A hole structure identification method and device

Abstract

The application provides a hole structure identification method, comprising: obtaining a pre-processed hole structure image; extracting parameters of the image; the parameters comprise noise parameters and binarization threshold parameters; adjusting the parameters based on multiple algorithms; and identifying the hole structure by using the image and the adjusted parameters. Thus, the current fuel cell chip production line lacks efficient and accurate hole structure identification methods, and the scanning electron microscope image with a large number of gray areas, a large micro-hole identification difficulty, a small background color difference, and a large gray scale distribution difference can be efficiently, accurately and automatically identified and calculated.

Description

Technical Field

[0001] This application relates to the technical field of image recognition, and in particular to a method for identifying hole structures. Background Technology

[0002] In recent years, fuel cell technology has continued to develop, enabling the efficient utilization of hydrogen energy. The large-scale hydrogen energy industry has placed higher demands on the core components of fuel cells. Among these, the catalyst layer, as a core component of the electrochemical reaction in a fuel cell system, requires precise identification and calculation of its pore structure, which has become a crucial factor affecting fuel cell performance.

[0003] Current methods for identifying pore structures primarily rely on mercury intrusion porosimetry (MIP) for measurement, with scanning electron microscopy (SEM) serving as an auxiliary observation method. Traditional MIP measurements on a single sample are time-consuming, require sophisticated experimental techniques, and are resource-intensive. SEM relies solely on visual observation to identify pore structures, resulting in high time consumption and low accuracy. Furthermore, SEM images exhibit characteristics such as small differences in background and target color, numerous and difficult-to-identify gray transition regions, significant image noise, and substantial variations in brightness and contrast between different SEM images. Traditional fixed-threshold binarization methods require simultaneous adjustment of multiple algorithm parameters, leading to low efficiency and poor accuracy. Traditional methods for maximizing inter-class variance (MOP) require iterating through all parameters, are time-consuming, and do not consider image noise during the adaptive process, resulting in poor identification capabilities for micropores and mesopores.

[0004] Therefore, how to improve the efficiency and accuracy of hole structure identification and reduce identification costs is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, embodiments of this application provide a method and apparatus for identifying hole structures, aiming to improve the efficiency and accuracy of hole structure identification and reduce identification costs.

[0006] In a first aspect, embodiments of this application provide a method for identifying hole structures, including:

[0007] Obtain the preprocessed pore structure image;

[0008] Extract parameters from the image; the parameters include noise parameters and binarization threshold parameters;

[0009] The parameters are adjusted based on multiple algorithms, including a first algorithm, a second algorithm, and a third algorithm. The first algorithm is used to generate a target fitness function to evaluate fitness using the parameters. The second algorithm is used to accelerate the optimization of the target fitness function. The third algorithm is used to increase randomness in the optimization process. The evaluation values ​​of the results of the second algorithm and the third algorithm are compared to obtain an adjustment scheme. The adjustment scheme is used to indicate the adjustment of the parameters.

[0010] The hole structure is identified using images and adjusted parameters.

[0011] Preferably, the identification of the hole structure using the image and adjusted parameters includes:

[0012] Using images and adjusted parameters, the topology of the holes is determined and the hole structure outline is marked.

[0013] Preferably, the identification of the hole structure using the image and adjusted parameters includes:

[0014] Calculate the hole area and equivalent hole diameter for all holes.

[0015] Preferably, before acquiring the preprocessed hole structure image, the method further includes:

[0016] Obtain scanning electron microscope (SEM) images of the cross-section;

[0017] Determine whether the dimension of the scanning electron microscope image is a preset value;

[0018] If so, convert to grayscale;

[0019] If not, import the original image.

[0020] Preferably, before obtaining the image and adjusted parameters to identify the hole structure, the method further includes:

[0021] The pore structure is expanded.

[0022] Secondly, embodiments of this application provide a hole structure identification device, comprising:

[0023] The first acquisition module is used to acquire the preprocessed hole structure image;

[0024] An image parameter extraction module is used to extract parameters from the image; the parameters include noise parameters and binarization threshold parameters.

[0025] An adaptive adjustment module is used to adjust the parameters based on multiple algorithms. The multiple algorithms include a first algorithm, a second algorithm, and a third algorithm. The first algorithm is used to generate a target fitness function to evaluate fitness using the parameters. The second algorithm is used to accelerate the optimization of the target fitness function. The third algorithm is used to increase randomness in the optimization and compare the evaluation values ​​of the results of the second algorithm and the third algorithm to obtain an adjustment scheme. The adjustment scheme is used to indicate the adjustment of the parameters.

[0026] The recognition module is used to identify hole structures using images and adjusted parameters.

[0027] Preferably, the identification module includes:

[0028] Hole structure segmentation units are used to determine the topology of holes and mark the hole structure outline using images and adjusted parameters.

[0029] Preferably, the identification module includes:

[0030] A pore size distribution and porosity calculation unit is used to calculate the pore area and equivalent pore size of all pores.

[0031] Thirdly, embodiments of this application provide a device including a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code to cause the device to perform the hole structure recognition method described in any of the first aspects above.

[0032] Fourthly, embodiments of this application provide a computer storage medium storing code, wherein when the code is executed, a device running the code implements the hole structure identification method described in any of the first aspects above.

[0033] This application provides a method for identifying pore structures. When executing the method, a preprocessed image of the pore structure is first acquired, and then parameters of the image are extracted. These parameters include noise parameters and a binarization threshold parameter. Then, the parameters are adjusted based on multiple algorithms. These multiple algorithms include a first algorithm, a second algorithm, and a third algorithm. The first algorithm is used to generate a target fitness function to evaluate fitness using the parameters. The second algorithm is used to accelerate the optimization of the target fitness function. The third algorithm is used to increase randomness in the optimization process. The evaluation values ​​of the results of the second and third algorithms are compared to obtain an adjustment scheme. This adjustment scheme is used to indicate the adjustment of the parameters. Finally, the image and the adjusted parameters are used to identify the pore structure. In this way, by adjusting the parameters through multiple algorithms, efficient and accurate pore structure identification is achieved. Thus, compared with the prior art, this application has the following beneficial effects: This application fills the technical gap of lacking a high-efficiency and high-precision pore structure identification method in fuel cell chip production lines. It can efficiently, accurately and automatically identify and measure scanning electron microscope images with many gray areas, great difficulty in identifying micro / mesopores, small background / target color difference and large gray distribution difference, which is of great significance to the large-scale development of fuel cell technology. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart of one method of the hole structure identification method provided in the embodiments of this application;

[0036] Figure 2 A flowchart illustrating the preprocessing of the hole structure provided in this application embodiment;

[0037] Figure 3 A flowchart illustrating another method for identifying hole structures provided in an embodiment of this application;

[0038] Figure 4 A scanning electron microscope image of sample 1 provided in the embodiments of this application;

[0039] Figure 5 Scanning electron microscope image of sample 2 provided in the embodiments of this application;

[0040] Figure 6 A grayscale image of sample 1 provided in the embodiments of this application;

[0041] Figure 7 A grayscale image of sample 2 provided in the embodiments of this application;

[0042] Figure 8 A schematic diagram of Gaussian blur operation provided for an embodiment of this application;

[0043] Figure 9 The grayscale image of Sample 1 after Gaussian blurring provided in the embodiments of this application;

[0044] Figure 10 The binarized and inverted grayscale image of Sample 1 provided in the embodiments of this application;

[0045] Figure 11 The grayscale image of Sample 2 after Gaussian blurring provided in the embodiments of this application;

[0046] Figure 12 The binarized and inverted grayscale image of sample 2 provided in the embodiments of this application;

[0047] Figure 13 The grayscale image of sample 1 after morphological operation of the dilated image provided in the embodiments of this application;

[0048] Figure 14 The grayscale image of sample 2 after morphological operation of the dilated image provided in the embodiments of this application;

[0049] Figure 15 A pore structure outline diagram of sample 1 provided in the embodiments of this application;

[0050] Figure 16 A pore structure outline diagram of sample 2 provided in the embodiments of this application;

[0051] Figure 17Pore ​​size distribution and pore area distribution diagram of sample 1 provided for embodiments of this application;

[0052] Figure 18 Pore ​​size distribution and pore area distribution diagram of sample 2 provided in the embodiments of this application;

[0053] Figure 19 This is a schematic diagram of a hole structure identification device provided in an embodiment of this application. Detailed Implementation

[0054] In recent years, with the large-scale development of fuel cell technology, its ecosystem has placed higher demands on the efficiency, precision, and low cost of its core components. The core components of a fuel cell mainly include structures such as catalytic layers, diffusion layers, and catalysts, which vary in brightness, contrast, size, grayscale distribution, materials used in their fabrication, and their proportions. Improving the efficiency and accuracy of identifying the pore structures of catalytic layers, diffusion layers, and catalysts while reducing costs will significantly enhance the overall fuel cell manufacturing process.

[0055] Existing technologies for pore structure identification include mercury intrusion porosimetry, gas adsorption, optical scattering, and scanning electron microscopy (SEM). However, all of these methods suffer from high experimental requirements and long computation times, making them unsuitable for mass production. SEM, in particular, requires visual observation of the pore structure, resulting in high costs and low efficiency. Furthermore, it struggles with the characteristics of catalytic layer SEM images, such as small differences in background and target color, numerous and difficult-to-identify gray transition areas, significant image noise, and large differences in brightness and contrast between different SEM images. Traditional fixed-threshold binarization methods require simultaneous adjustment of multiple algorithm parameters, leading to low efficiency and poor accuracy. Traditional methods for maximizing inter-class variance (MOP) require iterating through all parameters, are time-consuming, and do not consider image noise during the adaptive process, resulting in poor identification capabilities for micropores and mesopores.

[0056] Therefore, in view of the above-mentioned problems of electron microscopy scanning method, this application proposes a pore structure identification method that improves the efficiency and accuracy of pore structure identification and reduces the identification cost.

[0057] The method provided in this application embodiment is executed by a computer device and is used to identify hole structures.

[0058] Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0059] See Figure 1 , Figure 1 A flowchart of a method for identifying hole structures provided in this application embodiment includes:

[0060] Step S101: Obtain the preprocessed hole structure image.

[0061] Preprocessing of the pore structure images includes sample preparation using polishing equipment and scanning electron microscopy (SEM) imaging to obtain SEM images of the hierarchical structure cross-section. See also Figure 2 , Figure 2 A flowchart illustrating the preprocessing of the hole structure provided in this application embodiment.

[0062] Step S102: Extract the parameters of the image.

[0063] The image parameters include image noise parameters and binarization threshold parameters. The noise parameters define the noise reduction effect of the hierarchical interface scanning electron microscope image, while the binarization threshold parameters define the difference between the solid structure and pore structure of the catalyst in the binarized segmentation, improving the recognition accuracy of the pore structure. Preferably, Gaussian blur can be used to define the noise parameters.

[0064] Step S103: Adjust the parameters based on multiple algorithms.

[0065] First, the first algorithm uses the parameters to define a target fitness function, which is used to evaluate the fitness of the parameters. Second, the second algorithm accelerates the optimization of the target fitness function, preventing excessive time consumption from traversing all parameters. Finally, the third algorithm adds randomness to the optimization process, preventing accuracy degradation caused by local optimization, and compares the evaluation values ​​of the second and third algorithms to obtain the optimal image parameters.

[0066] Step S104: Identify the hole structure using the image and adjusted parameters.

[0067] Traverse the contours of the pore structure image, mark the topological structure between all pore contours, convert it into a three-channel image to outline the shape of all pore structures and number them. At the same time, the pore area and equivalent pore diameter of all pores can be calculated.

[0068] In the embodiments of this application, the above Figure 1 There are multiple possible implementations of steps S101-S104, which will be described below. It should be noted that the implementations given below are merely illustrative examples and do not represent all implementations of the embodiments of this application.

[0069] See Figure 3 The figure is a flowchart of another method for identifying hole structures provided in an embodiment of this application.

[0070] Step S301: Obtain the preprocessed hole structure image.

[0071] The preprocessing of the pore structure images includes sample preparation using polishing equipment and scanning electron microscopy (SEM) imaging to obtain SEM images of the cross-section of the hierarchical structure.

[0072] As one possible implementation, the `cvtColor` function from the OpenCV software library can be used to determine the dimension of the electron microscope image. If the dimension is a preset value, it is converted to a grayscale image; otherwise, the original image is imported. Preferably, the preset value can be 3. This reduces the error in hole structure recognition caused by discoloration errors in scanning electron microscopy, automatically identifies and statistically analyzes the grayscale value distribution of all pixels, and then uses this as the input image for subsequent steps.

[0073] For example, Figure 4 , Figure 5 The images shown are the original scanning electron microscope (SEM) images of Sample 1 and Sample 2, respectively. The original SEM images can be used to determine whether to convert them to grayscale images. Sample 1 and Sample 2 need to be converted. Figure 6 , Figure 7 These are the grayscale images of sample 1 and sample 2, respectively. As can be seen from the images, Figure 6 It exhibits a distinct bimodal characteristic, but the valley value between the two peaks is relatively high, resulting in a large gray transition area, making it difficult to identify; while Figure 7 The black areas, which do not exhibit obvious bimodal characteristics and have a gray level close to 0, are relatively concentrated in area. Figure 6 The differences are significant. However, all of the above samples can be accurately identified by this application, indicating that this application has good adaptability.

[0074] Step S302: Extract the parameters of the image.

[0075] The image parameters include image noise parameters and binarization threshold parameters. The noise parameters are used to define the noise reduction effect of the hierarchical interface scanning electron microscope image, and the binarization threshold parameters are used to define the difference between the binarized segmented cross-sectional structure and the pore structure, thereby improving the recognition accuracy of the pore structure.

[0076] As one possible implementation, Gaussian blurring can be used to eliminate noise after the noise parameters are extracted.

[0077] Specifically, first define the grayscale value I(x,y) for all pixels (x,y) in the grayscale image, where (x,y) represents the coordinates of a pixel in the grayscale image, and I(x,y) represents the grayscale value of the pixel at coordinates (x,y); then, construct a system with radius r g The side length is 2r g A Gaussian convolution kernel of +1 is used to construct a two-dimensional normally distributed weight matrix. Specifically, the elements G(u, v) in the matrix can be determined by the following formula:

[0078]

[0079] Where u and v are the u-th row and v-th column of the Gaussian convolution kernel, respectively, G(u,v) represents the weight value of the Gaussian convolution kernel on (u,v), and σ is the Gaussian standard deviation; finally, as Figure 8 As shown, the center point of the weight matrix is ​​sequentially mapped to the grayscale value of each pixel in the grayscale image. The inner product of the weight matrix and a pixel matrix of equal size is calculated, and this product is then assigned to the center point of the pixel matrix, thus obtaining the Gaussian blur value. Specifically, the Gaussian blur value can be determined using the following formula:

[0080]

[0081] in, Let I(x,y) represent the Gaussian blur value, where I(x,y) is the grayscale value of the pixel at coordinates (x,y) in the Gaussian blurred grayscale image. Its value is a value within the closed interval from 0 to 255. u and v are the u-th row and v-th column of the Gaussian convolution kernel, respectively. g Let G(u,v) be the radius of the Gaussian convolution kernel, and let G(u,v) represent the weight of the Gaussian convolution kernel on (u,v).

[0082] Repeat the above steps until all pixels of the original image are updated with Gaussian blur values. Achieve smoothing and continuity of catalyst and pore structure boundaries, and eliminate noise.

[0083] As one possible implementation, the extracted binarization parameter can be the grayscale value after Gaussian blurring. Figure 2 The binarization is inverted, defining pixel grayscale values ​​greater than the binarization threshold `thres` as 0, and vice versa as 255. This means that black holes and gray transition areas in the original image are defined as white, and white areas as black. Next, to achieve adaptive parameter tuning and determine the optimal value of the binarization threshold `thres` for adaptive recognition of gray transition areas, the probability and grayscale value of identifying the image as a catalyst and a pore structure after binarization are calculated, as shown in the following formula:

[0084] When i > thres, i = 0; when i ≤ thres, i = 255;

[0085]

[0086] Where thres is the binarization segmentation threshold, p i Let i be the probability of each pixel being updated to a grayscale value after binarization and inversion, and n be the probability of n being a value of i. i p represents the number of pixels with grayscale value i after binarization and inversion. catalyst,thres m represents the probability that a pixel is assigned to be identified as a catalyst. catalyst The average grayscale value of the pixel identified as the catalyst, m all This represents the average grayscale value of all pixels after binarization and inversion. Here, because parameters i and n...i Both are generated by Gaussian blur values, and Gaussian blur values ​​are generated by r. g It was decided, so m all m catalyst and p catalyst,thres The radius r of the Gaussian convolution kernel g It is determined together with the binarization threshold thres.

[0087] Step S303: Adjust the parameters based on multiple algorithms.

[0088] First, the first algorithm uses the parameter composite to define a target fitness function, which is used to evaluate the fitness of the parameters as they are adjusted. Second, the second algorithm accelerates the optimization of the target fitness function, preventing excessive time consumption from traversing all parameters. Finally, the third algorithm adds randomness to the optimization process, preventing accuracy degradation caused by local optimization, and compares the evaluation values ​​of the second and third algorithms to obtain the optimal image parameters.

[0089] As one possible implementation, the first algorithm can first establish the swarm size N of the particle swarm algorithm based on the initialized scanning electron microscope source image and grayscale histogram. pop The initial velocity v and position x are used to extract the Gaussian convolution kernel radius r from the image parameter extraction module. g The target fitness function f(r) is defined by combining the binarized threshold thres. g The target fitness function f(r) is evaluated. g A larger value for `thres` means a lower probability of misclassification and better overall performance. g Thres is a better solution.

[0090] Specifically, the target fitness function can be determined by the following formula:

[0091]

[0092] Where, r g p is the radius of the Gaussian convolution kernel, thres is the binarization threshold, and p is the value of p. catalyst,thres m represents the probability that a pixel is assigned to be identified as a catalyst. catalyst The average grayscale value of the pixel identified as the catalyst, m all This is the average grayscale value of all pixels after binarization and inversion.

[0093] As one possible implementation, the second algorithm can accelerate the image parameter optimization process using the PSO (Particle Swarm Optimization) algorithm. The optimal individual p of all particles is updated according to the PSO algorithm. best The optimal individual value is the global optimal value g of the cycle. bestBased on the individual's optimal p in the s-th cycle. best and global best g best Update the velocity and position of all particles.

[0094] Specifically, the velocity of a particle can be determined by the following formula:

[0095] v(s+1)=w(s)v(s)+c1r1[p best -x(s)]+c2r2[g best -x(s)],

[0096] The position of a particle can be determined by the following formula:

[0097] x(s+1)=x(s)+v(s+1),

[0098]

[0099] Where v and x are the particle's velocity and position, respectively, s is the s-th cycle, w(s) is the inertia weight of the s-th cycle, c1 and c2 are learning factors, r1 and r2 are random numbers in the open interval between 0 and 1, and p best and g best For the best individual and global outcome, w s and w e s represents the initial and maximum iteration weights, respectively. max This represents the maximum number of loops.

[0100] For example, in sample 1, c1 and c2 are both 2, r1 and r2 are random numbers in the open interval between 0 and 1, and p best and g best For the best individual and global outcome, w s and w e These are the inertia weights for the initial iteration and the iteration up to the maximum number of iterations, specifically 0.9 and 0.4, respectively. max The maximum number of iterations is 50. For sample 2, c1 and c2 are both 2, r1 and r2 are random numbers in the open interval between 0 and 1, and p... best and g best For the best individual and global outcome, w s and w e These are the inertia weights for the initial iteration and the iteration up to the maximum number of iterations, specifically 0.9 and 0.4, respectively. max This represents the maximum number of loops, specifically 50.

[0101] As one possible implementation, the third algorithm could be a GA genetic algorithm, which, after completing one PSO particle swarm algorithm update of the positions and velocities of all particles, uses a GA genetic algorithm to select, crossover, and mutate to generate new random particles.

[0102] First, select two particles, x1(s+1) and x2(s+1), from all the updated particles according to a certain probability. Specifically, the probability can be determined by the following formula:

[0103]

[0104] Among them, P selection (n pop ) is the nth pop The target fitness value of each particle, N pop The total number of particles in the particle swarm population is the population size.

[0105] Secondly, under the condition of satisfying the crossover and mutation probabilities, crossover and mutation are performed on the two selected particles.

[0106] As a preferred option, when r c <p crossover At that time, x1′(s+1)=a×x1(s+1)+(1-a)×x2(s+1);

[0107] When r c <p crossover At that time, x2′(s+1)=a×x2(s+1)+(1-a)×x1(s+1);

[0108] Where x1′(s+1) and x2′(s+1) are the positions of the two particles after the crossover, x1(s+1) and x2(s+1) are the positions after the selection operation, and a, b, r are the positions after the crossover. c and r m p are random numbers within the open interval from 0 to 1. crossover and p multation These represent the probabilities of crossover and mutation, respectively.

[0109] Finally, when r m <p crossover At that time, the particles are updated according to certain rules. Specifically, the rules can be determined by the following formula:

[0110] x′(s+1)=x(s+1)+0.1×b×[max(x(s+1)-min(x(s+1)],

[0111] Then, the Otsu fitness value is calculated based on the target fitness function, and compared with the individual best value of the PSO particle swarm optimization algorithm to update the individual and global best values. This process is repeated until the maximum number of iterations s is met. max Thus, the optimal image parameters that achieve the best denoising and binarization results can be found, i.e., the optimal Gaussian convolution kernel radius r is obtained. g And the binarization threshold thres.

[0112] For example, the total number of particles in the swarm population of sample 1 is 300, p crossover and p multation The probabilities of crossover and mutation are 0.6 and 0.04, respectively. The Gaussian convolution kernel radius of sample 1 is 4, and the effect is as follows: Figure 9 As shown, the contrast between the gray values ​​of the catalyst and the pore structure is more obvious after Gaussian blurring, eliminating noise in the scanning electron microscope image of the catalyst layer; the binarization threshold for sample 1 is 104, and the binarization effect after inverting the gray values ​​of the pore structure and the catalyst is as follows. Figure 10 As shown, most of the pore structures have been identified as white areas. The total number of particles in the particle swarm population of Sample 2 is 300, p crossover and p multation The probabilities of crossover and mutation are 0.6 and 0.04, respectively. The Gaussian convolution kernel radius of sample 2 is 3, and the effect is as follows: Figure 11 As shown, the contrast between the gray values ​​of the catalyst and the pore structure is more obvious after Gaussian blurring, eliminating noise in the scanning electron microscope image of the catalyst layer; the binarization threshold for sample 2 is 83, and the binarization effect after inverting the gray values ​​of the pore structure and the catalyst is as follows. Figure 12 As shown, most of the pore structures have been identified as white areas.

[0113] Step S304: Expand the pore structure.

[0114] Preferably, dilatational morphological operations can be used to partially expand the porous structure. The ratio of the actual size of the catalyst layer in the SEM image to the micropores and the ratio of the mesopores to the SEM image are within a fixed range. By using a dilatational convolution kernel of a fixed radius and iterating a fixed number of times, the ability to identify micropores and mesopores with an equivalent pore diameter less than 50 nm can be improved. This step can improve the identification of micropores with a pore size less than 2 nm and mesopores with a pore size greater than 2 nm but less than 50 nm.

[0115] For example, the effect of the dilation morphology operation on sample 1 is: Figure 13 The effect of the dilatation morphology operation on sample 2 is Figure 14 This allows us to see micropores with a pore size of less than 2 nanometers and mesopores with a pore size of more than 2 nanometers but less than 50 nanometers.

[0116] Step S305: Identify the hole structure using the image and adjusted parameters.

[0117] First, using the image and adjusted parameters, the topology of the hole is determined and the hole structure outline is marked.

[0118] Preferably, the module uses a tree-like topology to classify the pore structure into levels, traverses all contours of the expanded catalytic layer interface map, marks the topology between all pore contours, converts them into a three-channel image to outline the shape of all pore structures and number them n, with a total number of pores N.

[0119] For example, sample 1 has a total of 165 pores, and the pore structure outline is shown in the figure below. Figure 15 Sample 2 has a total of 223 pores, and the pore structure outline is shown in the figure below. Figure 16 .

[0120] Next, calculate the pore area and equivalent pore diameter of all pores.

[0121] As a preferred approach, the equivalent pore diameter is categorized by the nearest equivalent pore diameter array (observer) that needs to be observed in the catalyst layer pore structure. x The array length is X, which is the equivalent aperture D. n The observer is larger than the adjacent fixed equivalent hole diameter array. x-1 and observer x The average value, and less than the observer's x and observer x+1 The average value of the nth true equivalent hole diameter is categorized as observer. x The aperture area remains S n The method for calculating the equivalent pore diameter D(n) and porosity ε of the catalytic layer cross-section using scanning electron microscopy is as follows:

[0122]

[0123] Among them, D n and S n , respectively, represent the equivalent pore diameter and area of ​​pore number N, and l and h are the length and width of the actual dimensions of the catalyst layer in the scanning electron microscope image.

[0124] For example, Figure 17 The diagram shows the pore size distribution and pore area distribution of sample 1. Figure 18 The diagram shows the pore size distribution and pore area distribution of sample 2.

[0125] The above are some specific implementations of the hole structure identification method provided in the embodiments of this application. Based on this, this application also provides a corresponding device. The device provided in the embodiments of this application will be described below from the perspective of functional modularity.

[0126] See Figure 19 The schematic diagram of the hole structure recognition device 1900 shown includes a first acquisition module 1901, an image parameter extraction module 1902, an adaptive adjustment module 1903, and a recognition module 1904.

[0127] The first acquisition module 1901 is used to acquire the preprocessed hole structure image;

[0128] Image parameter extraction module 1902 is used to extract parameters of the image; the parameters include noise parameters and binarization threshold parameters;

[0129] The adaptive adjustment module 1903 is used to adjust the parameters based on multiple algorithms. The multiple algorithms include a first algorithm, a second algorithm, and a third algorithm. The first algorithm is used to generate a target fitness function to evaluate fitness using the parameters. The second algorithm is used to accelerate the optimization of the target fitness function. The third algorithm is used to increase randomness in the optimization and compare the evaluation values ​​of the results of the second algorithm and the third algorithm to obtain an adjustment scheme. The adjustment scheme is used to indicate the adjustment of the parameters.

[0130] The recognition module 1904 is used to identify hole structures using images and adjusted parameters.

[0131] The identification module includes:

[0132] Hole structure segmentation units are used to determine the topology of holes and mark the hole structure outline using images and adjusted parameters.

[0133] The identification module includes:

[0134] A pore size distribution and porosity calculation unit is used to calculate the pore area and equivalent pore size of all pores.

[0135] The device further includes:

[0136] The second acquisition module is used to acquire scanning electron microscope images of the cross-section;

[0137] The judgment module is used to determine whether the dimension of the scanning electron microscope image is a preset value;

[0138] If so, convert to grayscale;

[0139] If not, import the original image.

[0140] Expansion modules are used to expand the hole structure.

[0141] This application also provides corresponding devices and computer storage media for implementing the solutions provided in this application.

[0142] The device includes a memory and a processor. The memory stores instructions or code, and the processor executes the instructions or code to enable the device to perform the hole structure recognition method according to any embodiment of this application.

[0143] The computer storage medium stores code, and when the code is run, the device running the code implements the hole structure identification method described in any embodiment of this application.

[0144] In the embodiments of this application, the terms "first" and "second" (if they exist) are used only as name identifiers and do not represent the order of first and second.

[0145] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus a general-purpose hardware platform. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0146] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0147] The above description is merely an exemplary implementation of this application and is not intended to limit the scope of protection of this application.

Claims

1. A hole structure identification method characterized by, The method comprises: obtaining a pre-processed pore structure image; extracting parameters of the image; the parameters comprise a noise parameter and a binarization threshold parameter; adjusting the parameters based on multiple algorithms; the multiple algorithms comprise a first algorithm, a second algorithm and a third algorithm, the first algorithm is used to generate a target fitness function to evaluate fitness by using the parameters, the second algorithm is used to accelerate optimization of the target fitness function, and the third algorithm is used to increase random optimization, and evaluation values of results of the second algorithm and the third algorithm are compared to obtain an adjustment scheme, the adjustment scheme is used to indicate adjustment of the parameters; identifying a pore structure by using the image and the adjusted parameters; the pore structure is a pore structure of a fuel cell catalyst layer, a diffusion layer and a catalyst; The first algorithm establishes a group size N of the particle swarm algorithm according to an initialized scanning electron microscope source image and a gray scale histogram pop , an initial velocity v and a position x, and evaluates a target fitness function f(r g , thres) according to a Gaussian convolution kernel radius r g and a binary threshold value thres of the image parameter extraction module; when the value of the target fitness function f(r g , thres) is larger, it means that the probability of misclassification is smaller and the overall effect is better, and r g and thres at this time are better solutions Specifically, the target fitness function is determined by the following formula: ； where r g is the radius of the Gaussian kernel, thres is the threshold for binarization, p catalyst,thres is the probability of a pixel being classified as catalyst, m catalyst is the average gray value of the pixels classified as catalyst, m all is the average gray value of all pixels after binarization inversion. The second algorithm accelerates the image parameter optimization process by a PSO particle swarm algorithm; individual bests p best of all particles are updated according to the PSO particle swarm algorithm best ; speeds and positions of all particles are updated according to individual bests p best and global best g best of the s th round of circulation. The third algorithm is a GA genetic algorithm, and after updating the positions and speeds of all particles by the PSO particle swarm algorithm once, new random particles are generated by selection, crossover and mutation of the GA genetic algorithm; The evaluation values of the results of the second algorithm and the third algorithm are compared to obtain the adjustment scheme, and the adjustment scheme is used to indicate adjustment of the parameters, and specifically comprises: According to the target fitness function, the Otsu fitness value is calculated, and the individual and global optimal values are updated after being compared with the individual optimal value of the PSO particle swarm algorithm, and the loop is repeated until the maximum iteration number s is met max At this point, the image parameters that make the denoising and binarization effects optimal can be optimized, that is, the optimal Gaussian convolution kernel radius r g and the binarization threshold thres are obtained.

2. The method of claim 1, wherein, The identification of the pore structure by using the image and the adjusted parameters comprises: The topological structure of the pores is determined and the pore structure contour is marked by using the image and the adjusted parameters.

3. The method of claim 1, wherein, The identification of the pore structure by using the image and the adjusted parameters comprises: The pore area and the equivalent pore diameter of all pores are calculated.

4. The method of claim 1, wherein, Before obtaining the pre-processed pore structure image, the method further comprises: obtaining a scanning electron microscope image of a section; determining whether the dimension of the scanning electron microscope image is a preset value; if yes, converting the scanning electron microscope image into a gray-scale image; if no, importing the original image.

5. The method of claim 1, wherein, Before identifying the pore structure by using the image and the adjusted parameters, the method further comprises: dilating the pore structure.

6. A hole structure identification device characterized by comprising: The device comprises: a first obtaining module configured to obtain a pre-processed pore structure image; an image parameter extraction module configured to extract parameters of the image; the parameters comprise a noise parameter and a binarization threshold parameter; an adaptive adjustment module configured to adjust the parameters based on multiple algorithms; the multiple algorithms comprise a first algorithm, a second algorithm and a third algorithm, the first algorithm is used to generate a target fitness function to evaluate fitness by using the parameters, the second algorithm is used to accelerate optimization of the target fitness function, and the third algorithm is used to increase random optimization, and evaluation values of results of the second algorithm and the third algorithm are compared to obtain an adjustment scheme, the adjustment scheme is used to indicate adjustment of the parameters; an identification module configured to identify a pore structure by using the image and the adjusted parameters; the pore structure is a pore structure of a fuel cell catalyst layer, a diffusion layer and a catalyst; The first algorithm establishes a group size N of the particle swarm algorithm according to an initialized scanning electron microscope source image and a gray scale histogram pop , an initial velocity v and a position x, and evaluates a target fitness function f(r g , thres) according to a Gaussian convolution kernel radius r g and a binary threshold thres of the image parameter extraction module; when the value of the target fitness function f(r g , thres) is larger, it means that the probability of misclassification is smaller, and the overall effect is better, and r g and thres at this time are better solutions; Specifically, the target fitness function is determined by the following formula: ； where r g is the radius of the Gaussian kernel, thres is the threshold for binarization, p catalyst,thres is the probability of a pixel being classified as catalyst, m catalyst is the average gray value of the pixels classified as catalyst, m all is the average gray value of all pixels after binarization inversion. The second algorithm accelerates the image parameter optimization process using the PSO (Particle Swarm Optimization) algorithm; it updates the optimal individual p of all particles according to the PSO algorithm. best The optimal individual is the global optimal g in the s-th cycle. best Based on the individual's optimal p in the s-th cycle best and global best g best Update the velocity and position of all particles; The third algorithm is a GA genetic algorithm, and after updating the positions and speeds of all particles by the PSO particle swarm algorithm once, new random particles are generated by selection, crossover and mutation of the GA genetic algorithm; The evaluation values of the results of the second algorithm and the third algorithm are compared to obtain the adjustment scheme, and the adjustment scheme is used to indicate adjustment of the parameters, and specifically comprises: According to the target fitness function, the Otsu fitness value is calculated, and the individual and global optimal values are updated after being compared with the individual optimal value of the PSO particle swarm algorithm, and the loop is repeated until the maximum iteration number s is met max At this point, the image parameters that make the denoising and binarization effects optimal can be optimized, that is, the optimal Gaussian convolution kernel radius r g and the binarization threshold thres are obtained.

7. The apparatus of claim 6, wherein, The identification module comprises: A pore structure dividing unit is configured to determine the topological structure of the pores and mark the pore structure contour by using the image and the adjusted parameters.

8. The apparatus of claim 6, wherein, The identification module comprises: A pore size distribution and porosity calculating unit is configured to calculate the pore area and equivalent pore size of all the pores.

9. An apparatus, comprising: The device comprises a memory and a processor, the memory is configured to store instructions or codes, and the processor is configured to execute the instructions or codes to enable the device to perform the pore structure identification method according to any one of claims 1 to 5.

10. A computer storage medium, characterized in that, The computer storage medium stores codes, and when the codes are executed, a device running the codes implements the pore structure identification method according to any one of claims 1 to 5.

Images