Method and device for extracting structural parameters in optical scattering measurements

Abstract

The application relates to a method and device for extracting a structure parameter in optical scattering measurement, wherein the structure parameter is obtained through algorithm calculation; based on a machine learning algorithm, the structure parameters of a training set, a verification set and a test set corresponding to training set spectra, verification set spectra and test set spectra are respectively calculated; the structure parameters obtained through the algorithm and the machine learning algorithm are fused through a model fusion method for training, and finally, a target structure parameter is obtained. The method solves the deviation caused by system noise when the algorithm solves the structure parameter, and also solves the influence caused by random noise when the machine learning model solves the structure parameter, finally, the covariance coefficient, the mean absolute error and the maximum absolute error of the test result and the target result are improved, so that the extraction result of the structure parameter in the optical scattering measurement is more reasonable.

Description

Technical Field

[0001] This invention relates to the field of semiconductor materials technology, and specifically to a method and apparatus for extracting structural parameters in optical scattering measurements. Background Technology

[0002] In the semiconductor industry, the measurement and extraction of parameters of semiconductor nanostructures, including film thickness, linewidth, and sidewall angles, plays a crucial role in improving product quality and yield. Optical scattering measurement, also known as optical critical dimension (OCD) metrology, offers advantages over traditional atomic microscopes, scanning electron microscopes, and transmission electron microscopes in terms of speed, low cost, non-contact, non-destructive testing, and ease of integration, thus gaining widespread application in advanced manufacturing processes. Optical scattering measurement is not a "what you see is what you get" method; it requires extracting the parameters to be measured from the obtained spectrum. Ideally, a set of structural parameters should correspond to a fixed set of spectra. However, in actual measurements, the spectra obtained through optical scattering measurement are often affected by system noise and random noise, resulting in a certain deviation between the actual measured spectrum and the ideal spectrum, significantly impacting the extraction of correct structural parameters. Currently, the two main methods for optical scattering measurement are:

[0003] The LM algorithm can achieve good parameter extraction accuracy in noise-free conditions; however, its drawback is its reliance on the mapping relationships within the model. In practical applications, when the measured spectrum is significantly affected by noise, this algorithm cannot achieve accurate correspondences, leading to large deviations in the regression results for noisy spectra and thus affecting the matching of the results.

[0004] Machine learning methods utilize machine learning algorithms to eliminate the influence of system noise by eliminating biases in the model. However, they have few training samples and are greatly affected by random noise during the learning process, which affects the accuracy of the final structural parameter extraction. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method and apparatus for extracting structural parameters in optical scattering measurements. By using a model fusion method, the structural parameters obtained from the LM algorithm and machine learning algorithm are fused and trained to ultimately obtain the target structural parameters.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0007] In a first aspect, the present invention provides a method for extracting structural parameters in optical scattering measurements, comprising:

[0008] Acquire training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding true values ​​of structural parameters;

[0009] The first machine learning model, the second machine learning model, and the third machine learning model are trained using the training samples.

[0010] The structural parameters of the spectral data to be measured are extracted using the LM algorithm to obtain the first structural parameters corresponding to the spectral data to be measured.

[0011] The first machine learning model that has been trained is used to extract the structural parameters of the spectral data to be tested, so as to obtain the second structural parameters corresponding to the spectral data to be tested.

[0012] The first structural parameter is input into the pre-trained second machine learning model to obtain the first prediction result corresponding to the first structural parameter;

[0013] The second structural parameters corresponding to the spectral data to be tested are input into the trained third machine learning model to obtain the second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested.

[0014] The first prediction result and the second prediction result are weighted and summed to obtain the structural parameter result corresponding to the spectral data to be measured.

[0015] Furthermore, the training method for the first machine learning model includes:

[0016] The training samples are randomly divided into n equal parts, and one part is selected as the validation set, while the remaining parts are the training set.

[0017] Using the training set spectral data as input and the true values ​​of the structural parameters corresponding to the training set spectral data as output targets, the first machine learning model is trained independently n times to obtain n first machine learning models.

[0018] The n first machine learning models are validated using the spectral data of the validation set and the corresponding true values ​​of the structural parameters.

[0019] Furthermore, the pre-trained first machine learning model is used to extract structural parameters from the spectral data to be tested, thereby obtaining the second structural parameters corresponding to the spectral data to be tested. This includes: inputting the spectral data to be tested into n first machine learning models respectively to obtain n structural parameters corresponding to the spectral data to be tested.

[0020] Furthermore, the training method for the second machine learning model includes:

[0021] The LM algorithm is used to extract structural parameters from the spectral data of the training samples to obtain the first structural parameters corresponding to the spectral data of the training samples.

[0022] The second machine learning model is trained by taking the first structural parameter corresponding to the spectral data of the training sample as input and the true value of the structural parameter corresponding to the spectral data of the training sample as the output target.

[0023] Furthermore, the training method for the third machine learning model includes:

[0024] The validation set spectral data is used as input to n trained first machine learning models to predict n second structural parameters corresponding to the validation set spectral data.

[0025] Using the n second structural parameters corresponding to the validation set spectral data as input, and the true values ​​of the structural parameters corresponding to the validation set spectral data as the output target of the third machine learning model, the third machine learning model is trained to obtain n third machine learning models.

[0026] Furthermore, the second structural parameters corresponding to the spectral data to be tested are input into a pre-trained third machine learning model to obtain a second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested, including: inputting the second structural parameters corresponding to the spectral data to be tested into n pre-trained third machine learning models respectively to obtain n second prediction results corresponding to the second structural parameters corresponding to the spectral data to be tested.

[0027] Furthermore, the first prediction result and the second prediction result are weighted and summed to obtain the structural parameter result corresponding to the spectral data to be measured, including:

[0028] Result=α*A+β / n*(B_1+...+B_n)

[0029] Where A is the first prediction result, B_i is the second prediction result, 1≤i≤n, α+β=1, and more preferably, α=0.5, β=0.5.

[0030] In a second aspect, the present invention provides a device for extracting structural parameters in optical scattering measurements, comprising:

[0031] The data acquisition module acquires training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding true values ​​of structural parameters;

[0032] The training module uses the training samples to train the first machine learning model, the second machine learning model, and the third machine learning model;

[0033] The first extraction module uses the LM algorithm to extract the structural parameters of the spectral data to be tested, and obtains the first structural parameters corresponding to the spectral data to be tested.

[0034] The second extraction module uses the pre-trained first machine learning model to extract the structural parameters of the spectral data to be tested, and obtains the second structural parameters corresponding to the spectral data to be tested.

[0035] The first prediction module inputs the first structural parameters into the pre-trained second machine learning model to obtain the first prediction result corresponding to the first structural parameters;

[0036] The second prediction module inputs the second structural parameters corresponding to the spectral data to be tested into the pre-trained third machine learning model to obtain the second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested.

[0037] The parameter calculation module performs a weighted summation of the first prediction result and the second prediction result to obtain the structural parameter result corresponding to the spectral data to be measured.

[0038] Thirdly, the present invention provides an electronic device, comprising:

[0039] Memory, used to store computer software programs;

[0040] A processor is configured to read and execute the computer software program, thereby implementing a method for extracting structural parameters in optical scattering measurements as described in the first aspect of the present invention.

[0041] Fourthly, the present invention provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements a method for extracting structural parameters in optical scattering measurements as described in the first aspect of the present invention.

[0042] The beneficial effects of this invention are as follows: This invention calculates the structural parameters of the spectral data to be measured using both the traditional LM algorithm and machine learning algorithms. Finally, it fuses and trains the structural parameters obtained by the LM algorithm and machine learning algorithm using a model fusion method to obtain the target structural parameters. This solves the bias caused by system noise when the LM algorithm solves the structural parameters, and also solves the influence of random noise when the machine learning model solves the structural parameters. Ultimately, it improves the covariance coefficient, mean absolute error, and maximum absolute error between the test results and the target results, thus making the extraction results of structural parameters in optical scattering measurements more reasonable.

[0043] In summary, compared with the prior art, the method and apparatus for extracting structural parameters in optical scattering measurement provided by the present invention have the following advantages:

[0044] 1. This invention calculates the structural parameters of the spectral data under test using both the traditional LM algorithm and a machine learning algorithm. Finally, a model fusion method is used to fuse the structural parameters obtained from the LM and machine learning algorithms for training, ultimately yielding the target structural parameters. This solves the bias caused by system noise in the structural parameter calculation by the LM algorithm and the influence of random noise in the structural parameter calculation by the machine learning model. Ultimately, this improves the covariance coefficient, mean absolute error, and maximum absolute error between the test results and the target results, thus making the extracted structural parameters in optical scattering measurements more reasonable.

[0045] 2. This application uses n sets of training samples to perform n independent trainings, thereby obtaining n prediction results, which reduces the impact of random errors in the training samples on the prediction results. Attached Figure Description

[0046] Figure 1 This is a schematic flowchart of a method for extracting structural parameters in optical scattering measurement according to an embodiment of the present invention;

[0047] Figure 2 The sample structure diagram used in the experimental cases provided in the embodiments of the present invention;

[0048] Figure 3 This is a schematic diagram of the Mueller matrix ellipsomer.

[0049] Figure 4 R obtained through the LM algorithm 2 ;

[0050] Figure 5 This represents the deviation between the prediction result obtained through the LM algorithm and the target result.

[0051] Figure 6 R obtained through machine learning algorithms 2 ;

[0052] Figure 7 This represents the deviation between the predicted result obtained through machine learning algorithms and the target result.

[0053] Figure 8 R is a step in the fusion algorithm processing. 2 ;

[0054] Figure 9 The deviation between the prediction result and the target result at a certain step in the fusion algorithm processing.

[0055] Figure 10 The final R obtained through the fusion model 2 ;

[0056] Figure 11This represents the deviation between the final prediction obtained through the fusion model and the target result.

[0057] Figure 12 A schematic diagram of a device for extracting structural parameters in optical scattering measurement provided in an embodiment of the present invention;

[0058] Figure 13 A schematic diagram of an embodiment of the electronic device provided in this invention;

[0059] Figure 14 This is a schematic diagram of an embodiment of a computer-readable storage medium provided in this invention. Detailed Implementation

[0060] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0061] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0062] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0063] In practical optical scattering measurements, traditional structural parameter extraction methods often suffer from significant errors due to random and systemic noise. To obtain more reasonable structural parameters, new structural parameter extraction schemes are needed.

[0064] like Figure 1As shown, this embodiment of the invention provides a method for extracting structural parameters in optical scattering measurements, including:

[0065] S100, acquire training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding true values ​​of structural parameters.

[0066] S200, the first machine learning model, the second machine learning model, and the third machine learning model are trained using the training samples.

[0067] The first, second, and third machine learning models employ the same or different machine learning algorithms. Preferably, all three machine learning models are implemented using the ridge regression algorithm.

[0068] The training method for the first machine learning model includes:

[0069] The training samples are randomly divided into n equal parts, and one part is selected as the validation set, while the remaining parts are the training set.

[0070] Using the training set spectral data as input and the true values ​​of the structural parameters corresponding to the training set spectral data as output targets, the first machine learning model is trained independently n times to obtain n first machine learning models.

[0071] The n first machine learning models are validated using the spectral data of the validation set and the corresponding true values ​​of the structural parameters.

[0072] The training method for the second machine learning model includes:

[0073] The LM algorithm is used to extract structural parameters from the spectral data of the training samples to obtain the first structural parameters corresponding to the spectral data of the training samples.

[0074] The second machine learning model is trained by taking the first structural parameter corresponding to the spectral data of the training sample as input and the true value of the structural parameter corresponding to the spectral data of the training sample as the output target.

[0075] The training method for the third machine learning model includes:

[0076] The validation set spectral data is used as input to n trained first machine learning models to predict n second structural parameters corresponding to the validation set spectral data.

[0077] Using the n second structural parameters corresponding to the validation set spectral data as input, and the true values ​​of the structural parameters corresponding to the validation set spectral data as the output target of the third machine learning model, the third machine learning model is trained to obtain n third machine learning models.

[0078] S300, the LM algorithm is used to extract the structural parameters of the spectral data to be tested, and the first structural parameters corresponding to the spectral data to be tested are obtained.

[0079] To improve the convergence of the LM algorithm and make the structural parameters obtained by the LM algorithm more accurate, this embodiment of the invention also includes: optimizing the selection scheme of damping factor in the LM calculation process; optimizing delta_x used when calculating the partial derivative of floating parameters when calculating the Jacobian matrix in the LM calculation process; and optimizing the maximum and minimum value control scheme of floating parameters in the LM calculation process.

[0080] S400, the structure parameters of the spectral data to be tested are extracted using the pre-trained first machine learning model to obtain the second structure parameters corresponding to the spectral data to be tested; specifically, the spectral data to be tested is input into n first machine learning models respectively to obtain n structure parameters corresponding to the spectral data to be tested.

[0081] S500, the first structural parameters are input into the trained second machine learning model to obtain the first prediction result corresponding to the first structural parameters.

[0082] S600, the second structural parameters corresponding to the spectral data to be tested are input into a pre-trained third machine learning model to obtain a second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested. Specifically, the second structural parameters corresponding to the spectral data to be tested are input into n pre-trained third machine learning models to obtain n second prediction results corresponding to the second structural parameters corresponding to the spectral data to be tested.

[0083] S700, the first prediction result and the second prediction result are weighted and summed to obtain the structural parameter result corresponding to the spectral data to be measured.

[0084] Specifically, the weighted summation method is shown in the following formula:

[0085] Result=α*A+β / n*(B_1+...+Bn)

[0086] Where A is the first prediction result, B_i is the second prediction result, 1≤i≤n, α+β=1, and more preferably, α=0.5, β=0.5.

[0087] To provide a more detailed description of the present invention, the following specific examples will be used to further explain and illustrate the invention:

[0088] This case study uses a typical two-dimensional periodic structure as the sample, such as... Figure 2As shown, the sample structure consists of a silicon dioxide (SiO2) layer and a silicon nitride (SiN) layer filled within the silicon dioxide material. To reduce the complexity of the experimental model and improve the effectiveness of parameter extraction, some structural parameters that have little impact on the spectrum were set to fixed values. The main floating parameters were the top critical dimension (TCD), the height of the silicon nitride layer (H_SiN), the height of the silicon dioxide layer (H_OX), and the sidewall angle (SWA). The nominal dimensions of the sample were measured as follows: TCD = 56 nm, H_SiN = 31 nm, H_OX = 279 nm, and SWA = 87°.

[0089] The sample parameter to be extracted is H_OX, while other parameters H_SiN, TCD, and SWA fluctuate within a reasonable range. The ranges of the structural parameters TCD, H_SiN, H_OX, and SWA are 30-80 nm, 10-50 nm, 250 nm-300 nm, and 83-95°, respectively. Other experimental conditions are: measurement azimuth angle range of -90 to 90 degrees, incident angle fixed at 65 degrees, spectral wavelength range of 350 nm to 900 nm, and sampling at 1 nm intervals. A Mueller matrix ellipsometer is used to measure the spectrum corresponding to the sample, and its structure is shown below. Figure 3 As shown, the structural parameters of the samples are extracted using a combination of library search and the LM algorithm. Machine learning methods, on the other hand, extract structural parameters using the ridge regression algorithm. For clarity, we will denote the structural parameters extracted using the LM algorithm as the first structural parameter and the structural parameters extracted using the machine learning method as the second structural parameter.

[0090] It should be noted that the experimental objective is to maximize the covariance coefficient R between the test results and the target results. 2 The goal is to achieve an absolute bias of 0.9 or higher, and further minimize the values ​​of the mean absolute bias (mean_absolute_bias) and the maximum absolute bias (max_absolute_bias). The specific experimental steps are as follows:

[0091] First, the raw spectral data is acquired and then divided into training samples and test samples. The spectral data and corresponding structural parameters of the training samples and test samples are known.

[0092] Secondly, the LM algorithm is used to calculate and extract the structural parameters result_train_for and result_test_for corresponding to the training samples and test samples, respectively.

[0093] Third, the first machine learning model is used to calculate and extract the structural parameters result_train_inv and result_test_inv corresponding to the training samples and test samples.

[0094] Cross-validation is performed using training samples. The training samples are randomly divided into 5 equal parts. One part is selected as the validation set, and the remaining parts are the training set. The first machine learning model is trained 5 times to obtain the 5 structural parameters result_train_inv_i (1≤i≤5) corresponding to the training set and the 5 structural parameters result_val_inv_i (1≤i≤5) corresponding to the validation set.

[0095] After each training of the first machine learning model is completed, the test sample is input into the trained first machine learning model, and finally the five structural parameters result_test_inv_i corresponding to the test sample are obtained, where 1≤i≤5.

[0096] Fourth, the second machine learning model is trained using the structural parameter result_train_for and its corresponding true value result_train; the structural parameter result_test_for is used as the input to the trained second machine learning model to predict the first prediction result result_test_for_stack.

[0097] Fifth, the structural parameter result_val_inv_i (1≤i≤5) is used as the training set of the third machine learning model, and the true value of the structural parameter result_val_i (1≤i≤5) corresponding to result_val_inv_i (1≤i≤5) is used as the target set of the third machine learning model. The third machine learning model is trained to obtain 5 third machine learning models. The structural parameter result_test_inv_i (1≤i≤5) is used as the input of the 5 trained third machine learning models respectively, and the result_test_inv_stack_i (1≤i≤5) corresponding to the structural parameter result_test_inv_i (1≤i≤5) is predicted.

[0098] Finally, the final result is obtained by fixing the weights.

[0099] Compare the R values ​​obtained at each stage 2 The average error and maximum error were used to verify that the experimental method of model fusion can combine the shortcomings of positive and negative results to obtain the optimal model prediction results.

[0100] Figure 4 R is obtained directly from the forward model. 2 . Figure 5 This represents the deviation between the predicted result and the target result for each cell on the chip. As shown in the figure, the result predicted by the forward model, R... 2The value is 0.586, with a mean absolute error of 0.970 nm and a maximum absolute error of 3.005 nm. Its mean error is not large, but R... 2 The results are not ideal. This is because the measured spectra are significantly affected by noise, preventing them from fitting the spectra in the library well.

[0101] Figure 6 and Figure 7 R obtained through the inverse model 2 The deviation plot, its R 2 The R² value is 0.878, with a mean absolute error of 0.802 nm and a maximum absolute error of 2.616 nm. Compared to the results of the forward model, the performance is improved, but it still falls short of the expected R² value. 2 The value of 0.9 is somewhat different.

[0102] Next, the structural parameters are extracted using the method provided in this invention.

[0103] First, the collected raw spectral data is divided into training samples and test samples. This division can be randomized according to a certain ratio or done manually. In this embodiment, the raw spectral data is randomly divided in a 3:1 ratio. Then, the LM algorithm is used to calculate and extract the structural parameters result_train_for and result_test_for corresponding to the training and test samples, respectively. The structural parameter result_train_for is used as the training input of the ridge regression model, and the target result (i.e., the true value of the structural parameter) is used as the training objective to train the ridge regression model. Then, result_test_for is used as the test input and fed into the trained ridge regression model for prediction, and the prediction result is output. Figure 8 and Figure 9 R obtained through this method 2 The deviation plot. Its R... 2 The value is 0.917, with a mean absolute error of 1.718 nm and a maximum absolute error of 7.431 nm. Compared to the results of directly using the forward model, the results of the forward model learned through ridge regression show a significant improvement.

[0104] Secondly, using cross-validation, the training samples are randomly divided into 5 equal parts. Each time, 4 parts are selected as the training set and 1 part as the validation set. The ridge regression model is trained iteratively for 5 iterations. After each iteration, the trained model is used to predict the test samples. This method yields 5 training set results, 5 validation set results, and 5 test set results. The 5 validation set results are then used as the training set for a new ridge regression model, and the true values ​​of the structural parameters corresponding to the 5 validation sets are used as the target set for the new ridge regression model. This process trains the new ridge regression model, resulting in 5 trained ridge regression models. Finally, the 5 test set results are input into the corresponding trained ridge regression models for prediction. Compared to direct machine learning, the cross-validation results show a higher single-result R-value. 2 The difference is not significant.

[0105] Finally, the results of the LM algorithm learned by the ridge regression model and the results predicted by the second ridge regression model are each assigned a weight of 50%, that is, the weight of the LM algorithm result is 0.5, and the weight of each of the five ridge regression model learning results is 0.1, thus obtaining the final result. Figure 10 and Figure 11 This is the deviation plot obtained through this method. After fusing the results of the LM algorithm and the ridge regression algorithm, R0 2 The mean absolute error and maximum absolute error were 0.939 nm, 0.631 nm, and 1.592 nm, respectively. The mean absolute error and maximum absolute error were improved when extracted directly using machine learning methods. Furthermore, the mean absolute error and maximum absolute error were improved when extracted using the LM algorithm and the ridge regression algorithm. This indicates that the model integrates the results of the LM algorithm and machine learning algorithms, resulting in a more ideal outcome.

[0106] This invention trains a new method for extracting structural parameters by integrating structural parameters calculated by the LM algorithm and machine learning algorithm. It solves the bias caused by system noise when the LM algorithm solves structural parameters, and also solves the influence of random noise when the machine learning algorithm solves structural parameters. Ultimately, it improves the covariance coefficient, mean absolute error, and maximum absolute error between the measured result and the target result, thus making the extracted structural parameters in optical scattering measurement more reasonable.

[0107] This invention also provides a device for extracting structural parameters in optical scattering measurements, such as... Figure 12 As shown. The extraction device includes:

[0108] The data acquisition module acquires training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding true values ​​of structural parameters;

[0109] The training module uses the training samples to train the first machine learning model, the second machine learning model, and the third machine learning model;

[0110] The first extraction module uses the LM algorithm to extract the structural parameters of the spectral data to be tested, and obtains the first structural parameters corresponding to the spectral data to be tested.

[0111] The second extraction module uses the pre-trained first machine learning model to extract the structural parameters of the spectral data to be tested, and obtains the second structural parameters corresponding to the spectral data to be tested.

[0112] The first prediction module inputs the first structural parameters into the pre-trained second machine learning model to obtain the first prediction result corresponding to the first structural parameters;

[0113] The second prediction module inputs the second structural parameters corresponding to the spectral data to be tested into the pre-trained third machine learning model to obtain the second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested.

[0114] The parameter calculation module performs a weighted summation of the first prediction result and the second prediction result to obtain the structural parameter result corresponding to the spectral data to be measured.

[0115] Please see Figure 13 , Figure 13 This is a schematic diagram illustrating an embodiment of the electronic device provided in this invention. For example... Figure 13 As shown, this embodiment of the invention provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, it performs the following steps:

[0116] S100, acquire training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding structural parameter true values;

[0117] S200, the first machine learning model, the second machine learning model, and the third machine learning model are trained using the training samples;

[0118] S300, the LM algorithm is used to extract the structural parameters of the spectral data to be tested, and the first structural parameters corresponding to the spectral data to be tested are obtained;

[0119] S400, the pre-trained first machine learning model is used to extract the structural parameters of the spectral data to be tested, and the second structural parameters corresponding to the spectral data to be tested are obtained.

[0120] S500, the first structural parameters are input into the trained second machine learning model to obtain the first prediction result corresponding to the first structural parameters;

[0121] S600, the second structural parameters corresponding to the spectral data to be tested are input into the trained third machine learning model to obtain the second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested;

[0122] S700, the first prediction result and the second prediction result are weighted and summed to obtain the structural parameter result corresponding to the spectral data to be measured.

[0123] Please see Figure 14 , Figure 14 This is a schematic diagram illustrating an embodiment of a computer-readable storage medium provided by an embodiment of the present invention. For example... Figure 14 As shown, this embodiment provides a computer-readable storage medium 600, on which a computer program 611 is stored. When the computer program 611 is executed by a processor, it performs the following steps:

[0124] S100, acquire training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding structural parameter true values;

[0125] S200, the first machine learning model, the second machine learning model, and the third machine learning model are trained using the training samples;

[0126] S300, the LM algorithm is used to extract the structural parameters of the spectral data to be tested, and the first structural parameters corresponding to the spectral data to be tested are obtained;

[0127] S400, the pre-trained first machine learning model is used to extract the structural parameters of the spectral data to be tested, and the second structural parameters corresponding to the spectral data to be tested are obtained.

[0128] S500, the first structural parameters are input into the trained second machine learning model to obtain the first prediction result corresponding to the first structural parameters;

[0129] S600, the second structural parameters corresponding to the spectral data to be tested are input into the trained third machine learning model to obtain the second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested;

[0130] S700, the first prediction result and the second prediction result are weighted and summed to obtain the structural parameter result corresponding to the spectral data to be measured.

[0131] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0132] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting structural parameters in optical scattering measurements, characterized in that, The method includes: Acquire training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding true values ​​of structural parameters; The first machine learning model, the second machine learning model, and the third machine learning model are trained using the training samples. The structural parameters of the spectral data to be measured are extracted using the LM algorithm to obtain the first structural parameters corresponding to the spectral data to be measured. The first machine learning model that has been trained is used to extract the structural parameters of the spectral data to be tested, so as to obtain the second structural parameters corresponding to the spectral data to be tested. The first structural parameter is input into the pre-trained second machine learning model to obtain the first prediction result corresponding to the first structural parameter; The second structural parameters corresponding to the spectral data to be tested are input into the trained third machine learning model to obtain the second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested. The first prediction result and the second prediction result are weighted and summed to obtain the structural parameter result corresponding to the spectral data to be measured.

2. The method according to claim 1, characterized in that, The training method for the first machine learning model includes: The training samples are randomly divided into n equal parts, and one part is selected as the validation set, while the remaining parts are the training set. Using the training set spectral data as input and the true values ​​of the structural parameters corresponding to the training set spectral data as output targets, the first machine learning model is trained independently n times to obtain n first machine learning models. The n first machine learning models are validated using the spectral data of the validation set and the corresponding true values ​​of the structural parameters.

3. The method according to claim 2, characterized in that, The first machine learning model that has been trained is used to extract the structural parameters of the spectral data to be tested, so as to obtain the second structural parameters corresponding to the spectral data to be tested. This includes: inputting the spectral data to be tested into n first machine learning models respectively to obtain n structural parameters corresponding to the spectral data to be tested.

4. The method according to claim 1, characterized in that, The training method for the second machine learning model includes: The LM algorithm is used to extract structural parameters from the spectral data of the training samples to obtain the first structural parameters corresponding to the spectral data of the training samples. The second machine learning model is trained by taking the first structural parameter corresponding to the spectral data of the training sample as input and the true value of the structural parameter corresponding to the spectral data of the training sample as the output target.

5. The method according to claim 2, characterized in that, The training method for the third machine learning model includes: The validation set spectral data is used as input to n trained first machine learning models to predict n second structural parameters corresponding to the validation set spectral data. Using the n second structural parameters corresponding to the validation set spectral data as input, and the true values ​​of the structural parameters corresponding to the validation set spectral data as the output target of the third machine learning model, the third machine learning model is trained to obtain n third machine learning models.

6. The method according to claim 5, characterized in that, The second structural parameter corresponding to the spectral data to be tested is input into a pre-trained third machine learning model to obtain a second prediction result corresponding to the second structural parameter corresponding to the spectral data to be tested, including: inputting the second structural parameter corresponding to the spectral data to be tested into n pre-trained third machine learning models respectively to obtain n second prediction results corresponding to the second structural parameter corresponding to the spectral data to be tested.

7. The method according to claim 6, characterized in that, The first prediction result and the second prediction result are weighted and summed to obtain the structure parameter result corresponding to the spectral data to be measured, including: Where A is the first prediction result, B_i is the second prediction result, and 1≤i≤n. .

8. A device for extracting structural parameters in optical scattering measurements, characterized in that, include: The data acquisition module acquires training samples and spectral data to be tested, wherein the training samples include spectral data and their corresponding true values ​​of structural parameters; The training module uses the training samples to train the first machine learning model, the second machine learning model, and the third machine learning model; The first extraction module uses the LM algorithm to extract the structural parameters of the spectral data to be tested, and obtains the first structural parameters corresponding to the spectral data to be tested. The second extraction module uses the pre-trained first machine learning model to extract the structural parameters of the spectral data to be tested, and obtains the second structural parameters corresponding to the spectral data to be tested. The first prediction module inputs the first structural parameters into the pre-trained second machine learning model to obtain the first prediction result corresponding to the first structural parameters; The second prediction module inputs the second structural parameters corresponding to the spectral data to be tested into the pre-trained third machine learning model to obtain the second prediction result corresponding to the second structural parameters corresponding to the spectral data to be tested. The parameter calculation module performs a weighted summation of the first prediction result and the second prediction result to obtain the structural parameter result corresponding to the spectral data to be measured.

9. An electronic device, characterized in that, include: Memory, used to store computer software programs; A processor is configured to read and execute the computer software program, thereby implementing the method for extracting structural parameters in optical scattering measurements as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium, characterized in that, The storage medium stores a computer software program, which, when executed by a processor, implements a method for extracting structural parameters in optical scattering measurements as described in any one of claims 1-7.

Images