CN116768206A

CN116768206A - Graphite purifying process and system

Info

Publication number: CN116768206A
Application number: CN202310600615.7A
Authority: CN
Inventors: 叶水林
Original assignee: Ningbo Jinzhou New Material Technology Co ltd
Current assignee: Ningbo Jinzhou New Material Technology Co ltd
Priority date: 2023-05-25
Filing date: 2023-05-25
Publication date: 2023-09-19

Abstract

A graphite purifying process and a graphite purifying system acquire a reaction monitoring video of a preset time period acquired by a camera; and extracting time sequence change characteristic information of a reaction state of the reaction monitoring video by adopting an artificial intelligence technology based on deep learning, and fully expressing time sequence dynamic change characteristics of the reaction state so as to accurately carry out self-adaptive control on a reaction temperature value in real time based on the change condition of an actual reaction state, thereby ensuring the reaction rate and the purification effect.

Description

Graphite purifying process and system

Technical Field

The application relates to the technical field of intelligent purification, in particular to a graphite purification process and a graphite purification system.

Background

The main purpose of purifying graphite is to remove impurities therein and improve the purity and quality thereof so as to meet the application requirements of different fields.

Specifically, various amorphous substances, metal ions, oxides, and other impurities are often present in graphite, and these impurities may affect the properties of the graphite such as conductivity, mechanical properties, and chemical reactivity. For example, in the energy field of batteries, supercapacitors, and the like, it is necessary to use a high-purity graphite electrode material; in the fields of carbon fibers, composite materials and the like, graphite with low impurity content is required to be used as a raw material; in the fields of semiconductor manufacturing, coating materials, etc., it is also necessary to use graphite having a high purity as an additive or carrier.

Therefore, a process for purifying graphite is desired.

Disclosure of Invention

The present application has been made to solve the above-mentioned technical problems. The embodiment of the application provides a graphite purification process and a graphite purification system, which are used for acquiring a reaction monitoring video of a preset time period acquired by a camera; and extracting time sequence change characteristic information of a reaction state of the reaction monitoring video by adopting an artificial intelligence technology based on deep learning, and fully expressing time sequence dynamic change characteristics of the reaction state so as to accurately carry out self-adaptive control on a reaction temperature value in real time based on the change condition of an actual reaction state, thereby ensuring the reaction rate and the purification effect.

In a first aspect, there is provided a purification system for graphite comprising: the data acquisition module is used for acquiring a reaction monitoring video of a preset time period acquired by the camera; the key frame extraction module is used for extracting a plurality of reaction monitoring key frames from the reaction monitoring video; the reaction state feature extraction module is used for enabling the reaction monitoring key frames to pass through a convolutional neural network model comprising a depth feature fusion module respectively so as to obtain a plurality of reaction state feature matrixes; the feature optimization module is used for performing feature distribution optimization on the reaction state feature matrixes to obtain optimized reaction state feature matrixes; the characteristic difference representation module is used for calculating a transfer matrix between every two adjacent optimized reaction state characteristic matrixes in the optimized reaction state characteristic matrixes to obtain a plurality of transfer matrixes; the reaction state change feature global association module is used for expanding the plurality of transfer matrixes into a plurality of transfer feature vectors and then obtaining decoding feature vectors through a context encoder based on a converter; and the temperature value recommending module is used for enabling the decoding characteristic vector to pass through a decoder to obtain a decoding value, wherein the decoding value is used for representing a recommended reaction temperature value of the current time point.

In the above graphite purification system, the reaction state feature extraction module includes: the shallow layer extraction unit is used for extracting a shallow layer feature map from the shallow layer of the convolutional neural network model comprising the depth feature fusion module; the deep layer extraction unit is used for extracting a deep layer feature map from the deep layer of the convolutional neural network model comprising the deep and shallow feature fusion module; the fusion unit is used for fusing the shallow feature map and the deep feature map by using a depth feature fusion module of the convolutional neural network model comprising the depth feature fusion module so as to obtain a fusion feature map; and a pooling unit, configured to pool the fusion feature map along a channel dimension to obtain the multiple reaction state feature matrices.

In the above graphite purification system, the feature optimization module is configured to: performing Fourier-like scale domain probability correction on each reaction state feature matrix by using the following optimization formula to obtain a plurality of optimized reaction state feature matrices; wherein, the optimization formula is:wherein->Is the +.f. of the characteristic matrix of each reaction state>Characteristic value of the location- >And->The height and width of each reaction state characteristic matrix are respectively, and +.>And->For the superparameter for scale regulation, +.>Representing the calculation of a value of the natural exponent function raised to a power of a value, ">Is the +.f. of the feature matrix of each optimized reaction state>Characteristic values of the location.

In the above graphite purification system, the characteristic difference representation module is configured to: calculating a transfer matrix between every two adjacent optimized reaction state feature matrices in the optimized reaction state feature matrices according to the following transfer formula to obtain a plurality of transfer matrices; wherein, the transfer formula is:wherein->And->Respectively representing every two adjacent optimized reaction state characteristic matrixes in the optimized reaction state characteristic matrixes, and +.>Representing the plurality of transfer matrices, +.>Representing matrix multiplication.

In the above graphite purification system, the reaction state change feature global correlation module includes: a context coding unit, configured to perform global-based context semantic coding on the plurality of transfer feature vectors using the context encoder based on the converter to obtain a plurality of context feature vectors; and a concatenation unit, configured to concatenate the plurality of context feature vectors to obtain the decoded feature vector.

In the above graphite purification system, the context coding unit includes: a query vector construction subunit, configured to perform one-dimensional arrangement on the plurality of transfer feature vectors to obtain a global transfer feature vector; a self-attention subunit, configured to calculate a product between the global transfer feature vector and a transpose vector of each transfer feature vector in the plurality of transfer feature vectors to obtain a plurality of self-attention correlation matrices; the normalization subunit is used for respectively performing normalization processing on each self-attention correlation matrix in the plurality of self-attention correlation matrices to obtain a plurality of normalized self-attention correlation matrices; the attention calculating subunit is used for obtaining a plurality of probability values through a Softmax classification function by each normalized self-attention correlation matrix in the normalized self-attention correlation matrices; and an attention applying subunit configured to weight each of the plurality of transfer feature vectors with each of the plurality of probability values as a weight to obtain the plurality of context feature vectors, respectively.

In the above graphite purification system, the temperature value recommendation module is configured to: performing a decoding regression on the decoded feature vector using the decoder in the following formula to obtain the decoded value; wherein, the formula is: Wherein->Representing said decoded feature vector,/->Representing the decoded value->Representing a weight matrix, +.>Representing the bias vector +_>Representing a matrix multiplication.

In a second aspect, there is provided a process for purifying graphite comprising: acquiring a reaction monitoring video of a preset time period acquired by a camera; extracting a plurality of reaction monitoring key frames from the reaction monitoring video; the reaction monitoring key frames are respectively passed through a convolutional neural network model comprising a depth feature fusion module to obtain a plurality of reaction state feature matrixes; performing feature distribution optimization on the reaction state feature matrixes to obtain optimized reaction state feature matrixes; calculating transfer matrixes between every two adjacent optimized reaction state feature matrixes in the optimized reaction state feature matrixes to obtain a plurality of transfer matrixes; expanding the plurality of transfer matrixes into a plurality of transfer feature vectors, and then obtaining decoding feature vectors through a context encoder based on a converter; and passing the decoded feature vector through a decoder to obtain a decoded value, the decoded value being used to represent a recommended reaction temperature value for the current point in time.

In the above graphite purification process, the step of passing the plurality of reaction monitoring key frames through a convolutional neural network model including a depth feature fusion module to obtain a plurality of reaction state feature matrices includes: extracting a shallow feature map from a shallow layer of the convolutional neural network model comprising the deep and shallow feature fusion module; extracting a deep feature map from the deep layer of the convolutional neural network model comprising the deep-shallow feature fusion module; using a depth feature fusion module of the convolutional neural network model comprising a depth feature fusion module to fuse the shallow feature map and the deep feature map to obtain a fused feature map; and carrying out global mean pooling on the fusion feature map along the channel dimension to obtain the plurality of reaction state feature matrixes.

In the above graphite purification process, performing feature distribution optimization on the plurality of reaction state feature matrices to obtain a plurality of optimized reaction state feature matrices, including: to the following advantagesPerforming Fourier-like scale domain probability correction on each reaction state feature matrix by a formulation formula to obtain a plurality of optimized reaction state feature matrices; wherein, the optimization formula is:wherein->Is the +.f. of the characteristic matrix of each reaction state>Characteristic value of the location->And->The height and width of each reaction state characteristic matrix are respectively, and +.>Andfor the superparameter for scale regulation, +.>Representing the calculation of a value of the natural exponent function raised to a power of a value, ">Is the +.f. of the feature matrix of each optimized reaction state>Characteristic values of the location.

Compared with the prior art, the graphite purification process and the graphite purification system provided by the application have the advantages that the reaction monitoring video of a preset time period acquired by the camera is acquired; and extracting time sequence change characteristic information of a reaction state of the reaction monitoring video by adopting an artificial intelligence technology based on deep learning, and fully expressing time sequence dynamic change characteristics of the reaction state so as to accurately carry out self-adaptive control on a reaction temperature value in real time based on the change condition of an actual reaction state, thereby ensuring the reaction rate and the purification effect.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is an application scenario diagram of a graphite purification system according to an embodiment of the present application.

Fig. 2 is a block diagram of a graphite purification system according to an embodiment of the present application.

Fig. 3 is a block diagram of the reaction state feature extraction module in a graphite purification system according to an embodiment of the present application.

Fig. 4 is a block diagram of the global correlation module of the reaction state change characteristics in the graphite purification system according to the embodiment of the present application.

Fig. 5 is a block diagram of the context encoding unit in the graphite purification system according to an embodiment of the present application.

Fig. 6 is a flow chart of a process for purifying graphite according to an embodiment of the present application.

Fig. 7 is a schematic diagram of a system architecture of a graphite purification process according to an embodiment of the present application.

Detailed Description

The following description of the technical solutions according to the embodiments of the present application will be given with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

Unless defined otherwise, all technical and scientific terms used in the embodiments of the application have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to limit the scope of the present application.

In describing embodiments of the present application, unless otherwise indicated and limited thereto, the term "connected" should be construed broadly, for example, it may be an electrical connection, or may be a communication between two elements, or may be a direct connection, or may be an indirect connection via an intermediate medium, and it will be understood by those skilled in the art that the specific meaning of the term may be interpreted according to circumstances.

It should be noted that, the term "first\second\third" related to the embodiment of the present application is merely to distinguish similar objects, and does not represent a specific order for the objects, it is to be understood that "first\second\third" may interchange a specific order or sequence where allowed. It is to be understood that the "first\second\third" distinguishing objects may be interchanged where appropriate such that embodiments of the application described herein may be practiced in sequences other than those illustrated or described herein.

As described above, various amorphous substances, metal ions, oxides, and other impurities are often present in graphite, and these impurities may affect the properties of the graphite such as conductivity, mechanical properties, and chemical reactivity. For example, in the energy field of batteries, supercapacitors, and the like, it is necessary to use a high-purity graphite electrode material; in the fields of carbon fibers, composite materials and the like, graphite with low impurity content is required to be used as a raw material; in the fields of semiconductor manufacturing, coating materials, etc., it is also necessary to use graphite having a high purity as an additive or carrier. Therefore, a process for purifying graphite is desired.

Specifically, in the technical scheme of the application, a graphite purification system is provided, which comprises: adding sufficient amounts of an oxidizing agent including, but not limited to, hydrogen peroxide, nitric acid, etc., and a reducing agent including, but not limited to, disodium sulfite, etc., to a reaction vessel having an ionic liquid, wherein the ionic liquid is an imidozolium-based ionic liquid; adding graphite into a reaction vessel, and reacting the graphite with an oxidant and a reducing agent in the ionic liquid at a preset temperature to obtain a reaction product, thereby controlling oxidation or removing impurities on the surface of the graphite; the reaction product is washed with deionized water to remove ionic liquid and residual chemicals, thereby purifying the graphite.

Accordingly, considering that in the above-mentioned purification scheme of graphite, it is particularly important to control the temperature, since an appropriate temperature may increase the reaction rate and purification effect, but an excessively high temperature may cause degradation of the ionic liquid and structural damage of the graphite. Therefore, in the actual graphite purification process, the reaction temperature range should be determined according to the ionic liquid selected and the oxidizing agent and the reducing agent. Generally, the reaction temperature should be controlled below the melting point of the ionic liquid to avoid evaporation and decomposition of the solvent, typically between room temperature and 100 ℃.

Based on this, in the technical solution of the present application, in order to ensure that impurities can be effectively removed while maintaining the structural integrity of graphite, it is desirable to perform real-time precise control of the reaction temperature value through time-series variation analysis of the reaction state. However, because the time sequence change characteristic information of the reaction state in the time dimension is the hidden change characteristic information of the small scale, in the actual monitoring process, capturing and extracting the weak hidden change characteristic of the reaction state is difficult, and the control precision of the reaction temperature value is further reduced. Therefore, in this process, it is difficult to fully express the time sequence dynamic change characteristics of the reaction state, so as to accurately and in real time perform the adaptive control of the reaction temperature value based on the change condition of the actual reaction state, thereby ensuring the reaction rate and the purification effect.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. The development of deep learning and neural networks provides new solutions and schemes for mining the characteristic information of the time sequence dynamic change of the reaction state.

Specifically, in the technical scheme of the application, firstly, a reaction monitoring video of a preset time period is acquired through a camera. Next, it is considered that the time-series change characteristic of the reaction state due to the reaction of graphite with the oxidizing agent and the reducing agent in the ionic liquid in the reaction monitor video can be represented by the difference between the adjacent monitor frames in the reaction monitor video, that is, the change condition of the reaction state is represented by the image characterization of the adjacent image frames. However, since a large amount of data redundancy exists in consideration of the small difference between adjacent frames in the reaction monitor video, in order to reduce the amount of calculation and avoid adverse effects of data redundancy on detection, the reaction monitor video is key frame-sampled at a predetermined sampling frequency to extract a plurality of reaction monitor key frames from the reaction monitor video. Here, it is worth mentioning that the sampling frequency may be adjusted based on the application requirements of the actual scenario, instead of the default value.

Then, feature mining of each reaction monitoring key frame is performed by using a convolutional neural network model with excellent performance in terms of implicit feature extraction of images, and in particular, in order to accurately monitor dynamic implicit change feature information of the reaction state in time sequence in order to more accurately monitor hidden features of the reaction state in extracting hidden features of the reaction monitoring key frames, shallow features such as textures of reaction liquid in each reaction monitoring key frame are more focused, and have important significance for time sequence dynamic change monitoring of the reaction state. However, when the convolutional neural network is coded, as the depth of the convolutional neural network is deepened, shallow features become blurred and even submerged in noise.

Based on the above, in the technical scheme of the application, the convolutional neural network model comprising the depth feature fusion module is used for processing each reaction monitoring key frame to obtain a plurality of reaction state feature matrixes. It should be understood that, compared with a standard convolutional neural network model, the convolutional neural network model according to the present application can retain the shallow layer features and the deep layer features of the reaction liquid state of each reaction monitoring key frame, so that not only feature information is more abundant, but also features of different depths can be retained, so as to improve the accuracy of monitoring the reaction state. Meanwhile, the structure of the deep neural network is complex, a large amount of sample data is needed for training and adjusting, the training time of the deep neural network is long, and fitting is easy. Therefore, in the design of the neural network model, the combination of the shallow network and the deep network is generally adopted, and through depth feature fusion, the complexity of the network and the risk of overfitting can be reduced to a certain extent, and meanwhile, the feature extraction capability and the generalization capability of the model are improved.

Further, considering that the characteristic of the change of the reaction state of the reaction solution in the time dimension is weak, in order to accurately control the reaction temperature value in real time, it is necessary to accurately express the implicit dynamic change characteristic of the reaction state at every two time points. Specifically, in the technical scheme of the application, a transfer matrix between every two adjacent reaction state feature matrices in the plurality of reaction state feature matrices is calculated, so that reaction state feature difference information about reactants at two adjacent time points is extracted, namely time sequence change feature information about small scale reaction states at every two adjacent time points is obtained, and a plurality of transfer matrices are obtained.

Then, the change rule of the dynamic characteristic of the reaction state hidden characteristic of the reaction liquid in the time sequence global is considered, that is, the difference information of the characteristic of the reaction state of the reactant at every two adjacent time points has a time sequence dynamic association relation in the time dimension. Therefore, in the technical scheme of the application, after the transfer matrices are further developed into the transfer feature vectors, the transfer feature vectors are encoded in a context encoder based on a converter, so that the dynamic associated feature distribution information of the reaction state feature difference information about the reactant based on the time sequence global at every two adjacent time points is extracted, namely, the dynamic change feature information of the reaction state based on the time sequence global in the time dimension is extracted, and the decoding feature vector is obtained.

And then, further carrying out decoding regression on the decoding characteristic vector through a decoder to obtain a decoding value for representing the recommended reaction temperature value of the current time point. Namely, decoding is performed by using the dynamic correlation characteristic of the time sequence of the reaction state of the reactant, so that the time sequence change condition of the reaction state is accurately detected and judged, and the real-time accurate control of the reaction temperature value is performed, so that the reaction rate and the purification effect are ensured.

Particularly, in the technical scheme of the application, when the plurality of response monitoring key frames respectively obtain the plurality of response state feature matrices through the convolution neural network model comprising the depth feature fusion module, since the convolution neural network model performs feature extraction on two spatial dimensions of the image width and the height of the response monitoring frames, if the correlation degree of the semantic features of the image space of the convolution neural network model in the image space can be improved, the feature expression effect of the plurality of obtained response state feature matrices can be obviously improved, and finally the accuracy of the decoding value of the decoding feature vector obtained through a decoder can be improved.

Based on this, in the training process, each of the response state feature matrices obtained for the convolutional neural network model including the depth feature fusion module is, for example, written asThe fourier-like scale domain probability correction is expressed as:wherein->Is the characteristic matrix of each reaction stateIs>Characteristic value of the location->And->Each of the reaction state feature matrices is +.>Height and width of (2), and->And->Is a super parameter for scale adjustment.

Here, the fourier-like scale domain probability correction considers the homology of the high-dimensional feature distribution and the scale domain where the high-dimensional feature distribution is located, and can capture potential distribution association under the homologous space based on low-rank constraint of the scale space through fourier-like sparse low-rank transformation of the scale space, so that in the training process of the convolutional neural network model, joint space feature learning with feature integral scale coherence is realized while spatial dimension local image semantic association representation of feature values is obtained, and the feature expression effect of the obtained multiple reaction state feature matrixes is improved by improving the learning association degree of the model under the integral image semantic space scale, and finally the accuracy of the decoding values of the decoding feature vectors obtained by a decoder is improved. Therefore, the self-adaptive control of the reaction temperature value can be performed in real time and accurately based on the change condition of the actual reaction state, so that the reaction rate and the purification effect are ensured.

Fig. 1 is an application scenario diagram of a graphite purification system according to an embodiment of the present application. As shown in fig. 1, in the application scenario, first, a reaction monitor video (e.g., C as illustrated in fig. 1) of a predetermined period of time acquired by a camera is acquired; the acquired reaction monitor video is then input into a server (e.g., S as illustrated in fig. 1) deployed with a graphite purification algorithm, wherein the server is capable of processing the reaction monitor video based on the graphite purification algorithm to generate a decoded value representing a recommended reaction temperature value for the current point in time.

Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.

In one embodiment of the application, FIG. 2 is a block diagram of a graphite purification system according to an embodiment of the application. As shown in fig. 2, a graphite purification system 100 according to an embodiment of the present application includes: a data acquisition module 110 for acquiring a reaction monitoring video of a predetermined period acquired by the camera; a key frame extracting module 120, configured to extract a plurality of reaction monitoring key frames from the reaction monitoring video; the reaction state feature extraction module 130 is configured to pass the plurality of reaction monitoring key frames through a convolutional neural network model including a depth feature fusion module to obtain a plurality of reaction state feature matrices; the feature optimization module 140 is configured to perform feature distribution optimization on the plurality of reaction state feature matrices to obtain a plurality of optimized reaction state feature matrices; the feature difference representation module 150 is configured to calculate a transfer matrix between every two adjacent optimized reaction state feature matrices in the plurality of optimized reaction state feature matrices to obtain a plurality of transfer matrices; a global correlation module 160 for global correlation of the state change characteristics, which is configured to obtain a decoded feature vector by a context encoder based on a converter after expanding the plurality of transfer matrices into a plurality of transfer feature vectors; and a temperature value recommending module 170, configured to pass the decoded feature vector through a decoder to obtain a decoded value, where the decoded value is used to represent a recommended reaction temperature value at a current time point.

Specifically, in the embodiment of the present application, the data acquisition module 110 is configured to acquire the reaction monitoring video of the predetermined period acquired by the camera. As described above, various amorphous substances, metal ions, oxides, and other impurities are often present in graphite, and these impurities may affect the properties of the graphite such as conductivity, mechanical properties, and chemical reactivity. For example, in the energy field of batteries, supercapacitors, and the like, it is necessary to use a high-purity graphite electrode material; in the fields of carbon fibers, composite materials and the like, graphite with low impurity content is required to be used as a raw material; in the fields of semiconductor manufacturing, coating materials, etc., it is also necessary to use graphite having a high purity as an additive or carrier. Therefore, a process for purifying graphite is desired.

Specifically, in the technical scheme of the application, firstly, a reaction monitoring video of a preset time period is acquired through a camera.

Specifically, in the embodiment of the present application, the key frame extracting module 120 is configured to extract a plurality of reaction monitoring key frames from the reaction monitoring video. Next, it is considered that the time-series change characteristic of the reaction state due to the reaction of graphite with the oxidizing agent and the reducing agent in the ionic liquid in the reaction monitor video can be represented by the difference between the adjacent monitor frames in the reaction monitor video, that is, the change condition of the reaction state is represented by the image characterization of the adjacent image frames. However, since a large amount of data redundancy exists in consideration of the small difference between adjacent frames in the reaction monitor video, in order to reduce the amount of calculation and avoid adverse effects of data redundancy on detection, the reaction monitor video is key frame-sampled at a predetermined sampling frequency to extract a plurality of reaction monitor key frames from the reaction monitor video. Here, it is worth mentioning that the sampling frequency may be adjusted based on the application requirements of the actual scenario, instead of the default value.

Specifically, in the embodiment of the present application, the reaction state feature extraction module 130 is configured to obtain a plurality of reaction state feature matrices by passing the plurality of reaction monitoring key frames through a convolutional neural network model including a depth feature fusion module. Feature mining of the individual reaction monitoring key frames is then performed using a convolutional neural network model that has excellent performance in implicit feature extraction of images.

In particular, in order to accurately monitor the dynamic implicit variation characteristic information of the reaction state in time sequence when extracting the hidden characteristic about the reaction state in each reaction monitoring key frame, so as to accurately control the reaction temperature value, the shallow characteristics about the texture of the reaction liquid in each reaction monitoring key frame should be more focused, and these shallow characteristics have important significance for monitoring the time sequence dynamic variation of the reaction state. However, when the convolutional neural network is coded, as the depth of the convolutional neural network is deepened, shallow features become blurred and even submerged in noise.

Fig. 3 is a block diagram of the reaction state feature extraction module in the graphite purification system according to an embodiment of the present application, and as shown in fig. 3, the reaction state feature extraction module 130 includes: a shallow layer extracting unit 131, configured to extract a shallow layer feature map from a shallow layer of the convolutional neural network model including the depth feature fusion module; a deep layer extraction unit 132, configured to extract a deep layer feature map from the deep layer of the convolutional neural network model including the deep-shallow feature fusion module; a fusion unit 133, configured to fuse the shallow feature map and the deep feature map by using a depth feature fusion module of the convolutional neural network model including a depth feature fusion module to obtain a fused feature map; and a pooling unit 134, configured to pool the global average value of the fused feature map along the channel dimension to obtain the multiple reaction state feature matrices.

The convolutional neural network (Convolutional Neural Network, CNN) is an artificial neural network and has wide application in the fields of image recognition and the like. The convolutional neural network may include an input layer, a hidden layer, and an output layer, where the hidden layer may include a convolutional layer, a pooling layer, an activation layer, a full connection layer, etc., where the previous layer performs a corresponding operation according to input data, outputs an operation result to the next layer, and obtains a final result after the input initial data is subjected to a multi-layer operation.

The convolutional neural network model has excellent performance in the aspect of image local feature extraction by taking a convolutional kernel as a feature filtering factor, and has stronger feature extraction generalization capability and fitting capability compared with the traditional image feature extraction algorithm based on statistics or feature engineering.

Further, in encoding the plurality of reaction monitoring key frames using a convolutional neural network model, first a shallow feature map (e.g., the shallow layer refers to the first layer to the sixth layer) is extracted from a shallow layer of the convolutional neural network model, and a deep feature map (e.g., the last layer of the convolutional neural network model) is extracted from a deep layer of the convolutional neural network model, and then a feature representation including shallow features and deep features is obtained by fusing the shallow feature map and the deep feature map. In a specific encoding process, the extraction position of the shallow feature map is determined by the overall network depth of the convolutional neural network model, for example, when the network depth is 30, from the layer 3 of the convolutional neural network model, when the network depth is 40, from the layer 4 of the convolutional neural network model, which is not limited by the present application. Likewise, the extraction position of the deep feature map is not limited by the present application, and may be the last layer, the last but one layer, or the last but one layer and the last but one layer.

Specifically, in the embodiment of the present application, the feature optimization module 140 is configured to perform feature distribution optimization on the plurality of reaction state feature matrices to obtain a plurality of optimized reaction state feature matrices. Particularly, in the technical scheme of the application, when the plurality of response monitoring key frames respectively obtain the plurality of response state feature matrices through the convolution neural network model comprising the depth feature fusion module, since the convolution neural network model performs feature extraction on two spatial dimensions of the image width and the height of the response monitoring frames, if the correlation degree of the semantic features of the image space of the convolution neural network model in the image space can be improved, the feature expression effect of the plurality of obtained response state feature matrices can be obviously improved, and finally the accuracy of the decoding value of the decoding feature vector obtained through a decoder can be improved.

Based on this, in the training process, each of the response state feature matrices obtained for the convolutional neural network model including the depth feature fusion module is, for example, written asThe fourier-like scale domain probability correction is expressed as: performing Fourier-like scale domain probability correction on each reaction state feature matrix by using the following optimization formula to obtain a plurality of optimized reaction state feature matrices; wherein, the optimization formula is: / >Wherein->Is the +.f. of the characteristic matrix of each reaction state>Characteristic value of the location->And->The height and width of each reaction state characteristic matrix are respectively, and +.>And->For the superparameter for scale regulation, +.>Representing the calculation of a value of the natural exponent function raised to a power of a value, ">Is the +.f. of the feature matrix of each optimized reaction state>Characteristic values of the location.

Specifically, in the embodiment of the present application, the feature difference representation module 150 is configured to calculate a transfer matrix between every two adjacent optimized reaction state feature matrices in the plurality of optimized reaction state feature matrices to obtain a plurality of transfer matrices. Further, considering that the characteristic of the change of the reaction state of the reaction solution in the time dimension is weak, in order to accurately control the reaction temperature value in real time, it is necessary to accurately express the implicit dynamic change characteristic of the reaction state at every two time points. Specifically, in the technical scheme of the application, a transfer matrix between every two adjacent reaction state feature matrices in the plurality of reaction state feature matrices is calculated, so that reaction state feature difference information about reactants at two adjacent time points is extracted, namely time sequence change feature information about small scale reaction states at every two adjacent time points is obtained, and a plurality of transfer matrices are obtained.

Wherein the feature difference representation module 150 is configured to: calculating a transfer matrix between every two adjacent optimized reaction state feature matrices in the optimized reaction state feature matrices according to the following transfer formula to obtain a plurality of transfer matrices; wherein, the transfer formula is: Wherein->And->Respectively representing every two adjacent optimized reaction state characteristic matrixes in the optimized reaction state characteristic matrixes, and +.>Representing the plurality of transfer matrices, +.>Representing matrix multiplication.

Specifically, in the embodiment of the present application, the global correlation module 160 for the reaction state change feature is configured to obtain the decoded feature vector by using a context encoder based on a converter after expanding the plurality of transition matrices into a plurality of transition feature vectors. Then, the change rule of the dynamic characteristic of the reaction state hidden characteristic of the reaction liquid in the time sequence global is considered, that is, the difference information of the characteristic of the reaction state of the reactant at every two adjacent time points has a time sequence dynamic association relation in the time dimension.

Therefore, in the technical scheme of the application, after the transfer matrices are further developed into the transfer feature vectors, the transfer feature vectors are encoded in a context encoder based on a converter, so that the dynamic associated feature distribution information of the reaction state feature difference information about the reactant based on the time sequence global at every two adjacent time points is extracted, namely, the dynamic change feature information of the reaction state based on the time sequence global in the time dimension is extracted, and the decoding feature vector is obtained.

Fig. 4 is a block diagram of the global correlation module of the reaction state change characteristic in the graphite purification system according to the embodiment of the present application, and as shown in fig. 4, the global correlation module 160 of the reaction state change characteristic includes: a context encoding unit 161, configured to perform global-based context semantic encoding on the plurality of transfer feature vectors using the context encoder based on the converter to obtain a plurality of context feature vectors; and a concatenation unit 162, configured to concatenate the plurality of context feature vectors to obtain the decoded feature vector.

Further, fig. 5 is a block diagram of the context encoding unit in the graphite purification system according to an embodiment of the present application, and as shown in fig. 5, the context encoding unit 161 includes: a query vector construction subunit 1611, configured to perform one-dimensional arrangement on the plurality of transfer feature vectors to obtain a global transfer feature vector; a self-attention subunit 1612, configured to calculate a product between the global transfer feature vector and a transpose vector of each of the plurality of transfer feature vectors to obtain a plurality of self-attention correlation matrices; a normalization subunit 1613, configured to perform normalization processing on each of the plurality of self-attention correlation matrices to obtain a plurality of normalized self-attention correlation matrices; a degree of interest calculation subunit 1614, configured to obtain a plurality of probability values from each normalized self-attention correlation matrix in the plurality of normalized self-attention correlation matrices by using a Softmax classification function; and an attention applying subunit 1615, configured to weight each of the plurality of transferred feature vectors with each of the plurality of probability values as a weight to obtain the plurality of context feature vectors.

It should be appreciated that the context encoder is intended to mine for hidden patterns between contexts in a word sequence, optionally the encoder comprises: CNN (Convolutional Neural Network ), recurrent NN (RecursiveNeural Network, recurrent neural network), language Model (Language Model), and the like. The CNN-based method has a better extraction effect on local features, but has a poor effect on Long-Term Dependency (Long-Term Dependency) problems in sentences, so Bi-LSTM (Long Short-Term Memory) based encoders are widely used. The repetitive NN processes sentences as a tree structure rather than a sequence, has stronger representation capability in theory, but has the weaknesses of high sample marking difficulty, deep gradient disappearance, difficulty in parallel calculation and the like, so that the repetitive NN is less in practical application. The transducer has a network structure with wide application, has the characteristics of CNN and RNN, has a better extraction effect on global characteristics, and has a certain advantage in parallel calculation compared with RNN (RecurrentNeural Network ).

Specifically, in the embodiment of the present application, the temperature value recommending module 170 is configured to pass the decoded feature vector through a decoder to obtain a decoded value, where the decoded value is used to represent a recommended reaction temperature value at a current time point. And then, further carrying out decoding regression on the decoding characteristic vector through a decoder to obtain a decoding value for representing the recommended reaction temperature value of the current time point. Namely, decoding is performed by using the dynamic correlation characteristic of the time sequence of the reaction state of the reactant, so that the time sequence change condition of the reaction state is accurately detected and judged, and the real-time accurate control of the reaction temperature value is performed, so that the reaction rate and the purification effect are ensured.

Wherein, the temperature value recommending module 170 is configured to: performing a decoding regression on the decoded feature vector using the decoder in the following formula to obtain the decoded value; wherein, the formula is:wherein->Representing said decoded feature vector,/->Representing the decoded value->Representing a weight matrix, +.>Representing the bias vector +_>Representing a matrix multiplication.

In summary, a graphite purification system 100 according to an embodiment of the present application is illustrated, which acquires a reaction monitoring video of a predetermined period of time acquired by a camera; and extracting time sequence change characteristic information of a reaction state of the reaction monitoring video by adopting an artificial intelligence technology based on deep learning, and fully expressing time sequence dynamic change characteristics of the reaction state so as to accurately carry out self-adaptive control on a reaction temperature value in real time based on the change condition of an actual reaction state, thereby ensuring the reaction rate and the purification effect.

As described above, the graphite purification system 100 according to the embodiment of the present application may be implemented in various terminal devices, such as a server for purification of graphite, and the like. In one example, the graphite purification system 100 according to an embodiment of the present application may be integrated into the terminal equipment as a software module and/or hardware module. For example, the graphite purification system 100 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the graphite purification system 100 can also be one of a number of hardware modules of the terminal equipment.

Alternatively, in another example, the graphite purification system 100 and the terminal device may be separate devices, and the graphite purification system 100 may be connected to the terminal device through a wired and/or wireless network, and transmit interactive information in a agreed data format.

In one embodiment of the present application, fig. 6 is a flow chart of a process for purifying graphite according to an embodiment of the present application. As shown in fig. 6, the purification process of graphite according to an embodiment of the present application includes: 210, acquiring a reaction monitoring video of a preset time period acquired by a camera; 220, extracting a plurality of reaction monitoring key frames from the reaction monitoring video; 230, passing the plurality of reaction monitoring key frames through a convolutional neural network model comprising a depth feature fusion module to obtain a plurality of reaction state feature matrixes; 240, performing feature distribution optimization on the reaction state feature matrixes to obtain optimized reaction state feature matrixes; 250, calculating a transfer matrix between every two adjacent optimized reaction state feature matrices in the optimized reaction state feature matrices to obtain a plurality of transfer matrices; 260, expanding the plurality of transfer matrices into a plurality of transfer feature vectors, and then obtaining decoding feature vectors through a context encoder based on a converter; and, 270, passing the decoded feature vector through a decoder to obtain a decoded value, the decoded value being used to represent a recommended reaction temperature value for the current point in time.

Fig. 7 is a schematic diagram of a system architecture of a graphite purification process according to an embodiment of the present application. As shown in fig. 7, in the system architecture of the graphite purification process, first, a reaction monitoring video of a predetermined period of time acquired by a camera is acquired; then, extracting a plurality of reaction monitoring key frames from the reaction monitoring video; then, the reaction monitoring key frames are respectively passed through a convolutional neural network model comprising a depth feature fusion module to obtain a plurality of reaction state feature matrixes; then, performing feature distribution optimization on the reaction state feature matrixes to obtain optimized reaction state feature matrixes; then, calculating a transfer matrix between every two adjacent optimized reaction state feature matrices in the optimized reaction state feature matrices to obtain a plurality of transfer matrices; then, the plurality of transfer matrixes are unfolded into a plurality of transfer feature vectors, and then the transfer feature vectors are passed through a context encoder based on a converter to obtain decoding feature vectors; and finally, the decoding eigenvector is passed through a decoder to obtain a decoding value, wherein the decoding value is used for representing the recommended reaction temperature value of the current time point.

In a specific example, in the above graphite purification process, the step of passing the plurality of reaction monitoring key frames through a convolutional neural network model including a depth feature fusion module to obtain a plurality of reaction state feature matrices includes: extracting a shallow feature map from a shallow layer of the convolutional neural network model comprising the deep and shallow feature fusion module; extracting a deep feature map from the deep layer of the convolutional neural network model comprising the deep-shallow feature fusion module; using a depth feature fusion module of the convolutional neural network model comprising a depth feature fusion module to fuse the shallow feature map and the deep feature map to obtain a fused feature map; and carrying out global mean pooling on the fusion feature map along the channel dimension to obtain the plurality of reaction state feature matrixes.

In a specific example, in the above graphite purification process, performing feature distribution optimization on the plurality of reaction state feature matrices to obtain a plurality of optimized reaction state feature matrices, including: performing Fourier-like scale domain probability correction on each reaction state feature matrix by using the following optimization formula to obtain a plurality of optimized reaction state feature matrices; wherein, the optimization formula is: Wherein->Is the respective oppositeThe +.>Characteristic value of the location->And->The height and width of each reaction state characteristic matrix are respectively, and +.>And->For the superparameter for scale regulation, +.>Representing the calculation of a value of the natural exponent function raised to a power of a value, ">Is the +.f. of the feature matrix of each optimized reaction state>Characteristic values of the location.

In a specific example, in the above graphite purification process, calculating a transfer matrix between every two adjacent optimized reaction state feature matrices in the plurality of optimized reaction state feature matrices to obtain a plurality of transfer matrices includes: calculating a transfer matrix between every two adjacent optimized reaction state feature matrices in the optimized reaction state feature matrices according to the following transfer formula to obtain a plurality of transfer matrices; wherein, the transfer formula is:wherein->And->Respectively representing every two adjacent optimized reaction state characteristic matrixes in the optimized reaction state characteristic matrixes, and +.>Representing the plurality of transfer matrices, +.>Representing matrix multiplication.

In a specific example, in the above graphite purification process, the developing the plurality of transfer matrices into a plurality of transfer feature vectors and then passing through a context encoder based on a converter to obtain decoded feature vectors includes: performing global-based context semantic coding on the plurality of transfer feature vectors using the converter-based context encoder to obtain a plurality of context feature vectors; and concatenating the plurality of contextual feature vectors to obtain the decoded feature vector.

In a specific example, in the above graphite purification process, performing global-based context semantic encoding on the plurality of transfer feature vectors using the converter-based context encoder to obtain a plurality of context feature vectors, comprising: one-dimensional arrangement is carried out on the plurality of transfer feature vectors so as to obtain global transfer feature vectors; calculating the product between the global transfer feature vector and the transpose vector of each transfer feature vector in the plurality of transfer feature vectors to obtain a plurality of self-attention association matrices; respectively carrying out standardization processing on each self-attention correlation matrix in the plurality of self-attention correlation matrices to obtain a plurality of standardized self-attention correlation matrices; obtaining a plurality of probability values by using a Softmax classification function through each normalized self-attention correlation matrix in the normalized self-attention correlation matrices; and weighting each of the plurality of transfer feature vectors with each of the plurality of probability values as a weight to obtain the plurality of context feature vectors.

In a specific example, the graphite purification process described above Wherein, the decoding eigenvector is passed through a decoder to obtain a decoding value, the decoding value is used for representing a recommended reaction temperature value of the current time point, and the method comprises the following steps: performing a decoding regression on the decoded feature vector using the decoder in the following formula to obtain the decoded value; wherein, the formula is:wherein->Representing said decoded feature vector,/->Representing the decoded value->Representing a weight matrix, +.>Representing the bias vector +_>Representing a matrix multiplication.

It will be appreciated by those skilled in the art that the specific operation of the respective steps in the above-described graphite purification process has been described in detail in the above description of the graphite purification system with reference to fig. 1 to 5, and thus, repetitive description thereof will be omitted.

The present application also provides a computer program product comprising instructions which, when executed, cause an apparatus to perform operations corresponding to the above-described method.

In one embodiment of the present application, there is also provided a computer-readable storage medium storing a computer program for executing the above-described method.

It should be appreciated that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the forms of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects may be utilized. Furthermore, the computer program product may 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.) having computer-usable program code embodied therein.

Methods, systems, and computer program products of embodiments of the present application are described in the flow diagrams and/or block diagrams. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, 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, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The basic principles of the present application have been described above in connection with specific embodiments, however, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are merely examples and not intended to be limiting, and these advantages, benefits, effects, etc. are not to be considered as essential to the various embodiments of the present application. Furthermore, the specific details disclosed herein are for purposes of illustration and understanding only, and are not intended to be limiting, as the application is not necessarily limited to practice with the above described specific details.

The block diagrams of the devices, apparatuses, devices, systems referred to in the present application are only illustrative examples and are not intended to require or imply that the connections, arrangements, configurations must be made in the manner shown in the block diagrams. As will be appreciated by one of skill in the art, the devices, apparatuses, devices, systems may be connected, arranged, configured in any manner. Words such as "including," "comprising," "having," and the like are words of openness and mean "including but not limited to," and are used interchangeably therewith. The terms "or" and "as used herein refer to and are used interchangeably with the term" and/or "unless the context clearly indicates otherwise. The term "such as" as used herein refers to, and is used interchangeably with, the phrase "such as, but not limited to.

It is also noted that in the apparatus, devices and methods of the present application, the components or steps may be disassembled and/or assembled. Such decomposition and/or recombination should be considered as equivalent aspects of the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit embodiments of the application to the form disclosed herein. Although a number of example aspects and embodiments have been discussed above, a person of ordinary skill in the art will recognize certain variations, modifications, alterations, additions, and subcombinations thereof.

Claims

1. A graphite purification system, comprising: the data acquisition module is used for acquiring a reaction monitoring video of a preset time period acquired by the camera; the key frame extraction module is used for extracting a plurality of reaction monitoring key frames from the reaction monitoring video; the reaction state feature extraction module is used for enabling the reaction monitoring key frames to pass through a convolutional neural network model comprising a depth feature fusion module respectively so as to obtain a plurality of reaction state feature matrixes; the feature optimization module is used for performing feature distribution optimization on the reaction state feature matrixes to obtain optimized reaction state feature matrixes; the characteristic difference representation module is used for calculating a transfer matrix between every two adjacent optimized reaction state characteristic matrixes in the optimized reaction state characteristic matrixes to obtain a plurality of transfer matrixes; the reaction state change feature global association module is used for expanding the plurality of transfer matrixes into a plurality of transfer feature vectors and then obtaining decoding feature vectors through a context encoder based on a converter; and the temperature value recommending module is used for enabling the decoding characteristic vector to pass through a decoder to obtain a decoding value, wherein the decoding value is used for representing a recommended reaction temperature value of the current time point.

2. The graphite purification system of claim 1, wherein the reaction state feature extraction module comprises: the shallow layer extraction unit is used for extracting a shallow layer feature map from the shallow layer of the convolutional neural network model comprising the depth feature fusion module; the deep layer extraction unit is used for extracting a deep layer feature map from the deep layer of the convolutional neural network model comprising the deep and shallow feature fusion module; the fusion unit is used for fusing the shallow feature map and the deep feature map by using a depth feature fusion module of the convolutional neural network model comprising the depth feature fusion module so as to obtain a fusion feature map; and a pooling unit, configured to pool the fusion feature map globally along a channel dimension to obtain the multiple reaction state feature matrices.

3. The graphite purification system of claim 2, wherein the feature optimization module is configured to: performing Fourier-like scale domain probability correction on each reaction state feature matrix by using the following optimization formula to obtain a plurality of optimized reaction state feature matrices; wherein, the optimization formula is:wherein- >Is the +.f. of the characteristic matrix of each reaction state>Characteristic value of the location->And->The height and width of each reaction state characteristic matrix are respectively, and +.>And->For the superparameter for scale regulation, +.>Representing the calculation of a value of the natural exponent function raised to a power of a value, ">Is the +.f. of the feature matrix of each optimized reaction state>Characteristic values of the location.

4. A graphite purification system according to claim 3, wherein the characteristic difference representation module is configured to: calculating a transfer matrix between every two adjacent optimized reaction state feature matrices in the optimized reaction state feature matrices according to the following transfer formula to obtain a plurality of transfer matrices; wherein, the transfer formula is:wherein, the method comprises the steps of, wherein,and->Respectively representing every two adjacent optimized reaction state characteristic matrixes in the optimized reaction state characteristic matrixes, and +.>Representing the plurality of transfer matrices, +.>Representing matrix multiplication.

5. The graphite purification system of claim 4, wherein the reaction state change feature global correlation module comprises: a context coding unit, configured to perform global-based context semantic coding on the plurality of transfer feature vectors using the context encoder based on the converter to obtain a plurality of context feature vectors; and a concatenation unit configured to concatenate the plurality of context feature vectors to obtain the decoded feature vector.

6. The graphite purification system of claim 5, wherein the context coding unit comprises: a query vector construction subunit, configured to perform one-dimensional arrangement on the plurality of transfer feature vectors to obtain a global transfer feature vector; a self-attention subunit, configured to calculate a product between the global transfer feature vector and a transpose vector of each transfer feature vector in the plurality of transfer feature vectors to obtain a plurality of self-attention correlation matrices; the normalization subunit is used for respectively performing normalization processing on each self-attention correlation matrix in the plurality of self-attention correlation matrices to obtain a plurality of normalized self-attention correlation matrices; the attention calculating subunit is used for obtaining a plurality of probability values through a Softmax classification function by each normalized self-attention correlation matrix in the normalized self-attention correlation matrices; and an attention applying subunit configured to weight each of the plurality of transfer feature vectors with each of the plurality of probability values as a weight to obtain the plurality of context feature vectors, respectively.

7. The graphite purification system of claim 6, wherein the temperature value recommendation module is configured to: performing a decoding regression on the decoded feature vector using the decoder in the following formula to obtain the decoded value; wherein, the formula is: Wherein->Representing said decoded feature vector,/->Representing the decoded value->Representing a weight matrix, +.>Representing the bias vector +_>Representing a matrix multiplication.

8. A process for purifying graphite, comprising: acquiring a reaction monitoring video of a preset time period acquired by a camera; extracting a plurality of reaction monitoring key frames from the reaction monitoring video; the reaction monitoring key frames are respectively passed through a convolutional neural network model comprising a depth feature fusion module to obtain a plurality of reaction state feature matrixes; performing feature distribution optimization on the reaction state feature matrixes to obtain optimized reaction state feature matrixes; calculating transfer matrixes between every two adjacent optimized reaction state feature matrixes in the optimized reaction state feature matrixes to obtain a plurality of transfer matrixes; expanding the plurality of transfer matrixes into a plurality of transfer feature vectors, and then obtaining decoding feature vectors through a context encoder based on a converter; and passing the decoded feature vector through a decoder to obtain a decoded value, the decoded value being used to represent a recommended reaction temperature value for the current point in time.

9. The graphite purification process of claim 8, wherein passing the plurality of reaction monitoring key frames through a convolutional neural network model comprising a depth feature fusion module to obtain a plurality of reaction state feature matrices, respectively, comprises: extracting a shallow feature map from a shallow layer of the convolutional neural network model comprising the deep and shallow feature fusion module; extracting a deep feature map from the deep layer of the convolutional neural network model comprising the deep-shallow feature fusion module; using a depth feature fusion module of the convolutional neural network model comprising a depth feature fusion module to fuse the shallow feature map and the deep feature map to obtain a fused feature map; and carrying out global mean pooling on the fusion feature map along the channel dimension to obtain the plurality of reaction state feature matrixes.

10. The process for purifying graphite of claim 9, wherein optimizing the feature distribution of the plurality of reaction state feature matrices to obtain a plurality of optimized reaction state feature matrices comprises: performing Fourier-like scale domain probability correction on each reaction state feature matrix by using the following optimization formula to obtain a plurality of optimized reaction state feature matrices; wherein, the optimization formula is:wherein->Is the +.f. of the characteristic matrix of each reaction state>Characteristic value of the location->And->The height and width of each reaction state characteristic matrix are respectively, and +.>And->For use inSuper-parameters of scale regulation, ->Representing the calculation of a value of the natural exponent function raised to a power of a value, ">Is the +.f. of the feature matrix of each optimized reaction state>Characteristic values of the location.