An aps coating microscopic image porosity identification method

Abstract

The present application relates to a kind of APS coating microscopic image porosity identification method, belong to coating material microscopic structure image processing field.This method obtains the microscopic image of same APS coating sample under multiple magnifications, after gray scale normalization, calculate gray statistics along horizontal direction, identify and cut off the non-microscopic structure area of image bottom that meet dark area condition, obtain effective microscopic structure area;At least two threshold methods are used to generate pore candidate results, determine high-confidence pore area and uncertain area according to recognition consistency;Connected domain extraction is carried out to high-confidence pore area, calculate topographic parameter and complete pore classification, calculate single image modified porosity after removing noise or small pore;Again, according to magnification, the modified porosity and pore type distribution are calculated, and the comprehensive porosity, multiple consistent evaluation results and pore type distribution results are output.The present application can reduce pore misjudgment, improve the stability, accuracy and traceability of characterization.

Description

Technical Field

[0001] This invention relates to the field of microscopic image processing of coating materials, and in particular to a method for identifying porosity in microscopic images of APS coatings. Background Technology

[0002] Atmospheric Plasma Spraying (APS) coatings are widely used in wear-resistant, protective, heat-insulating, and functional coating systems. The microstructure of APS coatings typically includes structures such as circular or compact pores, elliptical pores, interlaminar gaps, crack-like pores, abnormally large pores, and irregular pores. Porosity and pore morphology affect the mechanical properties, thermophysical properties, service reliability, and failure behavior of the coating; therefore, stable and traceable quantitative characterization of porosity in microscopic images is of practical significance. Existing image-based porosity measurements typically binarize microscopic images and statistically analyze the ratio of pore pixel area to total area.

[0003] However, existing image-based porosity measurement methods have the following problems: SEM images often have instrument parameter bars, scale bars, text, or other non-microscopic tissue areas at the bottom. These areas may present large areas of dark color or high-contrast boundaries. If directly used in binarization, they are easily misjudged as pores, leading to an artificially high porosity; single threshold methods are sensitive to image brightness, contrast, sample preparation status, local shadows, and noise, and the pore areas obtained by different threshold methods may differ significantly; traditional methods output only a total porosity, making it difficult to simultaneously provide the distribution of types such as circular pores, lamellar pores, crack-type pores, and abnormally large pores; single magnification images are greatly affected by the field of view selection, making it difficult to take into account large-scale pores, small pores, and local tissue differences; deep learning segmentation methods usually require a large amount of pixel-level labeled data, which is not suitable for the rapid analysis of small batches of APS-coated samples in the laboratory. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method for identifying the porosity of APS coatings in microscopic images. This method can automatically exclude non-microscopic tissue areas in SEM images, integrate multiple threshold methods to reduce misjudgments, output quantitative results including the distribution of types such as circular pores, lamellar pores, crack-type pores, and abnormally large pores, and support image statistics at multiple magnifications, thereby achieving stable, accurate, and traceable characterization of the porosity of APS coatings.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] A method for identifying porosity in APS coating micrographs includes: Acquire microscopic images of the same APS-coated sample at multiple magnifications; The microscopic image is normalized to obtain the working image; Calculate the grayscale statistical value along the horizontal direction of the working image, identify the continuous row region located at the bottom of the image and whose grayscale statistical value meets the preset dark area condition, and cut off the continuous row region after determining it as a non-microscopic tissue region to obtain the effective microscopic tissue region. At least two thresholding methods were used to generate pore candidate results for the effective microstructure region; Based on the consistency of the pore candidate results, pixel regions that are identified as pores by at least two thresholding methods are defined as high-confidence pore regions, and pixel regions that are identified as pores by only one thresholding method are defined as uncertain regions. Connectivity extraction is performed on the high-confidence pore region to obtain multiple pore candidate regions; Calculate the morphological parameters for each of the candidate pore regions; The candidate pore regions are classified according to the morphological parameters. After removing candidate pore regions classified as noise or micropores, the ratio of the sum of the areas of the remaining candidate pore regions to the area of ​​the effective microstructure region is used as the corrected porosity of a single image. The porosity and pore type distribution are statistically corrected according to the multiple magnification factors, and the comprehensive porosity, multi-magnification consistency evaluation results and pore type distribution results of the APS coating are output.

[0007] Furthermore, in the above method, the step of normalizing the grayscale of the microscopic image to obtain the working image includes: Convert 16-bit grayscale microscopic images to 8-bit working images while preserving the original image files.

[0008] Furthermore, in the above method, the step of calculating grayscale statistical values ​​along the horizontal direction of the working image and identifying continuous row regions located at the bottom of the image whose grayscale statistical values ​​meet the preset dark area conditions includes: Calculate grayscale statistics along the row direction of the working image, wherein the grayscale statistics include at least one of the row grayscale median, mean, or lower quantile; Scan consecutive line regions from the bottom of the image upwards; When multiple consecutive rows of grayscale statistical values ​​are lower than a preset threshold, or when the grayscale decreases by a preset amount relative to the area above, the consecutive rows are identified as the SEM instrument parameter bar, scale bar, or text bar.

[0009] Furthermore, in the above method, the thresholding method includes at least two of the following: Otsu threshold, low grayscale percentile threshold, and mean minus multiple standard deviation threshold.

[0010] Furthermore, in the above method, the morphological parameters include area, aspect ratio, roundness, fill rate, and orientation angle.

[0011] Furthermore, in the above method, classifying the candidate pore regions based on the morphology parameters includes: When the area of ​​the candidate pore region is lower than the preset area threshold, it is classified as noise or micropore; When the aspect ratio of the candidate pore region is higher than the preset aspect ratio threshold, it is classified as a crack-type pore. When the area of ​​the candidate pore region is higher than the preset large pore threshold, it is classified as an abnormal large pore. When the roundness of a candidate pore region is higher than a preset roundness threshold, it is classified as a circular or compact pore.

[0012] Furthermore, in the above method, the classification types of the pore candidate regions also include elliptical pores or lamellar gaps and irregular pores.

[0013] Furthermore, in the above method, the step of statistically correcting the porosity and pore type distribution according to the multiple magnification factors includes: The corrected porosity, standard deviation, minimum value, maximum value, proportion of uncertain region, and pore type distribution under different magnification factors were statistically analyzed. Based on the statistical results at different magnifications, the comprehensive porosity and multi-magnification consistency evaluation results of the APS coating are output.

[0014] Furthermore, the above method also includes: A cropping diagram or a verification table is generated by selecting a portion of the candidate regions from the pore candidate regions; Real pores, pseudo-pores, uncertain regions, or pore types are manually labeled. The algorithm results are reviewed based on the results of manual labeling.

[0015] Furthermore, the above method also includes: The candidate regions that have been manually reviewed are used as a small training set. The classification model is trained using the region area, equivalent diameter, aspect ratio, roundness, fill rate, orientation angle, gray-scale statistical features, threshold consistency features, and magnification as input features, and manually labeled categories as labels. The classification model outputs the category or probability of a candidate region belonging to a real pore, a pseudo-pore, or a pore with uncertain boundaries, and performs pseudo-pore removal or porosity correction on subsequent candidate regions based on the output results.

[0016] In summary, compared with the prior art, the present invention has at least one of the following beneficial technical effects: This invention effectively improves the automation of image preprocessing by acquiring microscopic images at multiple magnifications and employing grayscale normalization processing, combined with methods for automatically identifying and cropping non-microscopic tissue regions. By generating candidate pore results using at least two thresholding methods and determining high-confidence pore regions and uncertain regions based on consistency, it significantly enhances the robustness and accuracy of pore identification. By extracting connected components and calculating morphological parameters to classify candidate pore regions, it achieves refined identification of different types of pores. By removing noise or micropores and calculating corrected porosity, and by statistically analyzing porosity and pore type distribution at multiple magnifications, it can output comprehensive porosity, multi-magnification consistency evaluation results, and pore type distribution results, providing more comprehensive, objective, and reliable technical support for APS coating quality evaluation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0018] Figure 1 This is a flowchart of the overall process for identifying the porosity of APS coating microscopic images according to the present invention.

[0019] Figure 2 This is a schematic diagram of the automatic extraction of effective microscopic tissue regions in this invention.

[0020] Figure 3 This is a schematic diagram of multi-threshold confidence segmentation in the present invention, wherein (a) is the effective microscopic tissue region, (b) is the threshold candidate region, (c) is the high-confidence retention region, and (d) is the recognition overlay image.

[0021] Figure 4 This is a schematic diagram of the calculation of connected domain morphology parameters and pore classification in this invention, wherein (a) is the high confidence retention interval, (b) is the identification overlay image, (c) is the sample pore classification, (i) is a circular or dense pore, (ii) is an elliptical or layered pore, (iii) is a crack-like pore, (iv) is an abnormally large pore, (v) is an irregular pore, and (vi) is a noisy or micro pore.

[0022] Figure 5 This is a schematic diagram illustrating manual review and small-sample machine learning correction.

[0023] Figure 6 This is a schematic diagram of the result of overlaying pore identification from a 500x magnified microscopic image.

[0024] Figure 7 This is a schematic diagram of the result of overlaying pore identification from a 1000x magnified microscopic image.

[0025] Figure 8 This is a schematic diagram of the result of overlaying pore identification from a 2000x magnified microscopic image.

[0026] Figure 9 The diagram shows the statistical results of porosity at multiple ratios. (a) represents the average corrected porosity, and (b) represents the average proportion of the uncertain region. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Furthermore, it should be understood that the specific embodiments described herein are only for illustration and explanation of this application and are not intended to limit this application.

[0028] It should be noted that the order of description of the following embodiments is not intended to limit the preferred order of the embodiments of this application. Furthermore, the descriptions of each embodiment in the following embodiments have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0029] The method steps described in this embodiment of the invention can be executed in the order described in the specific implementation, or the execution order of each step can be adjusted according to actual needs, provided that the technical problem can be solved. These are not listed one by one here.

[0030] like Figure 1As shown, the APS coating microscopic image porosity identification method includes image preprocessing, segmentation, feature extraction, classification, and statistical output stages. In the image preprocessing stage, multi-magnification microscopic images are used as input, and grayscale normalization converts 16-bit images into 8-bit working images. In the effective region extraction stage, grayscale statistics are calculated along the horizontal direction of the image, automatically identifying and cropping non-microscopic tissue areas such as the SEM instrument parameter bar, scale bar, and text bar to obtain the effective microscopic tissue areas. In the segmentation stage, multi-threshold confidence segmentation is performed on the effective microscopic tissue areas, using at least two threshold methods to generate porosity candidate results, and high-confidence porosity areas and uncertain areas are determined based on the consistency of the threshold results. In the feature extraction stage, connected component extraction is performed on the high-confidence porosity areas, and morphological parameters of each porosity candidate area are calculated. In the classification stage, the porosity candidate areas are classified according to the morphological parameters. In the statistical output stage, after removing noise or micropores, the corrected porosity is calculated. The corrected porosity and pore type distribution are statistically analyzed according to multiple magnifications, and the comprehensive porosity, multi-magnification consistency evaluation results and pore type distribution results of the APS coating are output.

[0031] This method is applicable to the microscopic characterization of porosity and pore morphology of APS coatings, thermal spray coatings, or similar porous coatings. It can be run on ordinary computers, does not rely on large-scale pixel-level manual annotation, and is suitable for small-batch laboratory samples, process comparison samples, and quality evaluation scenarios.

[0032] Acquiring microscopic images of the same APS-coated sample at multiple magnifications is the initial step in porosity identification methods. Input images are 16-bit grayscale PNG, TIF, or other microscopic image formats, with an image size of 1536 × 1420 pixels. In some embodiments, the total input images are 67, with the main data set including 36 images at 500x, 1000x, and 2000x magnification. Grayscale normalization is performed on the microscopic images to obtain working images. Grayscale normalization converts the 16-bit grayscale microscopic images into 8-bit working images while retaining the original image files. The working images are used for subsequent effective region extraction, threshold segmentation, and porosity statistical processing.

[0033] like Figure 2 As shown, the automatic extraction process of effective microscopic tissue regions processes the working image to crop out non-microscopic tissue regions. Figure 2 The left side shows the SEM microscopic image viewing interface. The top of the interface contains instrument parameter information such as file name and magnification, while the bottom displays the microscopic tissue image of the APS coating. Figure 2 The right side shows the microscopic tissue image after effective area extraction. The SEM instrument parameter bar, scale bar, and text bar at the bottom of the image have been automatically identified and cropped, leaving only the effective microscopic tissue area for subsequent pore identification processing.

[0034] Gray-level statistics are calculated along the horizontal direction of the working image. These statistics include at least one of the following: row gray-level median, mean, or low quantile, as well as combinations thereof. The row gray-level median is robust to abnormal pixels, the row gray-level mean reflects the average brightness level of the entire row, and the low quantile captures the distribution characteristics of dark pixels within a row. A continuous row region is scanned upwards from the bottom of the image. When the gray-level statistics of multiple consecutive rows are below a preset threshold, or when there is a preset decrease in gray-level value relative to the area above, this continuous row region is identified as a non-microscopic tissue area, such as the SEM instrument parameter bar, scale bar, or text bar.

[0035] The preset threshold is related to the overall grayscale distribution of the image. In some implementations, the preset threshold is set to a value between 50% and 70% of the average grayscale value of the working image. When the grayscale statistical value of a continuous row region is lower than this preset threshold, the region is determined to be a non-microscopic tissue region. In some implementations, the determination condition for the preset amplitude of grayscale decrease is that the grayscale statistical value of the continuous row region decreases by more than 30% to 50% relative to the average grayscale value of the microscopic tissue region above. Through the determination of the above preset dark area conditions, continuous row regions located at the bottom of the image that meet the preset dark area conditions can be automatically identified, and these continuous row regions are determined to be non-microscopic tissue regions and then cropped to obtain the effective microscopic tissue region.

[0036] Inclusions present in the microscopic images are segmented separately. Inclusions exhibit different grayscale and morphological characteristics from pores in the microscopic images. Inclusion regions are distinguished from pore regions using grayscale thresholds or morphological parameters, and inclusion regions are not included in porosity calculations. Microscopic images with scratches due to substandard polishing quality are removed from the analysis dataset to avoid scratches being misidentified as crack-type pores and affecting the accuracy of porosity statistics.

[0037] Subsequent porosity calculations all use the area of ​​the effective microstructure region as the denominator, rather than the area of ​​the original entire image. This processing method ensures that porosity calculations are limited to the effective microstructure region, avoiding the miscalculation of non-microstructure areas such as those in the SEM instrument parameter bar, scale bar, or text bar into the porosity calculation, and reducing the interference of non-structure areas on the porosity statistical results.

[0038] like Figure 3As shown, the multi-threshold confidence segmentation process uses at least two thresholding methods to generate pore candidate results for the effective microscopic tissue region. The thresholding methods include at least two of the following: Otsu thresholding, low gray-level percentile thresholding, and mean minus three standard deviation thresholding. Otsu thresholding automatically determines the segmentation threshold by maximizing the inter-class variance; low gray-level percentile thresholding determines the segmentation boundary based on the low percentile value of the image gray-level distribution; and mean minus three standard deviation thresholding uses the image gray-level mean minus three standard deviations as the segmentation threshold. Figure 3 (a) shows the original grayscale image of the effective microscopic tissue region. Figure 3 (b) shows the segmentation results of the threshold candidate region, with the white area marking the candidate pixels identified as pores by the thresholding method.

[0039] For each pixel in the effective microscopic tissue region, the number of times that pixel is identified as a pore by each thresholding method is counted. Based on the consistency of the pore candidate results, pixel regions that are identified as pores by at least two thresholding methods are determined as high-confidence pore regions, pixel regions that are identified as pores by only one thresholding method are determined as uncertain regions, and pixel regions that are not identified as pores by any thresholding method are determined as high-confidence non-pore regions. Figure 3 (c) shows the high-confidence retention region, which is the pixel region that is jointly identified as an aperture by at least two thresholding methods, compared to Figure 3 In (b) of the threshold candidate region, the area of ​​the high-confidence retention region is significantly reduced, indicating that a more reliable pore region was screened out through multi-threshold consistency judgment. Figure 3 (d) in the figure is the recognition overlay image, which is a visualization of the pore recognition results overlaid on the original image. Different color marks in the figure represent pore regions with different confidence levels.

[0040] By traversing all pixels within the effective microscopic tissue region and statistically analyzing the recognition results of each threshold method, a division between high-confidence pore regions and uncertain regions is formed. This multi-threshold confidence segmentation method avoids direct reliance on a single threshold result, reduces the sensitivity of single threshold methods to image brightness, contrast, and noise, and improves the robustness of pore recognition. Pixels in uncertain regions are recorded separately in subsequent processing to evaluate the uncertainty range of porosity calculation.

[0041] like Figure 4 As shown, the process of connected component extraction and morphological parameter calculation processes the high-confidence pore region to obtain multiple pore candidate regions and calculates the morphological parameters of each region. Figure 4 (a) shows a binarized image of the high confidence-preserving region, where the white area represents the connected region identified as a pore and the black background represents the non-pore region. Figure 4(b) shows the identification overlay, which overlays the pore identification results onto the original microstructure image, where different colored outlines mark different types of pore regions. Figure 4 (c) shows magnified images of six typical pore classification samples.

[0042] Connected component extraction is performed on high-confidence pore regions to obtain multiple pore candidate regions. Connected component extraction uses 8-neighbor connectivity, meaning that for each pore pixel in the binarized image, that pixel and its eight neighboring pixels (including horizontal, vertical, and diagonal neighbors) that are also identified as pores are grouped into the same connected component. Through 8-neighbor connected component extraction, interconnected pore pixels in the high-confidence pore region are aggregated into independent pore candidate regions, with each connected component corresponding to one pore candidate region. In some implementations, connected component extraction uses 4-neighbor connectivity, meaning that for each pore pixel in the binarized image, that pixel and its four neighboring pixels (only horizontal and vertical neighbors, excluding diagonal neighbors) that are also identified as pores are grouped into the same connected component. 4-neighbor connectivity is more stringent than 8-neighbor connectivity; for pore pixels connected diagonally, 4-neighbor connectivity identifies them as different connected components, while 8-neighbor connectivity identifies them as the same connected component.

[0043] Calculate the topographic parameters for each pore candidate region. These parameters include area, aspect ratio, roundness, fill rate, and orientation angle. In some implementations, the topographic parameters also include the width of the circumscribed rectangle, the height of the circumscribed rectangle, the perimeter, the equivalent diameter, the magnification, and the image number. The area represents the number of pixels contained in the pore candidate region or its corresponding actual area. The aspect ratio is defined as the ratio of the longer side to the shorter side of the circumscribed rectangle of the pore candidate region, used to characterize the elongation of the pore. Roundness is defined as... ,in The area of ​​the candidate pore region. The perimeter of the pore candidate region is represented by a roundness value closer to 1, indicating a more circular pore shape. The fill rate is defined as the ratio of the area of ​​the pore candidate region to the area of ​​its circumscribed rectangle, characterizing the degree to which the pore fills the circumscribed rectangle. The orientation angle represents the angle between the principal axis of the pore candidate region and the horizontal direction.

[0044] The width and height of the bounding rectangle represent the width and height of the minimum bounding rectangle of the pore candidate region, respectively. The perimeter represents the total length of the boundary of the pore candidate region. The equivalent diameter is defined as the diameter of a circle with the same area as the pore candidate region, calculated using the following formula: ,in This represents the area of ​​the candidate pore region. The magnification level records the magnification of the microscopic image containing the candidate pore region. The image number records the number of the microscopic image containing the candidate pore region, used for subsequent tracing and statistical analysis.

[0045] Based on the morphological parameters calculated above, each candidate pore region possesses a complete morphological feature description. Aspect ratio is used to identify crack-type pores or lamellar gaps; roundness is used to identify circular or compact pores; area is used to identify abnormally large pores or noisy micropores; and fill rate is used to distinguish between regular and irregular pores. These morphological parameters provide the data foundation for subsequent pore type classification and corrected porosity calculations.

[0046] Classifying candidate pore regions based on morphological parameters is the classification stage of porosity identification methods. For example... Figure 4 As shown in (c), the classification types of pore candidate regions include circular or compact pores, elliptical pores or lamellar gaps, crack-type pores, abnormally large pores, irregular pores, and noise or micropores. When the area of ​​a pore candidate region is lower than a preset area threshold, the pore candidate region is classified as noise or micropores, as shown in (vi) of (c). The area threshold is related to the magnification and pixel resolution of the microscopic image. In some embodiments, the area threshold is set to a number of pixels or the corresponding actual area value.

[0047] When the aspect ratio of a candidate pore region is higher than a preset aspect ratio threshold, the candidate pore region is classified as a crack-type pore. The aspect ratio threshold is set to 10, meaning that pores with an aspect ratio greater than 10 are defined as crack-type pores, as shown in (iii) of (c). Crack-type pores exhibit a high aspect ratio linear characteristic, extending along the coating lamellar direction or perpendicular to the lamellar direction.

[0048] When the area of ​​a candidate pore region exceeds a preset large pore threshold, the candidate pore region is classified as an abnormal large pore, as shown in (iv) of (c). The area of ​​an abnormal large pore is significantly larger than the average area of ​​other pores in the same field of view. In some embodiments, the large pore threshold is set as a preset percentage of the effective microstructure area or the absolute number of pixels.

[0049] When the roundness of a candidate pore region is higher than a preset roundness threshold, the candidate pore region is classified as a circular or compact pore, as shown in (i) of (c). The roundness threshold reflects the degree to which the pore shape is close to a circle; the closer the roundness value is to 1, the more regular the pore shape. In some embodiments, the roundness threshold is set to a value between 0.7 and 0.9.

[0050] The classification types of pore candidate regions also include elliptical pores or lamellar gaps and irregular pores. Elliptical pores or lamellar gaps exhibit an elongated shape extending along the lamellar direction, with an aspect ratio between circular pores and crack-type pores, such as the elliptical or lamellar pore sample shown in (ii) of (c). Irregular pores have irregular shapes and complex boundaries, and a low filling rate, such as the irregular pore sample shown in (v) of (c). In some embodiments, when the aspect ratio, roundness, and filling rate of a pore candidate region do not meet the above classification conditions, the pore candidate region is classified as an irregular pore.

[0051] After removing candidate pore regions classified as noise or micropores, the corrected porosity of a single image is calculated. The corrected porosity is defined as the ratio of the sum of the areas of the remaining candidate pore regions to the area of ​​the effective microstructure region, calculated using the following formula: ,in The sum of the areas of the remaining candidate pore regions after eliminating noise or micropores. The area of ​​the effective microstructure region is defined. By eliminating noise or micropores, the porosity correction eliminates the interference of image noise and extremely small pseudo-pores on the porosity calculation, making the porosity statistics more reflective of the true porosity characteristics of the coating.

[0052] like Figure 5 As shown, the manual review process involves interactive review and annotation of the pore candidate regions. A subset of candidate regions is selected to generate a cropped image or review table. The cropped image extracts and magnifies each pore candidate region from the original microscopic image, while the review table records the morphological parameters and location information of each candidate region. In some implementations, a cropped image summary table (contact sheet) is generated from the selected candidate regions. The contact sheet arranges the cropped images of multiple candidate regions on the same page in a grid format, facilitating quick manual browsing and batch annotation.

[0053] The process involves manual labeling of real pores, pseudo-pores, uncertain regions, and pore types. During manual review, operators determine whether a candidate region is a real pore, pseudo-pore, or uncertain region based on the candidate region image displayed in the cropped image or contact sheet, and label the pore type accordingly. Real pores include circular or compact pores, elliptical pores or lamellar gaps, crack-type pores, abnormally large pores, and irregular pores. Pseudo-pores refer to non-pore regions that are mistakenly identified as pores by the algorithm, including image noise, scratches, inclusion boundaries, or other non-pore structures. Uncertain regions refer to candidate regions that are difficult for humans to determine whether they are pores; these regions are recorded separately in subsequent statistics.

[0054] The algorithm results are reviewed based on the manually labeled results. The manually labeled results are used to evaluate the accuracy of the algorithm's identification and to correct the results. Candidate regions manually labeled as pseudo-pores are removed from the porosity statistics and do not participate in the final porosity calculation. Candidate regions manually labeled as true pores are retained in the porosity statistics, and the classification results are updated according to the manually labeled pore types. Candidate regions manually labeled as uncertain regions are recorded separately to assess the range of uncertainty in the porosity calculation.

[0055] In some implementations, 60 candidate pore regions were selected from representative images at 500x, 1000x, and 2000x magnification for manual verification. The verification results showed that 42 of the 60 candidate regions were manually identified as real pores or pore-like defects, 10 as pseudo-pores, and 8 as uncertain regions. Real pores or pore-like defects accounted for 70% of the candidate regions. After excluding uncertain samples, real pores accounted for 80.8%, pseudo-pores accounted for 16.7%, and uncertain samples accounted for 13.3%. This verification result indicates the presence of a certain proportion of pseudo-pores and pores with uncertain boundaries in the candidate regions. Manual verification can correct the algorithm results and improve the accuracy of porosity identification.

[0056] like Figure 5 As shown, the candidate regions, after manual review, serve as a small sample training set for training the classification model. The small sample training set contains manually labeled samples of real pores, pseudo-pores, and pores with uncertain boundaries. Each sample corresponds to the morphological parameters and manually labeled category of the pore candidate region. The input features of the classification model include region area, equivalent diameter, aspect ratio, roundness, fill rate, orientation angle, gray-level statistical features, threshold consistency features, and the magnification level. Gray-level statistical features include the mean gray-level, standard deviation gray-level, minimum gray-level, and maximum gray-level of pixels within the pore candidate region. Threshold consistency features indicate the number or proportion of times the pore candidate region is identified as a pore by various thresholding methods, reflecting the confidence level of the region in multi-threshold segmentation.

[0057] The classification model is trained using manually labeled categories. These categories include three types: true pores, pseudo-pores, and pores with uncertain boundaries. The classification model employs random forest, support vector machine, or gradient boosting tree classifiers. Random forest classifies by constructing multiple decision trees and voting or averaging their outputs, exhibiting good generalization ability for high-dimensional features and small sample data. Support vector machines separate samples of different categories by finding the maximum margin hyperplane, suitable for classification tasks with small sample sizes and high-dimensional features. Gradient boosting trees improve classification performance by iteratively training multiple decision trees and fitting the residuals of previous models, demonstrating strong modeling ability for nonlinear relationships between features.

[0058] The classification model outputs the category or probability of a candidate region belonging to a true pore, a pseudopore, or a pore with uncertain boundaries. When the model outputs a category, the candidate region is directly classified into the corresponding category. When the model outputs a probability, the candidate regions are prioritized based on their probability values, with those closer to the classification boundary receiving higher priority. Based on the model's output, pseudopores are removed or porosity is corrected for subsequent candidate regions. Candidate regions identified as pseudopores by the model are removed from the porosity statistics and do not participate in the porosity correction calculation. Candidate regions identified as pores with uncertain boundaries are recorded separately for assessing the uncertainty range of porosity calculations or for submission for manual review.

[0059] In some implementations, the statistical characteristics of the pore structure are analyzed in conjunction with the Weber probability distribution to characterize the uniformity of the coating microstructure or pores. The Weber distribution is a commonly used statistical model to describe the distribution of defect sizes and failure probabilities in materials, and its probability density function is: ,in For shape parameters, For scale parameters, The pore size is determined by fitting a Weber distribution to the area or equivalent diameter of the candidate pore regions to obtain shape and scale parameters. The shape parameter reflects the dispersion of the pore size distribution; a larger shape parameter indicates a more concentrated pore size distribution. The scale parameter reflects the characteristic value of the pore size; a larger scale parameter indicates a larger average pore size. These Weber distribution characteristic parameters provide quantitative indicators for evaluating coating uniformity and characterizing pore structure.

[0060] like Figure 6 As shown, the 500x magnification microscopic image overlay results demonstrate the effectiveness of APS coating after multi-threshold confidence segmentation and pore identification processing at this magnification. The image presents a grayish-white coating substrate background, with numerous marked and identified pore regions scattered throughout. The red, green, and other colored markers and regions in the image represent different types of pore candidate regions after algorithmic identification and classification. The red markers are densely and evenly distributed throughout the image area, representing the identified pore locations. The green marker areas show another type of pore or regions with specific morphological features. The image reveals a degree of randomness and non-uniformity in pore distribution; some areas have concentrated pores, while others are relatively sparse. The 500x magnification provides a large field of view, enabling the capture of the distribution characteristics of large-scale pores and the overall porosity level in the coating.

[0061] like Figure 7As shown, the 1000x magnification microscopic image overlay results demonstrate the porosity identification effect of the APS coating at higher magnification. The overall image has a gray-white background, representing the matrix structure of the coating. Numerous colored marked areas are distributed throughout the image, corresponding to different types of pore candidate regions identified by the algorithm. The green marked areas are numerous and widely distributed, mainly appearing as elongated strips or irregular shapes, representing areas identified as high-confidence pores, including lamellar gaps and crack-type pores. Small red dots are also scattered throughout the image; these red dots represent circular or compact pores, or specific types of pores marked after morphological parameter classification. Yellow and blue markers are sparsely distributed, corresponding to pore candidate regions of different confidence levels or types. The pore distribution is sparser on the right side of the image compared to the left side, indicating the non-uniformity of the coating structure. At 1000x magnification, various morphological features such as lamellar gaps, crack-type pores, and irregularly shaped pores can be observed more clearly.

[0062] like Figure 8 As shown, the overlay results of 2000x microscopic images demonstrate the pore recognition performance of the APS coating at high magnification. The image shows an overall gray-white coating matrix with numerous identified and marked pore regions. Three different colored markers are observed in the image: green outlines mark the boundaries of regions identified as high-confidence pores, red areas represent uncertain regions or pixel areas where thresholding methods yield inconsistent results, and dark gray or black areas represent original pores or dark tissue features in the coating. The image reveals that the pores are unevenly distributed in the coating, with some areas having denser pores and others relatively sparse. The pore shapes vary, including nearly circular compact pores, elongated lamellar gaps, and irregularly shaped pores. The overlay of green and red markers clearly demonstrates the algorithm's identification results for pore candidate regions, with the areas within the green boundaries being high-confidence pores confirmed by multiple thresholding methods, while the red areas require further judgment or manual verification. At a magnification of 2000x, the boundary morphology and microstructure of pores can be observed more precisely, providing higher resolution image data for pore morphology parameter calculation and type classification.

[0063] By comparison Figure 6 , Figure 7 and Figure 8The overlay results of pore identification from 500x, 1000x, and 2000x microscopic images show the differences in pore identification performance at different magnifications. Low-magnification images have a larger field of view, reflecting the overall distribution characteristics and macroscopic porosity level of the coating pores, but their ability to resolve fine pores is limited. High-magnification images have a smaller field of view, allowing for clearer observation of pore boundary morphology and microstructural features, but the statistical representativeness of a single image is significantly affected by the chosen field of view. The overlay display of pore identification results at multiple magnifications demonstrates that this method can identify and classify pores in APS coating microscopic images at different magnifications, distinguishing different types or confidence levels of pore regions by different colors, facilitating subsequent porosity calculations and statistical analysis of pore types.

[0064] like Figure 9 As shown, the statistical results of multi-magnification porosity demonstrate the method and results of statistically correcting the porosity and pore type distribution according to multiple magnifications. Figure 9 (a) shows the statistical results of the average corrected porosity at different magnifications. The horizontal axis represents the magnification, and the vertical axis represents the percentage of the average corrected porosity. Each bar chart has an error bar above it to indicate the standard deviation range. Figure 9 (b) shows the statistical results of the average uncertainty region proportion under different magnifications, with the horizontal axis representing the magnification and the vertical axis representing the average uncertainty region percentage.

[0065] The core step of the multi-magnification statistical method is to statistically analyze the corrected porosity and pore type distribution at multiple magnifications. For each magnification, the corrected porosity, standard deviation, minimum value, maximum value, proportion of uncertain regions, and pore type distribution of all images at that magnification are calculated. The statistical analysis of corrected porosity includes calculating the mean and standard deviation of the corrected porosity of multiple images at the same magnification. The standard deviation reflects the degree of porosity fluctuation between different fields of view at the same magnification. The minimum and maximum values ​​record the lower and upper limits of the corrected porosity at the same magnification, respectively, and are used to assess the distribution range of porosity. The proportion of uncertain regions is calculated as the percentage of pixel regions identified as pores by only one thresholding method within the effective microscopic tissue area. This proportion reflects the degree of uncertainty in the pore identification results.

[0066] In some implementations, the master data set includes 36 images at magnifications of 500x, 1000x, and 2000x. At 500x magnification, there are 15 images with a mean corrected porosity of 1.8458%, a standard deviation of 0.9578%, a minimum of 0.5635%, and a maximum of 3.7721%. At 1000x magnification, there are 9 images with a mean corrected porosity of 2.1622%, a standard deviation of 1.3553%, a minimum of 0.5022%, and a maximum of 4.2315%. At 2000x magnification, there are 12 images with a mean corrected porosity of 1.1115%, a standard deviation of 0.4630%, a minimum of 0.4024%, and a maximum of 1.8381%.

[0067] The porosity distribution statistics record the number of each type of pore at each magnification. In some embodiments, at 500x magnification, the number of circular or compact pores is 7192, the number of elliptical pores or lamellar gaps is 972, the number of irregular pores is 749, the number of crack-type pores is 53, and the number of abnormally large pores is 3. At 1000x magnification, the number of circular or compact pores is 2216, the number of elliptical pores or lamellar gaps is 297, the number of irregular pores is 422, the number of crack-type pores is 27, and the number of abnormally large pores is 2. At 2000x magnification, the number of circular or compact pores is 3375, the number of elliptical pores or lamellar gaps is 263, the number of irregular pores is 580, the number of crack-type pores is 5, and the number of abnormally large pores is 1.

[0068] Based on statistical results from different magnifications, the overall porosity and multi-magnification consistency evaluation results of the APS coating are output. The overall porosity is calculated by weighted averaging or arithmetic averaging the corrected porosity at multiple magnifications. In some embodiments, the overall average corrected porosity of 36 images at 500x, 1000x, and 2000x is 1.6801%. The multi-magnification consistency evaluation results assess the stability and reliability of porosity measurement by comparing the mean and standard deviation of the corrected porosity at different magnifications. When the mean of the corrected porosity at different magnifications is close and the standard deviation is small, it indicates that the porosity measurement results have good multi-magnification consistency. When the mean of the corrected porosity at different magnifications differs greatly or the standard deviation is large, it indicates that the porosity measurement results are greatly affected by the magnification and field of view selection, and a comprehensive evaluation combining statistical results from multiple magnifications is required.

[0069] The pore type distribution results output the quantity and proportion of different pore types at various magnifications. Through pore type distribution statistics, the relative content and distribution characteristics of different pore types in the APS coating, such as circular or compact pores, elliptical pores or lamellar gaps, crack-type pores, abnormally large pores, and irregular pores, can be understood. The pore type distribution results provide detailed information for evaluating the pore morphology and microstructure uniformity of the APS coating. Different pore types have varying impacts on the coating's mechanical properties, thermophysical properties, and service reliability; crack-type pores and abnormally large pores generally have a greater impact on coating performance than circular or compact pores.

[0070] The multi-magnification statistical method takes into account the influence of large-scale pores, small pores, and fluctuations between fields of view. Low-magnification images have a large field of view, which can capture large-scale pores and the overall pore distribution characteristics of the coating, but their ability to resolve small pores is limited. High-magnification images have a smaller field of view, which can distinguish small pores and pore boundary morphology, but the statistical representativeness of a single image is greatly affected by the choice of field of view. By statistically correcting the porosity and pore type distribution at multiple magnifications and outputting comprehensive porosity, multi-magnification consistency evaluation results, and pore type distribution results, this method can provide multi-scale, traceable evaluation results of the pore morphology of APS coatings.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying the porosity of APS coatings using microscopic images, characterized in that, include: Acquire microscopic images of the same APS-coated sample at multiple magnifications; The microscopic image is normalized to obtain the working image; Calculate the grayscale statistical value along the horizontal direction of the working image, identify the continuous row region located at the bottom of the image and whose grayscale statistical value meets the preset dark area condition, and cut off the continuous row region after determining it as a non-microscopic tissue region to obtain the effective microscopic tissue region. At least two thresholding methods were used to generate pore candidate results for the effective microstructure region; Based on the consistency of the pore candidate results, pixel regions that are identified as pores by at least two thresholding methods are defined as high-confidence pore regions, and pixel regions that are identified as pores by only one thresholding method are defined as uncertain regions. Connectivity extraction is performed on the high-confidence pore region to obtain multiple pore candidate regions; Calculate the morphological parameters for each of the candidate pore regions; The candidate pore regions are classified according to the morphological parameters. After removing candidate pore regions classified as noise or micropores, the ratio of the sum of the areas of the remaining candidate pore regions to the area of ​​the effective microstructure region is used as the corrected porosity of a single image. The porosity and pore type distribution are statistically corrected according to the multiple magnification factors, and the comprehensive porosity, multi-magnification consistency evaluation results and pore type distribution results of the APS coating are output.

2. The method according to claim 1, characterized in that, The step of normalizing the grayscale of the microscopic image to obtain the working image includes: Convert 16-bit grayscale microscopic images to 8-bit working images while preserving the original image files.

3. The method according to claim 1, characterized in that, The step of calculating grayscale statistical values ​​along the horizontal direction of the working image and identifying continuous row regions located at the bottom of the image whose grayscale statistical values ​​meet preset dark area conditions includes: Calculate grayscale statistics along the row direction of the working image, wherein the grayscale statistics include at least one of the row grayscale median, mean, or lower quantile; Scan consecutive line regions from the bottom of the image upwards; When multiple consecutive rows of grayscale statistical values ​​are lower than a preset threshold, or when the grayscale decreases by a preset amount relative to the area above, the consecutive rows are identified as the SEM instrument parameter bar, scale bar, or text bar.

4. The method according to claim 1, characterized in that, The thresholding method includes at least two of the following: Otsu threshold, low gray percentile threshold, and mean minus multiple standard deviation threshold.

5. The method according to claim 1, characterized in that, The morphological parameters include area, aspect ratio, roundness, fill rate, and orientation angle.

6. The method according to claim 1, characterized in that, The step of classifying the candidate pore regions according to the morphological parameters includes: When the area of ​​the candidate pore region is lower than the preset area threshold, it is classified as noise or micropore; When the aspect ratio of the candidate pore region is higher than the preset aspect ratio threshold, it is classified as a crack-type pore. When the area of ​​the candidate pore region is higher than the preset large pore threshold, it is classified as an abnormal large pore. When the roundness of a candidate pore region is higher than a preset roundness threshold, it is classified as a circular or compact pore.

7. The method according to claim 6, characterized in that, The classification types of the pore candidate regions also include elliptical pores or lamellar gaps and irregular pores.

8. The method according to claim 1, characterized in that, The step of statistically correcting the porosity and pore type distribution according to the multiple magnification factors includes: The corrected porosity, standard deviation, minimum value, maximum value, proportion of uncertain region, and pore type distribution under different magnification factors were statistically analyzed. Based on the statistical results at different magnifications, the comprehensive porosity and multi-magnification consistency evaluation results of the APS coating are output.

9. The method according to claim 1, characterized in that, Also includes: A cropping diagram or a verification table is generated by selecting a portion of the candidate regions from the pore candidate regions; Real pores, pseudo-pores, uncertain regions, or pore types are manually labeled. The algorithm results are reviewed based on the results of manual labeling.

10. The method according to claim 9, characterized in that, Also includes: The candidate regions that have been manually reviewed are used as a small training set. The classification model is trained using the region area, equivalent diameter, aspect ratio, roundness, fill rate, orientation angle, gray-scale statistical features, threshold consistency features, and magnification as input features, and manually labeled categories as labels. The classification model outputs the category or probability of a candidate region belonging to a real pore, a pseudo-pore, or a pore with uncertain boundaries, and performs pseudo-pore removal or porosity correction on subsequent candidate regions based on the output results.

Images