Multi-component gas telemetering laser echo signal processing system and method based on deep learning

Abstract

The invention discloses a multi-component gas telemetering laser echo signal processing system and method based on deep learning. The multi-component gas telemetering laser echo signal processing system comprises a laser emitting unit, a signal collecting unit and a signal processing unit. The laser emission unit generates and emits a laser signal with a specific wavelength, and detects multi-component gas in a target environment; the signal acquisition unit receives a laser signal returned after passing through a target environment; the signal processing unit is used for carrying out noise reduction and normalization processing on the initial signal captured by the acquisition unit, and realizing multi-component gas concentration inversion on the signal subjected to noise reduction and normalization processing by utilizing a deep learning model deployed by an upper computer. According to the invention, the interference of noise on signals is effectively suppressed, and the signal-to-noise ratio and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of analytical detection, and particularly relates to a multi-component gas telemetry laser echo signal processing system and method based on deep learning. Background Art

[0002] Laser demolition and rescue technology can accurately and efficiently control the geometric shape of the channels formed after the buried and collapsed body is stimulated, and improve the pore permeability of the channels and their surroundings. It is a new type of non-destructive demolition technology at present. This technology has the advantages of high efficiency, low cost, high safety and low pollution. Compared with traditional demolition technologies, this method is easier to quickly construct a rescue passage, with simple equipment, small floor area, long service life, a ceramic protective layer is formed on the high-temperature wellbore or channel, reducing the risk of secondary disasters, and it is relatively clean after drilling, reducing environmental pollution. Therefore, the laser demolition and rescue technology that can quickly construct a disaster accident rescue passage and make scientific rescue decisions has become the main development trend.

[0003] However, when accidents occur in mines, buildings, chemical industrial areas, and vehicles loaded with flammable and explosive gases, there may be extremely common flammable, explosive, and toxic dangerous chemical gases such as methane CH 4 , ammonia NH 3 , ethane C 2 H 4 etc. at the disaster accident site. Personnel may be trapped under the collapsed body, and the situation at the accident site is extremely complex. If rescue operations are carried out blindly, it is easy to trigger secondary disasters and cause harm to the trapped and rescue personnel.

[0004] The current multi-component gas telemetry laser echo signal processing methods mainly rely on traditional spectral demodulation and signal filtering technologies. These methods are easily affected by noise interference and background signals in complex environments, resulting in a decrease in detection accuracy. In addition, traditional methods usually have limited ability to identify and distinguish multi-component gases and are difficult to cope with the multiple absorption characteristics of mixed gases.

[0005] Therefore, it is necessary to study a multi-component gas telemetry laser echo signal processing method based on deep learning. Through deep learning technology, it can automatically identify and classify complex spectral signals, improve the signal-to-noise ratio and detection accuracy, enhance the ability to analyze multi-component gases, and provide auxiliary decision-making for on-site laser demolition rescue. Summary of the Invention

[0006] To overcome the above technical problems, the object of the present invention is to provide a multi-component gas telemetry laser echo signal processing system and method based on deep learning. This method can automatically learn the spectral characteristics of different gases, accurately identify and distinguish the mixed absorption spectra of multi-component gases in a complex gas environment, and greatly improve the detection adaptability and accuracy. At the same time, through the online update of the deep learning model, the dynamic adjustment of model parameters, adaptive signal preprocessing, active learning mechanism and non-linear feature extraction of the multi-layer neural network, the adaptive learning ability under different environmental conditions and gas concentration changes is realized, so that the system can automatically adjust the algorithm according to environmental changes and maintain high detection stability. In addition, the non-linear characteristics and adaptive learning ability of the multi-layer feedforward neural network are used to effectively suppress the interference of noise on the signal, and the signal-to-noise ratio and reliability are improved.

[0007] The technical solution adopted by the present invention is:

[0008] A multi-component gas telemetry laser echo signal processing system based on deep learning, comprising a laser emission unit 1, a signal acquisition unit 2, and a signal processing unit 3;

[0009] The laser emission unit 1 generates and emits a laser signal with a specific wavelength to detect multi-component gases in the target environment;

[0010] The signal acquisition unit 2 receives the laser signal returned after passing through the target environment;

[0011] The laser echo signal will change due to the interaction with gas molecules, such as intensity attenuation or phase change;

[0012] The signal processing unit 3 performs noise reduction and normalization processing on the initial signal captured by the acquisition unit 2, and uses the deep learning model deployed on the upper computer to perform inversion of the multi-component gas concentration on the signal after noise reduction and normalization processing;

[0013] The laser emission unit 1, the signal acquisition unit 2, and the signal processing unit 3 are connected by a BNC cable 17;

[0014] The laser emission unit 1 includes a laser controller 5, a laser driver 6, a laser emitter 7, a multi-channel laser generator 8, an optical fiber coupler 9, and a laser collimator 10;

[0015] Among them, the laser controller 5 and the laser driver 6 are connected in a double-line manner through a BNC cable 17 and a multi-mode optical fiber 18. The laser controller 5 is adjacent to the tunable laser 7 to drive it to generate a laser signal. The multi-channel laser generator 8 and the optical fiber coupler 9 are connected through a multi-mode optical fiber 18. The laser signal is collimated by the laser collimator 10. The multi-channel laser generator 8, the optical fiber coupler 9, and the laser collimator 10 are on the same horizontal axis to ensure that the beam directions are consistent;

[0016] In the laser emission unit 1, each gas has unique absorption characteristics within a certain specific wavelength range. By quickly tuning the laser wavelength, the characteristic absorption peaks of multiple target gases are scanned. The laser drive drives the tunable laser 7 to generate a laser beam, and the multi-channel laser generator 8 transmits the laser beam. The laser beam reaches the detection area through the fiber optic coupler 9 and the laser collimator 10.

[0017] The laser emission unit 1 modulates the wavelength of the laser light source according to the components of the detected gas.

[0018] The laser light source of the laser emitter 7 is a tunable laser diode. The laser emitter 7 in the laser emission unit 1 can adjust the wavelength to adapt to the absorption characteristics of different gases, realizing highly sensitive detection of multiple gas components.

[0019] The signal acquisition unit 2 includes a Fresnel lens 11 and a photodetector 12.

[0020] Among them, the Fresnel lens 11 and the photodetector 12 are on the same horizontal axis. The reflected laser signal is precisely focused on the photodetector 12 through the Fresnel lens 11. The unified horizontal axis where the multi-channel laser generator 8, the fiber optic coupler 9, and the laser collimator 10 are located is parallel to the same horizontal axis where the Fresnel lens 11 and the photodetector 12 are located, realizing parallel coaxial emission and reception.

[0021] The photodetector 12 captures and converts the optical signal into an electrical signal.

[0022] The signal processing unit 3 includes an analog-to-digital converter 13, a lock-in amplifier 14, a conversion adapter 15, and a host computer 16.

[0023] Among them, the photodetector 12 is connected to the analog-to-digital converter 13 through a BNC cable, and the analog-to-digital converter 14 is connected to the lock-in amplifier 14 through a shielded cable 4. The analog-to-digital converter 14 converts the electrical signal into a digital signal. The processed signal undergoes signal noise reduction and normalization processing through the lock-in amplifier 15. The analog-to-digital converter 13 connects the processed signal to the host computer through a standard data transmission interface, realizing the physical connection of the signal between the lock-in amplifier 14 and the host computer 16.

[0024] The Fresnel lens 11 is designed with a high-frequency annular structure, focusing the parallel beam or divergent beam to the focal point 11-2 position to form a light spot with a high energy density. The preparation material of the Fresnel lens 11 is acrylic material.

[0025] A broadband antireflection film 11-1 is provided on the surface of the Fresnel lens 11, which improves the light transmittance of the lens by reducing the reflection loss of light with different wavelengths, and optimizes the light transmission efficiency especially in a wide spectral range; at the same time, it effectively reduces the reflection on the surface of the Fresnel lens, enhances the utilization rate of the laser echo beam of the overall multi-component gas dynamic imaging monitoring system, and further improves the detection accuracy and sensitivity.

[0026] A signal processing method for a multi-component gas telemetry laser echo signal processing system based on deep learning includes the following steps:

[0027] Step 1: Collect the initial echo signal S(t) through the Fresnel lens 11 and the photodetector 12;

[0028] Step 2: Denoise the initial echo signal through the lock-in amplifier 14 of the signal processing unit 3 to obtain the processed signal S f (t):

[0029] Step 3: Normalize the denoised signal through the lock-in amplifier 14 of the signal processing unit 3 to obtain the signal S n (t).

[0030] In the said Step 2,

[0031] wherein, S f (t) is the denoised signal, S(t) is the initial echo signal, and RC is the time constant;

[0032] In the said Step 3,

[0033]

[0034] wherein, S n (t) is the signal after normalization processing, μ and σ are the mean and standard deviation of the filtered signal, and N is the number of signal samples;

[0035] Furthermore, μ and σ are obtained by statistically processing the denoised signal S n (t). The number of signal samples is directly related to the selection of the sampling frequency and time window of the initial signal. During the signal acquisition process, the photodetector 12 collects the optical signal through the Fresnel lens 11 and converts it into an electrical signal. The number of signal samples N is determined by the analog-to-digital converter 13 during the process of converting the analog signal into a digital signal through the set sampling frequency and time window.

[0036] In the said Step 3, the signal processing further includes:

[0037] Using the upper computer 16 to perform multi-component gas concentration inversion on the output signals after denoising and normalization to obtain different gas concentrations

[0038] C p = f θ* (S n (t))

[0039] where C p outputs the predicted gas concentration vector, and f θ* is a multi-component gas concentration inversion model, where θ represents the model parameters;

[0040]

[0041] where each is the predicted concentration of the corresponding gas;

[0042] Furthermore, the training and deployment of the multi-component gas concentration inversion model f θ* are both on the host computer 17. The model parameters θ are deployed on the host computer 17 and are continuously updated during the training process to minimize the error between the model prediction value and the actual value, thereby improving the accuracy of the inversion model;

[0043] The host computer 16 realizes the display of multi-component gas concentration data for the output predicted gas concentration vector through a simple character transmission protocol;

[0044] The host computer 16 internally deploys an imaging technology. According to the multi-component gas concentration, the absorption characteristics of each position are back-projected by back-projection, and the concentration distribution image of the gas is gradually constructed.

[0045] Advantages of the present invention:

[0046] 1. By introducing deep learning technology, the present invention can automatically learn and adapt to the spectral characteristics of different gases, accurately identify and distinguish the mixed absorption spectra of multi-component gases, and enhance the adaptability to complex gas environments;

[0047] 2. The present invention has an adaptive learning ability. Through the online update of the deep learning model, the dynamic adjustment of model parameters, adaptive signal preprocessing, active learning mechanism, and non-linear feature extraction of the multi-layer neural network, the adaptive learning ability under different environmental conditions and gas concentration changes is realized. It can automatically adjust and optimize the analysis algorithm under different environmental conditions and gas concentration changes, and maintain high detection performance and stability;

[0048] 3. By using the non-linear characteristics and adaptive learning ability of the multi-layer feedforward neural network, the present invention effectively reduces the influence of noise interference on signal processing, improves the signal-to-noise ratio, and improves the accuracy and reliability of gas detection;

[0049] The present invention relies on deep learning technology, reduces the dependence on manual parameter adjustment and complex algorithm design, simplifies the implementation process of the system, and reduces the maintenance cost and operation complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is the architecture diagram of the multi-component gas system of the present invention.

[0051] Figure 2 is the schematic structural diagram of the multi-component gas system of the present invention.

[0052] Figure 3 is the flowchart of the signal processing method of the present invention.

[0053] Figure 4 is the schematic structural diagram of the Fresnel lens of the present invention.

[0054] Figure 5 is the comparison diagram of the initial signal collected by the present invention and the signal after noise reduction and normalization processing.

[0055] Figure 6 is the comparison diagram of the retrieved concentration of the multi-component gas and the theoretical value concentration of the present invention.

[0056] REFERENCE SIGNS:

[0057] 1 - Laser emission unit; 2 - Signal acquisition unit; 3 - Signal processing unit; 4 - Shielded cable; 5 - Laser controller; 6 - Laser driver; 7 - Tunable laser; 8 - Multi-channel laser generator; 9 - Fiber optic coupler; 10 - Laser collimator; 11 - Fresnel lens; 11-1 - Broadband antireflection film; 11-2 - Focus; 12 - Photoelectric detector; 13 - Analog-to-digital converter; 14 - Phase-locked amplifier; 15 - Adapter; 16 - Host computer; 17 - BNC cable; 18 - Multimode optical fiber. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0059] As Figure 1 shown in Figure 2 the multi-component gas telemetry system includes a laser emission unit 1, a signal acquisition unit 2, and a signal processing unit 3;

[0060] The laser emission unit 1, the signal acquisition unit 2, and the signal processing unit 3 are connected through a BNC cable 17;

[0061] The laser emission unit 1 includes a laser controller 5, a laser driver 6, a tunable laser 7, a multi-channel laser generator 8, an optical fiber coupler 9, and a laser collimator 10;

[0062] The signal acquisition unit includes a coaxial Fresnel lens 11 and a photodetector 12;

[0063] The signal processing unit includes an analog-to-digital converter 13, a lock-in amplifier 14, an adapter 15, and a host computer 16;

[0064] Among them, the laser emission unit 1 generates and emits a laser signal of a specific wavelength to detect multi-component gases in the target environment. The laser light source of the tunable laser 7 is a tunable laser diode. The lasers in the emission unit can adjust the wavelength to adapt to the absorption characteristics of different gases, realizing high-sensitivity detection of multiple gas components; the signal acquisition unit 2 receives the laser signal returned after passing through the target environment. The laser echo signal will change due to the interaction with gas molecules, with intensity attenuation or phase change. The photodetector 12 of the acquisition unit 2 captures and converts the optical signal into an electrical signal; the signal processing unit 3 first performs noise reduction and normalization processing on the captured initial signal, and uses the deep learning model deployed on the host computer to perform inversion of the multi-component gas concentration on the signal after noise reduction and normalization processing;

[0065] The signal processing unit 3 first performs noise reduction processing on the captured initial signal. Since various background noises and interference factors may be mixed in the initial signal, the purpose of noise reduction processing is to remove these noises to ensure that the effective signals related to the target gas are retained; secondly, normalization processing is performed to eliminate the inconsistency in signal amplitude; finally, the deep learning model deployed on the host computer is used to perform inversion of the multi-component gas concentration on the signal after noise reduction and normalization processing. The deep learning model is trained with a large amount of training data and can automatically learn and adapt to the spectral characteristics of various gases, so as to accurately identify and distinguish the gas concentrations of different components in a complex environment. The application of the deep learning model can not only improve the detection accuracy, but also maintain high stability and adaptability under changing environmental conditions.

[0066] Furthermore, the laser emission unit 1 modulates the wavelength of the laser light source according to the components of the detected gas;

[0067] Furthermore, as Figure 4 shown, the Fresnel lens 11 is designed with a high-frequency annular structure to focus the parallel beam or divergent beam to the focal point 11-2 position, forming a light spot with a high energy density;

[0068] Among them, broadband antireflection films 12-1 are provided on the surfaces of the Fresnel lenses 12. Their function is to improve the light transmittance of the lenses by reducing the reflection loss of light with different wavelengths, especially to optimize the light transmission efficiency within a wide spectral range. At the same time, it effectively reduces the reflection on the surface of the Fresnel lenses, enhances the utilization rate of the laser echo beam of the overall multi-component gas dynamic imaging monitoring system, and further improves the detection accuracy and sensitivity.

[0069] As Figure 3 shown, a method for processing multi-component gas telemetry laser echo signals based on deep learning according to the present invention includes the following steps:

[0070] 1 Modulate the laser wavelength according to the detected target gas components and emit it to the detection area through the laser emission unit;

[0071] 2 Collect the initial echo signals through the Fresnel lens and the photodetector;

[0072] 3 Perform noise reduction processing on the initial echo signals through the lock-in amplifier of the signal processing unit to obtain the processed signal Sf(t):

[0073]

[0074] Wherein, S f (t) is the signal after noise reduction, S(t) is the initial echo signal, and RC is the time constant;

[0075] Perform normalization processing on the signal after noise reduction through the lock-in amplifier of the signal processing unit to eliminate the influence of signal amplitude changes on subsequent processing:

[0076]

[0077] Wherein, Sn(t) is the signal after normalization processing, μ and σ are the mean and standard deviation of the filtered signal, and N is the number of signal samples;

[0078] The signal processing further includes:

[0079] Use the upper computer to perform multi-component gas concentration inversion on the output signals after noise reduction and normalization;

[0080] C p =f θ* (S n (t))

[0081] Wherein, C p Output the predicted gas concentration vector, and f θ* is the multi-component gas concentration inversion model, where θ represents the model parameters;

[0082]

[0083] where each is the predicted concentration of the corresponding gas;

[0084] Furthermore, f θ* is modeled by a multi-layer feedforward neural network, with an input layer, a hidden layer, and an output layer, and there are L hidden layers;

[0085] Furthermore, from the input layer to the first hidden layer:

[0086] h (1) = g(W (1) S n (t) + b (1) )

[0087] Furthermore, from the a-th hidden layer to the (a + 1)-th hidden layer, a = 1, 2, …, L - 1:

[0088] h (a+1) = g(W (a+1) h a + b (a+1) )

[0089] Furthermore, the output layer inverses the gas concentration:

[0090] C p = h (L+1) = W (L+1) h (L) + b (L+1)

[0091] Furthermore, Sn(t) is the signal after noise reduction and normalization, W(l) and b(l) are the weight matrix and bias vector of the l-th layer respectively, and g(·) is a non-linear activation function;

[0092] The test set evaluates the gas concentration inversion signal through the training set and the validation set. The mean squared error MES emphasizes reducing large errors and is suitable for tasks that require high precision and are sensitive to large errors; the mean absolute error MAE emphasizes robustness and is suitable for average performance and tasks that are not sensitive to outliers. Therefore, MES and MAE are used to quantify the inversion ability of the model:

[0093] Set the ideal gas concentration vector where m is the number of gas types, and j represents the gas concentration vector of the j-th sample predicted by the model;

[0094] Use the model f θ * to evaluate each input signal S n (t) in the input test set;

[0095]

[0096] where, is the gas concentration vector of the j-th sample predicted by the model;

[0097] Therefore, MES and MAE can be calculated by the following equations:

[0098]

[0099] The values obtained from MES and MAE are used to evaluate the accuracy of the inverted gas concentration, further providing data support for optimizing the model;

[0100] Such as Figure 5 and Figure 6 As shown, the host computer realizes the display of multi-component gas concentration data for the output gas concentration vector through a simple character transmission protocol.

[0101] The above is only a detailed description of the preferred embodiments and principles of the present invention. For those of ordinary skill in the art, based on the idea provided by the present invention, there will be changes in the specific implementation manners, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. A multi-component gas telemetry laser echo signal processing system based on deep learning, characterized in that: It comprises a laser emitting unit (1), a signal collecting unit (2), and a signal processing unit (3); The laser emission unit (1) generates and emits a laser signal of a specific wavelength to detect multi-component gases in a target environment; The signal acquisition unit (2) receives the laser signal returned after passing through the target environment; The signal processing unit (3) performs noise reduction and normalization processing on the initial signal captured by the acquisition unit (2), and uses the deep learning model deployed by the host computer to realize multi-component gas concentration inversion on the signal after the noise reduction and normalization processing.

2. According to the multi-component gas remote sensing laser echo signal processing system based on deep learning in claim 1, it is characterized in that: The laser emitting unit (1), the signal collecting unit (2) and the signal processing unit (3) are connected via a BNC cable (17); The laser emitting unit (1) comprises a laser controller (5), a laser driver (6), a laser emitter (7), a multi-channel laser generator (8), a fiber coupler (9), and a laser collimator (10); The laser controller (5) and the laser driver (6) are connected in two lines via a BNC cable (17) and a multimode optical fiber (18); the laser controller (5) is adjacent to the tunable laser (7) to drive it to generate a laser signal; the multi-channel laser generator (8) and the optical fiber coupler (9) are connected via the multimode optical fiber (18); the laser signal is collimated via a laser collimator (10); the multi-channel laser generator (8), the optical fiber coupler (9) and the laser collimator (10) are on the same horizontal axis to ensure that the directions of the light beams are consistent.

3. According to claim 2, a multi-component gas remote sensing laser echo signal processing system based on deep learning is characterized in that: The laser light source of the laser emitter (7) is a tunable laser diode. The laser emitter (7) adjusts the wavelength to adapt to the absorption characteristics of different gases, thereby achieving high-sensitivity detection of multiple gas components.

4. According to the multi-component gas remote sensing laser echo signal processing system based on deep learning in claim 1, it is characterized in that: The signal collection unit (2) comprises a Fresnel lens (11) and a photoelectric detector (12); The Fresnel lens (11) and the photodetector (12) are located on the same horizontal axis, the reflected laser signal is precisely focused on the photodetector (12) through the Fresnel lens (11), and the same horizontal axis where the multi-channel laser generator (8), the optical fiber coupler (9) and the laser collimator (10) are located is parallel to the same horizontal axis where the Fresnel lens (11) and the photodetector (12) are located, thereby realizing parallel coaxial transmission and reception; The photodetector (12) captures and converts optical signals into electrical signals.

5. According to claim 4, a multi-component gas remote sensing laser echo signal processing system based on deep learning is characterized in that: The Fresnel lens (11) adopts a high-frequency annular structure design to focus a parallel light beam or a divergent light beam to a focal point (11-2) to form a light spot with a high energy density; the Fresnel lens (11) is made of acrylic material. The surface of the Fresnel lens (11) is provided with a broadband anti-reflection film (11-1), which improves the light transmittance of the lens by reducing the reflection loss of light of different wavelengths.

6. The multi-component gas telemetry laser echo signal processing system based on deep learning according to claim 1, characterized in that: The signal processing unit (3) comprises an analog-to-digital converter (13), a phase-locked amplifier (14), a conversion connector (15) and a host computer (16); The photodetector (12) and the analog-to-digital converter (13) are connected via a BNC cable, the analog-to-digital converter (14) and the phase-locked amplifier (14) are connected via a shielded cable (4), the analog-to-digital converter (14) converts the electrical signal into a digital signal, the processed signal is subjected to signal noise reduction and normalization processing via the phase-locked amplifier (15), and the analog-to-digital converter (13) connects the processed signal to a host computer via a standard data transmission interface, thereby realizing a physical signal connection between the phase-locked amplifier (14) and the host computer (16).

7. The signal processing method of a multi-component gas remote sensing laser echo signal processing system based on deep learning according to any one of claims 1 to 6, characterized in that: The following steps are involved: Step 1: Collecting the initial echo signal S(t) through the Fresnel lens (11) and the photodetector (12); Step 2: The initial echo signal is subjected to noise reduction processing by the phase-locked amplifier (14) of the signal processing unit (3) to obtain a processed signal S f (t): Step 3: The signal after noise reduction is normalized by the phase-locked amplifier (14) of the signal processing unit (3) to obtain a signal S n (t).

8. The signal processing method of the multi-component gas remote sensing laser echo signal processing system based on deep learning according to claim 7 is characterized in that: In the step 2, Among them, S f (t) is the signal after noise reduction, S(t) is the initial echo signal, and RC is the time constant; In step 3, Among them, S n (t) is the normalized signal, μ and σ are the mean and standard deviation of the filtered signal, and N is the number of signal samples; μ, σ are the noise-reduced signal S n (t) is statistically processed to obtain the number of signal samples, which is directly related to the sampling frequency of the initial signal and the selection of the time window. During the signal acquisition process, the photodetector (12) acquires the optical signal through the Fresnel lens (11) and converts it into an electrical signal. The number N of signal samples is determined by the analog-to-digital converter (13) in the process of converting the analog signal into a digital signal, through the set sampling frequency and time window.

9. The signal processing method of the multi-component gas remote sensing laser echo signal processing system based on deep learning according to claim 8 is characterized in that: In step 3, the signal processing further includes: The noise-reduced and normalized output signals of the host computer (16) are used to perform multi-component gas concentration inversion to obtain different gas concentrations. C p =f θ* (S n (t)) Among them, C p Output the predicted gas concentration vector, f θ* is the multi-component gas concentration inversion model, where θ represents the model parameters; Each of these is the predicted concentration of the corresponding gas.

10. The signal processing method of the multi-component gas remote sensing laser echo signal processing system based on deep learning according to claim 9 is characterized in that: Multi-component gas concentration inversion model θ* The training and deployment of are both on the host computer (17), and the model parameters θ are deployed on the host computer (17). θ is continuously updated during the training process to minimize the error between the model prediction value and the actual value, thereby improving the accuracy of the inversion model; The host computer (16) outputs the predicted gas concentration vector and realizes the display of multi-component gas concentration data through a simple character transmission protocol.

Images