CN112037181A

CN112037181A - 2D SAXS atlas analysis model training method and device

Info

Publication number: CN112037181A
Application number: CN202010808889.1A
Authority: CN
Inventors: 朱才镇; 唐正; 刘会超; 于佳立; 徐坚
Original assignee: Shenzhen University
Current assignee: Shenzhen University
Priority date: 2020-08-12
Filing date: 2020-08-12
Publication date: 2020-12-04
Anticipated expiration: 2040-08-12
Also published as: CN112037181B

Abstract

The embodiment of the application provides a two-dimensional small-angle X-ray scattering 2D SAXS atlas analytical model training method and a device, wherein the atlas analytical model training method is applied to electronic equipment and comprises the following steps: obtaining N first 2D SAXS spectra, wherein N is a positive integer; inputting the N first 2D SAXS maps into a deep learning model to obtain first model parameters, wherein the deep learning model is built on the basis of a deep learning framework and a deep convolution artificial neural network; and configuring the first model parameter into the deep learning model to obtain a first map analysis model. By adopting the embodiment of the application, the high-flux 2D SAXS spectrum analysis model can be obtained, and when the high-flux 2D SAXS spectrum analysis model is used for analyzing the distribution parameters of a single 2D SAXS spectrum, the analysis accuracy is high, the analysis speed is high (microsecond level), and therefore the analysis requirement of massive experimental data is met.

Description

2D SAXS atlas analysis model training method and device

Technical Field

The application relates to the technical field of artificial intelligence and small-angle X-ray scattering intersection, in particular to a method and a device for training a 2D SAXS atlas analysis model.

Background

Small Angle X-ray Scattering (SAXS) refers to a coherent Scattering phenomenon of electrons to X-rays within a Small Angle range near an original X-ray beam, and spatial geometrical information such as the shape, size, distribution and content of microstructures (including micro-nano particles, pore structures and the like) within a nano-scale (1-1000nm) range in a material can be effectively detected by analyzing X-ray Scattering intensity fluctuation caused by electron density difference between a matrix and the microstructures in a sample. Meanwhile, the SAXS technology has the characteristics of high penetrability, simplicity in sample preparation, nondestructive detection, rapidness in test, good statistics, wide application range and the like, is an indispensable microscopic-mesoscale key analysis characterization means in the current high-throughput characterization technology of the novel material nanoscale microstructure, and is widely applied to various research fields of alloys, suspensions, emulsions, colloids, polymer solutions, natural macromolecules, liquid crystals, films, polyelectrolytes, composites, nanomaterials and the like.

SAXS is simple to test, but data analysis is quite complex. Currently, SAXS data parsing methods mainly include one-dimensional SAXS (1D SAXS) and two-dimensional SAXS (2D SAXS) methods. The 1D SAXS method is used for realizing the analysis of a microstructure by converting a 2D SAXS spectrum into a one-dimensional integral curve, but the 1D SAXS method is not suitable for an anisotropic system with a scatterer having a highly preferred orientation; although the existing partial 2D SAXS spectrum analysis method can effectively calculate the 2D SAXS spectrum of an anisotropic system, the 2D SAXS spectrum analysis method mainly adopts direct fitting of the experimental 2D SAXS spectrum, needs to construct a reasonable and effective mathematical model and carry out rapid theoretical two-dimensional scattering spectrum calculation, and often needs to iterate thousands or even tens of thousands of times, so that the analysis speed of a single 2D SAXS spectrum cannot meet the analysis requirement of massive experimental data.

Disclosure of Invention

The embodiment of the application provides a deep learning-based 2D SAXS atlas analytical model training method and device, the atlas analytical model training method is based on a deep learning framework and a deep convolution artificial neural network to establish a deep learning model, and is based on a random uniform distribution function to establish a sample database of the deep learning model, so that the sample database is wide in range, the deep learning model is trained and optimized according to the sample database to obtain a target atlas analytical model, the accuracy of the target atlas analytical model in analyzing the distribution parameters of a single 2D SAXS is high, the analysis speed is high (microsecond level), and the analysis requirements of mass experimental data can be met.

A first aspect of an embodiment of the present application provides a method for training a 2D SAXS atlas resolution model, which is applied to an electronic device, and the method includes: obtaining N first 2D SAXS spectra, wherein N is a positive integer; inputting the N first 2D SAXS spectra into a deep learning model to obtain first model parameters; the deep learning model is established based on a deep learning framework and a deep convolution artificial neural network; and configuring the first model parameter into the deep learning model to obtain a first map analysis model.

It can be seen that, in the embodiment, the atlas resolution model training device inputs a plurality of 2D SAXS atlases into the deep learning model determined based on the deep learning framework and the artificial neural network for training, and the finally obtained first atlas resolution model can rapidly and accurately resolve the 2D SAXS atlases with anisotropy, so that the resolving requirements of massive 2D SAXS atlases are met.

With reference to the first aspect, in one possible embodiment, the obtaining N first 2D SAXS maps includes: acquiring the N groups of first distribution parameters, and determining N first 2D SAXS maps according to the N groups of first distribution parameters, wherein the N groups of first distribution parameters correspond to the N first 2D SAXS maps one by one.

It can be seen that, in this embodiment, the N sets of first distribution parameters correspond to the N first 2D SAXS maps one to one, so that the first map analysis model obtained based on the training of the plurality of 2D SAXS maps can quickly and accurately analyze the 2D SAXS maps with anisotropy, and the analysis requirement of the massive 2D SAXS maps is met.

With reference to the first aspect, in one possible embodiment, the method further includes: obtaining M second 2D SAXS spectra, wherein each second 2D SAXS spectrum in the M second 2D SAXS spectra comprises a set of second distribution parameters, and M is a positive integer; acquiring K groups of second model parameters, and determining a group of target model parameters from the K groups of second model parameters according to the M second 2D SAXS spectrums and the first spectrum analysis model, wherein K is a positive integer; and configuring the group of target model parameters to the first atlas analytical model to obtain a target atlas analytical model.

It can be seen that, in this embodiment, the 2D SAXS atlas resolution model training apparatus determines a set of target model parameters from the K sets of second model parameters based on the M second 2D SAXS atlases and the first atlas resolution model, and configures the set of target model parameters into the first atlas resolution model to obtain the target atlas resolution model, so that the target atlas resolution model has higher accuracy and higher resolution speed than the first atlas resolution model.

With reference to the first aspect, in one possible implementation, the determining, according to the M second 2D SAXS maps and the first map resolution model, a set of target model parameters from the K sets of second model parameters includes: performing the following steps for each of the K sets of second model parameters to obtain K first graphs: configuring a group of currently processed second model parameters to the first atlas analytical model to obtain a second atlas analytical model; randomly calling P second 2D SAXS spectrums from the M second 2D SAXS spectrums and inputting the P second 2D SAXS spectrums into the second spectrum analysis model to obtain P second prediction distribution parameters of the second spectrum analysis model, and determining the first curve graph of the second spectrum analysis model according to the P second prediction distribution parameters and second distribution parameters corresponding to the P second 2D SAXS spectrums, wherein K is a positive integer, and P is a positive integer smaller than M. And comparing the K first graphs, determining a target first graph, and determining a group of second model parameters corresponding to the target first graph as the group of target model parameters.

It can be seen that, in this embodiment, for the current second atlas resolution model, the 2D SAXS atlas training device obtains the first graph of the current second atlas resolution model by randomly calling P second 2D SAXS atlases from the M second 2D SAXS atlases and inputting the P second 2D SAXS atlases into the current second atlas resolution model, performs the above operation K times to obtain K first graphs, compares the K first graphs, and determines a target first graph, the second model parameters corresponding to the target first graph are determined as the set of target model parameters, the set of target model parameters is the second model parameters most suitable for the deep learning model, the target atlas analytical model configured with the set of target model parameters is thus of higher accuracy and speed of analysis relative to the first atlas analytical model.

With reference to the first aspect, in one possible embodiment, the method further includes: obtaining H third 2D SAXS spectra, wherein each of the H third 2D SAXS spectra comprises a set of third distribution parameters, H being a positive integer; inputting the H third 2D SAXS spectra into the target spectrum analysis model to obtain H groups of third prediction distribution parameters, wherein the H groups of third prediction distribution parameters are in one-to-one correspondence with the H third 2D SAXS spectra; and comparing each group of third predicted distribution parameters in the H groups of third predicted distribution parameters with the corresponding third distribution parameters in the third 2D SAXS atlas to evaluate the target atlas analytical model.

It can be seen that, in the present embodiment, the atlas resolution model training apparatus evaluates the target atlas resolution model based on the test set to analyze the effectiveness and the application range of the target atlas resolution model.

With reference to the first aspect, in one possible embodiment, the method further includes: obtaining a fourth 2D SAXS map of the metal nano-rod; inputting the fourth 2D SAXS atlas to the target atlas analytical model to obtain the fourth distribution parameter; and comparing the fourth distribution parameter with a preset distribution parameter to verify the target atlas analytical model, wherein the preset distribution parameter is obtained by processing the metal nanorods based on a Transmission Electron Microscope (TEM).

As can be seen, in this embodiment, the atlas resolution model training device analyzes the fourth 2D SAXS atlas of the gold nanorods based on the target atlas resolution model, so as to obtain a fourth distribution parameter of the gold nanorods; and comparing the fourth distribution parameter of the gold nanorods with the preset distribution parameter of the gold nanorods determined based on a Transmission Electron Microscope (TEM), thereby effectively verifying the accuracy of the target atlas analytical model.

In a second aspect, an embodiment of the present application provides an apparatus for implementing the method in the first aspect, including: an obtaining module, configured to obtain N first 2D SAXS spectra, where N is a positive integer; the input module is used for inputting the N first 2D SAXS maps into a deep learning model to obtain first model parameters, wherein the deep learning model is established on the basis of a deep learning framework and a deep convolution artificial neural network; and the configuration module is used for configuring the first model parameters into the deep learning model to obtain a first map analysis model.

With reference to the second aspect, in a possible implementation manner, the obtaining module is further configured to: acquiring N groups of first distribution parameters, and determining N first 2D SAXS maps according to the N groups of first distribution parameters, wherein the N groups of first distribution parameters correspond to the N first 2D SAXS maps one by one.

In a third aspect, an embodiment of the present application provides an electronic device, including: a processor and a memory; the processor is connected to a memory, wherein the memory is configured to store program code and the processor is configured to invoke the program code to perform the method according to any of the above first aspects.

In a fourth aspect, the present application provides a computer storage medium storing a computer program comprising program instructions that, when executed by a processor, perform the method according to any one of the first aspect.

Drawings

The drawings used in the embodiments of the present application are described below.

Fig. 1 is a flowchart of a method for training a 2D SAXS atlas resolution model according to an embodiment of the present disclosure;

FIG. 2 is a schematic diagram of another training method for a 2D SAXS atlas resolution model provided in an embodiment of the present application;

fig. 3 is a diagram illustrating an effect of short-axis mean parameter analysis on scatterers by the target atlas analysis model provided in the embodiment of the present application;

FIG. 4 is a diagram illustrating the effect of short-axis variance parameter analysis on scatterers by the target atlas analytical model provided in the present application;

fig. 5 is an effect diagram of long-axis mean parameter analysis performed on scatterers by the target spectrum analysis model according to the embodiment of the present application;

fig. 6 is a diagram illustrating an effect of long-axis variance parameter analysis on scatterers by the target atlas analytical model according to the embodiment of the present application;

fig. 7 is a diagram illustrating an effect of a zenith angle parameter analysis performed on a scatterer by the target spectrum analysis model according to the embodiment of the present application;

fig. 8 is a diagram illustrating an effect of the target atlas resolution model on vertex angle variance parameter resolution of scatterers according to the embodiment of the present application;

fig. 9 is a schematic structural diagram of a training apparatus of a 2D SAXS atlas resolution model according to an embodiment of the present application;

FIG. 10 is a schematic structural diagram of another training apparatus for a 2D SAXS atlas resolution model according to an embodiment of the present application;

fig. 11 is a schematic structural diagram of an electronic device for training a 2D SAXS atlas resolution model according to an embodiment of the present application.

Detailed Description

The embodiment of the application provides a method and a device for training a 2D SAXS atlas analytical model based on deep learning, which are applied to electronic equipment.

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

Referring to fig. 1, fig. 1 is a flowchart illustrating a method for training a 2D SAXS atlas resolution model according to the present application; the method for training the atlas resolution model comprises the following steps:

step S101: obtaining N first 2D SAXS spectra, wherein N is a positive integer.

Specifically, the N groups of first distribution parameters are generated according to a random uniform distribution function, and N first 2D SAXS maps are determined according to the N groups of first distribution parameters, where the N first 2D SAXS maps constitute a training set of the deep learning model, where the N first 2D SAXS maps correspond to the N groups of first distribution parameters one to one, each first 2D SAXS map in the N first 2D SAXS maps corresponds to a plurality of scatterers, each scatterer in the plurality of scatterers has a specific shape, size, and angle, and the first distribution parameter is a group of parameters describing shapes, sizes, and angles of the plurality of scatterers from a statistical perspective, that is, each first distribution parameter in the N groups of first distribution parameters corresponds to the plurality of scatterers.

For each of the N sets of first distribution parameters, respectively performing the following steps to obtain the N first 2D SAXS maps: the method comprises the steps of inputting size distribution parameters and angle distribution parameters in a set of currently processed first distribution parameters into corresponding distribution functions respectively to obtain size distribution and angle distribution of a plurality of scatterers corresponding to the set of first distribution parameters, substituting shapes, size distribution and angle distribution of the scatterers into a first calculation model, and obtaining a first 2D SAXS map of the scatterers based on GPU multi-thread parallel calculation, wherein the types and the number of the distribution functions are specifically determined by the shapes, the sizes and the angle distribution parameters in the set of first distribution parameters, and no specific limitation is made here.

For example, if the scatterers described by the set of first distribution parameters are ellipsoids, the size of each scatterer in the set of first distribution parameters includes a long axis and a short axis, at this time, the size distribution parameters in the set of first distribution parameters include a long axis parameter and a short axis parameter, the long axis parameter includes a long axis mean and a long axis variance, and the short axis parameter includes a short axis mean and a short axis variance; the distribution functions corresponding to the long axis and the short axis are lognormal distribution functions, the long axis parameters and the short axis parameters are respectively substituted into the lognormal distribution functions to obtain long axis distribution and short axis distribution of the scatterers, and the long axis distribution and the short axis distribution form size distribution of the scatterers; the angle of each scatterer in the plurality of scatterers comprises a zenith angle and a space azimuth angle, at this time, the angle distribution parameter in the group of first distribution parameters comprises a zenith angle parameter, and the zenith angle parameter comprises a zenith angle mean value and a zenith angle variance; the distribution function corresponding to the zenith angle is a circular distribution function, and the zenith angle mean value and the zenith angle variance are substituted into the circular distribution function to obtain zenith angle distribution of the scatterers; and the space azimuth angle is randomly generated according to the random uniform distribution function, and is in accordance with random uniform distribution without distribution parameters.

It will be understood that the number and type of the first distribution parameter is determined by the shape of the scatterers it describes, the distribution function is determined according to the type of the first distribution parameter, and when the scatterers are ellipsoids, the type of the first distribution parameter describing the scatterers comprises: major axis mean, major axis variance, minor axis mean, minor axis variance, zenith angle mean, and zenith angle variance; it should be understood that the shape of the scatterers is merely illustrated as an ellipsoid, and when the shape of the scatterers is a non-ellipsoid, the type and the number of the first distribution parameters describing the scatterers are different, so that the distribution functions corresponding to the first distribution parameters are also different, and the shape of the scatterers is not particularly limited herein.

Wherein, the random uniform distribution function can be a random.rand function in a numerical calculation function library NUMPY or a function similar to the random.rand function in the NUMPY function in other function libraries; wherein each of the N sets of first distribution parameters includes a shape, a size, and an angle of the scatterer.

The first calculation model is used for calculating a 2D SAXS spectrum corresponding to scatterers according to distribution parameters of the scatterers, the first calculation model comprises a double model and a single model, the double model is suitable for a scattering system with orientation scatterers and isotropic scatterers, and fitting calculation is carried out by adopting a spherical body model and an ellipsoid model to obtain the scattering system; the single model is suitable for a scattering system only with orientation scatterers, and is obtained by fitting calculation through an ellipsoid model, the specific type of the first calculation model is determined according to actual conditions, and no specific limitation is made here.

Step S102: inputting the N first 2D SAXS maps into a deep learning model to obtain first model parameters.

Specifically, according to a deep learning framework and a deep convolution artificial neural network, an input layer, a convolution layer, a pooling layer, a full-link layer and an output layer are respectively established, and the deep learning model is established; the N first 2D SAXS spectrums form a training set of the deep learning model, a certain number of first 2D SAXS spectrums are called randomly from the N first 2D SAXS spectrums in a circulating mode and input into the deep learning model, iteration is carried out on first model parameters of the deep learning model until the absolute value of the difference value between the first model parameters obtained in the current iteration and the first model parameters obtained in the last iteration is smaller than or equal to a threshold value A, the iteration process of the first model parameters is ended, and the first model parameters obtained in the current iteration are determined to be the first model parameters; the deep learning framework can be TensorFlow or PyTorch or other functionally similar deep learning framework, wherein N is a positive integer and A is a positive number.

Wherein, the single iteration process of the first model parameter specifically includes: randomly calling L first 2D SAXS spectrums from the N first 2D SAXS spectrums, and performing convolution, pooling and full-connection layer operation on the L first 2D SAXS spectrums and first model parameters obtained by the last iteration to obtain L groups of first prediction distribution parameters; determining a descending gradient of a first model parameter obtained by the current iteration according to the L groups of first predicted distribution parameters and the L groups of first distribution parameters corresponding to the L first 2D SAXS maps; if the first model parameter is iterated for the first time, determining an initial value of the first model parameter based on the random uniform distribution function; determining a first model parameter obtained by the current iteration according to the descending gradient and a back propagation algorithm (BP algorithm); when the absolute value of the difference value between the first model parameter obtained by the current iteration and the first model parameter obtained by the last iteration is less than or equal to a threshold value A, ending the iteration process of the first model parameter, and determining the first model parameter obtained by the current iteration as the first model parameter; otherwise, entering the next iteration process of the first model parameter; wherein L is a positive integer less than N.

The first model parameters include weights and offsets of the neural network, the number of the weights and the offsets is determined according to parameters of the deep convolutional artificial neural network, the parameters of the deep convolutional artificial neural network are determined according to actual conditions, and the parameters are not specifically limited herein, wherein the parameters of the neural network include the number of layers of the neural network and the number of neurons.

Step S103: and configuring the first model parameter into the deep learning model to obtain a first map analysis model.

Specifically, the first model parameter obtained through iteration in step S102 is configured in the deep learning model, so as to obtain the first atlas analytical model.

It can be seen that, in this embodiment, the 2D SAXS spectrum analysis model training method determines the first distribution parameters that are uniformly distributed through the random uniform distribution function, so that the first 2D SAXS spectra of the multiple scatterers are trained and optimized by using the first 2D SAXS spectra on the basis of a deep learning framework and an artificial neural network, so as to obtain a first spectrum analysis model, and the first spectrum analysis model can rapidly and preparatively analyze the 2D SAXS spectra with anisotropy, which can meet the analysis requirement of a large amount of 2D SAXS spectra.

In a possible embodiment, the method for training the 2D SAXS spectrum analysis model further includes: obtaining M second 2D SAXS spectra, wherein each of the M second 2D SAXS spectra comprises a set of second distribution parameters; acquiring K groups of second model parameters, and determining a group of target model parameters from the K groups of second model parameters according to the M second 2D SAXS spectrums and the first spectrum analysis model; and configuring the group of target model parameters to the first atlas analytical model to obtain a target atlas analytical model, wherein M and K are positive integers.

Specifically, the obtaining manner of the M second 2D SAXS maps is the same as the obtaining manner of the N first 2D SAXS maps in step S101, and details are not repeated here, where the M second 2D SAXS maps constitute a verification set of the deep learning model; the K groups of second model parameters are determined according to the random uniform distribution function, and each group of second model parameters in the K groups of second model parameters comprises a learning rate, a convolutional layer parameter, an optimizer, an activation function, the number of neural network layers and the number of neurons.

Executing the following steps aiming at each group of the K groups of second model parameters to obtain K first curve graphs: configuring a group of second model parameters currently processed to the first atlas analytical model to obtain a second atlas analytical model, randomly calling P second 2D SAXS atlases from the verification set to input the second atlas analytical model, determining P second prediction distribution parameters of the second atlas analytical model through convolution, pooling and full connection layer operation, and obtaining the first curve graph of the second atlas analytical model according to the P second prediction distribution parameters and second distribution parameters corresponding to the P second 2D SAXS atlases.

It is easy to understand that the K groups of second model parameters correspond to the K first curve graphs one by one, and the K first curve graphs correspond to K second map analytical models; and comparing the K first curve graphs to determine a target first curve graph, determining a group of second model parameters corresponding to the target first curve graph as the group of target model parameters, and configuring the group of target model parameters to the first atlas resolution model to obtain the target atlas resolution model, wherein K is a positive integer, and P is a positive integer smaller than M.

It can be seen that, in this embodiment, the 2D SAXS atlas resolution model training method determines a set of target model parameters from the K sets of second model parameters based on the validation set and the first atlas resolution model, and configures the set of target model parameters into the first atlas resolution model to obtain the target atlas resolution model, so that the target atlas resolution model has higher accuracy and higher resolution speed than the first atlas resolution model.

In a possible embodiment, the method for training the 2D SAXS spectrum analysis model further includes: obtaining H third 2D SAXS spectra, the H third 2D SAXS spectra constituting a test set of the deep learning model, wherein each of the H third 2D SAXS spectra comprises a set of third distribution parameters; inputting the H third 2D SAXS spectrums into the target spectrum analysis model, and determining H third prediction distribution parameters through convolution, pooling and full-connection operation, wherein the H groups of third prediction distribution parameters correspond to the H third 2D SAXS spectrums one to one; and comparing each of the H groups of third predicted distribution parameters with the corresponding third distribution parameter in the third 2D SAXS spectrum to evaluate the target spectrum analysis model, wherein H is a positive integer.

It can be seen that, in the present embodiment, the 2D SAXS atlas analytical model training method evaluates the target atlas analytical model based on the test set, so as to analyze the effectiveness and the application range of the target atlas analytical model.

In a possible embodiment, the method for training the 2D SAXS spectrum analysis model further includes: obtaining a fourth 2D SAXS map of the metal nano-rod; inputting the fourth 2D SAXS spectrum of the metal nanorod into the target spectrum analysis model to obtain a fourth distribution parameter of the metal nanorod; and comparing the fourth distribution parameter of the metal nanorods with a preset distribution parameter to verify the target atlas analytical model, wherein the preset distribution parameter is obtained by processing the metal nanorods based on a Transmission Electron Microscope (TEM).

Specifically, the type of the metal used by the metal nanorods can be selected according to the actual situation, and the details are described here by taking gold as an example: firstly, taking gold nanorods as scatterers, mixing the gold nanorods into super-high-elasticity polyurethane, and carrying out orientation through polymer drafting to prepare an orientation-controllable standard sample; then carrying out in-situ SAXS test on the gold nanorods by using a synchrotron radiation small-angle X-ray scattering experiment station to obtain a fourth 2D SAXS spectrum; inputting the fourth 2D SAXS spectrum into the target spectrum analysis model, and determining a fourth distribution parameter of the metal nanorod according to convolution, pooling and full-connection layer operation; and finally, comparing the fourth distribution parameter of the metal nano rod with the distribution parameter of the metal nano rod obtained based on a Transmission Electron Microscope (TEM) so as to verify the target atlas analytical model.

It is understood that, the types of the selected metals are different, and the process of preparing the standard sample by using the metal nanorods as scatterers, the materials used in the preparation process, and the process of the in-situ SAXS test in the experimental process also vary accordingly, and are not limited specifically herein.

As can be seen, in this embodiment, the 2D SAXS spectrum analysis model training method analyzes the fourth 2D SAXS spectrum of the gold nanorods based on the target spectrum analysis model, so as to obtain a fourth distribution parameter of the gold nanorods; and comparing the fourth distribution parameter of the gold nanorods with the preset distribution parameter of the gold nanorods determined based on a Transmission Electron Microscope (TEM), thereby effectively verifying the accuracy of the target atlas analytical model.

Referring to fig. 2, fig. 2 is a schematic diagram of another method for training a 2D SAXS atlas resolution model provided in the present application, and as shown in fig. 2, the method for training the atlas resolution model includes the following steps:

step S201: and establishing a sample database, wherein the sample database comprises a training set, a verification set and a test set.

Specifically, the sample database of the deep learning model includes the training set, the verification set, and the test set. The training set is composed of N first 2D SAXS maps in the embodiment of the method of fig. 1, each of the N first 2D SAXS maps comprising a set of first distribution parameters; the validation set is composed of M second 2D SAXS maps in the method embodiment of fig. 1, each of the M second 2D SAXS maps comprising a set of second distribution parameters; the test set consists of H second 2D SAXS profiles in the method embodiment of fig. 1, each of the H third 2D SAXS profiles comprising a set of third distribution parameters. Each of the N first 2D SAXS maps corresponds to a plurality of scatterers, each of the scatterers has a shape, a size, and an angle specific to the scatterer, and the set of first distribution parameters is a set of parameters that statistically describes the shape, the size, and the angle of the scatterer, that is, each of the N sets of first distribution parameters corresponds to the scatterers. The properties of the 2D SAXS atlases in the verification set and the test set are consistent with the properties of the first 2D SAXS atlases in the training set, and are not described herein again.

Taking the training set as an example, describing the establishment process of the sample database in detail, first, generating N sets of first distribution parameters in the training set according to the random uniform distribution function, and for each set of first distribution parameters in the N sets of first distribution parameters, respectively performing the following steps to obtain the N first 2D SAXS atlases: respectively inputting the size distribution parameters and the angle distribution parameters in a group of currently processed first distribution parameters into corresponding distribution functions to obtain the size distribution and the angle distribution of the scatterers corresponding to the group of first distribution parameters, substituting the shapes, the size distribution and the angle distribution of the scatterers into a theoretical 2D SAXS model, and performing multi-thread parallel calculation based on a GPU to obtain a first 2D SAXS atlas of the scatterers. The obtaining methods of the 2D SAXS maps in the verification set and the test set are consistent with the obtaining process of the first 2D SAXS map in the training set, and details are not repeated here.

Wherein the theoretical 2D SAXS model corresponds to the first calculation model in the embodiment of the method of fig. 1, and the random uniform distribution function and the theoretical 2D SAXS model are described in the embodiment of the method of fig. 1, and are not described herein again; the type and number of the distribution function are determined by the shape, size, and angle distribution parameters in the distribution parameters corresponding to each 2D SAXS spectrum in the sample database, which is not specifically limited herein, and the method embodiment shown in fig. 1 exemplifies the shapes of the scatterers as ellipsoids, which details the group of first distribution parameters corresponding to the scatterers under this condition, and the specific distribution function corresponding to the group of first distribution parameters, and are not described herein again.

Step S202: and establishing a deep learning model.

Specifically, according to a deep learning framework and a deep convolution artificial neural network, an input layer, a convolution layer, a pooling layer, a full-link layer and an output layer are respectively established, and the deep learning model is established; the deep learning model comprises the first model parameters to be determined and the target model parameters, the deep learning model is trained and determined by the first model parameters based on the sample database, and the deep learning model is optimized and determined by the target model parameters based on the sample database.

Wherein, the deep learning framework can be TensorFlow or PyTorch or other similar-function deep learning frameworks, which is not limited herein; the first model parameters comprise weights and offsets of the neural network, the number of the weights and the offsets is determined according to parameters of the deep convolution artificial neural network, the parameters of the deep convolution artificial neural network are determined according to actual conditions and are not specifically limited, and the parameters of the neural network comprise the number of layers of the neural network and the number of neurons; the second model parameters include learning rate, convolutional layer parameters, optimizer, activation function, number of neural network layers, and number of neurons.

Step S203: and training the deep learning model by using the training set, determining a first model parameter, configuring the first model parameter to the deep learning model, and obtaining the first atlas analytical model.

Specifically, a certain number of first 2D SAXS maps are called randomly from the training set in a loop and input into the deep learning model, so as to iterate first model parameters of the deep learning model, and when an absolute value of a difference value between a first model parameter obtained by current iteration and a first model parameter obtained by last iteration is smaller than or equal to a threshold value a, the iteration process of the first model parameters is ended, and the first model parameter obtained by current iteration is determined as the first model parameter; and A is a positive number.

Wherein, the single iteration process of the first model parameter specifically includes: randomly calling L first 2D SAXS spectrums from the training set, and performing convolution, pooling and full-connection layer operation on the L first 2D SAXS spectrums and the first model parameters obtained by the last iteration to obtain L groups of first prediction distribution parameters; determining a descending gradient of a first model parameter obtained by the current iteration according to the L groups of first predicted distribution parameters and the L groups of first distribution parameters corresponding to the L first 2D SAXS maps; if the first model parameter is iterated for the first time, determining an initial value of the first model parameter based on the random uniform distribution function; determining a first model parameter obtained by the current iteration according to the descending gradient and a back propagation algorithm (BP algorithm); when the absolute value of the difference value between the first model parameter obtained by the current iteration and the first model parameter obtained by the last iteration is less than or equal to a threshold value A, ending the iteration process of the first model parameter, and determining the first model parameter obtained by the current iteration as the first model parameter; otherwise, entering the next iteration process of the first model parameter; wherein L is a positive integer less than N.

Step S204: and optimizing the first atlas analytical model by using the verification set, determining a group of target model parameters, and configuring the group of target model parameters to the first atlas analytical model to obtain a target atlas analytical model.

In particular, the validation set is composed of the M second 2D SAXS maps, each of the M second 2D SAXS maps comprising a set of second distribution parameters; and determining K groups of second model parameters according to the random uniform distribution function, wherein each group of second model parameters in the K groups of second model parameters comprises a learning rate, a convolutional layer parameter, an optimizer, an activation function, the number of neural network layers and the number of neurons.

Step S205: and evaluating the target map analytical model by using the test set.

Specifically, the test set is composed of the H second 2D SAXS maps, each of the H third 2D SAXS maps includes a set of third distribution parameters, the H third 2D SAXS maps are input into the target map analysis model, H third predicted distribution parameters are determined through convolution, pooling and full-join operations, and the H set of third predicted distribution parameters are in one-to-one correspondence with the H third 2D SAXS maps; and comparing each of the H groups of third predicted distribution parameters with the corresponding third distribution parameter in the third 2D SAXS spectrum to evaluate the target spectrum analysis model, wherein H is a positive integer.

Step S206: and acquiring a fourth 2D SAXS spectrum of the metal nanorod, and inputting the fourth 2D SAXS spectrum into the target spectrum analysis model to perform experimental verification on the target spectrum analysis model.

Referring to fig. 3-8, fig. 3-8 are graphs obtained by analyzing the 2D SAXS spectrum of an ellipsoid by using a spectrum analysis model trained by the spectrum analysis model training method provided in the present application, and a scatter diagram of the predicted distribution parameters of the ellipsoid and the actual distribution parameters of the scatterer is obtained, and statistical distribution parameters and analysis time are given, where the distribution parameters of the ellipsoid include a major axis mean, a major axis variance, a minor axis mean, a minor axis variance, a zenith angle mean, and a zenith angle variance.

Referring to fig. 3, fig. 3 is a scatter diagram obtained by analyzing short axis Mean parameters of 200 400 × 400 pixel 2D SAXS images according to the target atlas analysis model and comparing the predicted values of the analyzed short axis Mean parameters with the labeled values of the ellipsoid (i.e., actual short axis Mean parameters), where the statistical Root Mean Square Error (RMSE) of the short axis Mean parameters is 0.190nm, the Mean Absolute Error (MAE) is 0.146nm, the R2 value is 0.9997, and the total analysis time is 304 μ s.

Referring to fig. 4, in fig. 4, a short-axis variance parameter analysis is performed on 200 400 × 400 pixel 2D SAXS images according to the target atlas resolution model, and a scatter diagram is obtained by comparing a predicted value of the short-axis variance parameter obtained through the analysis with a labeled value of the ellipsoid (i.e., an actual short-axis variance parameter), where a statistical root mean square error of the short-axis variance parameter is 1.656nm, an average absolute error is 1.204nm, an R2 value is 0.9991, and a total analysis time is 310 microseconds.

Referring to fig. 5, fig. 5 is a scatter diagram obtained by analyzing a long axis mean parameter of 200 400 × 400 pixel 2D SAXS images according to the target atlas analysis model and comparing a predicted value of the long axis mean parameter obtained by the analysis with a labeled value of the ellipsoid (i.e., an actual long axis mean parameter), where a statistical root mean square error of the long axis mean parameter is 0.764nm, an average absolute error is 0.594nm, an R2 value is 0.9998, and a total analysis time is 306 microseconds.

Referring to fig. 6, fig. 6 is a scatter diagram obtained by analyzing a long axis variance parameter of 200 400 × 400 pixel 2D SAXS images according to the target atlas analysis model and comparing a predicted value of the long axis variance parameter obtained by the analysis with a mark value of the ellipsoid (i.e., an actual long axis variance parameter), where a statistical root mean square error of the long axis variance parameter is 7.729nm, an average absolute error is 5.952nm, an R2 value is 0.9799, and a total analysis time is 340 microseconds.

Referring to fig. 7, fig. 7 is a scatter diagram obtained by analyzing vertex-angle mean parameters of 200 400 × 400 pixel 2D SAXS images according to the target atlas analysis model and comparing predicted values of the vertex-angle mean parameters obtained by the analysis with labeled values of the ellipsoid (i.e., actual vertex-angle mean parameters), where statistical root mean square errors of the vertex-angle mean parameters are 0.550 °, average absolute errors are 0.384 °, R2 values are 0.9995, and total analysis time is 300 microseconds.

Referring to fig. 8, fig. 8 is a scatter diagram obtained by analyzing zenith angle variance parameters of 200 400 × 400 pixel 2D SAXS images according to the target atlas analysis model and comparing the predicted values of the zenith angle variance parameters obtained by the analysis with the labeled values of the ellipsoid (i.e., actual zenith angle variance parameters), where the statistical root mean square error of the zenith angle variance parameters is 1.231 °, the average absolute error is 0.905 °, the value of R2 is 0.9982, and the total analysis time is 301 μ sec.

In summary, the 2D SAXS spectrum analysis model training method provided by the application establishes the deep learning model based on a deep learning framework and a deep convolution artificial neural network, and establishes the sample database of the deep learning model based on the random uniform distribution function, so that the sample database is wide in range, the deep learning model is trained and optimized according to the sample database to obtain the target spectrum analysis model, the accuracy of the target spectrum analysis model in analyzing the distribution parameters of the 2D SAXS of the scatterer is high, the analysis speed is high, and the analysis requirement of massive experimental data can be met.

Referring to fig. 9, fig. 9 is a schematic diagram of a training apparatus for a 2D SAXS atlas resolution model provided in the present application; as shown in fig. 9, the apparatus for training a atlas resolution model includes the following modules.

An obtaining module 901, configured to obtain N first 2D SAXS spectra, where N is a positive integer;

an input module 902, configured to input the N first 2D SAXS atlases into the deep learning model to obtain first model parameters, and configure the first model parameters into the deep learning model to obtain a first atlas resolution model, where the deep learning model is built based on a deep learning framework and a deep convolutional artificial neural network.

A configuration module 903, configured to configure the first model parameter to the deep learning model, so as to obtain a first atlas analysis model.

It can be seen that, in the embodiment, the 2D SAXS spectrum analysis model training device inputs a plurality of 2D SAXS spectra into the deep learning model determined based on the deep learning framework and the artificial neural network for training, and the finally obtained first spectrum analysis model can rapidly and accurately analyze the 2D SAXS spectra with anisotropy, so as to meet the analysis requirement of mass 2D SAXS spectra.

Optionally, as an implementation manner, the obtaining module 901 is specifically configured to obtain the N sets of first distribution parameters, and determine N first 2D SAXS maps according to the N sets of first distribution parameters, where the N sets of first distribution parameters are in one-to-one correspondence with the N first 2D SAXS maps.

Optionally, as an implementation manner, the obtaining module 901 is specifically configured to obtain M second 2D SAXS maps, and obtain K sets of second model parameters, where each second 2D SAXS map in the M second 2D SAXS maps includes a set of second distribution parameters, K is a positive integer, and M is a positive integer; the input module 902 is specifically configured to determine a set of target model parameters from the K sets of second model parameters according to the M second 2D SAXS atlases and the first atlas analytical model; the configuration module 903 is specifically configured to configure the set of target model parameters to the first atlas resolution model to obtain a target atlas resolution model.

Optionally, as an implementation manner, the obtaining module 901 is specifically configured to execute the following steps for each of the K sets of second model parameters, so as to obtain K first graphs: configuring a group of currently processed second model parameters to the first atlas analytical model to obtain a second atlas analytical model; randomly calling P second 2D SAXS spectrums from the M second 2D SAXS spectrums and inputting the P second 2D SAXS spectrums into a second spectrum analysis model to obtain P second prediction distribution parameters of the second spectrum analysis model, and determining the first curve graph of the second spectrum analysis model according to the P second prediction distribution parameters and second distribution parameters corresponding to the P second 2D SAXS spectrums, wherein K is a positive integer, and P is a positive integer smaller than M; the input module 902 is specifically configured to compare the K first graphs, determine a target first graph, and determine a set of second model parameters corresponding to the target first graph as the set of target model parameters.

Optionally, as an embodiment, the obtaining module 901 is specifically configured to obtain H third 2D SAXS maps, where each third 2D SAXS map in the H third 2D SAXS maps includes a group of third distribution parameters, and H is a positive integer; the input module 902 is specifically configured to input the H third 2D SAXS maps into the target map analysis model to obtain H groups of third predicted distribution parameters, where the H groups of third predicted distribution parameters are in one-to-one correspondence with the H third 2D SAXS maps; the configuration module 903 is specifically configured to compare each third predicted distribution parameter in the H groups of third predicted distribution parameters with a third distribution parameter in a third 2D SAXS spectrum corresponding thereto, so as to evaluate the target spectrum analysis model.

Optionally, as an embodiment, the obtaining module 901 is specifically configured to obtain a fourth 2D SAXS spectrum of the metal nanorods; the input module 902 is specifically configured to input the fourth 2D SAXS spectrum to the target spectrum analysis model, so as to obtain a fourth distribution parameter; the configuration module 903 is specifically configured to compare the fourth distribution parameter with a preset distribution parameter to verify the target atlas analysis model, where the preset distribution parameter is obtained by processing the metal nanorods based on a Transmission Electron Microscope (TEM).

It should be noted that, the implementation of the operations of the respective modules may also correspond to the corresponding description in the embodiment of the method shown in fig. 1.

Referring to fig. 10, fig. 10 is a schematic diagram of another training apparatus for 2D SAXS atlas resolution model provided in the present application; as shown in fig. 10, the atlas resolution model training apparatus 1000 includes the following modules.

A sample database establishing module 1001, configured to generate the N sets of first distribution parameters according to the random uniform distribution function, and determine the N first 2D SAXS atlases according to the N sets of first distribution parameters, where the N first 2D SAXS atlases constitute a training set of the deep learning model; the M groups of second distribution parameters are generated according to the random uniform distribution function, and the M second 2D SAXS maps are determined according to the M groups of second distribution parameters, wherein the M second 2D SAXS maps form a verification set of the deep learning model; the H groups of third distribution parameters are generated according to the random uniform distribution function, and the H third 2D SAXS maps are determined according to the H groups of third distribution parameters, wherein the H third 2D SAXS maps form a test set of the deep learning model; the training set, the verification set and the test set form a sample database of the deep learning model, wherein N, M and H are positive integers.

The deep learning model establishing module 1002 is configured to respectively create an input layer, a convolution layer, a pooling layer, a full-link layer and an output layer according to a deep learning framework and a deep convolution artificial neural network, and establish the deep learning model; the deep learning model comprises the first model parameters to be determined and the target model parameters, the deep learning model is trained and determined by the first model parameters based on the sample database, and the deep learning model is optimized and determined by the target model parameters based on the sample database.

A deep learning model training module 1003, configured to obtain the training set from the sample database establishing module 1001, and cyclically call a certain number of first 2D SAXS atlases from the training set at random and input the first 2D SAXS atlases into the deep learning model to iterate a first model parameter of the deep learning model until an absolute value of a difference between the first model parameter obtained through current iteration and the first model parameter obtained through previous iteration is less than or equal to a threshold a, end the first model parameter iteration process, and configure the first model parameter obtained through current iteration into the deep learning model to obtain a first atlas analysis model, where a is a positive number.

A deep learning model optimization module 1004, configured to obtain the verification set from the sample database creation module 1001, and determine K sets of second model parameters according to the random uniform distribution function; determining a set of target model parameters from the K sets of second model parameters according to the validation set and the first atlas analytical model; and configuring the group of target model parameters to the first atlas analytical model to obtain the target atlas analytical model, wherein K is a positive integer.

A deep learning model evaluation module 1005, configured to obtain the test set from the sample database establishing module 1001, and input H third 2D SAXS atlases in the test set into the target atlas analysis model to obtain H groups of third predicted distribution parameters, where the H groups of third predicted distribution parameters are in one-to-one correspondence with the H third 2D SAXS atlases; and comparing each third predicted distribution parameter in the H groups of third predicted distribution parameters with a third distribution parameter in a third 2D SAXS map corresponding to the third predicted distribution parameter so as to evaluate the target map analytical model.

The deep learning model verification module 1006 is configured to obtain a fourth 2D SAXS spectrum of the metal nanorod; inputting the fourth 2D SAXS atlas to the target atlas analytical model to obtain the fourth distribution parameter; and comparing the fourth distribution parameter with a preset distribution parameter to verify the target atlas analytical model, wherein the preset distribution parameter is obtained by processing the metal nanorods based on a Transmission Electron Microscope (TEM).

It should be noted that, the implementation of the operations of the respective modules may also correspond to the corresponding description in the embodiment of the method shown in fig. 2.

Referring to fig. 11, fig. 11 is a schematic structural diagram of an electronic device 1100 according to an embodiment of the present disclosure, and as shown in fig. 11, the electronic device 1100 includes a communication interface 1101, a processor 1102, a memory 1103, and at least one communication bus 1104 for connecting the communication interface 1101, the processor 1102, and the memory 1103.

The memory 1103 includes, but is not limited to, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), or a compact disc read-only memory (CD-ROM), and the memory 1103 is used for storing related instructions and data.

The communication interface 1101 is used for receiving and transmitting data.

The processor 1102 may be one or more Central Processing Units (CPUs), and in the case that the processor 1102 is one CPU, the CPU may be a single-core CPU or a multi-core CPU.

The processor 1102 in the electronic device 1100 is configured to read one or more program codes stored in the memory 1103, and perform the following operations: acquiring N first 2D SAXS maps, and inputting the N first 2D SAXS maps into a deep learning model to obtain first model parameters; and configuring the first model parameters into the deep learning model to obtain a first atlas analysis model, wherein the deep learning model is established based on a deep learning framework and a deep convolution artificial neural network, and N is a positive integer.

It should be noted that, implementation of each operation of the electronic device 1100 may also correspond to the corresponding description in the method embodiment in fig. 1.

An embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored, and when the computer program runs on a terminal, the method flows shown in the foregoing method embodiments are implemented.

The embodiment of the present application further provides a computer program product, and when the computer program product runs on a terminal, the method flows shown in the foregoing method embodiments are implemented.

It should be understood that the Processor mentioned in the embodiments of the present Application may be a Central Processing Unit (CPU), and may also be other general purpose processors, Digital Signal Processors (DSP), Application Specific Integrated Circuits (ASIC), Field Programmable Gate Arrays (FPGA) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components, and the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

It will also be appreciated that the memory referred to in the embodiments of the application may be either volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. The non-volatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable PROM (EEPROM), or a flash Memory. Volatile Memory can be Random Access Memory (RAM), which acts as external cache Memory. By way of example, but not limitation, many forms of RAM are available, such as Static random access memory (Static RAM, SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic random access memory (Synchronous DRAM, SDRAM), Double Data Rate Synchronous Dynamic random access memory (DDR SDRAM), Enhanced Synchronous SDRAM (ESDRAM), Synchronous link SDRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

It should be noted that when the processor is a general-purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, the memory (memory module) is integrated in the processor.

It should be noted that the memory described herein is intended to comprise, without being limited to, these and any other suitable types of memory.

It should be understood that, in the various embodiments of the present application, the sequence numbers of the above-mentioned processes do not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described device embodiments are merely illustrative, and for example, the division of the units is only one type of logical function division, and other division manners may be available in actual implementation, for example, a plurality of units or components may be combined or integrated into another device, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

The steps in the method of the embodiment of the application can be sequentially adjusted, combined and deleted according to actual needs.

The modules in the device can be merged, divided and deleted according to actual needs.

The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A two-dimensional small-angle X-ray scattering 2D SAXS atlas analytical model training method is applied to electronic equipment and is characterized by comprising the following steps:

obtaining N first 2D SAXS spectra, wherein N is a positive integer;

inputting the N first 2D SAXS spectra into a deep learning model to obtain first model parameters; the deep learning model is established based on a deep learning framework and a deep convolution artificial neural network;

and configuring the first model parameter into the deep learning model to obtain a first map analysis model.

2. The method of claim 1, wherein obtaining the N first 2D SAXS profiles comprises:

acquiring N groups of first distribution parameters;

determining N first 2D SAXS maps according to the N groups of first distribution parameters, wherein the N groups of first distribution parameters correspond to the N first 2D SAXS maps in a one-to-one mode.

3. The method according to any one of claims 1 or 2, further comprising:

obtaining M second 2D SAXS spectra, wherein each second 2D SAXS spectrum in the M second 2D SAXS spectra comprises a set of second distribution parameters, and M is a positive integer;

acquiring K groups of second model parameters, wherein K is a positive integer;

determining a set of target model parameters from the K sets of second model parameters according to the M second 2D SAXS atlases and the first atlas analytical model;

and configuring the group of target model parameters to the first atlas analytical model to obtain a target atlas analytical model.

4. The method of claim 3, wherein determining a set of target model parameters from the K sets of second model parameters according to the M second 2D SAXS atlases and the first atlas resolution model comprises:

performing the following steps for each of the K sets of second model parameters to obtain K first graphs: configuring a group of currently processed second model parameters to the first atlas analytical model to obtain a second atlas analytical model; randomly calling P second 2D SAXS spectrums from the M second 2D SAXS spectrums and inputting the P second 2D SAXS spectrums into a second spectrum analysis model to obtain P second prediction distribution parameters of the second spectrum analysis model, and determining the first curve graph of the second spectrum analysis model according to the P second prediction distribution parameters and second distribution parameters corresponding to the P second 2D SAXS spectrums, wherein K is a positive integer, and P is a positive integer smaller than M;

comparing the K first graphs to determine a target first graph;

and determining a set of second model parameters corresponding to the target first graph as the set of target model parameters.

5. The method of claim 3, further comprising:

obtaining H third 2D SAXS spectra, wherein each of the H third 2D SAXS spectra comprises a set of third distribution parameters, H being a positive integer;

inputting the H third 2D SAXS spectra into the target spectrum analysis model to obtain H groups of third prediction distribution parameters, wherein the H groups of third prediction distribution parameters are in one-to-one correspondence with the H third 2D SAXS spectra;

and comparing each group of third predicted distribution parameters in the H groups of third predicted distribution parameters with the corresponding third distribution parameters in the third 2D SAXS atlas to evaluate the target atlas analytical model.

6. The method of claim 3, further comprising:

obtaining a fourth 2D SAXS map of the metal nano-rod;

inputting the fourth 2D SAXS spectrum into the target spectrum analysis model to obtain a fourth distribution parameter;

and comparing the fourth distribution parameter with a preset distribution parameter to verify the target atlas analytical model, wherein the preset distribution parameter is obtained by processing the metal nanorods based on a Transmission Electron Microscope (TEM).

7. A2D SAXS atlas resolution model training apparatus, the apparatus comprising:

an obtaining module, configured to obtain N first 2D SAXS spectra, where N is a positive integer;

the input module is used for inputting the N first 2D SAXS maps into a deep learning model to obtain first model parameters, wherein the deep learning model is established on the basis of a deep learning framework and a deep convolution artificial neural network;

and the configuration module is used for configuring the first model parameters into the deep learning model to obtain a first map analysis model.

8. The apparatus of claim 7, wherein the obtaining module is configured to:

acquiring N groups of first distribution parameters;

9. An electronic device, comprising: a processor and a memory;

the processor is coupled to the memory, wherein the memory is configured to store program code and the processor is configured to call the program code to perform the method of any of claims 1-6.

10. A computer storage medium, characterized in that the computer storage medium stores a computer program comprising program instructions which, when executed by the processor, the processor performs the method according to any one of claims 1-6.