Gas concentration fast inversion method and system based on space optical convolution modulation

Abstract

The invention discloses a gas concentration fast inversion method and system based on space optical convolution modulation, and the method comprises the steps: determining a target gas and an absorption wave band thereof, designing a convolution kernel in advance, and deploying the convolution kernel to a gas concentration inversion device; the method comprises the following steps: sequentially filling target gases with different concentration typical values into a gas concentration inversion device, and obtaining convolution processing curves corresponding to standard concentrations through optical convolution processing, optical signal convergence, photoelectric conversion and electric signal operation; based on each standard concentration and the corresponding convolution processing curve, extracting an absorption characteristic, fitting a corresponding relation between the gas concentration and the absorption characteristic by using a least square method, and establishing a gas concentration inversion model; introducing to-be-calibrated gas into the gas concentration inversion device to obtain corresponding absorption characteristics, and calculating the concentration of the to-be-calibrated gas in combination with the gas concentration inversion model. According to the method, the thought of mask modulation in optical calculation is combined, time division multiplexing and dynamic caching technologies are utilized, and the response speed of the system is increased while the calculation precision is ensured.

Description

Technical Field

[0001] This invention relates to the field of spectroscopic analysis technology, and in particular to a method and system for rapid inversion of gas concentration based on spatial optical convolution modulation. Background Technology

[0002] Tunable semiconductor laser absorption spectroscopy (TDLAS) utilizes the wavelength scanning and current tuning characteristics of a laser diode. By changing the driving current, the wavelength of the output laser is controlled, thereby scanning the gas absorption spectrum. Analysis of the absorption spectrum yields gas parameter information, and the gas concentration is retrieved using the harmonic peaks after demodulation of the absorption spectrum. Traditional TDLAS methods directly utilize the peak and line shape characteristics of the absorption spectrum to retrieve gas concentration, but they are easily affected by background noise, baseline drift, and light intensity fluctuations in complex environments, limiting detection accuracy and stability.

[0003] To overcome these limitations, Spatial Optical Convolution Modulation (SOCM) emerged. This technique spatially filters the original absorption spectrum by introducing a specific convolution kernel function into the spectral spatial domain, effectively suppressing broadband background interference and enhancing the characteristics of the target absorption lines. SOCM utilizes the spatial coherence of absorption lines to extract weak absorption signals from complex backgrounds through multi-point weighted convolution operations, significantly improving the signal-to-noise ratio and detection sensitivity. However, as spectral analysis models become increasingly larger and the amount of data to be processed increases, traditional electronic computing has encountered bottlenecks. Traditional methods require acquiring all the data, fitting it, and then using a baseline-removing formula to calculate the processed curve. Summary of the Invention

[0004] The main objective of this invention is to overcome the aforementioned shortcomings of existing methods and propose a fast gas concentration inversion method and system based on spatial optical convolution modulation. Based on a photoelectric hybrid architecture, the convolution operation of the entire absorption spectrum is completed at high speed with extremely low latency, and most of the calculation process is completed in parallel, which significantly improves the calculation speed.

[0005] The present invention adopts the following technical solution:

[0006] A fast gas concentration inversion method based on spatial optical convolution modulation includes:

[0007] The target gas and its absorption band are determined, the convolution kernel is pre-designed, and the convolution kernel is deployed to the gas concentration inversion device;

[0008] The target gas with different typical concentrations is sequentially filled into the gas concentration inversion device, and the convolution processing curve corresponding to each standard concentration is obtained through optical convolution processing, optical signal convergence, photoelectric conversion and electrical signal calculation.

[0009] Based on the standard concentrations and corresponding convolutional processing curves, absorption features are extracted, and the least squares method is used to fit the correspondence between gas concentration and the absorption features to establish a gas concentration inversion model.

[0010] The gas to be calibrated is introduced into the gas concentration inversion device to obtain its corresponding absorption characteristics. Combined with the gas concentration inversion model, the concentration of the gas to be calibrated is calculated.

[0011] The convolution kernel is a one-dimensional tensor convolution kernel of length 2k+1, where k is an integer greater than 0; the convolution kernel uses a one-dimensional valid convolution method to extract information about the absorption signal in the depression on the ramp signal, and the weight w of the convolution kernel... i The sum satisfies as well as This makes the convolution kernel respond zero to linear ramp signals and significantly to nonlinear absorption signals from depressions, wherein... For discrete signal sequence indexing; when using the one-dimensional valid convolution method, starting from the acquisition of the 2k+1th signal, the convolution kernel obtains a convolutional value for each time it moves one grid.

[0012] The convolution kernel is designed with an odd number of columns, which has symmetry and a clear center point, which is the (k+1)th column, to achieve complete sampling of signal features and cover preamble information, subsequent information and center response.

[0013] The convolution kernel is deployed on the digital micromirror device of the gas concentration inversion device; the digital micromirror device acts as a spatial light modulator, controlling the reflectivity of the device by changing the number of open reflector mirrors, so as to modulate the intensity of reflected light and perform physical multiplication of light intensity × weight.

[0014] The specific process of the optical convolution processing is as follows: the change frequency of the digital micromirror device is set to 2k+1 times the macroscopic sampling frequency; the modulated image of the digital micromirror device at each transformation time corresponds to the weight of each column of the convolution kernel, and each column of the weight matrix is ​​mapped to the reflectivity of the digital micromirror device at 2k+1 transformation times in a proportional manner.

[0015] The specific process of optical signal convergence and photoelectric conversion is as follows: the optical signal modulated by the digital micromirror device is converged by a convex lens, and a clock signal synchronized with the conversion frequency of the digital micromirror device is set to receive the converged optical signal and convert it into an electrical signal. The receiving frequency is 2k+1 times the macroscopic sampling frequency, and the electrical signal is then processed using the macroscopic sampling frequency.

[0016] A sliding window caching mechanism is used to store the electrical signal in a buffer and perform operations on the electrical signal: several buffers are configured, the number of which corresponds to the number of tensors in the convolution kernel; let t n Representing the nth macroscopic moment, the data Buffer_n[j] at the jth position of the nth buffer satisfies:

[0017] Buffer_n[j] is the j-th weighted optical intensity electrical signal at the (n+j-1)-th macroscopic moment; where j=1,2,...,2k+1;

[0018] When any buffer is full of 2k+1 data, the convolution operation is automatically triggered. After the operation is completed, the memory of the buffer is cleared in order to receive new weighted light intensity electrical signals.

[0019] When the electrical signal is processed, a time-division multiplexing strategy is adopted, so that the spectral data obtained from a single measurement can be used multiple times. The spectral data at the same moment participates in multiple convolution calculations simultaneously, and the spatial convolution is unfolded through the time dimension.

[0020] The absorption feature is the absorption peak or peak-to-peak value of the convolutional curve; the linear correspondence between the gas concentration and the absorption feature is obtained by fitting with the least squares method, and the gas concentration inversion model is the linear correspondence model.

[0021] A fast gas concentration inversion system based on spatial optical convolution modulation includes:

[0022] The gas concentration inversion device is pre-deployed with convolution kernels, which are used to perform optical convolution processing, optical signal convergence, photoelectric conversion and electrical signal calculation on target gases with different typical concentrations that are sequentially filled in, to obtain the convolution processing curves corresponding to each standard concentration.

[0023] The gas concentration inversion model establishment device extracts absorption features based on standard concentrations and corresponding convolution processing curves, and uses the least squares method to fit the correspondence between gas concentration and the absorption features to establish a gas concentration inversion model.

[0024] The gas calibration device introduces the gas to be calibrated into the gas concentration inversion device, obtains its corresponding absorption characteristics, and calculates the concentration of the gas to be calibrated by combining it with the gas concentration inversion model.

[0025] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:

[0026] This invention achieves true light-speed processing through parallel convolution operations at the optical level. Unlike traditional serial electronic computing, optical convolution can complete parallel operations on multiple weights instantaneously during light propagation, resulting in a significantly faster detection speed. This optical parallel processing architecture fundamentally breaks through the speed bottleneck of electronic processing, providing a completely new technical path for rapid response to gas concentration information.

[0027] 2. This invention employs programmable optical modulation technology, enabling real-time adjustment of the shape, scale, and weighting coefficients of the convolution template to achieve dynamic adaptation to different gas absorption line shapes. Compared to traditional fixed electronic filters, optical convolution templates offer extremely high flexibility and reconfigurability, allowing for template switching in a very short time. This provides strong technical support for the simultaneous detection of multi-component gases and adaptive analysis in complex spectral environments.

[0028] 3. In this invention, the reflected light intensity is modulated multiple times simultaneously at the same macroscopic moment using a DMD within the optical domain, and then stored in a buffer to complete the weighted summation of the convolution weights. This avoids the complex process of first acquiring the original data and then performing digital convolution operations in traditional methods. After the optical signal is modulated by the DMD, the detector receives the result signal that has already undergone optical convolution processing, reducing the workload of subsequent digital signal processing steps, lowering the latency introduced by data transmission, storage, and computation, and realizing rapid inversion from optical signal to concentration result.

[0029] 4. In this invention, the parallel control capability of the mega-mirror array inherent in the DMD device is utilized to simplify the parallel processing that originally required complex circuits and multi-channel synchronous acquisition systems into spatial light modulation using a single DMD device. This greatly simplifies the system hardware structure, avoids crosstalk and synchronization errors between multi-channel circuits, significantly improves system reliability and stability, reduces manufacturing costs and maintenance difficulty, and facilitates industrial application. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the specific process of the method of the present invention;

[0031] Figure 2 A schematic diagram illustrating the buffer storage;

[0032] Figure 3 The results are based on simulations using the HITRAN database.

[0033] Figure 4 The actual result obtained after processing by the system of this invention;

[0034] Figure 5 A comparison chart of the three peak characteristics and their concentration relationships;

[0035] Figure 6 This is a diagram showing the composition of the gas concentration inversion device of the present invention;

[0036] Figure 7 A table comparing the three peak characteristics and their concentration relationships.

[0037] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Detailed Implementation

[0038] The present invention will be further described below through specific embodiments.

[0039] See Figure 1 A fast gas concentration inversion method based on spatial optical convolution modulation, comprising:

[0040] S1 determines the target gas and its absorption band, pre-designs the convolution kernel, and deploys the convolution kernel to the gas concentration inversion device.

[0041] In this step, the convolution kernel is a one-dimensional tensor convolution kernel of length 2k+1, where k is an integer greater than 0; the convolution kernel simulates the weights in a neural network, and uses a one-dimensional valid convolution method to extract information about the absorption signal in the depression on the ramp signal, and the weights w of the convolution kernel... i The sum satisfies as well as This makes the convolution kernel adaptable to linear ramp signals. The response is zero, and the response to the nonlinear absorption signal of the depression is significant, among which... Here, a and b are the indices of the discrete signal sequence, and a and b are constants. , The signal is a ramp signal. When using the one-dimensional valid convolution method, the convolution kernel performs convolution along the signal with a length of 2k+1. Starting from the acquisition of the 2k+1th signal, the convolution kernel obtains a convolutional value for each grid movement.

[0042] The design concept of the weight matrix (optical convolution kernel) is as follows: the designed weight matrix should be able to extract the depressions in the ramp signal. Therefore, when the signal is nearly linear, the value after optical convolution should be 0. When a depression signal (nonlinear relationship) occurs, the processed signal should not be 0, thereby realizing the extraction of the absorption peak signal. It is required that the response to linear signals be zero, while the response to nonlinear signals is strongest, highlighting peak and valley structures.

[0043] In this embodiment, the number of columns of the convolution kernel is designed to be an odd number. Odd numbers have symmetry and a clear center point, which is the (k+1)th column, so as to achieve complete sampling of signal features and cover preamble information, subsequent information and center response.

[0044] The convolution kernel is deployed on the Digital Micromirror Devices (DMDs) of the gas concentration inversion apparatus, and the convolution process is implemented optically. As a spatial light modulator, each micromirror unit on the DMD is an independent entity. By controlling the number of mirrors activated across the entire pixel array, the pixel reflectivity is controlled, thereby modulating the intensity of reflected light and performing a physical multiplication of intensity × weight. The high transformation frequency of the DMD allows for the acquisition of approximately 2k+1 data points at the same time, achieving a similar spatial parallel modulation (weight allocation). Combined with temporal sampling, continuous convolution operations are achieved, fully utilizing the device's spatiotemporal processing capabilities.

[0045] The specific process of optical convolution processing is as follows: The changing frequency of the digital micromirror device (DMD) is set to 2k+1 times the macroscopic sampling frequency. Within the allowable error range, it can be approximated that 2k+1 weight values ​​of light intensity are collected at the same time. The modulated image of the DMD at each transformation time corresponds to the weight of each column of the convolution kernel. By controlling the number of reflective micromirrors activated at each transformation time, a corresponding mode is set. The mode corresponds to the location where the reflective mirrors are activated, thereby controlling the reflectivity and achieving optical modulation with different weight values. Each column of the weight matrix is ​​mapped proportionally to the 2k+1 transformation times of the DMD, and the number of reflective micromirrors activated is reflected in different patterns on the DMD.

[0046] The gas concentration inversion device in this embodiment includes a laser, a gas chamber, a digital micromirror device (DMD), a convex lens, and a photodetector. The fiber laser serves as the light source unit, its core function being to emit laser light matching the absorption band of the target gas, providing the basic incident light signal for subsequent gas absorption and signal extraction. The gas chamber, as the gas interaction unit, provides an environment for the laser to absorb the target gas, creating a concave light signal as the laser passes through. The DMD is the core unit for spatial light modulation and optical convolution, responsible for deploying a pre-designed one-dimensional tensor convolution kernel. By controlling the number of micromirror reflection mirrors, the reflectivity is adjusted, mapping the convolution kernel weights to different modulation patterns. Simultaneously, it operates at a transformation frequency 2k+1 times the macroscopic sampling frequency, realizing the physical multiplication and optical convolution process of light intensity and weights.

[0047] A convex lens, also known as a focusing lens, serves as a light signal converging unit. It converges the diverging light signal reflected by the DMD, reducing light intensity attenuation and ensuring that the photodetector can efficiently capture the light signal. The photodetector (PD), as a photoelectric conversion unit, converts the converged light signal into an electrical signal under the control of a clock signal synchronized with the DMD's conversion frequency. Its receiving frequency is consistent with the DMD's conversion frequency, enabling the synchronous acquisition of 2k+1 weighted light intensity signals.

[0048] S2 sequentially fills the gas concentration inversion device with target gases of different typical concentrations, and obtains the convolution processing curves corresponding to each standard concentration through optical convolution processing, optical signal convergence, photoelectric conversion and electrical signal calculation.

[0049] In this step, the target concentration of the target gas is filled into the gas chamber of the gas concentration inversion device, and a laser is emitted directly through the gas chamber. The optical signal is then focused by the optical signal modulated by the digital micromirror device. A clock signal synchronized with the conversion frequency of the digital micromirror device is then set, and the focused optical signal is received and converted into an electrical signal. The receiving frequency is 2k+1 times the macroscopic sampling frequency. Subsequently, the electrical signal is calculated using the macroscopic sampling frequency.

[0050] In this step, further electrical signal calculations are performed through a computer terminal (PC). The frequency at which the PC processes the signal is the sampling frequency, which is 1 / (2k+1) of the PD receiving frequency. At this time, for the PC to process the signal, it receives 2k+1 modulated light intensity signals at each moment.

[0051] Furthermore, a time-division multiplexing sampling strategy is adopted, and the PD sequentially records the light intensity distribution after DMD modulation at 2k+1 micro-time points that are macroscopically approximating the same moment. Each time point contains 2k+1 weighted light intensity information, and the electrical signal information of the light intensity corresponding to these times is stored in the corresponding buffer for calculation.

[0052] A sliding window buffering mechanism is used to store electrical signals in a buffer and perform electrical signal operations: several buffers are configured, the number of buffers corresponding to the number of tensors in the convolution kernel; let t n Representing the nth macroscopic moment, the data Buffer_n[j] at the jth position of the nth buffer satisfies:

[0053] Buffer_n[j] is the j-th weighted optical intensity electrical signal at the (n+j-1)-th macroscopic moment; where j=1,2,...,2k+1;

[0054] When any buffer is full of 2k+1 data, the convolution operation is automatically triggered. After the operation is completed, the memory of the buffer is cleared in order to receive new weighted light intensity electrical signals.

[0055] like Figure 2 Record 2k+1 light intensity signals with different weights at each sampling time, and store the signals sequentially into the corresponding buffers. n This represents the nth sampling time, i.e., the macroscopic time mentioned above. The first data value at the first time t1 is stored in the first Buffer1, the second data value at the second time t2, the third data value at the third time t3, and so on up to the (2k+1)th time t. 2k+1 The 2k+1th data value is stored in the first Buffer (1); the first data value at the second time t2, the second data value at the third time t3, until t 2k+2 The value at time 2k+2 is stored in the second Buffer (2); the first data value at time t3 and the second data value at time t4 are stored until t3. 2k+3 The 2k+3rd data value at time t is stored in the third buffer (3); and so on until the last time t. n All weight values ​​are stored, and once each buffer is full, it will automatically perform calculations and then clear the memory to begin the next round of storage.

[0056] Another explanation is as follows:

[0057] The filling process of the first buffer is as follows: Buffer[n] represents the nth position of the data stored in the buffer.

[0058] Buffer1[1]←The first weighted light intensity at the first macroscopic moment

[0059] Buffer1[2]←The second weighted light intensity at the second macroscopic moment

[0060] Buffer1[3]←The third weighted light intensity at the third macroscopic moment

[0061] Buffer1[2k+1] ← The weighted light intensity at the (2k+1)th macroscopic moment.

[0062] The filling process of the second buffer:

[0063] Buffer2[1]←The first weighted light intensity at the second macroscopic moment

[0064] Buffer2[2]←The second weighted light intensity at the third macroscopic moment

[0065] Buffer2[3]←The third weighted light intensity at the fourth macroscopic moment

[0066] Buffer2[2k+1] ← The weighted light intensity at the (2k+2)th macroscopic moment (2k+1)th time.

[0067] The process of filling the nth buffer:

[0068] Buffer_n[j] ← The j-th weighted light intensity at the (n+j-1)th macroscopic moment (j=1,2,...,2k+1).

[0069] In this system, the number of buffers corresponds to the number of tensors in the convolution kernel. Whenever a buffer is full, containing all the data required for a complete convolution operation, all the data in the buffer are summed according to predetermined rules to obtain the result of the operation. Finally, these convolved values ​​are linearly combined sequentially to complete the linear convolution operation of the one-dimensional signal. After obtaining a value of the convolved curve, the internal data is cleared, and new data is stored to improve storage efficiency. In electrical signal processing, spectral data obtained from a single measurement can be used multiple times. Spectral data at the same moment participates in multiple convolution calculations simultaneously, achieving spatial convolution unfolding through the time dimension.

[0070] This embodiment employs a dynamic caching mechanism, allowing the system to continuously output results without waiting for a complete cycle, significantly improving data processing throughput.

[0071] S3 extracts absorption features based on standard concentrations and corresponding convolutional curves, and uses the least squares method to fit the correspondence between gas concentration and absorption features to establish a gas concentration inversion model.

[0072] In this step, target gases of different standard concentrations are used as samples to obtain convolution processing curves corresponding to each sample; absorption features are extracted from each convolution processing curve—the absorption peak or peak-to-peak value is selected as the feature parameter based on the curve shape (such as baseline stability and concave symmetry); the extracted absorption features are linearly fitted with the standard molar concentration of the corresponding sample using the least squares method to obtain the linear correspondence between the absorption features and the concentration, which is the gas concentration inversion model.

[0073] S4 introduces the gas to be calibrated into the gas concentration inversion device to obtain its corresponding absorption characteristics. Combined with the gas concentration inversion model, the concentration of the gas to be calibrated is calculated.

[0074] In this step, the actual gas concentration is inverted using the established model: for the actual gas to be calibrated, its convolution curve is first obtained according to the same process as the sample test, and then the same type of absorption feature as in the model building stage is extracted (to ensure feature consistency and accuracy); the absorption feature is substituted into the gas concentration inversion model, and the concentration of the actual gas can be inverted by calculating the linear relationship.

[0075] Based on this, this embodiment also proposes a fast gas concentration inversion system based on spatial optical convolution modulation, including:

[0076] The gas concentration inversion device is pre-deployed with convolution kernels, which are used to perform optical convolution processing, optical signal convergence, photoelectric conversion and electrical signal calculation on target gases with different typical concentrations that are sequentially introduced, so as to obtain the convolution processing curves corresponding to each standard concentration.

[0077] The gas concentration inversion model establishment device extracts absorption features based on standard concentrations and corresponding convolution processing curves, and uses the least squares method to fit the correspondence between gas concentration and absorption features to establish a gas concentration inversion model.

[0078] The gas calibration device introduces the gas to be calibrated into the gas concentration inversion device to obtain its corresponding absorption characteristics. Combined with the gas concentration inversion model, the concentration of the gas to be calibrated is calculated.

[0079] The gas concentration inversion device specifically includes a laser, a gas chamber, a digital micromirror device (DMD), a convex lens, and a photodetector (PD), and also integrates a clock synchronization unit, a buffer array, and a data processing subunit. The laser emits a laser beam that matches the absorption band of the target gas. The gas chamber holds target gases (including standard concentration gas and gas to be calibrated) of different typical concentrations, providing an environment for the absorption interaction between the laser and the gas. The DMD, as the core of spatial light modulation, has a conversion frequency set to 2k+1 times the macroscopic sampling frequency. By controlling the number of micromirrors activated, the reflectivity is adjusted, mapping the convolution kernel weights to different modulation patterns, realizing physical multiplication and optical convolution processing of light intensity and weights. The convex lens converges the divergent light signal reflected by the DMD, reducing light intensity attenuation. The PD... Under the control of the synchronous clock signal generated by the clock synchronization unit, the converged optical signal is converted into an electrical signal. The receiving frequency is consistent with the DMD conversion frequency, realizing the synchronous acquisition of 2k+1 weighted optical intensity signals. The number of buffers in the buffer array corresponds to the number of tensors in the convolution kernel. The sliding window buffering mechanism is used to store the electrical signal, satisfying the storage rule that Buffer_n[j] is the j-th weighted optical intensity electrical signal at the (n+j-1)-th macroscopic moment (j=1,2,...,2k+1). It also supports the parallel processing strategy of dynamic caching and spatiotemporal reuse. When the buffer is full, the convolution operation is automatically triggered. After the operation is completed, the memory is cleared to improve storage efficiency. The data operation subunit performs electrical signal operation according to the macroscopic sampling frequency. After linearly combining the convolution operation results, the convolution processing curves corresponding to the target gases of each standard concentration are output.

[0080] The gas concentration inversion model establishment device is signal-connected to the gas concentration inversion device and is used to receive the convolution processing curve. The device has a built-in feature extraction module and a model fitting module. The feature extraction module is used to select the absorption peak or peak-to-peak value as the absorption feature parameter according to the shape of the convolution processing curve (such as baseline stability and concave symmetry). The model fitting module is used to call the least squares method to linearly fit the extracted absorption features with the target gas molar concentration of the corresponding standard concentration, and establish a linear correspondence between the absorption features and the molar concentration. This linear correspondence is the gas concentration inversion model.

[0081] The gas calibration device is connected to the gas concentration inversion device and the gas concentration inversion model establishment device. It is used to introduce the gas to be calibrated into the gas chamber of the gas concentration inversion device and control the gas concentration inversion device to complete the laser emission, gas absorption, optical convolution processing, optical signal convergence, photoelectric conversion and electrical signal calculation in sequence according to the test process of standard concentration gas to obtain the convolution processing curve corresponding to the gas to be calibrated. Then, the feature extraction module of the gas concentration inversion model establishment device is called to extract the absorption features consistent with the type of the model establishment stage. Finally, the established gas concentration inversion model is called, the extracted absorption features are substituted into the linear correspondence to calculate the molar concentration of the gas to be calibrated, and the calibration result is output.

[0082] In this system, the gas concentration inversion device, the gas concentration inversion model building device, and the gas calibration device work together to complete the entire process of standard gas signal acquisition → inversion model construction → gas concentration calibration in sequence, which corresponds one-to-one with the steps of the aforementioned gas concentration rapid inversion method based on spatial optical convolution modulation.

[0083] Application Examples

[0084] Taking the design of a weight matrix with one row and five columns as an example, other feasible embodiments may also include the design of optical convolution kernels with more columns that are horizontally extended. This invention does not impose specific limitations on them.

[0085] Because it involves the design of optical convolution kernels, this invention obtains the absorption spectrum curve after convolution operation based on the baseline of the ramp signal removed using optical convolution technology. The principle is as follows:

[0086] First, design a weight matrix k=[1 -0.5 -1 -0.5 1] with one row and five columns. By deploying DMDs on optical devices, design different patterns to correspond to different reflectivities, thereby realizing the function of optical convolution kernels.

[0087] Next, the frequency of the laser ramp signal is set to 1Hz. Theoretically, the DMD's transformation frequency should be designed to be 5 times the sampling frequency, so that each sampling can be approximated as acquiring five modulated data points at the same macroscopic moment. However, this embodiment considers inserting some spacing between the weight matrices to optimize the restored signal. Also, in this embodiment, the duration of each image in the DMD can be customized; therefore, the duration of weight patterns of 1 and 0.5 is set to 200µs, while a weight of 0 (i.e., no reflection) with a duration of 800µs is inserted. The period of the DMD executing the five weight values ​​at the same macroscopic moment is 1800µs. Each transformed image modulates the light intensity signal through different reflectivities, and different reflectivities map to different coefficients in the convolution kernel. At each macroscopic sampling moment, the first transformed image maps to the value of the first column of the weight matrix, the second image maps to the value of the second column of the weight matrix, and so on. However, since the designed weight matrix only has two weights, 1 and 0.5, only two patterns are needed for deployment on the DMD. Additionally, the insertion spacing requires an extra all-white pattern (corresponding to 0 reflectivity). Therefore, only one weight corresponding to 1 (total reflection) needs to be designed, i.e., one all-black pattern (all-black corresponds to 1 reflectivity) and one half-black, half-white pattern with the 50% galvanometer in the ON state (0.5 reflectivity) corresponding to a weight of 0.5. After the light intensity signal is transmitted to the PC, the values ​​corresponding to the negative weights need to be negatively evaluated before calculation.

[0088] Light emitted from a fiber laser passes through the O2 gas to be tested stored in the gas chamber and is reflected by the DMD. A timing signal strictly synchronized with the DMD's conversion frequency controls the PD to collect the modulated light intensity signal. This can be controlled by a clock signal with a time interval of Δt. The light is focused onto the PD by a lens, and after photoelectric conversion by the PD, five different weighted modulated light intensities are recorded at the same macroscopic moment.

[0089] The collected data is then transmitted to a PC for parallel analysis. The intensity values ​​of the light after the first transformation of the DMD at time t1, the second transformation at time t1+Δt, the third pixel after the DMD at time t1+2Δt, and so on, until the fifth transformation at time t1+4Δt, are stored in the first buffer. A calculation is performed on the first buffer, and the resulting value is used as one value of the absorption peak curve after convolution. The buffer length is designed to be the tensor number of the convolution kernel (5). Therefore, after five clock cycles, the first buffer is full and the first value of the absorption peak curve is obtained. Each subsequent time step will have one buffer full for calculation. After the calculation is performed on the full buffer, the value is transmitted, and then the internal data is cleared to improve storage efficiency. Therefore, only five buffers are needed in this scheme. This process continues, taking the image from the first transformation at time t1+Δt, the image from the second transformation at time t1+2Δt, and so on up to the fifth transformation at time t1+5Δt. The light intensity values ​​are stored in the second buffer for processing to obtain the second value of the processed curve, and so on. Since each signal will participate in multiple convolution operations after acquisition, the modulated value is then stored in the corresponding buffer based on the DMD pattern. After t1+4Δt, each acquired signal fills a buffer before processing, meaning that a new convolution result is generated at each time point after t1+4Δt, and a value of the convolution-processed curve is obtained at each time point. This process continues until the data acquired at the last time point is stored in the buffer.

[0090] Simulations were conducted under relatively ideal conditions. This experiment used the HITRAN database to obtain O2 concentration and corresponding slope absorption signal curves, which were then processed by the system of this invention to obtain an optically convolutionally modulated curve. The concentration was linearly fitted to the peak value of the curve obtained after the system, and the concentration-peak-peak ratio was used to obtain the fitted curve. In this embodiment, typical concentration values ​​of 1%, 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, and 10% were selected. The fitted curve results are as follows... Figure 3 As shown, the feature of the curve extracted after simulation of the designed system has a high fitting degree of 0.9999 with the concentration.

[0091] According to the scheme in this embodiment, the experimental results are as follows: Figure 4 As shown in the figure. The actual results are compared with the simulation results above and the relationship between absorption peak and gas concentration obtained by removing the baseline using the traditional method based on the HITRAN database. The results are as follows. Figure 5 and Figure 7 As shown.

[0092] Finally, the O2 concentration was deduced from the data collected in the actual optical and electrical systems. In practical systems, the lack of spacing between tensor kernels can lead to poor signal extraction quality and a low signal-to-noise ratio. Therefore, for noisy or low-quality signals, it is advisable to insert spacing directly into each tensor within the convolution kernel to achieve high-quality signal extraction.

[0093] The present invention also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist alone and not assembled into the electronic device.

[0094] The aforementioned computer-readable medium carries one or more programs that, when executed by an electronic device, cause the electronic device to implement the methods described in the above embodiments.

[0095] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0096] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this disclosure.

[0097] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein.

[0098] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A method for rapid inversion of gas concentration based on spatial optical convolution modulation, characterized in that, include: The target gas and its absorption band are determined, the convolution kernel is pre-designed, and the convolution kernel is deployed to the gas concentration inversion device; The target gas with different typical concentrations is sequentially filled into the gas concentration inversion device, and the convolution processing curve corresponding to each standard concentration is obtained through optical convolution processing, optical signal convergence, photoelectric conversion and electrical signal calculation. Based on the standard concentrations and corresponding convolutional processing curves, absorption features are extracted, and the least squares method is used to fit the correspondence between gas concentration and the absorption features to establish a gas concentration inversion model. The gas to be calibrated is introduced into the gas concentration inversion device to obtain its corresponding absorption characteristics. Combined with the gas concentration inversion model, the concentration of the gas to be calibrated is calculated.

2. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 1, characterized in that, The convolution kernel is a one-dimensional tensor convolution kernel of length 2k+1, where k is an integer greater than 0; the convolution kernel uses a one-dimensional valid convolution method to extract information about the absorption signal in the depression on the ramp signal, and the weight w of the convolution kernel... i The sum satisfies as well as This makes the convolution kernel respond zero to linear ramp signals and significantly to nonlinear absorption signals from depressions, wherein... For discrete signal sequence indexing; when using the one-dimensional valid convolution method, starting from the acquisition of the 2k+1th signal, the convolution kernel obtains a convolutional value for each time it moves one grid.

3. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 1, characterized in that, The convolution kernel is designed with an odd number of columns, which has symmetry and a clear center point, which is the (k+1)th column, to achieve complete sampling of signal features and cover preamble information, subsequent information and center response.

4. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 1, characterized in that, The convolution kernel is deployed on the digital micromirror device of the gas concentration inversion device; the digital micromirror device acts as a spatial light modulator, controlling the reflectivity of the device by changing the number of open reflector mirrors, so as to modulate the intensity of reflected light and perform physical multiplication of light intensity × weight.

5. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 4, characterized in that, The specific process of the optical convolution processing is as follows: the change frequency of the digital micromirror device is set to 2k+1 times the macroscopic sampling frequency; the modulated image of the digital micromirror device at each transformation time corresponds to the weight of each column of the convolution kernel, and each column of the weight matrix is ​​mapped to the reflectivity of the digital micromirror device at 2k+1 transformation times in a proportional manner.

6. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 4, characterized in that, The specific process of optical signal convergence and photoelectric conversion is as follows: the optical signal modulated by the digital micromirror device is converged by a convex lens, and a clock signal synchronized with the conversion frequency of the digital micromirror device is set to receive the converged optical signal and convert it into an electrical signal. The receiving frequency is 2k+1 times the macroscopic sampling frequency, and the electrical signal is then processed using the macroscopic sampling frequency.

7. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 6, characterized in that, A sliding window caching mechanism is used to store the electrical signal in a buffer and perform operations on the electrical signal: several buffers are configured, the number of which corresponds to the number of tensors in the convolution kernel; let t n Representing the nth macroscopic moment, the data Buffer_n[j] at the jth position of the nth buffer satisfies: Buffer_n[j] is the j-th weighted optical intensity electrical signal at the (n+j-1)-th macroscopic moment; where j=1,2,...,2k+1; When any buffer is full of 2k+1 data, the convolution operation is automatically triggered. After the operation is completed, the memory of the buffer is cleared in order to receive new weighted light intensity electrical signals.

8. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 1, characterized in that, When the electrical signal is processed, a time-division multiplexing method is used, so that the spectral data obtained from a single measurement can be used multiple times. The spectral data at the same moment participates in multiple convolution calculations simultaneously, and the spatial convolution is unfolded through the time dimension.

9. The method for fast inversion of gas concentration based on spatial optical convolution modulation as described in claim 1, characterized in that, The absorption feature is the absorption peak or peak-to-peak value of the convolutional curve; the linear correspondence between the gas concentration and the absorption feature is obtained by fitting with the least squares method, and the gas concentration inversion model is the linear correspondence model.

10. A fast gas concentration inversion system based on spatial optical convolution modulation, characterized in that, include: The gas concentration inversion device is pre-deployed with convolution kernels, which are used to perform optical convolution processing, optical signal convergence, photoelectric conversion and electrical signal calculation on target gases with different typical concentrations that are sequentially filled in, to obtain the convolution processing curves corresponding to each standard concentration. The gas concentration inversion model establishment device extracts absorption features based on standard concentrations and corresponding convolution processing curves, and uses the least squares method to fit the correspondence between gas concentration and the absorption features to establish a gas concentration inversion model. The gas calibration device introduces the gas to be calibrated into the gas concentration inversion device, obtains its corresponding absorption characteristics, and calculates the concentration of the gas to be calibrated by combining it with the gas concentration inversion model.

Images