Raman-near infrared spectrum combined intelligent sensor and detection method

Abstract

The invention discloses a Raman-near infrared spectrum combined intelligent sensor and a detection method, and relates to the technical field of spectrum intelligent sensing detection, a Raman laser light source is excited by a constant current pulse driving module to generate pulse laser, and a near infrared light source is combined to form Raman-near infrared mixed light; the method comprises the following steps of: performing dispersion according to wavelength by an MOEMS light splitting component, performing shunt detection, dynamically amplifying two paths of electric signals through a dynamic gain adjustment method, removing cross crosstalk through dual-band notch filtering, performing synchronous analog-to-digital conversion and performing wavelet transform noise reduction, and establishing characteristic peak space correlation mapping based on a molecular vibration-rotation coupling principle. Raman spectrum comparison qualitation and corrected partial least square method quantification are combined, trace components are identified through characteristic peak expansion, and finally accurate qualitative and quantitative detection of the sample is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent spectral sensing and detection technology, specifically to an intelligent sensor and detection method using Raman-near-infrared spectroscopy. Background Technology

[0002] In the field of intelligent spectral sensing and detection technology, although the combined use of Raman spectroscopy and near-infrared spectroscopy can complement each other to achieve qualitative and quantitative analysis of substances, and integrated intelligent sensors have become the mainstream development direction of spectral detection, traditional Raman-near-infrared combined intelligent sensors still have the following technical bottlenecks: First, the sensor's signal processing unit uses a fixed-gain amplifier circuit, which cannot match the differences between the transient, narrow pulse, and large amplitude fluctuation of Raman signals and the steady-state and low-fluctuation characteristics of near-infrared signals. This easily causes pulse waveform distortion or DC signal amplitude instability. Furthermore, the sensor's filtering module often relies on empirically set frequency bands, resulting in inaccurate suppression of crosstalk between the two types of signals. Residual crosstalk can interfere with the extraction of spectral characteristic peaks, leading to sensor... First, the sensing accuracy is reduced. Second, the data analysis unit of the sensor lacks cross-spectral correlation verification logic, relying solely on Raman or near-infrared spectral comparison for qualitative and quantitative analysis. This easily leads to misjudgment of sample component groups due to overlapping characteristic peaks or background interference. The quantitative model does not incorporate the correlation characteristics of the two types of spectra for correction, resulting in large systematic errors. At the same time, it is difficult to identify trace components that are masked by the characteristic peaks of the main components, leading to substandard total content calculations and failing to meet the high-precision and high-reliability detection requirements of smart sensors. Third, the spectroscopic modules of traditional sensors mostly adopt fixed grating structures, resulting in poor spectroscopic flexibility and an inability to adapt to the wide-band dispersion requirements of Raman-near-infrared mixed light, which contradicts the trend of integration and miniaturization of smart sensors. Summary of the Invention

[0003] The purpose of this invention is to provide a Raman-near-infrared spectroscopy combined intelligent sensor and detection method, which achieves the goal of suppressing crosstalk and realizing efficient and accurate qualitative and quantitative analysis of samples through MOEMS precise spectroscopy, differential signal optimization and cross-spectral feature correlation analysis.

[0004] The technical solution to achieve the objective of this invention is as follows:

[0005] On the one hand, a smart detection method using Raman-near-infrared spectroscopy includes the following steps:

[0006] Acquire the Raman and near-infrared signals output by the optical sensing module;

[0007] The two electrical signals are dynamically amplified by dynamic gain adjustment, and filtered and separated by dual-band filtering to obtain pure Raman signal and pure near-infrared signal. The pure Raman signal and pure near-infrared signal are then synchronously converted by ADC and denoised by wavelet transform to obtain Raman digital signal and near-infrared digital signal.

[0008] Based on the principle of molecular vibration-rotation coupling, spatial correlation analysis is performed on the characteristic peaks of Raman digital signals and near-infrared digital signals to establish a mapping relationship of spatial correlation between Raman and near-infrared characteristic peaks. The sample component groups are determined by comparing Raman characteristic peaks. The component content is calculated by partial least squares method based on the mapping relationship correction. The characteristic peaks are expanded by the characteristic peak half-width expansion algorithm and sidelobe information is extracted to identify trace components. The qualitative component groups and qualitative component contents are confirmed iteratively.

[0009] Furthermore, a mapping relationship for the spatial correlation of Raman-near-infrared characteristic peaks is established, including:

[0010] Preprocessing is performed on time-synchronized Raman and near-infrared digital signals. For the Raman digital signal, an adaptive iterative reweighted penalized least squares method is used to subtract fluorescence background to eliminate baseline interference, and then a Gaussian function is used for smoothing to weaken the jagged peak distortion caused by random noise. For the near-infrared digital signal, a moving average method is used to correct the baseline, and a Savitzky-Golay filtering algorithm is used to shape the peak shape to avoid baseline tilting of characteristic peaks caused by low-frequency fluctuations. After preprocessing, characteristic peak parameters of the two types of spectra are extracted respectively. The wavenumber, peak intensity and full width at half maximum (FWHM) of the characteristic peaks are extracted from the Raman spectrum to form a Raman characteristic peak parameter set, and the wavelength, peak intensity and FWHM of the characteristic peaks are extracted from the near-infrared spectrum to form a near-infrared characteristic peak parameter set.

[0011] The correlation degree is calculated based on the principle of molecular vibration-rotation coupling. First, the conversion coefficient between the wavenumber of Raman characteristic peaks and the wavelength of near-infrared characteristic peaks is determined. Then, the spatial position matching term is calculated, and the relative intensity matching term is calculated. The two results are multiplied to obtain the correlation degree. Characteristic peak pairs that meet the correlation degree threshold are selected. The Raman characteristic peak wavenumber, near-infrared characteristic peak wavelength, corresponding molecular component identification, and correlation degree information of each pair of peaks are integrated to finally establish a Raman-near-infrared characteristic peak spatial correlation mapping table.

[0012] Furthermore, the dynamic gain adjustment amplifies the two electrical signals, including:

[0013] A dual-channel adaptive gain amplifier circuit was constructed. A programmable dynamic gain adjustment method was adopted to preset the gain adjustment range and step logic. For the transient and narrow pulse characteristics of Raman signals, a sliding time window method was used to monitor the amplitude. The window length was dynamically adjusted according to the pulse duration to avoid peak missed detection caused by too short a window and interference from multiple pulse amplitude superposition caused by too long a window. The pulse peak value and pulse effective value in each window were extracted simultaneously. For the steady-state and low fluctuation characteristics of near-infrared signals, a continuous sampling averaging algorithm was used to monitor the amplitude. Multiple sets of DC signal amplitudes collected within a unit time were accumulated and averaged to obtain the DC average value. At the same time, the maximum and minimum values ​​of the signal amplitude within the time period were extracted, and the difference between the two was calculated as the fluctuation amplitude.

[0014] Targeted amplitude ranges are set for the two signals respectively. The effective amplitude range of the Raman signal is determined based on the minimum effective input voltage and saturation voltage of the subsequent synchronous ADC conversion. The stable amplitude range of the near-infrared signal is determined based on the allowable fluctuation range of its steady-state characteristics. The gain is adjusted in fine steps. For the Raman signal, if the effective value of the pulse is detected to be lower than the lower limit or the peak value is higher than the upper limit, the gain is adjusted in steps until both fall into the range. For the near-infrared signal, if the mean value is lower than the lower limit or the fluctuation amplitude exceeds the preset fluctuation threshold, the mean value is stabilized within the range through gain feedback adjustment. Finally, a standard signal with the same characteristics as the two signals is used to input the circuit. The error between the amplified signal amplitude and the theoretical gain calculation value is compared to calibrate the amplification accuracy. The amplified Raman signal and near-infrared signal are then output.

[0015] Furthermore, filtering and separation are performed using a dual-band filtering method, including the following steps:

[0016] First, the frequency characteristics of the two amplified electrical signals are analyzed by fast Fourier transform. From the frequency spectrum of the Raman signal, the Raman characteristic frequency band is identified based on the molecular vibration characteristics. At the same time, the near-infrared crosstalk frequency band mixed in due to incomplete spectral dispersion is also identified.

[0017] From the frequency spectrum of the near-infrared signal, the near-infrared characteristic frequency band is identified based on the molecular rotation characteristics. At the same time, the mixed Raman crosstalk frequency band is identified. Differential filtering is performed on the two signals. For the Raman signal, notch processing logic based on a double quadratic filter structure is applied. The center frequency is set to the near-infrared crosstalk frequency band and the quality factor is set to a preset value. The crosstalk signal is suppressed. Then, a two-stage all-pass filter is cascaded to perform phase compensation to ensure that the phase of the filtered signal is consistent with the original signal, thus obtaining the preliminary filtered Raman signal.

[0018] For near-infrared signals, a notch filtering logic based on a Butterworth low-pass filter is applied. The cutoff frequency is set as the lower limit of the Raman crosstalk frequency band to suppress Raman crosstalk in a directional manner. At the same time, DC component compensation is performed. The original baseline of the signal before filtering is first sampled and calculated. After filtering, the offset baseline is calculated again by sampling. The baseline is returned to the original level by superimposing the compensation component to obtain the preliminary filtered near-infrared signal.

[0019] Calculate the crosstalk suppression ratio of the two signals. If it is lower than the preset crosstalk suppression threshold, then correct the notch center frequency and bandwidth and re-filter until the target is met. Finally, perform low-pass smoothing filtering on the two preliminary filtered signals to remove circuit thermal noise and high-frequency noise left over from FFT operation, and output a pure Raman signal and a pure near-infrared signal without crosstalk.

[0020] Furthermore, synchronous ADC conversion and wavelet noise reduction include:

[0021] After synchronous ADC conversion of the pure Raman signal and the pure near-infrared signal, the original Raman digital signal and the original near-infrared digital signal are generated and time synchronization calibration is performed.

[0022] For two types of synchronous digital signals, an appropriate wavelet basis is selected, the corresponding decomposition level is set, the noise standard deviation is calculated, and the coefficients of each high-frequency subband are processed by a soft threshold function. Inverse wavelet transform is performed to verify the effectiveness of the denoised signal, and the denoised Raman digital signal and near-infrared digital signal are output.

[0023] Further, trace component identification includes:

[0024] After initially determining the sample component groups and their contents, the sum of the contents of all components in the component content set is calculated, and this sum is compared with the preset component content threshold.

[0025] If the total content does not reach the threshold, trace component identification is carried out based on the principle of multi-component spectral superposition. For each characteristic peak in the Raman and near-infrared characteristic peak parameter sets, its full width at half maximum (FWHM) is expanded to a preset multiple of the original FWHM to form an expanded characteristic peak interval. The expanded characteristic peak interval is then scanned to extract the sidelobe signals within the interval. By calculating the signal-to-noise ratio (SNR) of the sidelobe signals, interference signals with SNRs lower than a preset value are eliminated to obtain a set of suspected trace component characteristic peaks. The set of suspected characteristic peaks is then compared with the standard Raman database and the standard near-infrared database to screen out candidate trace components with matching characteristic peak parameters. Finally, by combining the Raman-near-infrared characteristic peak spatial correlation mapping table, it is verified whether the associated characteristic peaks corresponding to the candidate trace components exist simultaneously in the Raman and near-infrared characteristic peak parameter sets. After verification, the type of trace component is determined.

[0026] Further, the calculation of component content includes:

[0027] Based on the Raman-near-infrared characteristic peak spatial correlation mapping table, the associated near-infrared characteristic peaks corresponding to each substance in the sample component group are extracted to form a subset of near-infrared characteristic peaks specific to each preliminary substance;

[0028] A partial least squares method was used to construct an initial correlation model between characteristic peak intensity and component content. Standard samples of the substance with known concentration gradients were selected, and near-infrared characteristic peak intensity data of the standard samples at each gradient were collected. An initial linear model was fitted with concentration as the dependent variable and intensity as the independent variable. The correlation degree was introduced as a correction coefficient to correct the initial correlation model, resulting in a corrected initial correlation model. The peak intensity of each characteristic peak in the near-infrared characteristic peak subset was substituted into the corrected initial correlation model to calculate the content of each preliminary substance, forming a component content set.

[0029] Further, time-synchronized sampling includes:

[0030] A dual-channel analog-to-digital converter is used, and a phase-locked loop circuit is used to achieve co-source locking with an external reference clock to generate a synchronous sampling clock with frequency accuracy that meets the preset requirements, ensuring the consistency of the time base for the conversion of the two signals. A pure Raman signal is input, and the start time of the Raman pulse is identified by a preset pulse rising edge threshold to generate a synchronous trigger signal, which serves as a unified start sampling command for the dual-channel analog-to-digital converter.

[0031] The sampling rate parameters are configured based on the characteristic differences between the two signals. For the narrow pulse and high transient characteristics of the pure Raman signal, the sampling rate is set as a multiple of the inverse of its pulse width. For the steady-state and low fluctuation characteristics of the pure near-infrared signal, its sampling rate is set to be an integer multiple of the Raman signal sampling rate. This ensures that the sampling points of the two signals have an integer correspondence on the time axis, which is convenient for subsequent time synchronization. When the synchronization trigger signal is generated, the dual-channel analog-to-digital converter starts sampling simultaneously. The first channel continuously samples the pure Raman signal at high speed to generate a raw Raman digital signal containing the absolute timestamp and quantization amplitude of each sampling point. The second channel synchronously samples the pure near-infrared signal to generate a raw near-infrared digital signal containing the corresponding absolute timestamp and quantization amplitude. Time synchronization calibration is performed on the two raw digital signals to ensure that the timestamp sequence is completely consistent. After wavelet transform denoising, the time-synchronized Raman digital signal and near-infrared digital signal are obtained.

[0032] Further, the preliminary determination of the sample composition group includes the following steps:

[0033] First, retrieve the standard Raman database, which contains the standard parameter set of characteristic peaks of known substances in the specific band of Raman spectrum. For each characteristic peak in the Raman characteristic peak parameter set, calculate its relative intensity and complete the normalization process of the Raman characteristic peak parameter set to obtain the normalized Raman characteristic peak parameter set.

[0034] For each substance in the standard Raman database, extract its characteristic peak standard wavenumber set, compare it with the wavenumber set of the normalized Raman characteristic peak parameter set, calculate the positional fit, count the number of characteristic peak pairs that satisfy the wavenumber deviation less than or equal to the preset wavenumber deviation threshold, divide the number of pairs by the maximum value of the sample Raman peak number and the standard Raman peak number to obtain the positional fit.

[0035] Extract the standard relative intensity set of the substance and compare it with the relative intensity set of the normalized Raman characteristic peak parameter set to calculate the relative intensity similarity. First, calculate the absolute deviation of the relative intensity of each pair of matching peaks, then calculate the average of all deviations. Subtract the average value from 1 to obtain the relative intensity similarity. Set the positional fit weight and the relative intensity similarity weight, and multiply the positional fit and the relative intensity similarity by their respective weights and sum them to obtain the comprehensive matching degree.

[0036] Substances with a comprehensive matching degree greater than or equal to a preset comprehensive matching degree threshold are selected and included in the sample candidate substance list. Based on the Raman-near-infrared characteristic peak spatial correlation mapping table, for each substance in the sample candidate substance list, it is verified whether its corresponding associated near-infrared characteristic peak exists in the near-infrared characteristic peak parameter set. If it exists, cross-validation is completed, and the substance is determined as the preliminary substance of the sample. Finally, all the validated substances are integrated to form the sample component group.

[0037] Secondly, a Raman-near-infrared spectroscopy combined intelligent sensor includes an optical sensing module, a dual-channel sensing unit, a signal processing unit, and a data analysis unit:

[0038] The optical sensing module separates the Raman-near-infrared mixed light according to wavelength dispersion using the MOEMS beam splitter, and the corresponding detectors convert it into Raman and near-infrared signals.

[0039] The dual-channel sensing unit acquires the Raman signal and near-infrared signal output by the optical sensing module;

[0040] The signal processing unit dynamically amplifies the two electrical signals using dynamic gain adjustment, filters and separates them using dual-band filtering to obtain pure Raman and pure near-infrared signals, performs synchronous ADC conversion on the pure Raman and pure near-infrared signals and uses wavelet transform for noise reduction to obtain Raman digital signals and near-infrared digital signals.

[0041] Based on the principle of molecular vibration-rotation coupling, the data analysis unit performs spatial correlation analysis on the characteristic peaks of Raman and near-infrared digital signals, establishes a mapping relationship of spatial correlation between Raman and near-infrared characteristic peaks, determines the sample component group based on Raman characteristic peak comparison, calculates the component content using partial least squares method based on mapping relationship correction, expands the characteristic peaks through characteristic peak half-width expansion algorithm and extracts sidelobe information to identify trace components, and iteratively confirms the qualitative component group and qualitative component content.

[0042] Compared with the prior art, the significant advantages of this invention are:

[0043] 1. By employing dynamic gain adjustment and dual-band filtering, the characteristics of the two types of signals are accurately matched, crosstalk is effectively suppressed, and a pure spectral signal without phase distortion and baseline shift is obtained.

[0044] 2. Based on the principle of molecular vibration-rotation coupling, a spatial correlation mapping of Raman-near-infrared characteristic peaks is established. Combined with cross-validation and modified partial least squares method, the accuracy of qualitative analysis and quantitative analysis of substances are greatly improved. Attached Figure Description

[0045] Figure 1 A flowchart of a Raman-near-infrared spectroscopy-coupled detection method;

[0046] Figure 2 This is a schematic diagram of the optical sensing module of the intelligent sensor in this invention;

[0047] Figure 3 This is a flowchart illustrating the filtering and separation process of Raman and near-infrared signals in this invention.

[0048] Figure 4 This is a flowchart of the identification and content analysis of the sample to be tested in this invention.

[0049] Explanation of reference numerals in the attached figures:

[0050] 1. Constant current pulse drive module; 2. Raman laser source; 3. Near-infrared source; 4. Fiber optic probe; 5. Sample to be tested; 6. Entrance slit; 7. Collimating mirror; 8. MOEMS beam splitter; 9. Near-infrared converging mirror; 10. Raman converging mirror; 11. Near-infrared exit slit; 12. Near-infrared high-pass filter; 13. Near-infrared detector; 14. Raman exit slit; 15. Raman filter; 16. Raman detector. Detailed Implementation

[0051] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0052] Example 1

[0053] like Figure 1As shown, this invention discloses a Raman-near-infrared spectroscopy coupled detection method, comprising the following steps:

[0054] Acquire Raman and near-infrared signals output by the optical sensing module;

[0055] The two electrical signals are dynamically amplified by dynamic gain adjustment and then filtered and separated by dual-band filtering to obtain a pure Raman signal and a pure near-infrared signal without crosstalk. The pure Raman signal and the pure near-infrared signal are synchronously converted by ADC and denoised by wavelet transform to obtain a time-synchronized Raman digital signal and a near-infrared digital signal.

[0056] Based on the principle of molecular vibration-rotation coupling, spatial correlation analysis is performed on the characteristic peaks of Raman and near-infrared digital signals to establish a mapping relationship between Raman and near-infrared characteristic peaks. The Raman digital signal is compared with a standard Raman database to identify characteristic peaks corresponding to molecular chemical bond vibrations within the dedicated band of the Raman spectrum. Cross-validation of the mapping relationship is performed, and the comprehensive matching degree is calculated by the similarity of characteristic peak positions and relative intensity to determine the sample composition group. The characteristic peaks corresponding to the sample components in the near-infrared digital signal are extracted. An initial correlation model between characteristic peak intensity and component content is established using the partial least squares method based on the mapping relationship correction. The content of each preliminary substance is calculated, and it is determined whether the sum of the contents of each component meets the component content threshold. If not, based on the principle of multi-component spectral superposition, the characteristic peak range is expanded by characteristic peak half-width extension and sidelobe information extraction algorithms. After supplementing the identification of trace component characteristic peaks, the content is recalculated until the threshold is met. Finally, the sample composition group is confirmed a second time by combining the spatial correlation mapping relationship between Raman and near-infrared characteristic peaks. The finally determined sample composition group and corresponding content are used as qualitative and quantitative results and transmitted to the display screen for visualization.

[0057] like Figure 2 As shown, further, in the Raman-near-infrared integrated smart sensor, the optical sensing module is the core beam-splitting component of the smart sensor. Through light source illumination, mixed light transmission, MOEMS dispersion, split detection, and signal separation, Raman and near-infrared signals are obtained and transmitted to the dual-channel sensing unit of the smart sensor, including the following steps:

[0058] When the sample 5 is being tested, the constant current pulse drive module 1 outputs a constant current pulse current, which drives the Raman laser source 2 to generate a narrow bandwidth pulse laser with the same frequency as the pulse current. The pulse laser is transmitted to the fiber optic probe 4 through the beam splitter fiber, and then irradiates the surface of the sample 5 in a directional manner, exciting the sample molecules to generate pulsed Raman scattered light in the characteristic wavelength range corresponding to Raman scattering. The scattered light has the same frequency as the laser pulse and belongs to the high frequency AC optical signal. At the same time, the near-infrared source 3 is activated and outputs continuous light. The continuous light is reflected by the sample surface to form continuous near-infrared reflected light, which belongs to the low frequency DC optical signal. The pulsed Raman scattered light and the continuous near-infrared reflected light are mixed on the sample surface and transmitted back to the beam splitter fiber through the fiber optic probe 4 to form Raman-near-infrared mixed light.

[0059] The split fiber transmits the returned Raman-near-infrared mixed light to the entrance slit 6 to filter the mixed light, resulting in narrow spot mixed light, while filtering out wide-band stray light. Subsequently, the collimating mirror 7 reflects the narrow spot mixed light into parallel mixed light and ensures that the parallel light is incident on the MOEMS beam splitter 8 in a center-aligned manner, thereby avoiding the dispersion shift problem caused by the divergence of the optical signal.

[0060] The grating of the MOEMS beam splitter 8 rotates at a preset frequency to disperse the incident parallel mixed light according to wavelength. Raman monochromatic light in the characteristic wavelength range corresponding to Raman scattering is dispersed to the optical path direction of Raman converging mirror 10 during the rotation of the grating, while near-infrared monochromatic light in the characteristic wavelength range corresponding to near-infrared is dispersed to the optical path direction of near-infrared converging mirror 9. The dispersion of the mixed light across the entire wavelength range can be achieved by the grating completing a single full-cycle rotation.

[0061] The Raman monochromatic light, dispersed by the MOEMS beam splitter 8, is directed to the Raman converging mirror 10, where it is focused into a concentrated spot and transmitted to the Raman exit slit 14 for sieving, resulting in pure Raman wavelength light. Subsequently, the Raman filter 15 filters out the laser background and near-infrared stray light, outputting pure Raman pulse light to the Raman detector 16, which converts the pure Raman pulse light into a high-frequency AC Raman signal. Similarly, the near-infrared monochromatic light, dispersed by the MOEMS beam splitter 8, is directed to the near-infrared converging mirror 9, where it is focused into a concentrated spot and transmitted to the near-infrared exit slit 11 for sieving, resulting in pure near-infrared wavelength light. Subsequently, the near-infrared high-pass filter 12 filters out Raman stray light, outputting pure near-infrared continuous light to the near-infrared detector 13, which converts the pure near-infrared continuous light into a low-frequency DC near-infrared signal. The Raman signal and the near-infrared signal are then input into the subsequent circuit system for detection.

[0062] like Figure 3As shown, further, Raman and near-infrared signals are acquired, dynamically amplified using dynamic gain adjustment, and then filtered and separated using a dual-band filtering method to obtain a clean signal without crosstalk. This process includes the following steps:

[0063] The signal output terminal of the MOEMS spectroscopic Raman-near-infrared spectroscopy coupling optical structure is connected to the output Raman signal. Near-infrared signals Each signal is connected to a pre-built dual-channel adaptive gain amplifier circuit. This circuit uses a programmable dynamic gain adjustment method and achieves adaptive matching of amplitude fluctuations of the two signals through a preset gain adjustment range and step logic.

[0064] For the accessed Raman signal Near-infrared signals Amplitude monitoring is performed based on the characteristics of the Raman signal. For the transient and narrow pulse characteristics of the Raman signal, a sliding time window method is used for amplitude monitoring. The window duration is dynamically adjusted according to the pulse duration, and the pulse peak value within each window is extracted. With pulse RMS value This approach avoids peak detection errors caused by excessively short windows and prevents interference from multiple pulse amplitude superposition caused by excessively long windows, achieving accurate capture of pulse signal amplitude changes. Targeting the steady-state and low-fluctuation characteristics of near-infrared signals, a continuous sampling averaging algorithm is used to calculate the DC mean and fluctuation amplitude. Multiple near-infrared signal amplitude data collected within a unit of time are accumulated, and the sum is divided by the number of data points to obtain the DC mean within that unit of time. Simultaneously, the maximum and minimum values ​​of the collected near-infrared signal amplitude data are extracted within a unit of time, and the difference between the two values ​​is the fluctuation amplitude. Through DC average and fluctuation amplitude Characterizes the amplitude reference and stability of the DC signal;

[0065] Effective amplitude ranges and stable amplitude ranges are set for Raman and near-infrared signals, respectively. The preset effective amplitude range for the Raman signal is... The lower limit The minimum effective input voltage corresponding to the subsequent digital signal conversion process, and the upper limit The voltage does not exceed the saturation voltage of the conversion process. By targeting the range, the signal integrity and conversion effectiveness are balanced. The preset stable amplitude range of the near-infrared signal is... The difference between its upper and lower limits matches the allowable fluctuation range of its DC steady-state characteristics, ensuring that the amplitude stability of the amplified signal meets the accuracy requirements of subsequent spectral characteristic peak extraction, and avoiding errors in component content calculation due to excessive fluctuations. In the dynamic gain adjustment logic, for Raman signals, if detected... The control gain is increased gradually in fine steps, and monitoring is performed again after each increase. and until and To avoid focusing solely on the effective value while ignoring waveform distortion caused by peak values, if detected... Then, the gain is gradually reduced in fine steps to ensure the integrity of the pulse waveform. For near-infrared signals, if detected... or ,in A preset fluctuation threshold is set, which is obtained based on the 3σ principle to statistically analyze the normal fluctuation range of a pure near-infrared signal. This threshold is then adjusted dynamically to achieve... Stable landing Simultaneously suppressing gain feedback This solves the problem that traditional fixed gain cannot cope with small fluctuations in DC signals;

[0066] When calibrating the amplification accuracy of electrical signals and outputting qualified signals, a standard pulse signal with the same characteristics as Raman signals and a known amplitude, and a standard DC signal with the same characteristics as near-infrared signals are selected and connected to the two channels of the dual-channel adaptive gain amplifier circuit, respectively. The amplitude of the amplified signal is acquired, and the result of multiplying the amplitude of the standard signal by the theoretical gain is compared with the result of multiplying the amplitude of the standard signal by the theoretical gain. The difference between the two is calculated to obtain the amplitude error of the two amplified signals. ,like ,in A preset error threshold is set based on ADC quantization accuracy and industry standards, representing the maximum amplification error that will not affect subsequent signal processing. The step accuracy and feedback coefficient are adjusted during gain adjustment, and the amplification process is re-executed. After confirming that the amplification accuracy meets the requirements, the amplified Raman signal is obtained. and amplified near-infrared signals ;

[0067] The amplified part was subjected to Fast Fourier Transform. and Frequency response analysis was performed to obtain the frequency spectra of the two signals. and ,from Identifying the characteristic frequency bands of Raman signals This frequency band is determined by the molecular vibrational characteristics of Raman spectroscopy, as well as the near-infrared crosstalk frequency band mixed in due to incomplete spectral dispersion caused by the optical structure. ,from Characteristic frequency bands for identifying near-infrared DC signals This frequency band is determined by the molecular rotational properties of the near-infrared spectrum, as well as the Raman crosstalk frequency band mixed in within it. By analyzing the frequency spectrum, precise frequency targets are provided for subsequent notch filtering, avoiding the blindness of traditional filtering that relies on experience to set frequency bands.

[0068] With near-infrared crosstalk frequency band Raman crosstalk frequency band As the target, the amplified Raman signal and amplified near-infrared signals Perform differential filtering and distortion compensation on the amplified Raman signal. When performing filtering, The input is based on the first notch processing logic of a double quadratic filter structure, and the center frequency is set to... Quality Factor The band-stop filter parameters are preset values ​​to suppress... To address the pulse signal phase distortion problem easily caused by traditional notch filtering for near-infrared crosstalk signals in the frequency band, a linear phase filter is used. A second-stage all-pass filter is cascaded after a double quadratic filter. Phase compensation through the all-pass network ensures that the phase response of the filtered signal is consistent with... Consistent, outputting the initial filtered Raman signal In the amplified near-infrared signal When performing filtering, Input the second notch filtering logic based on the Butterworth low-pass filter, and set the cutoff frequency to [value missing]. The lower limit, targeted inhibition To address Raman crosstalk in the frequency band and baseline shift that easily occurs after DC signal filtering, synchronous DC component compensation is performed. Before performing the filtering operation, the signal is continuously sampled, and the DC mean value during this period is calculated. This value is used as the original baseline. After the filtering operation is completed, the newly output preliminary filtered signal is immediately sampled for the same duration to calculate its DC mean value, thus obtaining the offset baseline. By calculating the difference between the original baseline and the offset baseline, the DC component that needs to be compensated is determined. This compensated component is then dynamically superimposed onto the filtered signal to bring the signal baseline back to its original level before filtering, and the preliminary filtered near-infrared signal is output. ;

[0069] To verify the filtering effect, a crosstalk suppression ratio (CSR) test was performed. First, the preliminary filtered Raman signal was calculated. Crosstalk suppression ratio The calculation formula is as follows:

[0070] ,

[0071] in, Near-infrared crosstalk frequency band before filtering amplitude, Near-infrared crosstalk frequency band after filtering The amplitude; calculate the preliminary filtered near-infrared signal. Crosstalk suppression ratio The calculation formula is as follows:

[0072] ,

[0073] in, Raman crosstalk frequency band before filtering amplitude, Raman crosstalk frequency band after filtering The amplitude, if or ,in, A preset crosstalk suppression threshold is set. This threshold is determined by combining the noise level of the detection system with filtering experiments to achieve the minimum suppression ratio without crosstalk interference. If crosstalk remains in the signal, the center frequency is corrected based on the frequency shift of the remaining crosstalk. , And adjust the corresponding notch bandwidth according to the residual amplitude. , And re-execute the filtering operation, so that and This ensures that crosstalk has been completely filtered out before proceeding to the subsequent signal purification stage;

[0074] To address the residual high-frequency noise that may be introduced during notch filtering, including circuit thermal noise and FFT operation noise, and Low-pass smoothing filtering is performed separately to remove residual high-frequency noise without affecting the amplitude and phase characteristics of the effective signal, ultimately obtaining a pure Raman signal without crosstalk, phase distortion, or baseline shift. and pure near-infrared signals .

[0075] Furthermore, the pure Raman signal and pure near-infrared signal are synchronously converted by ADC and denoised by wavelet transform to obtain time-synchronized Raman digital signal and near-infrared digital signal, including the following steps:

[0076] Synchronous analog-to-digital conversion is performed using a dual-channel analog-to-digital converter, coupled with a synchronous clock generation and trigger edge detection mechanism. The synchronous clock generation achieves co-location locking with an external reference clock via a phase-locked loop, generating an adjustable-frequency synchronous sampling clock CLK. Its frequency accuracy is controlled within a preset range to ensure time reference consistency between the two signal conversion paths. The trigger edge detection mechanism incorporates a pure Raman signal. The pulse start time is identified by a preset pulse rise edge threshold, and a synchronous trigger signal TRIG is generated as the start sampling command for dual-channel analog-to-digital conversion, thus solving the time alignment problem between pure Raman signals and pure near-infrared signals.

[0077] For pure Raman signals With pure near-infrared signals The features of this configuration allow for the allocation of core parameters for dual-channel analog-to-digital conversion. Because of its narrow pulse and high transient characteristics, its sampling rate is set. It is several times the reciprocal of the pulse width to ensure complete reconstruction of the pulse waveform and accuracy of amplitude quantization. Because it is a steady-state DC signal with a low-frequency fluctuation, its sampling rate is set accordingly. and Maintain an integer multiple relationship to ensure that the two sampling points have an integer correspondence on the time axis;

[0078] Pure Raman signal The first channel of the dual-channel analog-to-digital converter provides a pure near-infrared signal. The second channel is connected, and the CLK output of the synchronous clock drives the sampling clock terminals of both channels. When the edge detection mechanism is triggered and the TRIG signal is output, both channels start sampling simultaneously. The first channel... Continuous high-speed sampling is performed to generate a raw Raman digital signal containing complete time-domain characteristics of the pulse. Its data format is timestamp-amplitude sequence ,in The sampling point number, For that moment Quantization amplitude, The absolute time corresponding to this point is calculated using the following formula:

[0079] ,

[0080] in, To synchronize the timing of signal generation For pure Raman signal The sampling rate, the second channel synchronously... Sampling is performed to generate raw near-infrared digital signals. Its data format is timestamp-amplitude sequence ,in The sampling point number, For that moment Quantization amplitude, The absolute time corresponding to this point is calculated using the following formula:

[0081] ,

[0082] in, To synchronize the timing of signal generation For pure near-infrared signals The sampling rate is ensured through synchronous triggering of the TRIG signal. and The time deviation does not exceed 1 CLK cycle, achieving preliminary time alignment of the original digital signal;

[0083] Raman raw digital signal and near-infrared raw digital signals Perform time synchronization calibration, and construct a time mapping relationship based on the timestamp sequences of the two. Each sampling point in ,exist Search for timestamps satisfy sampling points ,in, A preset synchronization threshold is set to half the CLK period. If a unique matching point exists, then a synchronization is established. The correspondence is such that if multiple matching points exist, the one with the closest timestamp is selected. As a matching point, if no matching point exists, linear interpolation is used to... Supplementary sampling points are generated between adjacent sampling points to ensure... Each sampling point has a corresponding The sampling points ultimately yield a pair of synchronized digital signals that are strictly aligned with the time axis. ,in, For Raman synchronous digital signals, It is a near-infrared synchronous digital signal, and its timestamp sequence is completely identical;

[0084] For synchronous digital signal pairs Targeted wavelet transform denoising processes are performed separately for Raman synchronous digital signals. Because it contains high-frequency details such as steep pulse edges and characteristic peaks of molecular vibrations, the db6 wavelet basis is used, and the number of decomposition layers of the db6 wavelet basis is set. The highest effective frequency of the Raman signal The decision is made, and the calculation formula is as follows:

[0085] ,

[0086] in, For pure Raman signal The sampling rate is determined by this formula, which ensures that the highest frequency subband after decomposition mainly contains noise components, while the effective signal is concentrated in the low-frequency subband and the 1st to 3rd order high-frequency subbands. The coefficients of the highest frequency subband after decomposition are selected, and the noise standard deviation is calculated using the median absolute deviation method. ,right The high-frequency subband coefficients are processed using a soft thresholding function, and the threshold value is... The calculation formula is:

[0087] ,

[0088] in, for The number of sampling points, through statistics The total number of timestamp-amplitude sequences is obtained. The operation rule of the soft threshold function is as follows: for sub-band coefficients... When the absolute value of the coefficient When, output ,in, For symbolic functions, Take 1 at time. When -1 is taken, When the signal is zero, the output is 0 to suppress noise while preserving the abrupt change characteristics of the pulse edge. An inverse db6 wavelet transform is then performed on each sub-band coefficient after processing to obtain the denoised Raman digital signal. ;

[0089] For near-infrared synchronous digital signals Because it is a smooth DC signal (containing slowly varying fluctuations of component characteristic peaks), the sym4 wavelet basis is selected, and the number of decomposition levels is set. The highest effective frequency of near-infrared signals The decision is made, and the calculation formula is as follows:

[0090] ,

[0091] in, For pure near-infrared signals The sampling rate ensures that noise is mainly concentrated in... High-frequency subbands of the first order, selected after decomposition The coefficients of the high-frequency subbands are adopted using the same... The same MAD method is used to calculate the noise standard deviation. ,right The high-frequency subband coefficients are processed using a hard thresholding function, and the threshold value is... The calculation formula is:

[0092] ,

[0093] in, for The number of sampling points, the operation rule of the hard threshold function is, when the absolute value of the coefficient is... When, output ,when When the signal is zero, the output is 0 to remove minor baseline fluctuation noise while preserving the gradual variation trend of the component characteristic peaks. An inverse sym4 wavelet transform is then performed on each sub-band coefficient after processing to obtain the denoised near-infrared digital signal. ;

[0094] The effectiveness of the denoised signal is verified by calculating the denoised Raman digital signal. Raman synchronous digital signal before noise reduction Raman signal-to-noise ratio Raman signal-to-noise ratio The calculation formula is as follows:

[0095] ,

[0096] in, for , signal power for and The difference in power, required The baseline drift and the baseline drift respectively satisfy the preset Raman signal noise reduction threshold and drift threshold. Similarly, the calculation is performed. and signal-to-noise ratio and baseline drift, where baseline drift is the absolute difference between the mean baselines before and after filtering, and is required to be... The baseline drift amount meets the preset near-infrared signal denoising threshold and drift amount threshold, respectively. The Raman signal denoising threshold and near-infrared signal denoising threshold are determined based on the inherent characteristics of the signal and the accuracy requirements of subsequent qualitative and quantitative analysis. The drift amount threshold is calibrated by standard sample experiments and set with reference to industry specifications. If the above indicators are not met, the wavelet basis type, decomposition layer number or threshold function of the corresponding signal is adjusted, and the denoising process is re-executed until the requirements are met.

[0097] Those that pass the verification Raman digital signals defined as time synchronization Qualified Defined as time-synchronized near-infrared digital signal The timestamp sequences of the two are completely identical, providing a unified spectral data foundation for subsequent spatial correlation analysis of characteristic peaks.

[0098] like Figure 4As shown, further, based on the principle of molecular vibration-rotation coupling, a spatial correlation mapping relationship of Raman-near-infrared characteristic peaks is established. Component group identification and content calculation are completed sequentially. Trace components are supplemented through iterative optimization until the total content meets the threshold. Finally, qualitative and quantitative results are output for visualization. The process includes the following steps:

[0099] Construct a spatial correlation mapping relationship between Raman and near-infrared characteristic peaks for time-synchronized Raman digital signals. Near-infrared digital signals synchronized with time Baseline correction and peak normalization are performed on time-synchronized Raman digital signals. An adaptive iterative reweighted penalized least squares method is used to subtract the baseline, eliminating the interference of fluorescence background on the characteristic peaks. Then, a Gaussian function is applied to smooth the signal, weakening the jagged peak distortion caused by random noise. This method is suitable for time-synchronized near-infrared digital signals. The moving average method is used for baseline correction, and the peak shape is shaped by combining the Savitzky-Golay filtering algorithm to avoid the characteristic peak baseline tilt caused by low frequency fluctuations.

[0100] Preprocessed time-synchronized Raman digital signal The key parameters of all characteristic peaks within their specific bands are extracted to form a Raman characteristic peak parameter set. Its expression is as follows:

[0101] ,

[0102] in, For the first The wavenumber of each Raman characteristic peak, , The total number of Raman characteristic peaks within the band. For the first The peak intensity of each Raman characteristic peak, For the first The full width at half maximum (FWHM) of each Raman characteristic peak; for the preprocessed time-synchronized near-infrared digital signal. Extract the key parameters of its characteristic peaks to form a near-infrared characteristic peak parameter set. Its expression is as follows:

[0103] ,

[0104] in, For the first The wavelength of the near-infrared characteristic peak , This represents the total number of near-infrared characteristic peaks within the band. The peak intensity of this characteristic peak. The full width at half maximum (FWHM) of this characteristic peak;

[0105] Based on the principle of molecular vibration-rotation coupling, the vibrational and rotational motions of the same molecule exhibit an energy coupling effect, resulting in a fixed physical correlation between their corresponding Raman and near-infrared characteristic peaks. Using this correlation as the core, the degree of correlation between the Raman and near-infrared characteristic peaks is calculated. The calculation formula is as follows:

[0106] ,

[0107] in, The conversion factor between the wavenumber of Raman characteristic peaks and the wavelength of near-infrared characteristic peaks is determined by fundamental constants of molecular spectroscopy. It is used to unify the metrological scales of Raman characteristic peaks and near-infrared characteristic peaks, ensuring their comparability in spectral position. For the first The wavenumber of each Raman characteristic peak, The peak intensity of the Raman characteristic peak. For the first The wavelength of the near-infrared characteristic peak The peak intensity of the near-infrared characteristic peak. Raman characteristic peak parameter set The mean of all peak intensities, Near-infrared characteristic peak parameter set The mean of all peak intensities, the first term in the formula This is the spatial location matching term for the correlation degree. This value decreases as the location deviation increases and increases as the deviation decreases. It is used to quantify the correlation contribution of two characteristic peaks in spectral position. Adding 1 to the denominator avoids a denominator of 0 when the location deviation is 0, ensuring the mathematical validity of the formula. The second term in the formula... The relative strength matching term for correlation degree, based on the exponential function property, has a maximum value of 1, and this value decreases as the strength deviation increases. It is used to quantify the correlation contribution of two characteristic peaks in the relative strength distribution. By multiplying the spatial location matching term and the relative intensity matching term, the correlation between Raman characteristic peaks and near-infrared characteristic peaks is comprehensively quantified, providing a quantitative basis for cross-spectral characteristic peak matching of the same molecular component, and setting a correlation threshold. This threshold is verified and calibrated using multiple sets of standard samples to ensure the selection of transspectral characteristic peak pairs corresponding to the same molecule. At that time, the judgment of the first The first Raman characteristic peak and the second If each near-infrared characteristic peak belongs to the same molecular component, a correlation table is established by integrating all characteristic peak pairs that meet the correlation conditions to create a spatial correlation table for Raman-near-infrared characteristic peaks. The table contains the wavenumber of Raman characteristic peaks, the wavelength of near-infrared characteristic peaks, the corresponding molecular component identifiers, and the correlation degree. This mapping table provides a basis for subsequent qualitative cross-validation of components and correction of the content calculation model.

[0108] The composition of the sample was determined by Raman spectroscopy comparison. A standard Raman database was retrieved, which includes a set of standard parameters for characteristic peaks of various substances in their respective Raman spectral bands, covering characteristic peak wavenumbers, relative intensity percentages, and full width at half maximum (FWHM) information. This database was used to compare the time-synchronized Raman digital signals. The corresponding Raman characteristic peak parameter set Normalization is performed, and the relative intensity of each characteristic peak is calculated. ,in, The peak intensity of the Raman characteristic peak. for The highest peak intensity is obtained to obtain the normalized Raman characteristic peak parameter set. For each substance in the standard Raman database, extract its standard characteristic peak wavenumber set. , Given the total number of characteristic peaks of the standard substance, calculate the sum of this set and... Locational fit of the wavenumber set of characteristic peaks The calculation formula is as follows:

[0109] ,

[0110] in, To meet The number of characteristic peaks, where For the target reference material in the standard Raman database The wavenumber of each Raman characteristic peak, , The preset wavenumber deviation threshold is determined by comparison with standard samples, based on the inherent accuracy of the Raman spectrometer. Indicates taking and The larger value in the formula is used as the denominator in the positional agreement calculation to avoid distortion of the agreement results due to differences in the number of characteristic peaks between the sample and the standard material, ensuring the objectivity and comparability of the quantitative indicators, and extracting the relative intensity set of the characteristic peaks of the standard material. ,and The relative intensity sets in the data are compared to calculate the relative intensity similarity. The calculation formula is as follows:

[0111] ,

[0112] in, for The relative intensity of the matching feature peaks, For the relative intensities of the corresponding matching characteristic peaks in the set of relative intensities of the standard substance characteristic peaks, a positional fit weight is set. and relative strength similarity weight ,and Calculate the overall matching degree The calculation formula is as follows:

[0113] ,

[0114] Among them, the positional fit weight and relative strength similarity weight The initial range of values ​​was determined based on the substance specificity and experimental condition sensitivity of Raman characteristic peak positions, followed by batch verification and calibration of the optimal combination using standard material samples. This was then dynamically adjusted and verified using multi-component sample variance analysis, and a comprehensive matching threshold was set. When a certain substance When this happens, it is added to the sample candidate substance list. For substances in the sample candidate substance list, a mapping relationship table is used. To verify whether the associated near-infrared characteristic peaks corresponding to the candidate substances exist in If the substance is found, cross-validation is performed, and the substance is identified as a preliminary component of the sample. Finally, all substances that have passed cross-validation are integrated to form the sample composition group. ;

[0115] The partial least squares method based on mapping relationship correction is used to calculate the component content for sample component groups. The Middle Preliminary substances , ,in, The total amount of all preliminary substances, based on the mapping table. From the near-infrared characteristic peak parameter set Extract the associated near-infrared characteristic peaks corresponding to the substance to form a subset of near-infrared characteristic peaks specific to each substance. A partial least squares method was used to construct an initial correlation model between the characteristic peak intensity and the component content, and a known concentration gradient was selected. For various material standard samples, near-infrared characteristic peak intensity data were collected, and an initial correlation model was established. To compensate for the systematic error of quantitative analysis using single near-infrared spectroscopy, a mapping relationship table was used. correlation in The initial correlation model is revised by introducing a correction coefficient. The corrected model expression is:

[0116] ,

[0117] Among them, the correction coefficient , For ingredient content, The characteristic peak intensity, coefficient and constant term This method, using partial least squares (PLS) based on the near-infrared characteristic peak intensities of multiple sets of standard samples with known content, involves data preprocessing followed by fitting a linear model to minimize the error between the predicted content and the actual content of the standard samples. Each preliminary substance... Corresponding near-infrared feature peak subset Substituting the peak intensity into the corrected initial correlation model, the content of each preliminary substance was calculated. , forming a set of component contents ;

[0118] Calculate the set of component contents The sum of the contents of all components ,Will With component content threshold Perform a comparison, if If the preliminary content calculation result is qualified, then the result is deemed acceptable. If the sample contains unidentified trace components, a trace component supplementation identification process is required. Based on the principle of multi-component spectral superposition, the characteristic peaks of trace components are easily masked by the characteristic peaks of principal components. Therefore, a characteristic peak half-width at half-maximum (FWHM) expansion algorithm is used to... and For each characteristic peak, its full width at half maximum (FWHM) is expanded to a preset multiple of the original FWHM to form an expanded characteristic peak range. ,in, The preset expansion coefficient, The wavenumber or wavelength at the center of the characteristic peak. Using the original half-width at half-maximum (HWHM), a signal scan is performed on the expanded characteristic peak region to extract sidelobe signals within the region. These sidelobe signals are characteristic peak signals of trace components. The extracted sidelobe signals are then filtered by signal-to-noise ratio (SNR) to remove interference signals with SNR below a preset value, resulting in a set of suspected trace component characteristic peaks. This set is then compared with a standard Raman database and a standard near-infrared database, and combined with a mapping table. Verification is performed to determine the types of trace components. For the identified trace components, the model construction and content calculation process is repeated to obtain the content of the trace components. The trace component content is then added to the component content set and updated to obtain a new total content. Repeat until ;

[0119] The sample composition was reconfirmed and the results were visualized. Preliminary substances and supplementary identified trace components were integrated to form a complete candidate substance set, based on a mapping table. The correlation between the Raman characteristic peaks and near-infrared characteristic peaks of each substance in the set was verified one by one. After confirming that there were no contradictory correlations, the substance was confirmed as an actual component in the sample, thus forming a qualitative component group that characterizes the final set of sample components. and qualitative component content ,right and The data is formatted to generate a standardized data report containing the substance name, chemical identifier, corresponding characteristic peak parameters, content values, and measurement confidence level. The measurement confidence level is determined by a comprehensive matching degree. and correlation The weighted calculations yielded a standardized data report, which was then transmitted to a display terminal for visualization.

[0120] Example 2

[0121] This invention discloses a Raman-near-infrared spectroscopy combined smart sensor, comprising an optical sensing module, a dual-channel sensing unit, a signal processing unit, and a data analysis unit:

[0122] The optical sensing module separates the Raman-near-infrared mixed light according to wavelength dispersion using the MOEMS beam splitter, and the corresponding detectors convert it into Raman and near-infrared signals.

[0123] The dual-channel sensing unit acquires the Raman signal and near-infrared signal output by the optical sensing module;

[0124] The signal processing unit uses dynamic gain adjustment to dynamically amplify the two electrical signals, and then uses dual-band filtering to separate them, obtaining a pure Raman signal and a pure near-infrared signal without crosstalk. The pure Raman signal and the pure near-infrared signal are then converted by synchronous ADC and denoised by wavelet transform to obtain a time-synchronized Raman digital signal and a near-infrared digital signal.

[0125] Based on the principle of molecular vibration-rotation coupling, the data analysis unit performs spatial correlation analysis on the characteristic peaks of Raman and near-infrared digital signals, establishing a mapping relationship for the spatial correlation of Raman-near-infrared characteristic peaks. The Raman digital signal is compared with a standard Raman database, identifying characteristic peaks corresponding to molecular chemical bond vibrations within the dedicated bands of the Raman spectrum. Cross-validation of the mapping relationship is performed, and a comprehensive matching degree is calculated based on the similarity of characteristic peak positions and relative intensity to determine the sample composition group. Characteristic peaks corresponding to sample components are extracted from the near-infrared digital signal. An initial correlation model between characteristic peak intensity and component content is established using a partial least squares method based on mapping relationship correction. The content of each preliminary substance is calculated, and it is determined whether the sum of the component contents meets the component content threshold. If not, based on the principle of multi-component spectral superposition, the characteristic peak range is expanded using characteristic peak half-width extension and sidelobe information extraction algorithms. Trace component characteristic peaks are identified, and the content is recalculated until the threshold is met. Finally, the sample composition group is reconfirmed based on the Raman-near-infrared characteristic peak spatial correlation mapping relationship. The final determined sample composition group and corresponding content are presented as qualitative and quantitative results and displayed on a screen for visualization.

[0126] This invention discloses a Raman-near-infrared spectroscopy combined intelligent sensor and detection method. The overall process revolves around light source excitation, mixed light dispersion, signal optimization processing, and cross-spectral qualitative and quantitative analysis. First, a constant current pulse drives a Raman laser source to generate a narrow-bandwidth pulsed laser at the same frequency, while a near-infrared source outputs continuous light, creating a differentiated excitation mode. Combined with a MOEMS dispersion component, precise dispersion of the Raman-near-infrared mixed light according to wavelength is achieved. Then, a dynamic gain adjustment method is used to dynamically match signal characteristics. Combined with dual-band notch filtering based on FFT to accurately locate crosstalk frequencies, and targeted noise reduction using db6 and sym4 wavelet bases, crosstalk-free and distortion-free signal optimization is achieved. Finally, a spatial correlation mapping of Raman-near-infrared characteristic peaks is established based on the molecular vibration-rotation coupling principle. The correlation degree of cross-spectral characteristic peaks is quantified to achieve qualitative cross-validation. Preliminary qualitative analysis is achieved by Raman spectral comparison, and precise quantitative analysis is achieved by partial least squares method with correlation degree as correction coefficient. The algorithm for characteristic peak half-width expansion and sidelobe information extraction is used to identify trace components, ultimately achieving high-precision detection of sample component groups and contents. This method specifically addresses the pain points of traditional combined solutions, such as poor spectroscopic flexibility, difficulty in suppressing signal crosstalk, and low qualitative and quantitative accuracy, providing an efficient and reliable technical solution for the analysis of material composition in multiple fields.

[0127] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A Raman-near-infrared spectroscopy coupled detection method, characterized in that, Includes the following steps: Acquire the Raman and near-infrared signals output by the optical sensing module; The two electrical signals are dynamically amplified by dynamic gain adjustment, and filtered and separated by dual-band filtering to obtain pure Raman signal and pure near-infrared signal. The pure Raman signal and pure near-infrared signal are then synchronously converted by ADC and denoised by wavelet transform to obtain Raman digital signal and near-infrared digital signal. Based on the principle of molecular vibration-rotation coupling, spatial correlation analysis is performed on the characteristic peaks of Raman digital signals and near-infrared digital signals to establish a mapping relationship of spatial correlation between Raman and near-infrared characteristic peaks. The sample component group is determined by comparing Raman characteristic peaks. The component content is calculated by partial least squares method based on mapping relationship correction. The characteristic peaks are expanded by characteristic peak half-width extension algorithm and sidelobe information is extracted to identify trace components. The qualitative component group and qualitative component content are confirmed iteratively.

2. The Raman-near-infrared spectroscopy detection method as described in claim 1, characterized in that, Establish a spatial correlation mapping relationship between Raman and near-infrared characteristic peaks, including: Baseline correction and peak shape normalization are performed on the time-synchronized Raman digital signal and near-infrared digital signal, respectively, and Raman characteristic peak parameter set and near-infrared characteristic peak parameter set are extracted. Based on the principle of molecular vibration-rotation coupling, the spatial position matching term and relative intensity matching term of Raman characteristic peaks and near-infrared characteristic peaks are calculated and multiplied to obtain the correlation degree. Characteristic peak pairs that meet the correlation degree threshold are selected and the information of correlated characteristic peaks is integrated to establish a spatial correlation mapping table of Raman-near-infrared characteristic peaks.

3. The Raman-near-infrared spectroscopy detection method as described in claim 1, characterized in that, Dynamic gain adjustment amplifies two electrical signals, including: A dual-channel adaptive gain amplifier circuit was constructed. For Raman signals, the sliding time window method was used to extract the pulse peak value and pulse effective value. For near-infrared signals, the continuous sampling mean algorithm was used to calculate the DC mean and fluctuation amplitude. Set the amplitude range for the two signals, adjust the gain in fine steps and calibrate the amplification accuracy, and output the amplified Raman signal and near-infrared signal.

4. The Raman-near-infrared spectroscopy detection method as described in claim 1, characterized in that, Filtering and separation are performed using a dual-band filtering method, including: The amplified electrical signal was analyzed for frequency characteristics using Fast Fourier Transform to identify the characteristic frequency bands and crosstalk frequency bands of Raman and near-infrared signals. For the Raman signal, a double quadratic notch filter and a cascaded all-pass filter are used for phase compensation. For the near-infrared signal, a Butterworth low-pass filter is used and DC component compensation is performed. Low-pass smoothing filters are applied to both signals to output a pure Raman signal and a pure near-infrared signal without crosstalk.

5. The Raman-near-infrared spectroscopy detection method as described in claim 1, characterized in that, Synchronous ADC conversion and wavelet noise reduction, including: After synchronous ADC conversion of the pure Raman signal and the pure near-infrared signal using a dual-channel analog-to-digital converter, time-synchronized Raman raw digital signal and near-infrared raw digital signal are generated. Based on the two types of digital signals, the appropriate wavelet basis and decomposition level are selected, the noise standard deviation is calculated, and the high-frequency subband coefficients of each order are processed by the soft threshold function. The inverse wavelet transform is performed, and the denoised Raman digital signal and near-infrared digital signal are output.

6. The Raman-near-infrared spectroscopy detection method as described in claim 1, characterized in that, Trace component identification, including: After initially determining the sample composition groups and their contents, the total content of each component is calculated and compared with the component content threshold. Based on the principle of multi-component spectral superposition, the full width at half maximum (FWHM) of Raman and near-infrared characteristic peaks was extended. Sidelobe signals were extracted by scanning the extended range. The types of trace components were determined by comparison with a standard Raman database and verification by spatial correlation mapping of characteristic peaks.

7. A Raman-near-infrared spectroscopy detection method as described in claim 1 or 2, characterized in that, Component content calculation includes: Based on the Raman-near-infrared characteristic peak spatial correlation mapping table, the associated near-infrared characteristic peaks corresponding to the preliminary substances are extracted to form a subset of near-infrared characteristic peaks for each preliminary substance. A partial least squares method was used to construct an initial correlation model between the intensity of characteristic peaks and the content of components. The correlation degree was introduced as a correction coefficient to correct the initial correlation model. The peak intensity of the near-infrared characteristic peak subset was substituted into the corrected initial correlation model to calculate the content of each preliminary substance.

8. The Raman-near-infrared spectroscopy detection method as described in claim 5, characterized in that, Time synchronization of Raman raw digital signals and near-infrared raw digital signals, including: A dual-channel analog-to-digital converter and a phase-locked loop are used to generate a synchronous sampling clock, which detects the rising edge of the Raman signal pulse and generates a synchronous trigger signal. The sampling rate is configured based on the characteristics of the two signals, and pure Raman signals and pure near-infrared signals are acquired simultaneously to generate raw Raman digital signals and raw near-infrared digital signals containing timestamps and amplitude sequences.

9. The Raman-near-infrared spectroscopy detection method as described in claim 6, characterized in that, The sample composition groups have been preliminarily determined, including: The standard Raman database is retrieved and the Raman characteristic peak parameter set is normalized. The positional similarity and relative intensity similarity between the Raman characteristic peak and the standard Raman characteristic peak are calculated and weighted to obtain the comprehensive matching degree. Candidate substances whose comprehensive matching degree meets the matching degree threshold are screened. By combining the spatial correlation mapping table of Raman-near-infrared characteristic peaks, the existence of the associated near-infrared characteristic peaks corresponding to the candidate substances is verified, and the sample composition group is determined.

10. A Raman-near-infrared spectroscopy coupled intelligent sensor, used to implement the Raman-near-infrared spectroscopy coupled detection method of any one of claims 1-9, characterized in that, It includes an optical sensing module, a dual-channel sensing unit, a signal processing unit, and a data analysis unit: The optical sensing module separates the Raman-near-infrared mixed light according to wavelength dispersion using the MOEMS beam splitter, and the corresponding detectors convert it into Raman and near-infrared signals. The dual-channel sensing unit acquires the Raman signal and near-infrared signal output by the optical sensing module; The signal processing unit dynamically amplifies the two electrical signals using dynamic gain adjustment, filters and separates them using dual-band filtering to obtain pure Raman and pure near-infrared signals, performs synchronous ADC conversion on the pure Raman and pure near-infrared signals and uses wavelet transform for noise reduction to obtain Raman digital signals and near-infrared digital signals. Based on the principle of molecular vibration-rotation coupling, the data analysis unit performs spatial correlation analysis on the characteristic peaks of Raman and near-infrared digital signals, establishes a mapping relationship of spatial correlation between Raman and near-infrared characteristic peaks, determines the sample component group based on Raman characteristic peak comparison, calculates the component content using partial least squares method based on mapping relationship correction, expands the characteristic peaks through characteristic peak half-width expansion algorithm and extracts sidelobe information to identify trace components, and iteratively confirms the qualitative component group and qualitative component content.

Images