Open Access
Add to BibSonomy
Abstract
Existing non-contact flame temperature measuring methods depend on complex, bulky and expensive optical instruments, which make it difficult for portable applications and high-density distributed networking monitoring. Here, we demonstrate a flame temperature imaging method based on a perovskite single photodetector. High-quality perovskite film epitaxy grows on the SiO2/Si substrate to fabricate the photodetector. Duo to the Si/MAPbBr3 heterojunction, the light detection wavelength is extended from 400 nm to 900 nm. Then, a perovskite single photodetector spectrometer has been developed using the deep-learning method for spectroscopic measurement of flame temperature. In the temperature test experiment, the spectral line of doping element K+ has been selected to measure the flame temperature. The photoresponsivity function of the wavelength was learned based on a commercial standard blackbody source. The spectral line of element K+ has been reconstructed using the photocurrents matrix by the regression solving photoresponsivity function. As a validation experiment, the “NUC” pattern is realized by scanning the perovskite single-pixel photodetector. Finally, the flame temperature of adulterated element K+ has been imaged with the error of 5%. It provides a way to develop high precision, portable, low-cost flame temperature imaging technology.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Highly accurate measurement technology of flame temperature is important for combustion mechanism analysis, explosion damage assessment, propellant formulation optimization and fire safety warning. It has a wide application prospect in the field of aeroengine, fuel booster, military, thermal power engineering, metallurgical industry, fire safety and so on [1–3].
Generally, the mainly measurement methods of flame temperature are contact-type [4,5] and non-contact-type [6–8]. Contact measure has high accuracy but point survey and interference by intrusion. Non-contact measure has wider range, non-invasive, fast response, high accuracy, anti-interference and measurement continuity. Currently, Non-contact-type flame temperature measurement methods are laser spectroscopy and radiation spectroscopy. Typical laser spectroscopy methods include interferometric spectroscopy [9,10], scattering spectroscopy [11] and absorption spectroscopy [12,13]. These methods have high accuracy and non-invasive, but it is difficult to achieve remote measurement. Traditional radiation spectroscopy methods include monochromatic [14], two-color [15] and hyperspectral [16,17]. These methods are achievable remote measurement. In has been widely used in a variety of high temperature measurement industries. Cai reports a hybrid method to combines three-dimensional fiber optic imaging with two-color pyrometry to simultaneous image the 3D thermometry [18]. And, they further present an approach for high-speed four-dimensional (3D + t) thermometry of a weakly turbulent diffusion flame [19]. However, the performance of non-contact-type methods depends on the complex, bulky and expensive optical instruments, which is difficult to make portable applications. Especially, it is difficult for high-density distributed networking to monitor the 3D flame temperature.
The single-pixel detector imaging is a portable, low-cost method to be used in many fields. Cai reports a method of flame imaging with a cost-effective single-pixel UV camera [20]. Ji reports a novel wide-angle FOV camera is developed by combining a single perovskite photodetector using computational technology [21–23]. Hasan review that the single-pixel detector has been used as the miniaturization optical spectrometers based on the spectral response learning and reconstruction method [24]. It can capture images with a single-pixel detector, rather than a CCD/CMOS array. It provides a low-cost, high spatial resolution method to image the temperature of flame. Meanwhile, perovskite photodetectors have exhibited excellent performance with high sensitivity, fast response speeds, and easy wavelength selection. The peak responsivity of the Si/MAPbBr3 heterojunction photodetector is up to 11.5 mA/W [25]. And that, the spectral wavelength of the doped element in the flame is in the visible light band. Perovskite photodetectors are the appropriate choice to be used to image the two-dimensional flame temperature.
In this work, we present a high precision, portable, low-cost flame temperature im-aging method based on a perovskite single photodetector. Perovskite film epitaxy grows on the Si wafer with a SiO2 dielectric layer forming a photodetector structure. Duo to the high light responsivity in the range from 400 nm to 900 nm, a perovskite single photodetector spectrometer has been developed using the deep-learning method. The photoresponsivity of different wavelength is depend on the voltage at both ends of structure, which can be learned as a photoresponsivity matrix for the deep-learning system based on a commercial standard blackbody source. Then, based on the doping element spectroscopic measurement of flame temperature method, the spectral line of element K+ have been selected to measure the flame temperature, which also acted as the characteristic spectrum has been learned by the perovskite single photodetector spectrometer. Using the deep-learning method, only the spectral line of element K+ has been reconstructed using the corresponding photocurrent vector in a multispectral experimental environment. As a validation experiment, typical gray scale pictures with letters imaging is realized by scanning the perovskite single-pixel photodetector. Finally, the flame temperature has been imaged and the accuracy of temperature measurement is smaller than 5% whit verifying by the infrared thermal imaging technology. It provides a potential route to develop high precision, portable, low-cost flame temperature imaging technology.
2. Materials and methods
CH3NH3Br (≥99.5%) was purchased from Ningbo Borun New Material Technology Co., Ltd. PbBr2 (>99.9%) was purchased from Aladdin and N, N-dimethylformamide with molecular sieves (DMF, 99.8%) and CH2Cl2 (Analytical Reagent) were purchased from Sigma-Aldrich. SiO2/Si substrate was obtained from the Dongguan Senshuo Technology Co., Ltd.
The MAPbBr3 solution was prepared to grow perovskite thin film on the surface of SiO2/Si substrate. With equal molar ratios, MAPbBr3 solution was prepared by mixing CH3NH3Br (0.1344 g) and PbBr2 (0.4405 g) in DMF (1 ml) solvent in a glass bottle. After sealing and stirring at room temperature for 12 h, the solution was then filtered using a 0.45 µm disposable needle filter.
Perovskite photodetector was fabricated on SiO2/Si substrate. Single-crystal MAPbBr3 perovskite was directly grown on SiO2/Si substrate, and then gold electrode of 50 nm was sputtered onto the surface of the MAPbBr3 perovskite. The gold electrode and the Si substrate were connected using the silver wires. The photoelectrical performance of the detector was measured using a semiconductor parameter analyzer system (Keithley 4200-SCS) under ambient conditions.
Potassium sulfate (K2SO4) was purchased from Shanghai Maclin Biochemical Technology Co. Potassium sulfate powder was grinded in a bowl and pressed into tablets by automated X-Press (Model: PP-30S from Tianjin Jiaxinhai Machinery Equipment Co.). The tablets were used to measure the spectral information of K+ element by the Laser-induced breakdown spectroscopy (LIBS) (Model: ChemReveal from American TSI Co.) [26].
A commercial blackbody source (Model: LS-3000 from CI Co., Israel) was used to train the photoresponsivity model in a dark room. The temperature range of blackbody source is 50◦C-3000◦C and the spectral emissivity is 0.997 (Wavelength range from 0.2 µm to 11 µm). The imaging photodetector was placed at 0.5 m from the blackbody source.
The accuracy of flame temperature was verified by thermocouples and infrared thermograph. The temperature upper limit of K-type Pt-Rh thermocouples is 1600 K with the error of 1%. The temperature range of infrared thermograph (Model: 858X-Ex from Shanghai Hot Image Technology Co., Ltd) is 253.2 K-1973.2 K with the error of 2%.
3. Results and discussion
3.1 Flame temperature imaging scheme and measurement principle
Traditional spectral imaging measurement method of flame temperature as shown in Fig. 1. The chemical elements of medium in the flame can radiate the characteristic spectral lines as the electrons of chemical elements jump from the ground state to the excited state and back to ground state pumped by the local high temperature. Generally, the elements of Gu+ and K+ are usually used as the doping element in the flame. In our previous work [27–29], the imaging grayscale of the flame with doping K+ element is much higher than that with the doping Gu+ element. Hence, the K+ element have been selected in this work.
Fig. 1. The measure principle of the single-detector spectral imaging measurement method of flame temperature. (a) Schematic diagram of the LIBS. (b) Spectrum lines of sample (K2SO4) (c) Traditional method. (d) Single photodetector method. (e) Schematic of the learning process by tuning the bias voltage. (f) Responsivity matrix RV, λ of the perovskite single photodetector. (g) Sampling and reconstruction of the spectrum of unknown flame based on the photocurrent vector IV.
Firstly, spectral signature of K+ element in the K2SO4 powder have been tested using the Laser-induced breakdown spectroscopy (LIBS) system as shown in Fig. 1(a). The laser passes through the objective lens to hit the K2SO4 powder, and the excited light carrying spectral information reaches the spectrometer. The strong spectral line of element K (404.4 nm, 691.1 nm, 693.8 nm, 766.5 nm, 769.9 nm) was shown in Fig. 1(b). By contrast, the intensity of spectral line at 766.5 nm and 769.9 nm is extremely stronger than others. Hence, this wavelength band was selected to be learned and reconstructed by perovskite photo-detector to image the temperature of flame.
Based on the method of spectroscopic measurement of temperature [30–32], the spectral signals at any temperature node of flame can be detected by the camera with a narrow band filter, which only transfers the wavelengths signals of chemical elements of medium in the flame as shown in Fig. 1(c). Then, the flame temperature can be calculated using the transfer function between the spectrum and temperature based on the Planck's law, Blackbody radiation law.
Compere to the traditional photodetection system, a microscale perovskite single photodetector with the deep-learning algorithm has been designed to replace the photodetection system in this work as illustrated in Fig. 1(d). The spectrum of chemical elements in the flame will be detected using the perovskite single-photodetector by comprising three steps: learning, sampling, and reconstruction, as shown in Fig. 1(f) and 1(g). The photoresponsivity (R) of the perovskite single photodetector could be tuned by varying the bias voltage (V), as shown in Fig. 1(e). R is a function of both the light wavelength (λ) of chemical elements and the bias voltage (V).
The learning process is as follows. The photoresponsivity function R (V, λ) can be built into a matrix RV, λ for the neural network algorithm by learning the bias voltage and wavelength, as shown in Fig. 1(f). The photoresponsivity row vector for Vi (RVi, λ1, RVi, λ2, …, RVi, λn) in the matrix RV, λ was learned from spectrum of chemical elements at the different bias voltage.
As illustrated in Fig. 1(g), the photocurrent of a spectrum of unknown flame was measured and sampled at different bias voltages (V1-Vn) to build the response vector IV. The spectrum of unknown flame can be reconstructed based on the responsivity matrix RV, λ. Then, the temperature of unknown flame can be calculated based on the Planck's law.
3.2 Single photodetector design and optical properties
Single crystal perovskite was heterogeneously grown on the SiO2/Si substrate using the optimized antisolvent vapor-assisted crystallization method, as shown in Fig. 2(a). The SiO2/Si wafer was placed into the MAPbBr3/DMF solution in a beaker, and which was placed in a big sealed bottle with CH2Cl2. Then, the saturated CH2Cl2 vapor diffuses into the MAPbBr3 solution to precipitate the MAPbBr3 crystal on the SiO2/Si substrate. The photo image of Si/SiO2/perovskite as shown in Fig. 2(b), the MAPbBr3 crystal film is orange with a hexagon shape. The size of the two-dimensional perovskite is about 10 µm × 10 µm and the thickness is about 3 µm.
Fig. 2. Fabrication and characterization of the MAPbBr3 film. (a) Schematic diagram for growing MAPbBr3 on the SiO2/Si substrate. (b) Photo image of MAPbBr3. (c) Cross-sectional image of the interface between the Si and MAPbBr3. (d) XRD mapping of the single-crystal and two-dimensional MAPbBr3. (e) Photoluminescence properties of MAPbBr3.
The cross-sectional image of the Si/SiO2/MAPbBr3 interface have been observed using the high-resolution TEM (HRTEM) as shown in Fig. 2(c). The clear lattice fringes of both Si and MAPbBr3 can be observed to prove the perovskite crystalline quality is very high. The X-ray diffraction (XRD) test results further proves the excellent single crystal properties of MAPbBr3. The XRD spectrum shows the sharp peaks at 2θ = 14.9, 30.1 and 45.8°, which are corresponding to the (001), (002) and (003) planes of the crystal structure of MAPbBr3, respectively (Fig. 2(d)). Meanwhile, the photoluminescence (PL) spectrum of the MAPbBr3 has been detected as illustrated in Fig. 2(e), a clear absorption peak at the wavelength of ∼542 nm has been obtained, which is consistent with other reported values. Based on the above test results, it is proved that the high-crystal quality perovskite has been grown on Si/SiO2 substrate.
The structure of the Si/SiO2/perovskite/Au photodetector is shown in Fig. 3(a). The spectral response characteristic was measured under the wavelength from 400 nm to 900 nm (Fig. 3(b)). Generally, the spectral response range of perovskite materials does not exceed 700 nm. In our work, the photogenerated carriers are generated in the Si/MAPbBr3 heterojunction as the wavelengths of 700 nm to 900 nm. At the bias voltage, the built-in electric field induced a photocurrent under the light illumination [33].
Fig. 3. Photodetector schematic and optical performance. (a) Schematic diagram of the Si/SiO2/perovskite/Au photodetector. (b) Spectral response performance in the wavelength range of 400 nm-900 nm. (c) Current-voltage (I-V) curves of the photodetector under different wavelengths. (d) I-V curves of the photodetector under different light intensities at a wavelength of 550 nm. (e) Responsivity and Detectivity as functions of the power density at 3 V bias. (f) The transient response of the device at a wavelength of 550 nm under a 1kHz pulse light.
Figure 3(c) displays the I-V curves of the perovskite photodetector under the wavelengths of 590 nm and 770 nm with light power density of 1 mW/cm2. The well-defined relationship between the photocurrent and the bias voltage and wavelength is essential for the realization of spectral reconstruction. In the forward bias, as the bias voltage increases from 0 V to 1 V, the photocurrent increases to 0.5 µA under the wavelength of 770 nm and 2.6 µA under the wavelength of 590 nm. However, the photocurrent at 770 nm is smaller than the dark current of 1.1 µA. It reveals that our perovskite photodetector is difficult to using at 770 nm in the forward bias. In the reverse bias, as the bias voltage increases from 0 V to -1 V, the photocurrent under the wavelengths of 590 nm and 770 nm is far larger than dark current (0.2 µA). The photocurrent is 4.8 µA and can continue to increase with the input light intensity and bias voltage. The responsivity/photocurrent matrix R (Vbias, λ) related to wavelength and bias voltage is obtained for spectral reconstruction algorithm.
Meanwhile, Fig. 3(d) displays the I-V curves of the perovskite photodetector under the light intensity at the wavelengths of 550 nm. The photocurrent increases gradually under the different orders of magnitude of light intensity. The photocurrent can be increased more than 0.1 mA under the light power density of 10 mW/cm2 at a bias of 3 V, which is far larger than dark current (0.2 µA). And then, the calculated detectivity can be up to 5.22 × 1010 Jones at 3 V bias as shown in Fig. 3(e). The responsivity and detectivity reduced with the increase in the light intensity. The responsivity and detectivity are as high as 13.45 A/W and 8.56 × 1012 Jones, respectively at the light intensity of 1 µW/cm2. The reason was that more photogenerated carriers recombined with the photogenerated hole in the device as increasing the light intensity.
Figure 3(f) shows the response speed of the perovskite photodetector. The rise time is defined as the time from 10% to 90% and the decay time as the time 90% to 10% of the maximum optical signals. The rise time is determined by the narrow width of the space charge region, and the decay time is determined by the lifetime of nonequilibrium minority carriers, which is proportional to semiconductor conductivity. The rise time is about 5.04 µs and the decay time is 65.85 µs under a 1 kHz pulse light. Therefore, our perovskite photodetector can fast response to light signals. The stability of our detector has been tested as shown in Fig. S1 in the Supplement 1.
3.3 Spectral response learning process and reconstruction
In order to reconstruct the doping element spectrum using the machine learning algorithm, we use a blackbody source to train the photocurrents responsivity model and acquire the spectral responsivity matrix. In the learning process, a 768 nm filter lens with the bandwidth of 20 nm was placed on the surface of our perovskite photodetector photodetector. At given bias voltage, the photocurrents (I) of our photodetector have been learned by the blackbody source at different temperature Ti. Hence, the photocurrents matrix I at different the bias voltage Vi is depending on the blackbody source temperature Ti and the spectral responsivity R(λ) as I = f (T, R(λ)), where f is an unknown non-linear function to be used to reconstruct spectrum. The function f can be mapped to the space of spectral responsivity R(λ) and the incident power density P (T, λ) of blackbody source to establish the non-linear regression equation. R(λ) can be regression analyzed by capturing the non-linear relationship between photocurrents and temperatures. And P (T, λ) was calculated by the wavelength λ and the temperature T based on Planck’s Law.
As shown in Fig. 4(a), the power densities (P) of the blackbody source at three temperatures of Tx, Ty and Tz is corresponding to the wavelength. The unknown responsivity R(λ) at the given bias voltage Vi is the blue shaded area in Fig. 4(a). For a given temperature Ti, the photocurrent I(T) of our photodetector is an integral of the product of the power density and the responsivity over the entire wavelength range of interests from λmin to λmax as follow:
Fig. 4. Element spectral responsivity matrix learning. (a) Power spectral densities (P) of a blackbody source at three different temperatures (Tx (red), Ty (blue) and Tz (green)) and the unknown responsivity curve of perovskite photodetector at the wavelength of 758-778 nm to be learned at a given biasing voltage (black dotted line). (b) The photoresponsivity spectrum for each of bias voltage (Vi) is reconstructed from the photocurrents (Integration of the photoresponse spectrum at different temperatures) and incident light spectra from 400 nm to 900 nm.
Hence, based on the Eq. (1), the photocurrent measurements at the blackbody temperatures from T1 to Tn has been decomposed into a matrix equation to train the responsivity model by discretization as follow: [34]
The responsivity vector RVi at bias voltage Vi can be reconstructed by the Eq. (2) and the corresponding curves as shown in Fig. 4(b).
In order to find the best solution to minimize the squared error ‖R × PT, λ−IT‖2 between the measurements (IT) and calculated results, R × PT, λ. A regularization method has been used to improve the robustness by minimizing the cost function: [35,36]
Here, α×w(R) is the penalty term in the cost function, where w(R)=‖R‖2 in the Tikhonov regularization. α is the penalty term coefficient. Besides, the cutoff wavelength λc has been selected to avoid over-fitting.
In the learning process, these two parameters have been adaptively chosen to be a specific dataset to minimize mean squared errors (MSEs) in a cross-validation. The dataset was divided into a training set to train responsivity model, and a test set to validated the least MSEs as a function of these two parameters. The parameters λc and α can result in under-fitting or over-fitting and a large MSEs, and even lead to the failure of generalization. Thus, we chose the adaptively parameters for training the responsivity model. This model has been verified to reconstruct the spectrum of doping element in this work.
To verify the capability of the trained responsivity model to reconstruct varied spectrum, the doping element K+ spectrum has been reconstructed by leveraging Tikhonov regularization regression (as shown in the Supplement 1). As shown in Fig. 5(a), three spectral response curves at three different bias voltages (Vi, Vj and Vk) has been used to reconstruct the unknown spectrum (the blue dashed line). The photocurrents of these three spectral response curves was expressed in an integral at each bias voltage as follow:
Based on the Eq. (4), the photocurrent of each bias voltage was integrated by responsivity curves and the incident spectrum as shown in Fig. 5(b) (red, blue, and green crosses). The unknown spectrum can be reconstructed by fitting these photocurrent data as shown in the right of Fig. 5(b). In the reconstruction process, like the learning process, a penalty coefficient α of regularizations in the regression was used to avoid overfitting. The continuous integral equations at these bias voltages were discretized into a matrix equation as follow:
Fig. 5. Spectroscopy reconstruction. (a) Photoresponsivity spectra at three bias voltage, Vi (red), Vj (blue) and Vk (green), and an unknown doping element light spectrum (blue dashed line). (b) Unknown spectrum was reconstructed by the photocurrents as a function of Vi, Vj and Vk. (c) The spectral signature of K+ element is measured and reconstructed as a function of V, from V = 0 to -1 V (500 points).
A confirmatory experiment was performed to measure the spectrum of the doping element K+ as shown in Fig. 5(c). In this measurement, the bias voltage is from 0 V to 1 V and the number of sampling points is n = 500, as the red hollow ring in the Fig. 5(c). By fitting the red hollow ring, the reconstructed spectrum is the blue dashed curve, which is agrees well with the results measured by the LIBS system as shown in Fig. 1(b). It proves our method successfully reconstruct the spectrum of doping element K+ in the flame.
3.4 Two-dimensional (2-D) flame temperature image and verification
Similar to the imaging method as shown in Fig. S2 in the Supplement 1, an oxygen-alcohol diffusion flame with the height of 60 mm and a diameter of 35 mm was used as the imaging target. In order to reduce the influence of flame flow on test results. The imaging time needs to be as short as possible. And there is no airflow in the testing environment. In the experiments, the scanning step is 60 µm, and scanning speed is 700 mm/s. The single photodetector area is about 10 µm × 10 µm. The size of the flame is about 60 mm × 35 mm. The total imaging time is less than 60 s. Small amount of potassium sulfate powder (K2SO4) was intermittently sprayed along the direction of the flame using the feeding structure. The perovskite photodetector was used to capture the spectral information of the doping element K+ in the flame.
The single photodetector is used to scan the flame point by point, and obtain the photocurrent corresponding to the spectrum of the K+ element in each scanning point. Record the photocurrent of each point, and then plot these photocurrents of each point with software to map the spectral imaging image. According to the calibration function between the photocurrents of our photodetector and emissivity calibration function [27], the spectral emissivity distribution of flame was calibrated by the doping element energy spectrum as shown in Fig. 6(a). Then, based on the Wien's displacement law, the temperature can be calculated by the equation T = 2.9 × 10−3/λ (m. K). Where, λ is the wavelength of the K+ element. The temperature distribution of 2-D flame can be calculated by the 2-D emissivity matrix and the radiation luminance matrix of flame. As shown in Fig. 6(b), the flame temperature is from 1050 K to 1150 K.
Fig. 6. Measurement results of flame with element doping and the error analysis. (a) Emissivity distribution of flame. (b) Temperature distribution of flame. (c) Division of original image of flame. (d) Infrared thermal image. (e) Comparison between calculated temperature and standard temperature and relative errors.
In order to verify the results of single photodetector imaging method, the thermocouples and infrared thermograph have been used to measure the temperature distribution of the flame. Firstly, due to the high accuracy, 10 thermocouples were arranged on the outer layer in the direction of 2-D flame diffusion as shown in Fig. 6(c). When the infrared thermal imaging temperature is equal to the thermocouple measured temperature at a certain point of unfiltered and undoped flame, the infrared emissivity of the flame can be obtained to image the actual temperature of flame by the infrared thermograph. As shown in Fig. 6(d), the actual temperature of flame is increasing from 1005.5 K to 1125.5 K.
As shown in Fig. 6(e), there is a good agreement between infrared thermal imaging and our perovskite spectral imaging. Compare to the actual temperature of each point along the flame diffusion direction, the measurement result of perovskite photodetector is higher. But the variation trend is same. The maximum relative error is smaller than 5%. There are two possible reasons. One is the integral error to calculate the temperature field. Another is the effect of doped element powder on the distribution of primordial elements and force fields in the flame. Nonetheless, our approach is proved to be a way to image the spectral temperature based on the single photodetector.
4. Conclusions
In conclusions, we report a high precision, portable, low-cost flame temperature imaging scheme based on a perovskite single photodetector. Based on deep-learning method, this single photodetector has been designed as a spectrometer does not depend on the complex, expensive optical components of movable gratings, dispersion module, interferometers, or tunable lasers. Following the method of spectroscopic measurement of flame temperature, the spectral line of doping element K+ in flame has been learned by this spectrometer to train the photoresponsivity model. And the flame temperature was reconstructed and imaged with the error of 5% by the regression solving photoresponsivity function. It provides a way to develop on-chip, low-cost flame temperature imaging technology.
Funding
National Natural Science Foundation of China (52075504); Shanxi Key Laboratory of Signal Capturing & Processing (ISPT2020-10); Natural Science Foundation of Shanxi Province of China (201901D111162) ; Shanxi 1331 Project Key Subject Construction.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
Supplemental document
See Supplement 1 for supporting content.
References
1. S.T. Lazar, T.J. Kolibaba, and J.C. Grunlan, “Flame-retardant surface treatments,” Nat. Rev. Mater. 5(4), 259–275 (2020). [CrossRef]
2. T. Pinheiro Araújo, C. Mondelli, M. Agrachev, T. S. Zou, P. O. Willi, K. M. Engel, R. N. Grass, W. J. Stark, O. V. Safonova, G. Jeschke, S. Mitchell, and J. Pérez-Ramírez, “Flame-made ternary Pd-In2O3-ZrO2 catalyst with enhanced oxygen vacancy generation for CO2 hydrogenation to methanol,” Nat. Commun. 13, 5610 (2022). [CrossRef]
3. T. G. Osimitz, S. Kacew, and A. W. Hayes, “Assess flame retardants with care,” Science 365(6457), 992–993 (2019). [CrossRef]
4. C. Xiong, Y. Liu, H. Fan, X. Huang, and Y. Nakamura, “Fluctuation and extinction of laminar diffusion flame induced by external acoustic wave and source,” Sci. Rep. 11(1), 14402 (2021). [CrossRef]
5. S. Wang, D. Zhang, W. Wang, J. Zhong, K. Feng, Z. Wu, B. Du, J. He, Z. Li, L. He, W. Sun, D. Yang, and G. A. Ozin, “Grave-to-cradle upcycling of Ni from electroplating wastewater to photothermal CO2 catalysis,” Nat. Commun. 13(1), 5305 (2022). [CrossRef]
6. T. Chen, X. Guo, J. Jia, and J. Xiao, “Frequency and Phase Characteristics of Candle Flame Oscillation,” Sci. Rep. 9(1), 342 (2019). [CrossRef]
7. A. Aouali, S. Chevalier, A. Sommier, E. Abisset-Chavanne, J.-C. Batsale, and C. Pradere, “3D infrared thermospectroscopic imaging,” Sci. Rep. 10(1), 22310 (2020). [CrossRef]
8. N. Hansen, R. S. Tranter, K. Moshammer, J. B. Randazzo, J. P. A. Lockhart, P. G. Fugazzi, T. Tao, and A. L. Kastengren, “2D-imaging of sampling-probe perturbations in laminar premixed flames using Kr x-ray fluorescence,” Combust. Flame 181, 214–224 (2017). [CrossRef]
9. X. Xiao, C. W. Choi, and I. K. Puri, “Temperature measurements in steady two-dimensional partially premixed flames using laser interferometric holography,” Combust. Flame 120(3), 318–332 (2000). [CrossRef]
10. G. Sharma, A. M. Shrivastav, A. Kumar, and R. Jha, “Non-graphene two-dimensional nanosheets for temperature sensing based on microfiber interferometric platform: Performance analysis,” Sens. Actuators, A 289, 180–187 (2019). [CrossRef]
11. H. Tang, C. Yang, G. Wang, T. F. Guiberti, and G. Magnotti, “Raman spectroscopy for quantitative measurements of temperature and major species in high-pressure non-premixed NH3/H2/N2 counterflow flames,” Combust. Flame 237, 111840 (2022). [CrossRef]
12. E. Boigné, P. Muhunthan, D. Mohaddes, Q. Wang, S. Sobhani, W. Hinshaw, and M. Ihme, “X-ray computed tomography for flame-structure analysis of laminar premixed flames,” Combust. Flame 200, 142–154 (2019). [CrossRef]
13. J. H. Frank, A. Shavorskiy, H. Bluhm, B. Coriton, E. Huang, and D. L. Osborn, “In situ soft x-ray absorption spectroscopy of flames,” Appl. Phys. B 117(1), 493–499 (2014). [CrossRef]
14. H. Wang, D. J. Kline, and M. R. Zachariah, “In-operando high-speed microscopy and thermometry of reaction propagation and sintering in a nanocomposite,” Nat. Commun. 10(1), 3032 (2019). [CrossRef]
15. N. J. Kempema and M. B. Long, “Effect of soot self-absorption on color-ratio pyrometry in laminar coflow diffusion flames,” Opt. Lett. 43(5), 1103–1106 (2018). [CrossRef]
16. M. J. Montgomery, H. Kwon, A. L. Kastengren, L. D. Pfefferle, T. Sikes, R. S. Tranter, Y. Xuan, and C. S. McEnally, “In situ temperature measurements in sooting methane/air flames using synchrotron x-ray fluorescence of seeded krypton atoms,” Sci. Adv. 8(17), eabm7947 (2022). [CrossRef]
17. I. Obreht, D. De Vleeschouwer, L. Wörmer, M. Kucera, and K. Hinrichs, “Last Interglacial decadal sea surface temperature variability in the eastern Mediterranean,” Nat. Geosci. 15(10), 812–818 (2022). [CrossRef]
18. Q. Wang, J. Huang, H. Liu, Z. Qin, and W. Cai, “3D particle sizing, thermometry and velocimetry of combusting aluminized propellants,” Combust. Flame 247, 112500 (2023). [CrossRef]
19. T. Yu, F. J. Bauer, F. J. Huber, S. Will, and W. Cai, “4D temperature measurements using tomographic two-color pyrometry,” Opt. Express 29(4), 5304–5315 (2021). [CrossRef]
20. J. Zhang, J. Dai, Q. Wang, and W. Cai, “Demonstration of a cost-effective single-pixel UV camera for flame chemiluminescence imaging,” Appl. Opt. 58(19), 5248–5256 (2019). [CrossRef]
21. Z. Ji, Y. Liu, and X. Chen, “Mosaic-free compound eye camera based on multidirectional photodetectors and single-pixel imaging,” Opt. Lett. 47(24), 6349–6352 (2022). [CrossRef]
22. Z. Ji, Y. Liu, C. Zhao, Z. L. Wang, and W. Mai, “Perovskite wide-angle field-of-view camera,” Adv. Mater. 34(41), 2206957 (2022). [CrossRef]
23. Z. Ji, Y. Liu, M. Yao, Z. Zhang, J. Zhong, and W. Mai, “Achieving 256 × 256-pixel color images by perovskite-based photodetectors coupled with algorithms,” Adv. Funct. Mater. 31(37), 2104320 (2021). [CrossRef]
24. Z. Yang, T. Albrow-Owen, W. Cai, and T. Hasan, “Miniaturization of optical spectrometers,” Science 371(6528), eabe0722 (2021). [CrossRef]
25. M. Cao, J. Tian, Z. Cai, L. Peng, L. Yang, and D. Wei, “Perovskite Heterojunction Based on H3NH3PbBr3 Single Crystal for HighSensitive Self-Powered Photodetector,” Appl. Phys. Lett. 109(23), 233303 (2016). [CrossRef]
26. S. Maurice, B. Chide, N. Murdoch the SuperCam team, et al., “In situ recording of Mars soundscape,” Nature 605(7911), 653–658 (2022). [CrossRef]
27. Y. Yang, X. Hao, L. Zhang, and L. Ren, “Application of Scikit and Keras Libraries for the Classification of Iron Ore Data Acquired by Laser-Induced Breakdown Spectroscopy (LIBS),” Sensors 20(5), 1–11 (2020). [CrossRef]
28. X. Liu, X. Hao, B. Xue, B. Tai, and H. Zhou, “Two-Dimensional Flame Temperature and Emissivity Dis-tribution Measurement Based on Element Doping and Energy Spectrum Analysis,” IEEE Access 8, 63–74 (2020). [CrossRef]
29. B. Xue, X. Hao, X. Liu, Z. Han, and H. Zhou, “Simulation of an NSGA-III Based Fireball Inner-Temperature-Field Reconstructive Method,” IEEE Access 8, 8–19 (2020). [CrossRef]
30. M. S. Fang, H. G. Gu, Z. F. Guo, J. M. Liu, L. S. Huang, and S. Y. Liu, “Temperature and thickness dependent dielectric functions of MoTe2 thin films investigated by spectroscopic ellipsometry,” Appl. Surf. Sci. 605, 154813 (2022). [CrossRef]
31. J. Dedina, A. Dulivo, L. Lampugnani, T. Matousek, and R. Zamboni, “Selenium hydride atomization, fate of free atoms and spec-troscopic temperature in miniature diffusion flame atomizer studied by atomic absorption spectrometry,” Spectrochim. Acta, Part B 53(13), 1777–1790 (1998). [CrossRef]
32. D. U. Kim, K. S. Park, C. B. Jeong, G. H. Kim, and K. S. Chang, “Quantitative temperature measurement of multi-layered semiconductor devices using spectroscopic thermoreflectance microscopy,” Opt. Express 24(13), 13906–13916 (2016). [CrossRef]
33. X. Geng, F. Wang, H. Tian, Q. Feng, H. Zhang, R. Liang, Y. Shen, Z. Ju, G. Gou, N. Deng, Y. Li, J. Ren, D. Xie, Y. Yang, and T. Ren, “Ultrafast Photodetector by Integrating Perovskite Directly on Silicon Wafer,” ACS Nano 14(3), 2860–2868 (2020). [CrossRef]
34. S. Yuan, D. Naveh, K. Watanabe, T. Takashi, and F. Xia, “A wavelength-scale black phosphorus spectrometer,” Nat. Photonics 15(8), 601–607 (2021). [CrossRef]
35. T. Paul, P. W. Chi, P. M. Wu, and M. K. Wu, “Computation of distribution of relaxation times by Tikhonov regularization for Li ion batteries: usage of L-curve method,” Sci. Rep. 11(1), 12624 (2021). [CrossRef]
36. H. H. Yoon, H. A. Fernandez, F. Nigmatulin, W. Cai, Z. Yang, H. Cui, F. Ahmed, X. Cui, M. G. Uddin, E. D. Minot, H. Lipsanen, K. Kim, P. Hakonen, T. Hasan, and Z. Sun, “Miniaturized spectrometers with a tunable van der Waals junction,” Science 378(6617), 296–299 (2022). [CrossRef]
