A method, device and storage medium for extracting optical cavity ring-down information

Abstract

The present application discloses a method, device, and storage medium for extracting optical cavity ring-down information. The method includes: generating a first preset number of ring-down time constant data; inputting all the ring-down time constant data into a ring-down signal mathematical model, and using the ring-down signal mathematical model to generate corresponding simulated cavity ring-down signals; superimposing random noise on each group of simulated cavity ring-down signals to obtain a second preset number of one-dimensional simulated cavity ring-down signals; dividing the one-dimensional simulated cavity ring-down signals into two sets, namely a training set and a test set; constructing an initial convolutional neural network (CNN) architecture; training the initial convolutional neural network (CNN) architecture using the training set; inputting the test set into a converged neural network model for testing, and adjusting and optimizing the neural network model based on the test results; processing the measured cavity ring-down signals using the adjusted neural network model, and outputting the corresponding ring-down time constant.

Description

Technical Field

[0001] The present application relates to the field of spectrum detection technology, and in particular to a method, device and storage medium for extracting optical cavity ring-down information. Background Art

[0002] Cavity ring-down spectroscopy (CRDS) is a high-precision, high-sensitivity optical detection technique based on a high-finesse optical resonant cavity. Since the 1970s, CRDS has become a popular technique in modern optics, chemical engineering, medicine, and other research fields, particularly in high-reflectivity measurements, due to its advantages such as unaffected measurement results from incident light intensity fluctuations and a long effective absorption path. It is particularly well-suited for high-reflectivity measurements, being the only method currently capable of precisely measuring reflectivity. It is currently widely used in trace gas monitoring, combustion diagnostics, spectral measurements, and chemical reaction kinetics. However, due to noise interference from the optical system and electronic circuits, only a low signal-to-noise ratio (SNR) cavity ring-down signal can be obtained, severely limiting the accuracy of ring-down time extraction and affecting the stability and reliability of CRS measurements. The noise interference of the ring-down signal and the inherent limitations of traditional ring-down time constant algorithms have become significant factors limiting the accuracy of ring-down time constant extraction.

[0003] In terms of signal filtering, software-based filtering technology is widely used in the field of signal filtering due to its simple implementation and low cost. However, in the field of ring-down spectrum signal filtering, only a few can achieve real-time filtering, and they all inevitably have certain defects. For example, the multi-signal mean filter (MAF) can reduce white noise by 1 / N by averaging N spectra. Although widely used, this method is very time-consuming and can only suppress white noise; Savitzky-Golay (SG) requires the selection of smoothing window size and polynomial order, and is only suitable for filtering high-frequency noise; the Kaltman filter (KF) algorithm may cause serious distortion of signals in nonlinear systems, etc. The neural network-based cavity ring-down signal filtering method in the present invention can effectively avoid the above problems and also has better performance in terms of filtering effect.

[0004] Four typical methods exist for extracting the cavity ring-down time constant in cavity ring-down technology: nonlinear least squares (NLS), linear least squares (LLS), linear regression ofsum (LRS), and frequency domain analysis. However, these algorithms require data fitting based on their respective fitting functions, and their fitting results are significantly affected by parameter selection and noise fluctuations. Summary of the Invention

[0005] In order to solve the above technical problems, the present application provides a method, device and storage medium for extracting optical cavity ring-down information.

[0006] The technical solution provided in this application is described below. In a first aspect, this application provides a method for extracting optical cavity ring-down information, the method comprising:

[0007] Generate a first preset number of ring-down time constant data according to a preset generation rule;

[0008] Inputting all the ring-down time constant data into a ring-down signal mathematical model, and using the ring-down signal mathematical model to generate a corresponding simulated cavity ring-down signal;

[0009] superimposing random noise on each group of simulated cavity ring-down signals to obtain a second preset number of one-dimensional simulated cavity ring-down signals, where the second preset number is greater than the first preset number;

[0010] Dividing the one-dimensional simulated cavity ring-down signal into two sets, namely a training set and a test set;

[0011] Build the initial convolutional neural network CNN architecture;

[0012] The initial convolutional neural network (CNN) architecture is trained using the training set to obtain a converged neural network model.

[0013] Inputting the test set into the converged neural network model for testing, and adjusting and optimizing the neural network model according to the test results;

[0014] The adjusted neural network model is used to process the measured cavity ring-down signal and output the corresponding ring-down time constant.

[0015] Optionally, the using the training set to train the initial convolutional neural network (CNN) architecture to obtain a converged neural network model includes:

[0016] Using the corresponding simulated ring-down time as a label, a label set is generated, and each group of simulated cavity ring-down signals in the training set is prepared into a training data set;

[0017] Pack the training data set and the label set into npy format files;

[0018] Read the npy file and train the constructed initial convolutional neural network (CNN) architecture;

[0019] The training process includes:

[0020] The number of convolutional layers and fully connected layers, random inactivation rate, learning rate, and number of epoch layers are continuously adjusted until a converged neural network model is obtained.

[0021] Optionally, inputting the test set into a converged neural network model for testing, and adjusting and optimizing the neural network model according to the test results includes:

[0022] Each set of simulated cavity ring-down signals in the test set is input into the converged neural network model for testing, so that the corresponding test ring-down time constant is output;

[0023] Comparing the test ring-down time constant with the simulated time constant corresponding to each group of simulated cavity ring-down signals in the test set to obtain a time constant comparison result;

[0024] The parameters of the converged neural network model are adjusted according to the time constant comparison result.

[0025] Optionally, when dividing the training set and the test set, the training set accounts for 70% and the test set accounts for 30%.

[0026] Optionally, a filtering module is provided in the neural network model, and after the measured cavity ring-down signal is processed by the adjusted neural network model, a filtering result of the cavity ring-down signal is also output.

[0027] Optionally, the generation rules include:

[0028] Conduct statistics and analysis on the existing ring-down time constant data to obtain a statistical model;

[0029] A first preset number of ring-down time constant data is generated using the statistical model.

[0030] Optionally, superimposing random noise on each group of simulated cavity ring-down signals includes:

[0031] Generate noise data through a pseudo-random number generation algorithm or collect real noise data;

[0032] Adjusting the intensity and spectral characteristics of the noise data to make it consistent with the actual environment;

[0033] The generated noise data is superimposed on the simulated cavity ring-down signal in a certain ratio.

[0034] A second aspect of the present application provides an optical cavity ring-down information extraction device, comprising:

[0035] A time constant generating module generates a first preset number of ring-down time constant data according to a preset generating rule;

[0036] a signal modeling module, inputting all the ring-down time constant data into a ring-down signal mathematical model, and generating a corresponding simulated cavity ring-down signal using the ring-down signal mathematical model;

[0037] a noise adding module for superimposing random noise on each group of simulated cavity ring-down signals to obtain a second preset number of one-dimensional simulated cavity ring-down signals, where the second preset number is greater than the first preset number;

[0038] a data partitioning module, which divides the one-dimensional simulated cavity ring-down signal into two sets, namely a training set and a test set;

[0039] CNN model building module, building the initial convolutional neural network CNN architecture;

[0040] The training module uses the training set to train the initial convolutional neural network (CNN) architecture to obtain a converged neural network model.

[0041] A testing and optimization module inputs the test set into the converged neural network model for testing, and adjusts and optimizes the neural network model according to the test results;

[0042] The inference module processes the measured cavity ring-down signal using the adjusted neural network model and outputs the corresponding ring-down time constant.

[0043] A third aspect of the present application provides a device for extracting optical cavity ring-down information, the device comprising:

[0044] processor, memory, input and output units, and buses;

[0045] The processor is connected to the memory, the input and output unit, and the bus;

[0046] The memory stores a program, and the processor calls the program to execute the first aspect and any optional method in the first aspect.

[0047] In a fourth aspect, the present application provides a computer-readable storage medium, on which a program is stored. When the program is executed on a computer, the program executes the first aspect and any optional method in the first aspect.

[0048] It can be seen from the above technical solutions that this application has the following advantages:

[0049] This method uses a neural network to learn the cavity ring-down signal. In terms of signal filtering, the cavity ring-down signal filter proposed by this method shows the best performance compared with several widely used filtering algorithms. At the same time, compared with other deep learning models, this model has a more accurate and stable performance in processing attenuated signals.

[0050] In terms of ring-down time extraction, compared with the four typical methods in this technical field: nonlinear least squares fitting method, linear least squares fitting algorithm, linear regression summation method (continuous integration method) and frequency domain analysis method, the neural network model can directly identify the input cavity ring-down signal and output the corresponding ring-down time without the need for signal fitting and other operations performed by conventional ring-down time extraction. There are significant improvements in extraction precision, extraction accuracy and computational complexity.

[0051] This method uses a neural network to obtain high signal-to-noise ratio cavity ring-down spectrum filtering data while simultaneously obtaining information such as the signal amplitude and time constant of the measured cavity ring-down spectrum data, thereby improving the accuracy and efficiency of the cavity ring-down signal filtering technology.

[0052] This method simulates the noise and interference present in real-world applications by superimposing random noise on the simulated cavity ring-down signal. This helps improve the robustness of the model, making it more adaptable to real-world applications in different noise environments.

[0053] This method can be applied to the need for cavity ring-down information extraction in various optical fields, such as optical sensing, optical communications, and photonics. By adjusting the generation rules and model architecture, it can be customized for different application scenarios and problems, and can also be extended to similar signal processing problems in other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in this application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0055] Figure 1 A flow chart of an embodiment of the optical cavity ring-down information extraction method provided in this application;

[0056] Figure 2 This is a schematic diagram of the measured filtering results of the method of this application;

[0057] Figure 3 A schematic structural diagram of an embodiment of the optical cavity ring-down information extraction device provided in this application;

[0058] Figure 4 This is a structural diagram of an embodiment of the optical cavity ring-down information extraction device provided in this application. DETAILED DESCRIPTION

[0059] It should be noted that the method provided in this application can be applied to a terminal, a system, or a server. For example, the terminal can be a smartphone, a computer, a tablet computer, a smart TV, a smartwatch, a portable computer terminal, or a fixed terminal such as a desktop computer. For ease of explanation, this application uses a terminal as the execution subject for example.

[0060] The embodiments provided in this application are described in detail below:

[0061] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of a method for extracting optical cavity ring-down information provided in this application. The method includes:

[0062] S101, generating a first preset number of ring-down time constant data according to a preset generation rule;

[0063] This step involves generating ring-down time constant data based on pre-defined generation rules. These rules can be based on previous research, experimental data, or simulation models. Based on these rules, a certain amount of ring-down time constant data is generated, serving as input for subsequent steps. For example, the ring-down time constant data can be generated based on physical equations or statistical models.

[0064] S102, inputting all the ring-down time constant data into a ring-down signal mathematical model, and generating a corresponding simulated cavity ring-down signal using the ring-down signal mathematical model;

[0065] In this step, the generated ring-down time constant data is input into a predefined mathematical model of the ring-down signal. This model can be based on physical principles, mathematical models, or empirical rules. By applying the model, a simulated cavity ring-down signal corresponding to the input ring-down time constant data is generated.

[0066] A mathematical model of a ring-down signal is a mathematical expression or algorithm used to describe the ring-down behavior within a cavity. Based on optical physics principles, electromagnetic wave propagation models, or other empirical laws, it can convert the input ring-down time constant into a simulated cavity ring-down signal. By applying the mathematical model to the ring-down time constant data as input and calculating and processing it, a cavity ring-down signal corresponding to the time constant is generated. This allows for the numerical simulation of a ring-down signal with similar characteristics. The following are some implementations based on mathematical models:

[0067] a. Transfer function model: Transfer function models, such as difference equations and differential equations, can be used to describe the cavity ring-down behavior. The ring-down time constant is used as the input variable, and the corresponding cavity ring-down signal is calculated based on the transfer function expression and parameters. The mathematical model can also be an exponential decay form, eliminating the need for differential equations and difference equations.

[0068] b. Fourier transform-based: The Fourier transform converts a signal from the time domain to the frequency domain and generates a cavity ring-down signal based on the amplitude and phase information of the frequency components. In this method, the ring-down time constant can be converted into a frequency response or frequency-domain filter characteristic, which is then converted back to the time domain using an inverse Fourier transform.

[0069] c. Based on statistical models: Statistical models, such as autoregressive models (AR) and hidden Markov models (HMM), can be used to establish the ring-down signal generation process. By learning the statistical characteristics of known ring-down time constant data, a simulated cavity ring-down signal can be generated based on the model parameters.

[0070] d. Machine Learning-Based Approaches: Utilizing machine learning techniques such as generative adversarial networks (GANs) and recurrent neural networks (RNNs), the generation pattern of the ring-down signal can be learned directly from the ring-down time constant data. By training the model, a simulated cavity ring-down signal with a time constant corresponding to the input can be generated.

[0071] Ignoring the effects of noise, the cavity ring-down signal can be approximated as an exponentially decaying signal, with the ring-down time constant being related to the decay exponent. Therefore, the simulated ring-down time constant can be used to create a simulated ring-down time signal, which can then be used to construct a training dataset. Based on this principle, 2,000 sets of ring-down time constant data can be simulated. Furthermore, using the mathematical model of the ring-down signal, 2,000 sets of original simulated cavity ring-down signal data can be simulated.

[0072] S103, superimposing random noise on each group of simulated cavity ring-down signals to obtain a second preset number of one-dimensional simulated cavity ring-down signals, where the second preset number is greater than the first preset number;

[0073] In this step, random noise is superimposed on each set of simulated cavity ring-down signals. This addition of random noise simulates the noise and interference present in actual application scenarios. A second preset number of one-dimensional simulated cavity ring-down signals is generated, ensuring that the second preset number is greater than the first preset number. This allows the number of generated training samples to be controlled based on actual needs.

[0074] In this step, random noise refers to undesired interference introduced randomly into the signal. In practical applications, the cavity ring-down signal is often affected by various noise sources, such as electromagnetic interference and thermal noise. To more accurately simulate these interferences, random noise needs to be added to the simulated cavity ring-down signal.

[0075] When superimposing noise, the noise data generated by the algorithm can be superimposed, or the noise data obtained by actual collection can be superimposed. An embodiment of superimposing noise is:

[0076] Noise data is generated using a pseudo-random number generation algorithm or real noise data is collected. The intensity and spectral characteristics of the noise data are adjusted to match the actual environment. The generated noise data is then superimposed with the simulated cavity ring-down signal at a certain ratio. For each set of data, different noise data can be superimposed to obtain multiple sets of data. For example, random noise can be superimposed on each of the 2,000 simulated original cavity ring-down signals to obtain 10,000 one-dimensional simulated cavity ring-down signals.

[0077] To generate noise data algorithmically, you can use the following algorithm:

[0078] a. Additive Gaussian Noise: You can generate random noise samples using a Gaussian distribution and then add them to the simulated cavity ring-down signal. Gaussian noise has zero mean and a specified standard deviation. The strength of the noise can be controlled by adjusting the standard deviation.

[0079] b. Uniformly distributed noise: You can use a uniform distribution to generate random noise samples and then add them to the simulated cavity ring-down signal. The amplitude range of the uniformly distributed noise can be adjusted according to actual needs.

[0080] c. White noise: A white noise signal can be used. This is a random signal with an average power spectral density. By adding white noise to the simulated cavity ring-down signal, a wideband noise interference can be simulated.

[0081] It’s worth noting that signals in practical applications are often affected by various noise sources, such as electromagnetic interference and thermal noise. By superimposing random noise, these interferences can be simulated more accurately, making the generated simulated cavity ring-down signal closer to the actual situation.

[0082] The introduction of random noise can also increase the complexity and diversity of the data. The randomness of the noise makes the generated simulated cavity ring-down signal more detailed and varied, allowing the model to more comprehensively learn and understand the characteristics of the ring-down signal in different noise environments.

[0083] S104, dividing the one-dimensional simulated cavity ring-down signal into two sets, namely a training set and a test set;

[0084] In this step, the generated one-dimensional simulated cavity ring-down signal is divided into two sets: a training set and a test set. This division can be random or chronological. The training set is used to train the neural network model, while the test set is used to evaluate model performance and perform adjustments and optimizations. A typical split ratio is 70% for training and 30% for testing.

[0085] S105. Build the initial convolutional neural network (CNN) architecture;

[0086] This step involves building the initial convolutional neural network (CNN) architecture. In this application, to mitigate overfitting in the CNN network, each convolutional layer was randomly dropped out (dropout). Three fully connected layers were then added to map the feature space calculated by the previous layers to the sample label space. The model was optimized by continuously adjusting parameters such as the number of convolutional and fully connected layers, the dropout rate, the learning rate, and the number of epochs, using final training accuracy and training time as criteria.

[0087] S106, using the training set to train the initial convolutional neural network (CNN) architecture to obtain a converged neural network model;

[0088] In this step, the initial CNN neural network architecture is trained using the training set. Through forward and backpropagation algorithms, the network's weights and biases are updated based on the training set's inputs and expected outputs. After multiple iterations of training, the neural network model gradually converges, meaning it learns the mapping relationship between inputs and outputs.

[0089] During the training process, the simulated cavity ring-down signal from the training set is used as input, and the network's predicted output is calculated through forward propagation. Forward propagation is the process of passing the input signal from the network's input layer to the output layer. Each layer performs a series of linear and nonlinear operations (such as convolution, pooling, activation functions, etc.).

[0090] Backpropagation is a key step in training neural networks. By calculating the partial derivatives of the loss function with respect to the network parameters (weights and biases), the gradients are propagated from the output layer to the input layer using the chain rule to update the network parameters. This process gradually brings the model's predictions closer to the expected output of the training set, minimizing the loss function.

[0091] For optimization, an optimization algorithm is used to update the network's parameters based on the gradient information obtained through backpropagation. Common optimization algorithms include stochastic gradient descent (SGD), momentum method, and Adam. These algorithms adjust the update step size and direction of the parameters to enable the network to converge to the optimal solution more quickly.

[0092] A specific training process embodiment includes:

[0093] a. Using the corresponding simulated ring-down time as a label, generating a label set, and preparing each group of simulated cavity ring-down signals in the training set into a training data set;

[0094] According to a preset generation rule, a simulated ring-down time corresponding to each group of simulated cavity ring-down signals in the training set is generated as a label set. Each label in the label set corresponds one-to-one to each group of simulated cavity ring-down signals in the training set.

[0095] b. Pack the training data set and the label set into npy format files;

[0096] Each set of simulated cavity ring-down signals and their corresponding label sets in the training set are packaged into an npy file. The npy file is a file format used to store multidimensional array data, making it easy to read and process data.

[0097] c. Read the npy file and train the constructed initial convolutional neural network (CNN) architecture;

[0098] The training process includes: continuously adjusting the number of convolutional layers and fully connected layers, random inactivation rate, learning rate and number of epoch layers until a converged neural network model is obtained.

[0099] Read the packaged npy file and train the initial convolutional neural network (CNN) architecture. During training, adjust the number of convolutional and fully connected layers, the dropout rate, the learning rate, and the number of epochs until a converged neural network model is achieved. Adjusting the number of convolutional and fully connected layers can increase or decrease the complexity of the neural network to suit the task complexity and the model's representational capabilities. Adjusting the dropout rate can reduce the risk of overfitting and improve the model's generalization. Adjusting the learning rate and the number of epochs can control the parameter update step size and number of iterations during training, achieving better convergence.

[0100] S107, inputting the test set into the converged neural network model for testing, and adjusting and optimizing the neural network model according to the test results;

[0101] In this step, the generated test set is input into the converged neural network model for testing. Through the forward propagation process, the neural network model processes the input signal and outputs the corresponding prediction results. Based on the test results, the model performance is evaluated and adjustments and optimizations are made to improve the model's accuracy and generalization ability. During the testing process, the simulated cavity ring-down signal of the test set is input into the converged neural network model. Through the forward propagation process, the signal is passed from the input layer to the output layer of the network. The model processes the input signal according to the learned parameters and outputs the prediction results. The neural network model generates corresponding prediction results based on the characteristics of the input signal and the learned weights. These prediction results can be estimates of the ring-down time constant or other parameters related to the cavity ring-down signal. The model's performance can be evaluated by comparing the model's prediction results with the true values ​​in the test set. Common evaluation metrics include mean square error (MSE), mean absolute error (MAE), and accuracy. These metrics can help understand the model's performance and error on the test set.

[0102] Based on the test results, the neural network model can be adjusted and optimized to improve the accuracy and generalization ability of the model. Some specific optimization methods include:

[0103] Adjust the network structure: You can increase or decrease the number of network layers, adjust the number of nodes in each layer, change the activation function, etc. to improve the representation ability of the model.

[0104] Adjust the learning rate: The learning rate determines the degree of change of the model parameters at each update. The learning rate can be adjusted according to the performance of the model to obtain better convergence performance.

[0105] Regularization: By introducing regularization terms, such as L1 regularization or L2 regularization, the complexity of the model can be controlled, the risk of overfitting can be reduced, and the generalization ability of the model can be improved.

[0106] Batch Normalization: Using batch normalization technology can accelerate the convergence of the network, reduce the gradient vanishing problem, and improve the stability and generalization ability of the model.

[0107] S108. Process the measured cavity ring-down signal using the adjusted neural network model and output a corresponding ring-down time constant.

[0108] In the final step, the measured cavity ring-down signal is processed using the tuned and optimized neural network model. By inputting the measured signal into the neural network model, the model outputs the corresponding ring-down time constant as a result. This method can extract the ring-down time constant information from the measured cavity ring-down signal.

[0109] In practical applications, the cavity ring-down signal obtained by actual measurement is input into the adjusted and optimized neural network model for processing. The measured cavity ring-down signal can be an actual signal collected from experimental measurements or sensors.

[0110] The measured cavity ring-down signal is fed into the neural network model through a forward propagation process. The model processes the input signal using the learned weights and biases and outputs the corresponding ring-down time constant. After forward propagation, the neural network model generates an output: the ring-down time constant corresponding to the input measured cavity ring-down signal. This output can be used as an estimate of the ring-down time constant in the measured signal.

[0111] The trained and optimized model reads measured or simulated one-dimensional cavity ring-down signal data. The model identifies the data features, finds the corresponding label, and then outputs the raw cavity ring-down signal (noise-free) corresponding to the label. It also outputs the neural network filtering results and the ring-down time extraction results, which can be used to implement subsequent technologies such as gas concentration inversion and fluorescence lifetime analysis.

[0112] See Figure 2 , Figure 2 The figure shows the measured filtering results of the method of the present application. The statistically obtained cavity ring-down signal noise influence is significantly reduced, and the signal-to-noise ratio is improved by 27 times. It can be seen that the cavity ring-down signal filtering and ring-down time extraction method of the basic neural network are effective in filtering the low signal-to-noise ratio cavity ring-down signal.

[0113] This method uses a neural network to learn the cavity ring-down signal. In terms of signal filtering, the cavity ring-down signal filter proposed by this method shows the best performance compared with several widely used filtering algorithms. At the same time, compared with other deep learning models, this model has a more accurate and stable performance in processing attenuated signals.

[0114] In terms of ring-down time extraction, compared with the four typical methods in this technical field: nonlinear least squares fitting method, linear least squares fitting algorithm, linear regression summation method (continuous integration method) and frequency domain analysis method, the neural network model can directly identify the input cavity ring-down signal and output the corresponding ring-down time without the need for signal fitting and other operations performed by conventional ring-down time extraction. There are significant improvements in extraction precision, extraction accuracy and computational complexity.

[0115] This method uses a neural network to obtain high signal-to-noise ratio cavity ring-down spectrum filtering data while simultaneously obtaining information such as the signal amplitude and time constant of the measured cavity ring-down spectrum data, thereby improving the accuracy and efficiency of the cavity ring-down signal filtering technology.

[0116] This method simulates the noise and interference present in real-world applications by superimposing random noise on the simulated cavity ring-down signal. This helps improve the robustness of the model, making it more adaptable to real-world applications in different noise environments.

[0117] This method can be applied to the need for cavity ring-down information extraction in various optical fields, such as optical sensing, optical communications, and photonics. By adjusting the generation rules and model architecture, it can be customized for different application scenarios and problems, and can also be extended to similar signal processing problems in other fields.

[0118] The method provided in this application can use the output filtering results and ring-down time constant to perform gas concentration inversion and fluorescence lifetime analysis.

[0119] By extracting the ring-down time constant data, information related to the cavity ring-down signal can be obtained. The characteristics of the cavity ring-down signal are affected by gas concentration, so the extracted ring-down time constant can be used to invert the gas concentration using a known relationship model. Based on the existing relationship model between gas concentration and the ring-down time constant, an inversion model can be established. By inputting the extracted ring-down time constant, the inversion model can be used to calculate the corresponding gas concentration.

[0120] There is a certain correlation between the ring-down time constant and the fluorescence lifetime of a substance. The fluorescence lifetime is the duration of fluorescence photons emitted by a molecule after being excited, while the ring-down time constant provides the time scale of the cavity ring-down signal. Based on the known relationship between the fluorescence lifetime and the ring-down time constant, fluorescence lifetime analysis can be performed. The extracted ring-down time constant can be used to estimate or analyze the fluorescence lifetime of a substance.

[0121] Using this scheme for gas concentration inversion and fluorescence lifetime analysis has many advantages, such as:

[0122] Non-invasive: This approach is based on the extraction and analysis of cavity ring-down signals, without requiring direct contact or interference with the substance being measured. This makes it suitable for gas concentration inversion and fluorescence lifetime analysis of sensitive samples or samples that cannot be directly measured.

[0123] High Precision: By using deep learning models such as convolutional neural networks, this solution can extract key information such as the ring-down time constant from the cavity ring-down signal, enabling high-precision gas concentration inversion and fluorescence lifetime analysis. Neural networks have strong fitting and generalization capabilities, and can learn complex nonlinear relationships, improving the accuracy of analysis results.

[0124] Real-time performance: Because this solution utilizes neural networks for signal processing and analysis, it can achieve real-time or near-real-time gas concentration inversion and fluorescence lifetime analysis. This is extremely valuable for applications requiring immediate monitoring and control.

[0125] Adaptability: This solution can be adjusted and optimized based on the specific needs and sample characteristics of real-world applications. By adjusting the network structure, training parameters, and data preprocessing methods, it can adapt to the analysis requirements of different gases and fluorescence lifetime characteristics, improving the adaptability and generalization of the analysis.

[0126] Data diversity: Because this approach uses simulated cavity ring-down signals for training and testing, it can generate diverse datasets covering a wide range of gas concentrations and fluorescence lifetimes. This helps improve the robustness and adaptability of the model, enabling it to provide accurate and reliable analysis results under various conditions.

[0127] The above content describes the optical cavity ring-down information extraction method in this application. The following describes the optical cavity ring-down information extraction device and storage medium provided in this application:

[0128] See Figure 3 , the present application provides an optical cavity ring-down information extraction device, comprising:

[0129] 301 a time constant generating module, generating a first preset number of ring-down time constant data according to a preset generating rule;

[0130] 302 signal modeling module, inputting all the ring-down time constant data into a ring-down signal mathematical model, and generating a corresponding simulated cavity ring-down signal using the ring-down signal mathematical model;

[0131] 303 a noise adding module, superimposing random noise on each group of simulated cavity ring-down signals to obtain a second preset number of one-dimensional simulated cavity ring-down signals, where the second preset number is greater than the first preset number;

[0132] 304, a data partitioning module, which divides the one-dimensional simulated cavity ring-down signal into two sets, namely a training set and a test set;

[0133] 305CNN model building module, building the initial convolutional neural network CNN architecture;

[0134] 306 training module, using the training set to train the initial convolutional neural network CNN architecture to obtain a converged neural network model,

[0135] 307 testing and optimization module, inputting the test set into the converged neural network model for testing, and adjusting and optimizing the neural network model according to the test results;

[0136] The inference module 308 processes the measured cavity ring-down signal using the adjusted neural network model and outputs the corresponding ring-down time constant.

[0137] Optionally, the training module is specifically used to:

[0138] Using the corresponding simulated ring-down time as a label, a label set is generated, and each group of simulated cavity ring-down signals in the training set is prepared into a training data set;

[0139] Pack the training data set and the label set into npy format files;

[0140] Read the npy file and train the constructed initial convolutional neural network (CNN) architecture;

[0141] The training process includes:

[0142] The number of convolutional layers and fully connected layers, random inactivation rate, learning rate, and number of epoch layers are continuously adjusted until a converged neural network model is obtained.

[0143] Optional, testing and optimization modules are specifically used to:

[0144] Each set of simulated cavity ring-down signals in the test set is input into the converged neural network model for testing, so that the corresponding test ring-down time constant is output;

[0145] Comparing the test ring-down time constant with the simulated time constant corresponding to each group of simulated cavity ring-down signals in the test set to obtain a time constant comparison result;

[0146] The parameters of the converged neural network model are adjusted according to the time constant comparison result.

[0147] Optionally, a filtering module is provided in the neural network model, and after the measured cavity ring-down signal is processed by the adjusted neural network model, a filtering result of the cavity ring-down signal is also output.

[0148] Optionally, an inversion and analysis module 309 is also included for performing gas concentration inversion and fluorescence lifetime analysis using the output filtering results and ring-down time constant.

[0149] Optionally, the noise adding module is specifically used to:

[0150] Generate noise data through a pseudo-random number generation algorithm or collect real noise data;

[0151] Adjusting the intensity and spectral characteristics of the noise data to make it consistent with the actual environment;

[0152] The generated noise data is superimposed on the simulated cavity ring-down signal in a certain ratio.

[0153] See Figure 4 , the present application also provides an optical cavity ring-down information extraction device, comprising:

[0154] Processor 401, memory 402, input and output unit 403, bus 404;

[0155] The processor 401 is connected to the memory 402, the input and output unit 403 and the bus 404;

[0156] The memory 402 stores a program, and the processor 401 calls the program to execute any of the above methods.

[0157] The present application also relates to a computer-readable storage medium on which a program is stored, wherein when the program is run on a computer, the computer is caused to execute any of the above methods.

[0158] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0160] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0162] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk, and other media that can store program code.

Claims

1. A method for extracting optical cavity ring-down information, characterized in that: The method comprises: Generate a first preset number of ring-down time constant data according to a preset generation rule; Inputting all the ring-down time constant data into a ring-down signal mathematical model, and using the ring-down signal mathematical model to generate a corresponding simulated cavity ring-down signal; superimposing random noise on each group of simulated cavity ring-down signals to obtain a second preset number of one-dimensional simulated cavity ring-down signals, where the second preset number is greater than the first preset number; Dividing the one-dimensional simulated cavity ring-down signal into two sets, namely a training set and a test set; Build the initial convolutional neural network CNN architecture; Using the training set to train the initial convolutional neural network (CNN) architecture to obtain a converged neural network model; Inputting the test set into the converged neural network model for testing, and adjusting and optimizing the neural network model according to the test results; The adjusted neural network model is used to process the measured cavity ring-down signal and output the corresponding ring-down time constant.

2. The method for extracting cavity ring-down information according to claim 1, wherein: The initial convolutional neural network (CNN) architecture is trained using the training set to obtain a converged neural network model, including: Using the corresponding simulated ring-down time as a label, a label set is generated, and each group of simulated cavity ring-down signals in the training set is prepared into a training data set; Pack the training data set and the label set into npy format files; Read the npy file and train the constructed initial convolutional neural network (CNN) architecture; The training process includes: The number of convolutional layers and fully connected layers, random inactivation rate, learning rate, and number of epoch layers are continuously adjusted until a converged neural network model is obtained.

3. The method for extracting cavity ring-down information according to claim 1, wherein: Inputting the test set into the converged neural network model for testing, and adjusting and optimizing the neural network model according to the test results includes: Each set of simulated cavity ring-down signals in the test set is input into the converged neural network model for testing, so that the corresponding test ring-down time constant is output; Comparing the test ring-down time constant with the simulated time constant corresponding to each group of simulated cavity ring-down signals in the test set to obtain a time constant comparison result; The parameters of the converged neural network model are adjusted according to the time constant comparison result.

4. The method for extracting cavity ring-down information according to claim 1, wherein: When dividing the training set and the test set, the training set accounts for 70% and the test set accounts for 30%.

5. The method for extracting cavity ring-down information according to claim 1, wherein: The neural network model is provided with a filtering module. After the measured cavity ring-down signal is processed by the adjusted neural network model, a filtering result of the cavity ring-down signal is also output.

6. The method for extracting cavity ring-down information according to claim 1, wherein: The generation rules include: Conduct statistics and analysis on the existing ring-down time constant data to obtain a statistical model; A first preset number of ring-down time constant data is generated using the statistical model.

7. The method for extracting cavity ring-down information according to claim 1, wherein: The superimposing random noise on each group of simulated cavity ring-down signals comprises: Generate noise data through a pseudo-random number generation algorithm or collect real noise data; Adjusting the intensity and spectral characteristics of the noise data to make it consistent with the actual environment; The generated noise data is superimposed on the simulated cavity ring-down signal in a certain ratio.

8. An optical cavity ring-down information extraction device, characterized in that: include: A time constant generating module generates a first preset number of ring-down time constant data according to a preset generating rule; a signal modeling module, inputting all the ring-down time constant data into a ring-down signal mathematical model, and generating a corresponding simulated cavity ring-down signal using the ring-down signal mathematical model; a noise adding module for superimposing random noise on each group of simulated cavity ring-down signals to obtain a second preset number of one-dimensional simulated cavity ring-down signals, where the second preset number is greater than the first preset number; a data partitioning module, which divides the one-dimensional simulated cavity ring-down signal into two sets, namely a training set and a test set; CNN model building module, building the initial convolutional neural network CNN architecture; The training module uses the training set to train the initial convolutional neural network (CNN) architecture to obtain a converged neural network model. A testing and optimization module inputs the test set into the converged neural network model for testing, and adjusts and optimizes the neural network model according to the test results; The inference module processes the measured cavity ring-down signal using the adjusted neural network model and outputs the corresponding ring-down time constant.

9. An optical cavity ring-down information extraction device, characterized in that: The device comprises: processor, memory, input and output units, and buses; The processor is connected to the memory, the input and output unit, and the bus; The memory stores a program, and the processor calls the program to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a program stored thereon, wherein the program, when executed on a computer, performs the method according to any one of claims 1 to 7.

Images