Method and system for detecting quality of finished product of micromolecular polysilicon pore throat dredging agent

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a small molecule polysilicon pore throat dredging agent finished product quality detection method and system, and the method comprises the steps: SEM image acquisition parameters are set, the SEM image acquisition parameters comprise magnification times and resolution, and SEM images of pores of different sizes before and after dredging are acquired; extracting multi-scale pore structure features before and after dredging according to the SEM image; calculating a quantitative descriptor according to the SEM image, and carrying out feature splicing on the multi-scale pore structure features and the quantitative descriptor to obtain comprehensive pore structure representation; the comprehensive pore structure expression is used as input, the qualitative index and the quantitative index of the quality detection of the dredging agent are predicted based on the dredging effect prediction model, and the quality detection precision of the finished product of the micromolecular polysilicon pore throat dredging agent is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method and system for detecting the quality of a finished product of a small molecule polysilicon pore-throat dredging agent. Background Art

[0002] Polysilicon materials are widely used in many fields due to their unique properties, such as waterproof and anticorrosive coatings, biomedical materials, electronic packaging materials, etc. However, during the use of polysilicon materials, their pores may be blocked by impurities, affecting the performance of the material. Using small molecule compounds as unclogging agents can penetrate into the pores of polysilicon materials, react chemically with the blockages or dissolve them, thereby unclogging the pores and restoring the performance of the material. Compared with the use of large molecule unclogging agents, small molecule unclogging agents have the advantages of small size and strong permeability.

[0003] Traditional methods for detecting the quality of unclogging agents mainly rely on experimental tests and empirical judgments, lacking quantitative and intelligent means, resulting in insufficient detection accuracy. Summary of the invention

[0004] The present application aims to solve at least one of the technical problems in the related art to a certain extent. To this end, one purpose of the present application is to propose a method and system for quality detection of finished products of small molecule polysilicon pore throat dredging agent, thereby improving the accuracy of quality detection of finished products of small molecule polysilicon pore throat dredging agent.

[0005] One aspect of the present application provides a method for testing the quality of a finished product of a small molecule polysilicon pore throat dredging agent, comprising: Step S100: setting SEM image acquisition parameters, wherein the SEM image acquisition parameters include magnification and resolution, and acquiring SEM images of pores of different scales before and after dredging; Step S200: extracting multi-scale pore structure characteristics before and after dredging based on the SEM image; Step S300: Calculate the quantitative descriptor according to the SEM image, perform feature splicing on the multi-scale pore structure features and the quantitative descriptor to obtain a comprehensive pore structure representation; Step S400: using the comprehensive pore structure representation as input, based on the dredging effect prediction model, predicting the qualitative and quantitative indicators of the dredging agent quality detection; The specific method for setting the SEM image acquisition parameters is: Step S110: dividing the pores of the sample before and after dredging into three scales according to the pore size and geometric shape, including micron-scale pores, mesopores and macropores; Step S120: Obtaining the pore size range of the pores in the sample before dredging And the range of pore size after dredging ,in, and are the minimum and maximum pore diameters before dredging, and They are the minimum and maximum pore diameters after dredging, respectively; Step S130: Setting the magnification before dredging The value range is such that the pixel diameter of the smallest aperture before dredging in the SEM image is Equal to the minimum aperture before dredging and the actual size of each pixel The ratio of the magnification before dredging Greater than or equal to the pixel diameter threshold of the minimum pore size before dredging in the SEM image ; Step S140: Setting the pre-dredging resolution The value range of , the resolution before dredging is less than or equal to half of the minimum aperture before dredging; Step S150: Setting the magnification after dredging The value range is such that the pixel diameter of the smallest aperture after dredging in the SEM image is Equal to the minimum aperture after dredging and the actual size of each pixel ratio, and the magnification after dredging The pixel diameter threshold greater than or equal to the minimum pore size after dredging in the SEM image ; Step S160: Setting the post-unblocking resolution The value range of , the resolution after dredging is less than or equal to half of the minimum aperture after dredging; Step S170: Preset optimal pore pixel diameter , according to different pore sizes d and optimal pore pixel diameters, the magnifications of different pore sizes are calculated, and the magnifications satisfy the corresponding magnifications before dredging The value range or magnification after dredging The value range of Step S180: Defining resolution according to different aperture sizes , and the resolution satisfies the corresponding range of values ​​of the resolution before dredging or the range of values ​​of the resolution after dredging; The specific method for extracting the multi-scale pore structure characteristics before and after dredging based on the SEM image is: Step S210: For micron-sized pores, extract the local texture features of the silicone structure before and after dredging based on Gabor filters , the parameters of the Gabor filter Including scale parameters , Direction parameters and frequency parameters ; Step S220: For mesopores and macropores, extract their morphological features through morphological operations and region segmentation algorithms and region segmentation features ; Step S230: The local texture features of micron-scale pores are combined with the morphological features and regional segmentation features of mesopores and macropores by feature splicing to obtain multi-scale pore structure features. ; The scale parameters are set by obtaining the resolution r and magnification m of the SEM image and the pixel size range of the siloxane structure. , then the corresponding physical size range is ;in, , are the minimum and maximum values ​​of the physical size range respectively; Within the physical size range, a logarithmic sampling method is used to select scale parameter ,in, For the scale parameters, the i-th scale parameter is ; The specific method for extracting the local texture features of the siloxane structure is: For each set of Gabor filter parameters , calculate the Gabor filter on the SEM image The convolution response on , get a set of Gabor feature maps; Concatenate the Gabor feature maps obtained by all Gabor filter parameters into a long vector , the long vector serves as a local texture feature of the siloxane structure under the micron-scale pores; The specific method of calculating the quantitative descriptor based on the SEM image and combining the multi-scale pore structure characteristics with the quantitative descriptor to obtain the comprehensive pore structure representation is: Step S310: Calculate the quantitative descriptor D of the pores before and after dredging based on the SEM image. The quantitative descriptor includes the specific surface area Kong Rong and pore size distribution The BET method was used to calculate the specific surface area of ​​the pores, the mercury intrusion method was used to measure the pore volume, and the BJH method was used to calculate the pore size distribution of the pores; Step S320: performing feature splicing on the multi-scale pore structure features and the quantitative descriptors to obtain a comprehensive pore structure representation X; The specific method of predicting the qualitative and quantitative indicators of the dredging agent quality detection based on the dredging effect prediction model with the comprehensive pore structure representation as input is as follows: Step S410: obtaining a quality inspection training sample of a finished dredging agent, wherein the training sample includes a comprehensive pore structure representation of the pores before and after dredging and corresponding qualitative index labels and quantitative index labels; Step S420: Design a multi-scale convolutional neural network, wherein the multi-scale convolutional neural network comprises n convolutional branches, each of which is used to extract pore structure feature maps of different scales. ; Step S430: Design an attention mechanism module, introduce the attention mechanism, and calculate the attention weight of the pore structure feature map at each scale , obtain the weighted pore structure feature map of each scale, fuse the weighted pore structure feature map to obtain the weighted fusion feature z; Step S440: Design a pore structure evolution modeling module, use a long short-term memory network, take the weighted fusion features at the current moment as input, and output the hidden state at the current moment ; Step S450: constructing a dredging effect prediction module, using the hidden state output by the pore structure evolution modeling module as input through a fully connected layer, to predict the qualitative and quantitative indicators of the dredging agent quality detection; Step S460: Define the cross entropy loss function of qualitative indicators And the mean square error loss function of quantitative indicators , the weighted sum of the cross entropy loss function and the mean square error loss function is used to obtain the total loss function ; Step S470: A dredging effect prediction model is constructed by a multi-scale convolutional neural network, an attention mechanism module, a pore structure evolution modeling module and a dredging effect prediction module. Training is performed based on training samples, with minimizing the total loss function as the training goal. The training ends when the total loss function converges. The qualitative and quantitative indicators of the quality inspection of the new dredging agent finished product are predicted based on the trained dredging effect prediction model.

[0006] One aspect of the present application provides a finished product quality detection system of a small molecule polysilicon pore throat dredging agent, comprising: An SEM image acquisition module is used to set SEM image acquisition parameters, including magnification and resolution, and to acquire SEM images of pores of different scales before and after dredging; Multi-scale feature extraction module, used to extract multi-scale pore structure features before and after dredging based on SEM images; A comprehensive feature representation module is used to calculate quantitative descriptors based on SEM images, and to perform feature splicing of multi-scale pore structure features and quantitative descriptors to obtain a comprehensive pore structure representation; The quality detection output module is used to predict the qualitative and quantitative indicators of the dredging agent quality detection based on the dredging effect prediction model with the comprehensive pore structure representation as input.

[0007] Compared with the prior art, the finished product quality detection method and system of small molecule polysilicon pore throat dredging agent proposed in this application has the following advantages: This application introduces multi-scale pore structure characteristics and deep learning models to extract the multi-scale characteristics of micron-level pores, mesopores and macropores before and after dredging, comprehensively characterizes the changes in pore structures at different scales, provides richer and more comprehensive information for comprehensive evaluation of the dredging effect of the dredging agent, and improves detection accuracy.

[0008] For micron-sized pores, the present application uses Gabor filters to extract local texture features, which can effectively capture the changes in the silicone structure before and after dredging and reflect the influence of the dredging agent on the microscopic pores. The Gabor filter can adaptively extract texture features of different sizes and directions by adjusting the scale, direction and frequency parameters, thereby improving the flexibility and adaptability of feature extraction.

[0009] This application comprehensively describes the changes in pore structure before and after dredging by fusing multi-scale features, thereby improving the expressive power and generalization performance of the prediction model.

[0010] This application automatically learns key features at different scales through a multi-scale convolutional structure, without the need for manual feature design, thereby improving the adaptability and intelligence of feature extraction. By introducing an attention mechanism, it dynamically focuses on key areas and features at different scales according to the differences in dredging effects, thereby improving the adaptability and robustness of the prediction model. A long short-term memory network is used to capture the temporal dependency of pore morphology and quantitative indicators before and after dredging, making full use of the dynamic information in the dredging process.

[0011] This application effectively solves the problems of lack of quantitative and intelligent means and insufficient detection accuracy in traditional unclogging agent quality detection methods through multi-scale pore structure feature extraction and deep learning model construction. It comprehensively describes the changes in pore structures at different scales, provides rich and comprehensive information, and greatly improves the accuracy, efficiency and intelligence level of unclogging agent quality detection, providing reliable technical support for the research and development and application of unclogging agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A method flow chart of the quality detection method of the finished product of the small molecule polysilicon pore throat dredging agent provided in this application; Figure 2 A flow chart of a method for obtaining multi-scale pore structure characteristics provided in this application; Figure 3 A flow chart of the method for obtaining the comprehensive pore structure representation provided for this application; Figure 4 This is a functional module diagram of the finished product quality detection system of the small molecule polysilicon pore throat unclogging agent provided in this application. DETAILED DESCRIPTION

[0013] In order to better understand the present application, various aspects of the present application will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of exemplary embodiments of the present application and do not limit the scope of the present application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0014] In the accompanying drawings, the size, dimensions and shapes of the elements have been slightly adjusted for ease of illustration. The drawings are for illustration only and are not strictly drawn to scale. As used herein, the terms "substantially", "approximately" and similar terms are used as terms of approximation, not as terms of degree, and are intended to illustrate the inherent deviations in measurements or calculations that will be recognized by those of ordinary skill in the art. In addition, in this application, the order in which the steps are described does not necessarily represent the order in which these processes occur in actual operation, unless otherwise specified or can be derived from the context.

[0015] It should also be understood that expressions such as "including", "comprising", "having", "containing" and / or "comprising" are open rather than closed expressions in this specification, which indicate the presence of the stated features, elements and / or components, but do not exclude the presence of one or more other features, elements, components and / or combinations thereof. In addition, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features rather than just the individual elements in the list. In addition, when describing embodiments of the present application, "may" is used to mean "one or more embodiments of the present application". And, the term "exemplary" is intended to refer to an example or illustration.

[0016] Unless otherwise defined, all words (including engineering terms and scientific and technological terms) used in this article have the same meaning as those commonly understood by ordinary technicians in the field to which this application belongs. It should also be understood that unless there is a clear explanation in this application, words defined in commonly used dictionaries should be interpreted as having the same meaning as their meaning in the context of the relevant technology, and should not be interpreted in an idealized or overly formal sense.

[0017] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0018] Example 1

[0019] like Figure 1 As shown, the quality detection method of the finished product of the small molecule polysilicon pore throat dredging agent provided in this application includes: Step S100: setting SEM image acquisition parameters, wherein the SEM image acquisition parameters include magnification and resolution, and acquiring SEM images of pores of different scales before and after dredging; The specific method for setting the SEM image acquisition parameters is: Step S110: dividing the pores of the sample before and after dredging into three scales according to the pore size and geometric shape, including micron-scale pores, mesopores and macropores; The classification rules of micron-sized pores, mesopores and macropores are as follows: Pores with a pore size greater than or equal to 1 micron and irregular geometric shapes are marked as micron-sized pores; the micron-sized pores can be directly observed in the SEM image, have obvious morphological characteristics, and have irregular geometric shapes, such as slits, cracks, and dendrites; Pores with a pore size greater than or equal to 2 nanometers and less than or equal to 50 nanometers are marked as mesopores; the mesopores usually require a higher magnification to be distinguished in SEM images, and their morphological features are more refined than those of micron-sized pores; Pores with a pore size greater than 50 nanometers and regular geometric shapes are marked as macropores; the macropores have regular geometric shapes, such as spherical, cylindrical, honeycomb, etc., the orientation and arrangement of the pores are relatively orderly, and the pore walls are relatively smooth; Step S120: Obtaining the pore size range of the pores in the sample before dredging And the range of pore size after dredging ,in, and are the minimum and maximum pore diameters before dredging, and They are the minimum and maximum pore diameters after dredging, respectively; Step S130: Setting the magnification before dredging The value range is such that the pixel diameter of the smallest aperture before dredging in the SEM image is Equal to the minimum aperture before dredging and the actual size of each pixel The ratio of the magnification before dredging Greater than or equal to the pixel diameter threshold of the minimum pore size before dredging in the SEM image ; The pixel diameter threshold of the minimum pore size before dredging in the SEM image is preset by a person skilled in the art based on experience; preferably, the value range of the pixel diameter threshold is 5 to 10; The pixel diameter of the minimum aperture before unblocking satisfy: ; From the above formula, we can get the magnification before dredging. The value range of satisfies: ; Step S140: Setting the pre-dredging resolution The value range of , the resolution before dredging is less than or equal to half of the minimum aperture before dredging; The resolution before dredging is: ; Step S150: Setting the magnification after dredging The value range is such that the pixel diameter of the smallest aperture after dredging in the SEM image is Equal to the minimum aperture after dredging and the actual size of each pixel ratio, and the magnification after dredging The pixel diameter threshold greater than or equal to the minimum pore size after dredging in the SEM image ; The pixel diameter threshold of the minimum pore size after dredging in the SEM image is preset by a person skilled in the art based on experience; preferably, the value range of the pixel diameter threshold is 10-20; The pixel diameter of the minimum aperture after dredging satisfies: ; From the above formula, we can get the magnification after dredging. The value range of satisfies: ; Step S160: Setting the post-unblocking resolution The value range of , the resolution after dredging is less than or equal to half of the minimum aperture after dredging; The resolution after dredging is: ; Step S170: Preset optimal pore pixel diameter , according to different pore sizes d and optimal pore pixel diameters, the magnifications of different pore sizes are calculated, and the magnifications satisfy the corresponding magnifications before dredging The value range or magnification after dredging The value range of The calculation formula for the magnification of the different aperture sizes is: ; The optimal pore pixel diameter The value of is set by those skilled in the art according to the aperture and imaging requirements. Preferably, the optimal aperture pixel diameter is The value range of is 10~50; The pore size The smaller, the higher the magnification The larger the size, the more detailed the pores. Step S180: Defining resolution according to different aperture sizes , and the resolution satisfies the corresponding range of values ​​of the resolution before dredging or the range of values ​​of the resolution after dredging; The resolution The calculation formula is: , where q is the resolution factor; Preferably, the value of the resolution coefficient is 2-5, and the specific value is selected according to the imaging requirements; The smaller the aperture, the higher the resolution. The lower the resolution, the more fine structure of the pores can be resolved.

[0020] The resolution refers to the minimum resolvable distance of the instrument. When the resolution factor is set to 2, for a one-micron pore, the required resolution is 0.5, which means that the smaller the pore, in order to ensure the same resolution factor, the instrument needs to have a higher resolution capability, that is, a smaller R value.

[0021] The specific method for collecting SEM images of pores of different scales before and after dredging is: SEM images of pores of different scales are collected according to the corresponding magnification and resolution, and SEM images of micron-scale pores are subjected to wavelet transform denoising and contrast adjustment; For the SEM images of mesopores and macropores, the pore area was smoothed to enhance the contrast between the pore area and the matrix; The above steps set the SEM image acquisition parameters and adaptively determine the magnification and resolution according to the pore size range of pores of different scales before and after dredging, ensuring the acquisition of high-quality pore structure images, laying the foundation for subsequent feature extraction and analysis.

[0022] Step S200: extracting multi-scale pore structure characteristics before and after dredging based on the SEM image; like Figure 2 As shown, the specific method for extracting the multi-scale pore structure characteristics before and after dredging based on the SEM image is: Step S210: For micron-sized pores, extract the local texture features of the silicone structure before and after dredging based on Gabor filters , the parameters of the Gabor filter Including scale parameters , Direction parameters and frequency parameters ; The scale parameters are set by obtaining the resolution r and magnification m of the SEM image and the pixel size range of the siloxane structure. , then the corresponding physical size range is ;in, , are the minimum and maximum values ​​of the physical size range respectively; The pixel size range of the siloxane structure is obtained based on experimental measurements; Within the physical size range, a logarithmic sampling method is used to select scale parameter ,in, For the scale parameters, the i-th scale parameter is ; The calculation formula of the scale parameter is: ,in, ; The scale parameter is used to control the size of the filter, which is adapted to the size of the siloxane structure.

[0023] The method for setting the direction parameters is: select the Direction parameter value ,in, Indicates Direction parameters; The calculation formula of the direction parameter is: ,in, ; The direction parameter is used to control the direction of the filter. Including 4, 6, 8, etc., to capture texture features in different directions; The method for setting the frequency parameters is: obtaining the frequency range of the siloxane structure texture ; The frequency range of the siloxane structure texture is obtained by performing Fourier transform on the SEM image, analyzing the frequency spectrum characteristics of the siloxane structure texture, and determining the frequency range. ; In the frequency range, the sampling method of logarithmic equal interval is used to select Frequency parameters ;in, For the frequency parameters; The calculation formula of the frequency parameter is: ,in, .

[0024] In summary, the parameter setting of the Gabor filter can be expressed as a triple .

[0025] The specific method for extracting the local texture features of the siloxane structure is: For each set of Gabor filter parameters , calculate the Gabor filter on the SEM image The convolution response on , get a set of Gabor feature maps; Concatenate the Gabor feature maps obtained by all Gabor filter parameters into a long vector , the long vector serves as a local texture feature of the siloxane structure under the micron-scale pores; The Gabor filter in the SEM image The convolution response on is calculated as: , where (x, y) represents the pixel coordinates in the Gabor feature map, represents the pixel coordinates on the SEM image, is the mother function of the Gabor filter; The calculation formula of the mother function of the Gabor filter is: , where j* is an imaginary unit, s, , f are scale, direction and frequency respectively; Step S220: For mesopores and macropores, extract their morphological features through morphological operations and region segmentation algorithms and region segmentation features ; The morphological characteristics include, but are not limited to, pore size, geometry, and pore orientation; The regional segmentation features include but are not limited to the number of pores, pore connectivity, pore distribution, and pore area; The morphological operation includes corrosion operation, expansion operation, opening operation and closing operation; the corrosion operation is used to remove noise and small protrusions at the pore edges of mesopores and macropores, the expansion operation is used to fill small holes inside the pores to make the pore shape more complete, and the opening operation and closing operation are used to smooth the pore boundaries and eliminate small connected areas; The region segmentation algorithm adopts a threshold-based segmentation algorithm, and according to the grayscale difference between the pores of the mesopores and macropores and the matrix of the sample material to be dredged, the SEM image is segmented into two parts, the pores and the matrix, by a preset threshold; The above steps extract multi-scale pore structure features and adopt different feature extraction methods for pores of different scales. For example, Gabor filtering is used to extract texture features for micron-scale pores, and morphological operations and regional segmentation are used to extract morphological and regional features for mesopores and macropores. Feature fusion is then performed to comprehensively characterize the changes in pore structure before and after dredging.

[0026] Step S230: The local texture features of micron-scale pores are combined with the morphological features and regional segmentation features of mesopores and macropores by feature splicing to obtain multi-scale pore structure features. .

[0027] Step S300: Calculate the quantitative descriptor according to the SEM image, perform feature splicing on the multi-scale pore structure features and the quantitative descriptor to obtain a comprehensive pore structure representation; like Figure 3 As shown, the specific method of calculating the quantitative descriptor based on the SEM image, and combining the multi-scale pore structure characteristics with the quantitative descriptor to obtain the comprehensive pore structure representation is: Step S310: Calculate the quantitative descriptor D of the pores before and after dredging based on the SEM image. The quantitative descriptor includes the specific surface area Kong Rong and pore size distribution The BET method was used to calculate the specific surface area of ​​the pores, the mercury intrusion method was used to measure the pore volume, and the BJH method was used to calculate the pore size distribution of the pores; The quantitative descriptor can be expressed as: ; The specific surface area is calculated by using the BET equation, which is: , through linear fitting right The relationship between the monolayer adsorption capacity and and BET constant The value of is used to calculate the specific surface area; where v is the volume of the adsorbed gas, and are the equilibrium pressure and saturated vapor pressure, respectively. is the monolayer adsorption amount, c is the BET constant; The calculation formula of the specific surface area is: ;in, is Avogadro's constant, S is the cross-sectional area of ​​the adsorbate molecule, and V is the volume of the sample to be dredged; The saturated vapor pressure is obtained by looking up the table according to the adsorbate and temperature; The volume and equilibrium pressure of the adsorbed gas are obtained through experimental measurement; The cross-sectional area of ​​the adsorbate molecule is obtained by looking up the type of adsorbate in the table. The cross-sectional area of ​​the commonly used adsorbate nitrogen is 0.162 nm 2 ; The volume of the sample to be dredged is measured by a fluid displacement method.

[0028] The pore volume is calculated by using a mercury intrusion method, which is based on the Washburn equation and measures the volume of a mercury column at different pressures to obtain the pore volume. The equation for the Washburn equation is: ; Where P is the mercury injection pressure, is the surface tension of mercury, is the contact angle between mercury and the pore wall, is the aperture size; The calculation formula of the pore volume is: ;in, The aperture size is The volume distribution function of the pores; The mercury injection pressure and pore volume distribution function are obtained by measuring the mercury injection experiment; The contact angle between the mercury and the pore wall is related to the pore wall material. Preferably, the contact angle is generally 130° to 140°. The pore size distribution is calculated by using the BJH method, which is based on the Kelvin equation to calculate the pore size distribution of pores; The equation of the Kelvin equation is: ;in, is the molar volume of the adsorbate, is the ideal gas constant, is the absolute temperature, is the surface tension of the adsorbate, is the contact angle between the adsorbate and the pore wall; The calculation formula of the pore size distribution is: ; The surface tension and molar volume of the adsorbate are obtained by looking up a table according to the type of the adsorbate; The contact angle between the adsorbate and the pore wall is set to 0; Step S320: performing feature splicing on the multi-scale pore structure features and the quantitative descriptors to obtain a comprehensive pore structure representation X; The comprehensive pore structure is expressed as: ; The above steps form a comprehensive pore structure representation by combining quantitative descriptors reflecting pore physical properties with multi-scale features reflecting pore morphology and distribution, enhancing the expressive power and complementarity of features.

[0029] Step S400: using the comprehensive pore structure representation as input, based on the dredging effect prediction model, predicting the qualitative and quantitative indicators of the dredging agent quality detection; The specific method of predicting the qualitative and quantitative indicators of the dredging agent quality detection based on the dredging effect prediction model with the comprehensive pore structure representation as input is as follows: Step S410: obtaining a quality inspection training sample of a finished dredging agent, wherein the training sample includes a comprehensive pore structure representation of the pores before and after dredging and corresponding qualitative index labels and quantitative index labels; The qualitative index refers to a quality grade index that is qualitatively evaluated by those skilled in the art; The quantitative index refers to the conductivity of the small molecule polysilicon pore throat after dredging, which is measured experimentally and is used to evaluate the degree of improvement of the dredging effect; Step S420: Design a multi-scale convolutional neural network, wherein the multi-scale convolutional neural network comprises n convolutional branches, each of which is used to extract pore structure feature maps of different scales. ; The multi-scale convolutional neural network consists of multiple parallel convolution branches, each of which uses a different convolution kernel size to capture features of different scales; The expression of the pore structure characteristic diagram is: ,in, is the pore structure characteristic diagram of the ath scale, is the mapping function of the convolution branch of the ath scale, X represents the comprehensive pore structure representation of the input, represents the convolution kernel weight of the ath scale, represents the convolution kernel bias of the a-th scale, is the activation function; The output of the multi-scale convolutional neural network is recorded as ,in is the pore structure characteristic map of the ath scale; Step S430: Design an attention mechanism module, introduce the attention mechanism, and calculate the attention weight of the pore structure feature map at each scale , obtain the weighted pore structure feature map of each scale, fuse the weighted pore structure feature map to obtain the weighted fusion feature z; The calculation formula of the attention weight is: ,in, is the attention weight vector of the ath scale The transpose of is the attention weight vector of the g-th scale The transpose of is the pore structure characteristic map of the g-th scale; The calculation formula of the weighted fusion feature is: ; Step S440: designing a pore structure evolution modeling module, using a long short-term memory network, taking the weighted fusion features at the current moment as input, and outputting the hidden state at the current moment; The state transfer formula of the long short-term memory network is: ; ; ; ; ; ; in, , , , They are input gate, forget gate, output gate, and candidate memory unit states. , , , are the weight matrices of the input gate, forget gate, output gate, and candidate memory unit states, respectively. , , , are the bias vectors of the input gate, forget gate, output gate, and candidate memory cell states, respectively. is the memory unit state, , are the hidden states of the current moment and the previous moment respectively, is the weighted fusion feature at the current moment, is the sigmoid activation function, is element-wise multiplication, is the hyperbolic tangent activation function; Step S450: constructing a dredging effect prediction module, using the hidden state output by the pore structure evolution modeling module as input through a fully connected layer, to predict the qualitative and quantitative indicators of the dredging agent quality detection; The prediction formula of the qualitative index is: ,in, and is the weight matrix and bias vector of the fully connected layer used to predict qualitative indicators; The prediction formula of the quantitative index is: ,in, and is the weight matrix and bias vector of the fully connected layer used to predict quantitative indicators; Step S460: Define the cross entropy loss function of qualitative indicators And the mean square error loss function of quantitative indicators , the total loss function is obtained by weighted summing the cross entropy loss function and the mean square error loss function; The calculation formula of the total loss function is: ,in, , are the weight coefficients of the cross entropy loss function and the mean square error loss function respectively; The weight coefficients of the cross entropy loss function and the mean square error loss function are set by those skilled in the art based on experience. and The sum is 1; The calculation formula of the cross entropy loss function is: ,in, , are the true qualitative indicator label and predicted qualitative indicator of the ni-th training sample, respectively, and N represents the number of training samples; The calculation formula of the mean square error loss function is: ,in, , are the true quantitative indicator label and predicted quantitative indicator of the ni-th training sample respectively; Step S470: A dredging effect prediction model is constructed by a multi-scale convolutional neural network, an attention mechanism module, a pore structure evolution modeling module and a dredging effect prediction module. Training is performed based on training samples, with minimizing the total loss function as the training goal. The training ends when the total loss function converges. The qualitative and quantitative indicators of the quality inspection of the new dredging agent finished product are predicted based on the trained dredging effect prediction model.

[0030] The above steps use a multi-scale convolutional neural network to extract feature maps of different scales, introduce an attention mechanism to adaptively adjust the importance of features at each scale, use a long short-term memory network to model the dynamic evolution of the pore structure, and use multi-task learning to simultaneously predict qualitative and quantitative indicators, thereby improving the prediction performance and generalization ability of the model.

[0031] Example 2

[0032] like Figure 4 As shown, the finished product quality detection system of the small molecule polysilicon pore throat dredging agent provided in this application includes: An SEM image acquisition module is used to set SEM image acquisition parameters, including magnification and resolution, and to acquire SEM images of pores of different scales before and after dredging; Multi-scale feature extraction module, used to extract multi-scale pore structure features before and after dredging based on SEM images; A comprehensive feature representation module is used to calculate quantitative descriptors based on SEM images, and to perform feature splicing of multi-scale pore structure features and quantitative descriptors to obtain a comprehensive pore structure representation; The quality detection output module is used to predict the qualitative and quantitative indicators of the dredging agent quality detection based on the dredging effect prediction model with the comprehensive pore structure representation as input.

[0033] The parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0034] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for testing the quality of a finished product of a small molecule polysilicon pore throat dredging agent, characterized in that: include: Setting SEM image acquisition parameters, including magnification and resolution, to acquire SEM images of pores of different scales before and after dredging; Extract multi-scale pore structure characteristics before and after dredging based on SEM images; Quantitative descriptors are calculated based on SEM images, and multi-scale pore structure features are feature spliced ​​with quantitative descriptors to obtain a comprehensive pore structure representation; Taking the comprehensive pore structure representation as input, based on the dredging effect prediction model, the qualitative and quantitative indicators of the dredging agent quality detection are predicted.

2. The finished product quality detection method of small molecule polysilicon pore throat dredging agent according to claim 1, characterized in that: The specific method for setting the SEM image acquisition parameters includes: The pores of the samples before and after dredging were divided into three scales according to the pore size and geometric shape, including micron-scale pores, mesopores, and macropores; Obtain the pore size range of the pores in the sample before dredging And the range of pore size after dredging ,in, and are the minimum and maximum pore diameters before dredging, and are the minimum and maximum pore diameters after dredging, respectively; Set the magnification before dredging The value range is such that the pixel diameter of the smallest aperture before dredging in the SEM image is Equal to the minimum aperture before dredging and the actual size of each pixel The ratio of the magnification before dredging Greater than or equal to the pixel diameter threshold of the minimum pore size before dredging in the SEM image ; Set the pre-dredge resolution The value range of , the resolution before dredging is less than or equal to half of the minimum aperture before dredging; Set the magnification after dredging The value range is such that the pixel diameter of the smallest aperture after dredging in the SEM image is Equal to the minimum aperture after dredging and the actual size of each pixel ratio, and the magnification after dredging The pixel diameter threshold greater than or equal to the minimum pore size after dredging in the SEM image ; Set the resolution after unblocking The value range of the resolution after dredging is less than or equal to half of the minimum aperture after dredging.

3. The finished product quality detection method of small molecule polysilicon pore throat dredging agent as claimed in claim 2, characterized in that, The specific method for setting the SEM image acquisition parameters also includes: Preset optimal pore pixel diameter , according to different pore sizes d and optimal pore pixel diameters, the magnifications of different pore sizes are calculated, and the magnifications satisfy the corresponding magnifications before dredging The value range or magnification after dredging The value range of Defining resolution based on different aperture sizes , and the resolution satisfies the corresponding value range of the resolution before dredging or the value range of the resolution after dredging.

4. The finished product quality detection method of small molecule polysilicon pore throat dredging agent as claimed in claim 3, characterized in that: The specific method for extracting the multi-scale pore structure characteristics before and after dredging based on the SEM image is: For micron-scale pores, the local texture features of the siloxane structure before and after dredging are extracted based on Gabor filters. , the parameters of the Gabor filter Including scale parameters , Direction parameters and frequency parameters ; For mesopores and macropores, their morphological features are extracted through morphological operations and region segmentation algorithms. and region segmentation features ; The local texture features of micron-scale pores are combined with the morphological features and regional segmentation features of mesopores and macropores by feature splicing to obtain multi-scale pore structure features. .

5. The finished product quality detection method of small molecule polysilicon pore throat dredging agent as claimed in claim 4, characterized in that, The scale parameters are set by obtaining the resolution r and magnification m of the SEM image and the pixel size range of the siloxane structure. , then the corresponding physical size range is ;in, , are the minimum and maximum values ​​of the physical size range respectively; Within the physical size range, a logarithmic sampling method is used to select scale parameter ,in, For the scale parameters, the i-th scale parameter is .

6. The finished product quality detection method of the small molecule polysilicon pore throat dredging agent as claimed in claim 5, characterized in that: The specific method of calculating the quantitative descriptor based on the SEM image and combining the multi-scale pore structure characteristics with the quantitative descriptor to obtain the comprehensive pore structure representation is: The quantitative descriptor D of the pores before and after dredging is calculated based on the SEM images. The quantitative descriptor includes the specific surface area Kong Rong and pore size distribution The BET method was used to calculate the specific surface area of ​​the pores, the mercury intrusion method was used to measure the pore volume, and the BJH method was used to calculate the pore size distribution of the pores; The multi-scale pore structure features are concatenated with the quantitative descriptors to obtain the comprehensive pore structure representation X.

7. The finished product quality detection method of the small molecule polysilicon pore throat dredging agent according to claim 6, characterized in that: The specific method of predicting the qualitative and quantitative indicators of the dredging agent quality detection based on the dredging effect prediction model by taking the comprehensive pore structure representation as input includes: Obtaining training samples for quality inspection of the finished product of the unclogging agent, wherein the training samples include a comprehensive pore structure representation of the pores before and after unclogging and corresponding qualitative index labels and quantitative index labels; A multi-scale convolutional neural network is designed, wherein the multi-scale convolutional neural network comprises n convolutional branches, each of which is used to extract pore structure feature maps of different scales. ; Design an attention mechanism module, introduce the attention mechanism, and calculate the attention weight of the pore structure feature map at each scale , obtain the weighted pore structure feature map of each scale, fuse the weighted pore structure feature map to obtain the weighted fusion feature z; Design a pore structure evolution modeling module, use a long short-term memory network, take the weighted fusion features at the current moment as input, and output the hidden state at the current moment ; A dredging effect prediction module is constructed, and the hidden state output by the pore structure evolution modeling module is used as input through the fully connected layer to predict the qualitative and quantitative indicators of the output dredging agent quality detection.

8. The finished product quality detection method of small molecule polysilicon pore throat dredging agent as claimed in claim 7, characterized in that: The specific method of predicting the qualitative and quantitative indicators of the dredging agent quality detection based on the dredging effect prediction model by taking the comprehensive pore structure representation as input also includes: Defining the cross entropy loss function for qualitative indicators And the mean square error loss function of quantitative indicators , the weighted sum of the cross entropy loss function and the mean square error loss function is used to obtain the total loss function ; The dredging effect prediction model is composed of a multi-scale convolutional neural network, an attention mechanism module, a pore structure evolution modeling module and a dredging effect prediction module. Training is performed based on training samples, with minimizing the total loss function as the training goal. The training ends when the total loss function converges. The qualitative and quantitative indicators of the quality inspection of the new dredging agent finished product are predicted based on the trained dredging effect prediction model.

9. A finished product quality inspection system for small molecule polysilicon pore throat dredging agent, which is used to implement the finished product quality inspection method for small molecule polysilicon pore throat dredging agent according to any one of claims 1 to 8, characterized in that: include: An SEM image acquisition module is used to set SEM image acquisition parameters, including magnification and resolution, and to acquire SEM images of pores of different scales before and after dredging; Multi-scale feature extraction module, used to extract multi-scale pore structure features before and after dredging based on SEM images; A comprehensive feature representation module is used to calculate quantitative descriptors based on SEM images, and to perform feature splicing of multi-scale pore structure features and quantitative descriptors to obtain a comprehensive pore structure representation; The quality detection output module is used to predict the qualitative and quantitative indicators of the dredging agent quality detection based on the dredging effect prediction model with the comprehensive pore structure representation as input.

Images