traditional methods often aim to eliminate or correct scattering effects to improve detection accuracy.11−13 In the field of complex solution detection, Li et al. proposed the “M + N″ theory,14 demonstrating that increasing the dimensionality of effective information without increasing the dimensionality of interference information plays a crucial role in improving the accuracy of solution component detection. Consequently, some researchers have sought to utilize, rather than eliminate, scattering information in turbid media to enhance the accuracy of component analysis under scattering effects. Lenz et al. proposed a method based on the Kubelka–Munk light propagation model to spatially characterize the scattering and absorption properties of turbid media.15 However, while scattering effects are utilized, data redundancy is also introduced, leading to lower data utilization, reduced signal-to-noise ratio, and decreased model performance. Hence, there is a need to explore an information reconstruction method to extract effective information from scattering.

Our research group has long been engaged in improving the accuracy of turbid media detection and has previously found that wedge-shaped sample cells can effectively enhance the utilization of scattered spectra. By leveraging scattering effects, a multipath model can be constructed within the wedge-shaped sample cell, allowing the light intensity distribution on the transmission surface to contain scattering information on the turbid media. Furthermore, our group has proposed several data processing and modeling methods, such as multidimensional spectral fusion, multidimensional radial distance method, and feature region method,16−18 achieving significant progress in scattering information extraction. In addition, we have investigated the optimal angles of wedge-shaped sample cells.19 However, the optimal wedge angle is influenced by the optical properties of the sample and a fixed wedge angle is not suitable for all turbid media. Therefore, there is a need to explore a more universal concentration detection method for turbid media.

Therefore, this paper proposes a multipath combination fusion analysis method based on near-infrared optical detection technology. This method models the transmission light intensity distribution information collected under multiple optical path combinations, thereby increasing the dimensionality of the effective information for modeling. This turbid media chemical analysis technique is simple in process and can effectively improve concentration detection accuracy while enhancing the universality and robustness of the model.

2. Materials and Methods

2.1. Simulation Model for Component Detection Using Wedge-Shaped Sample Cells at Various Angles

When light is perpendicularly incident on a homogeneous, purely absorptive material and it is assumed that there is no interaction between the absorbing molecules, the relationship between the incident and transmitted light follows the Beer–Lambert law. However, in actual turbid media, light attenuation involves both absorption and scattering effects. This means that the photon path length parameter in the Beer–Lambert law is no longer simply the sample thickness but rather the actual distance traveled by the photons in the measured liquid, which is significantly greater than the thickness of the liquid. In the quantitative analysis of turbid media, we generally consider only Rayleigh scattering and assume that only elastic scattering occurs. That is, when light enters a turbid medium, the scattering particles within the medium scatter or refract the incident photons, altering the distance traveled by the light within the medium. Based on this, a wedge-shaped model is proposed, as shown in Figure 1.

Figure 1.

Figure 1

Open in a new tab

Wedge-shaped sample cell model.

The longitudinal direction of the horizontal transmission surface of the wedge-shaped sample cell is defined as the gradient direction (GD), while the lateral direction is defined as the equal thickness direction (TD). In the figure, D1, D2, and D3 represent the side length parameters of the wedge-shaped sample cell, O denotes the photon incidence point, Δx and Δy indicate the horizontal and vertical distances that photons travel within the turbid medium, respectively, and θ is the angle between the plane containing the photon incidence point and the horizontal transmission surface. When photons are vertically incident from point O and pass through turbid media with different optical path combinations, they undergo scattering and generate light intensity distributions with characteristic differences on the transmission surface. These distribution characteristics form light spot images with varying morphologies on the transmission surface. Preliminary studies have revealed that the angle in the wedge model causes variations in the thickness of the medium within the sample cell, leading to symmetric light spots in the equal thickness direction and asymmetric light spots in the gradient direction. Therefore, compared to the planar model, the wedge model extends the optical path dimension and effectively increases the information content of the light spots through multipath effects.

As shown in Figure 1, with the variation of the wedge angle θ, each angle corresponds to a Δx and Δy. Therefore, the distance that photons travel within the turbid medium changes continuously with the angle, as expressed in eq 1.

2.1. 1

Here, lθ represents the actual path length traveled by the photons at an angle θ, in centimeters, f(θ) is a function related to the angle of the wedge-shaped sample cell, n represents different angles (n = 1,2,3, ···, 8), corresponding to the eight different angles of the wedge-shaped sample cell used in the experiment (i.e., 10, 15, 20, 25, 30, 35, 40, and 45°). Therefore, the expression for the emitted light intensity, I, is given by eq 2. Here, I0 represents the incident light intensity, ελ is the absorption coefficient at wavelength λ, and ci is the concentration of the ith component.

2.1. 2

Although the light spot images obtained through the wedge model contain absorption and scattering information from the turbid medium, the relatively fixed wedge angle results in limited information from single acquisitions and a lower signal-to-noise ratio (SNR). To investigate the impact of different optical path combinations in the wedge model on the analysis of turbid media components, a Monte Carlo simulation was conducted to model the transmission light intensity distribution through turbid media under varying optical path combinations.

In the simulation, the edge lengths D1 and D2 were set to 6 cm, and the angle θ was varied by continuously changing D3. The photon emission position was fixed at the center of the incident surface O. The scattering coefficient was controlled within the range of 0.05 to 0.30 mm–1, while the absorption coefficient was set at 0.05 mm–1, the anisotropic factor g = 0.01, and the refractive index n = 1.37 for the turbid medium in the Monte Carlo simulation. In the simulation, the parameters n and g remained constant, while the scattering coefficient was varied in a stepwise manner. The number of emitted photons was set to 108, and the wedge angle was continuously changed within the range of 10 to 45° with a gradient of 5°. To assess the effectiveness and robustness of this model in real experimental environments, white noise was introduced into the light intensity distribution on the transmission surface of the wedge-shaped sample cell to simulate various experimental errors, including environmental errors, systematic errors, and random errors.

Figure 2 illustrates the relative errors associated with the regression of the light intensity distributions obtained at four fixed wedge angles against concentration using the partial least squares regression (PLSR) model. It can be observed that the errors are relatively dispersed when regression is performed for solution concentrations with different scattering coefficients at various wedge angles. Therefore, for turbid media with different scattering coefficients, the optimal optical path combination that expresses their compositional information will exhibit subtle differences.

Figure 2.

Figure 2

Open in a new tab

Variation of PLSR regression error with scattering coefficient at a fixed wedge angle.

Therefore, this study designs a multipath combination and fusion approach to maximize the capture of scattering information in turbid media. On this basis, the multipath combination and fusion strategy are thoroughly analyzed and validated. Additionally, this fusion strategy is expected to increase the information dimension while reducing the required sample data, thus improving the computational efficiency of the model.

2.2. Experimental Design

The experimental acquisition system employs a 785 nm single-wavelength laser (FL-785–100-M-B, Max-Ray Photonics) as the light source and is composed of an industrial area scan camera (MV-CA020–20GM, HIK VISION), a wedge-shaped sample cell, an experimental platform, a computer, and optical fibers, as illustrated in Figure 3. During the experiment, the laser is vertically incident on the incident surface of the wedge-shaped sample cell. As photons propagate through the phantom solution within the sample cell, they undergo interactions such as absorption and scattering, ultimately forming an oval-like light spot on the transmission surface of the sample cell. By rotating the wedge-shaped sample cell about its axis, the wedge angle can be altered, thereby changing the multipath combination. Finally, the industrial area scan camera captured the light spot images vertically above the transmission surface of the wedge-shaped sample cell.

Figure 3.

Figure 3

Open in a new tab

Experimental data collection system.

The phantom solution used in the experiment was prepared by mixing distilled water, Indian Ink, and Intralipid-20%. The phantom samples were formulated according to the optical parameters set in the Monte Carlo simulation. The absorption coefficient was fixed by maintaining the concentration of Indian Ink at 0.022%, while the scattering coefficient was varied stepwise by adjusting the concentration of Intralipid-20% within the range of 1.4 to 3.9%. A total of 26 sample groups were prepared.

When the transmission surface light intensity is collected, first turn on the laser and configure the relevant parameters. Simultaneously, the industrial area-scan camera was connected, setting the acquisition frame rate to 11 fps and the exposure time to 0.03 s. Then, the prepared sample solution was added to the wedge-shaped sample cell and images of the light spots were captured at a 10° wedge angle. A total of 20 light spot images, each with a resolution of 1300 × 1023 pixels, are collected. After the image acquisition was completed at the current angle, the wedge angle was adjusted incrementally by 5 up to 45°. For each sample, light spot images are captured at 8 different angles, resulting in a total of 160 images per sample.

Once the data for one sample group are fully acquired, turn off the laser and thoroughly clean the sample cell with anhydrous ethanol. Repeat the above steps until all 26 sample groups with varying concentrations have been imaged at eight different angles.

2.3. Data Processing

After obtaining light spot images of different concentrations, this study first averaged the 20 light spot images collected at each wedge angle for each sample group. The intensity distribution curve along the gradient direction corresponding to the maximum value of the light spot is then extracted. Subsequently, the intensity distribution curves obtained at the eight wedge angles are fused, as shown in Figure 4. The specific fusion steps are as follows:

  • (1)

    Extract the eight intensity distribution curves corresponding to different wedge angles at the same concentration, resulting in a total of eight curves.

  • (2)

    Concatenate these eight intensity distribution curves along the gradient direction for the same concentration.

  • (3)

    This process is repeated until the intensity distribution curves for all 26 different concentrations are extracted and combined according to the aforementioned method, resulting in a data set of size 26 × 8 × 1023.

Figure 4.

Figure 4

Open in a new tab

Data fusion methods.

From the 26 rows of data, 19 rows are randomly selected as the training set, while 7 rows are designated as the prediction set.

3. Results and Discussion

Theoretically, from the perspective of sample cell model construction, the wedge model differs from the traditional flat model in that there is an angle between the incident light beam surface and the light spot exit surface. Consequently, there is a shift between the light exit point and the incident point. We analyzed the light spot images obtained at the eight angles. As the angle of the wedge-shaped sample cell changes, the light path combinations also vary, significantly affecting the intensity distribution on the transmission surface. As shown in Figure 5, it can be observed that as the wedge angle increases, the peak of the light spot continually shifts, with the maximum light intensity moving a certain distance toward the direction of the thinner section of the sample cell. Moreover, the maximum light intensity gradually decreases as the thickness of the sample cell increases.

Figure 5.

Figure 5

Open in a new tab

Variation of spot peak value and offset with increasing angle of the wedge model.

The above research indicates that the asymmetric intensity distribution information in the gradient direction contains not only the maximum light intensity value but also the degree of shift of the maximum light intensity. Changes in the chemical composition within the turbid medium or variations in the light path combinations of the wedge-shaped sample cell result in corresponding changes in the transmitted light intensity distribution. Therefore, we studied the fusion of the intensity distribution curves obtained from different light path combinations of the wedge model light spot images in the gradient direction.

A partial least squares regression (PLSR) model was utilized to analyze the concentration of the fused intensity distribution curve data. PLSR is widely applied in data analysis due to its modeling efficiency and high physical interpretability.20,21 The model evaluation metrics include the correlation coefficient of the calibration set (Rc), the root-mean-square error of calibration (RMSEC), the correlation coefficient of the prediction set (Rp), and the root-mean-square error of prediction (RMSEP). The regression results for the concentration of light intensity data based on multipath extension fusion are presented in Table 1.

Table 1. Regression Results of the Wedge Model for Turbid Medium Concentrations at Different Wedge Angles.

wedge angles data dimensions Rc RMSEC Rp RMSEP
multi-angle combinations 26 × 8184 0.9996 0.0079 0.9950 0.0279
10° 26 × 8184 0.9902 0.0382 0.9867 0.0457
15° 26 × 8184 0.9935 0.0328 0.9897 0.0357
20° 26 × 8184 0.9984 0.0163 0.9862 0.0414
25° 26 × 8184 0.9759 0.0608 0.9712 0.0676
30° 26 × 8184 0.9715 0.0651 0.9873 0.0446
35° 26 × 8184 0.9903 0.0367 0.9919 0.0389
40° 26 × 8184 0.9674 0.0696 0.9897 0.0401
45° 26 × 8184 0.9923 0.0642 0.9811 0.0473
Open in a new tab

As shown in Table 1, the regression results for concentrations based on multipath extension fusion data are superior compared to those using a single angle, achieving an Rp of 0.9950 and an RMSEP of only 0.0279. Table 1 also indicates that the data after multipath extension fusion captures the subtle differences and unique characteristics expressed by the different angles regarding the components of the turbid medium. This approach effectively integrates the concentration information on the turbid medium contained in various optical path combinations, increasing the information dimensionality while reducing the data dimensionality. This ensures computational efficiency while enhancing the robustness and observability of the model.

Figure 6 compares the fitting results for the training and prediction sets for concentrations at 10°, 20°, 30°, 40°, and the multiangle combination. From Figure 6, it is evident that the predicted values of the training and prediction sets in the multipath extension fusion approach are closer to the actual values.

Figure 6.

Figure 6

Open in a new tab

Comparison of concentration regression fitting results at different angles and multiangle combinations. (a) Regression fitting results for the training set. (b) Regression fitting results for the prediction set.

The above results indicate that the features after multipath extension fusion significantly increase the information dimension of the turbid medium. By a combination of the advantages of different optical path combinations for concentration regression, the applicability of this method has been greatly enhanced. Therefore, the detection method based on multipath extension fusion can significantly improve the accuracy of chemical analysis of turbid media.

4. Conclusions

Due to the influence of scattering effects in turbid media, concentration analysis methods based on the Lambert–Beer law struggle to achieve high-precision concentration detection. Therefore, our research group proposed a wedge-shaped sample cell designed to utilize the scattering effect to increase the information dimension of the transmitted light intensity distribution. The multipath effect brought about by scattering in the wedge-shaped sample cell allows the transmitted light intensity distribution to include the scattering information on the turbid medium. However, the single optical path combination corresponding to a fixed wedge angle is not applicable to all turbid media, necessitating further exploration of a more universal concentration detection method. Changes in the optical path combination of the wedge-shaped sample cell can be caused by variations in the angle, with different wedge angles corresponding to different optical path combinations.

Therefore, this study has enhanced the accuracy of concentration detection in turbid media by introducing an innovative multipath extension fusion method. This method integrates the advantages of various angles for the analysis of components in turbid media, taking into account the scattering information from different optical path combinations. It not only optimizes the measurement results but also provides a new analytical strategy for achieving high-precision chemical analysis of other scattering media in the future. Furthermore, this method increases the dimensionality of effective information while improving analytical accuracy, thereby enhancing the robustness of the model and providing a new theoretical foundation and experimental approach for research in related fields.

Acknowledgments

This work described in this paper was supported by National Natural Science Foundation of China (grant no. 62105242).

Author Contributions

The manuscript was written through the contributions of all authors. All authors have given approval to the final version of the manuscript. Z.Z. and X.D. provided experimental ideas and method design. Y.Y. and R.Z. collected and analyzed the data and wrote the original script. H.W. and J.W. supervised the experiment and reviewed and revised the first draft. G.H. coordinated the running of the experimental process.

The authors declare no competing financial interest.

References

  1. Choudhari A.; Rube M.; Sadli I.; Sebeloue M.; Tamarin O.; Dejous C. Love-Wave Acoustic Sensors Behavior in Complex Liquids: Multiparameter Sensing Using Acoustic and Electrical Signals. IEEE Sensors Journal 2024, 24 (14), 22300–22306. 10.1109/JSEN.2024.3401452. [DOI] [Google Scholar]
  2. Huang J.; Zhu Y.; Li Y.; Lam E. Y. Snapshot Polarization-Sensitive Holography for Detecting Microplastics in Turbid Water. ACS Photonics 2023, 10 (12), 4483–4493. 10.1021/acsphotonics.3c01350. [DOI] [Google Scholar]
  3. Zhang H.; Gong L.; Li X.; Liu F.; Yin J. An Underwater Imaging Method of Enhancement via Multi-Scale Weighted Fusion. Front. Mar. Sci. 2023, 10, 1150593. 10.3389/fmars.2023.1150593. [DOI] [Google Scholar]
  4. Huang P.-L.; Lo Y.-L.; Chen Y.-R.; Liu C.-Y. Mueller Matrix Polarimetry Approach for Noninvasive Glucose Sensing with Absorbance and Anisotropic Parameters on Fingertips. Journal of Biophotonics 2024, 17 (9), e202400052 10.1002/jbio.202400052. [DOI] [PubMed] [Google Scholar]
  5. Tsai C.-C.; Lo Y.-L.; Chang C.-M.; Huang P.-L. Characterization of Mean Absorbance of Anisotropic Turbid Media Using Stokes–Mueller Matrix Polarimetry Approach. IEEE Transactions on Instrumentation and Measurement 2023, 72, 1–10. 10.1109/TIM.2023.3296117.37323850 [DOI] [Google Scholar]
  6. Li R.; Peng T.; Bai C.; Wang P.; Zhou M.; Yu X.; Min J.; Yao B. Characterization of the Angular Memory Effect of Dynamic Turbid Media. Opt. Express 2023, 31 (17), 27594–27603. 10.1364/OE.495970. [DOI] [PubMed] [Google Scholar]
  7. Maluf M. M.; Bauab K.; Boettger B. C.; Pignatari A. C. C.; Carvalhaes C. G. Evaluation of XGEN Multi Sepsis Flow Chip Molecular Assay for Early Diagnosis of Bloodstream Infection. Curr. Microbiol. 2023, 80 (7), 231. 10.1007/s00284-023-03325-w. [DOI] [PubMed] [Google Scholar]
  8. Huang J.; Wu X.; Feng C.; Yang Y. Application of the Method of Melanin Bleaching After Immunohistochemical Staining of Melanin-Containing Tissues. Appl. Immunohistochem. 2022, 30 (7), 526–530. 10.1097/PAI.0000000000001047. [DOI] [PubMed] [Google Scholar]
  9. Hu X.-B.; Liu C.-X.; Zhu Y.-C.; Chen R.-P.; Zhao B.; Wu F.-M.; Rosales-Guzmán C. Inhomogeneous Enantiomeric Solution Concentration Measurement Harnessing Vectorial Structured Light. ACS Photonics 2024, 11, 4533. 10.1021/acsphotonics.4c00652. [DOI] [Google Scholar]
  10. Mallet A.; Tsenkova R.; Muncan J.; Charnier C.; Latrille É.; Bendoula R.; Steyer J.-P.; Roger J.-M. Relating Near-Infrared Light Path-Length Modifications to the Water Content of Scattering Media in Near-Infrared Spectroscopy: Toward a New Bouguer-Beer-Lambert Law. Anal. Chem. 2021, 93 (17), 6817–6823. 10.1021/acs.analchem.1c00811. [DOI] [PubMed] [Google Scholar]
  11. Mishra P.; Lohumi S. Improved Prediction of Protein Content in Wheat Kernels with a Fusion of Scatter Correction Methods in NIR Data Modelling. Biosystems Engineering 2021, 203, 93–97. 10.1016/j.biosystemseng.2021.01.003. [DOI] [Google Scholar]
  12. Mishra P.; Marini F.; Biancolillo A.; Roger J.-M. Improved Prediction of Fuel Properties with Near-Infrared Spectroscopy Using a Complementary Sequential Fusion of Scatter Correction Techniques. Talanta 2021, 223, 121693 10.1016/j.talanta.2020.121693. [DOI] [PubMed] [Google Scholar]
  13. Mishra P.; Nordon A.; Roger J.-M. Improved Prediction of Tablet Properties with Near-Infrared Spectroscopy by a Fusion of Scatter Correction Techniques. J. Pharm. Biomed. Anal. 2021, 192, 113684 10.1016/j.jpba.2020.113684. [DOI] [PubMed] [Google Scholar]
  14. Wan X.; Li G.; Li T.; Yan W.; He G.; Lin L. A Review on M + N Theory and Its Strategies to Improve the Accuracy of Spectrochemical Composition Analysis of Complex Liquids. Appl. Spectrosc. Rev. 2018, 55, 87–18. 10.1080/05704928.2018.1517361. [DOI] [Google Scholar]
  15. Lenz A. J. M.; Clemente P.; Climent V.; Lancis J.; Tajahuerce E. Imaging the Optical Properties of Turbid Media with Single-Pixel Detection Based on the Kubelka-Munk Model. Opt. Lett. 2019, 44 (19), 4797–4800. 10.1364/OL.44.004797. [DOI] [PubMed] [Google Scholar]
  16. Zhao Z.; Yin H.; Wang Y.; Yan W.; Wang J.; Wang H. Improving the Detection Accuracy of Complex Solution Components Based on Multi-Dimensional Spectroscopy Fusion Method. Infrared Phys. Techn. 2019, 102, 103062 10.1016/j.infrared.2019.103062. [DOI] [Google Scholar]
  17. Zhao Z.; Wang Y.; Yin H.; Yue C.; Wang H.; Han G.; Wang J.; Miao J.; Zhang G.; Yu M.; Chen F. Research on Quantitative Analysis of Turbid Media Based on Multi-Dimension Radial Distance Method. Infrared Phys. Techn. 2020, 111, 103512 10.1016/j.infrared.2020.103512. [DOI] [Google Scholar]
  18. Zhao Z.; Yue C.; Fan W.; Wang Y.; Zhao W.; Han G.; Wang H. Hyperspectral Image Feature Region of Solution Composition Analysis Method Based on Multidimensional Spectra. Infrared Phys. Techn. 2022, 123, 104196 10.1016/j.infrared.2022.104196. [DOI] [Google Scholar]
  19. Zhao Z.; Zhu R.; Yang Y.; Wang H.; Wang J.; Han G. Optical Analysis Method for Turbid Media Based on Wedge-Shaped Cells at Different Angles. Acs Omega 2024, 9 (16), 18119–18126. 10.1021/acsomega.3c09525. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Yao K.; Sun J.; Cheng J.; Xu M.; Chen C.; Zhou X. Monitoring S-Ovalbumin Content in Eggs during Storage Using Portable NIR Spectrometer and Multivariate Analysis. Infrared Physics & Technology 2023, 131, 104685 10.1016/j.infrared.2023.104685. [DOI] [Google Scholar]
  21. Wei X.-L.; Jiang L.; Shi Q.-L.; Mo Z.-H. Machine-Learning-Assisted SERS Nanosensor Platform toward Chemical Fingerprinting of Baijiu Flavors. Microchim Acta 2023, 190 (6), 207. 10.1007/s00604-023-05794-z. [DOI] [PubMed] [Google Scholar]

Articles from ACS Omega are provided here courtesy of American Chemical Society

RESOURCES