Method for predicting medium expansion rate of high-purity metal based on image processing

Abstract

The invention relates to the technical field of measurement, in particular to a high-purity metal expansion rate prediction method based on image processing, which comprises the following steps: acquiring gray images of grating images of a metal surface at different temperatures, and obtaining a spectrogram of the gray image corresponding to any temperature; determining an optimal filtering radius of a filter of the spectrogram; filtering the spectrogram according to the optimal filtering radius to obtain an optimal spectrogram; obtaining an equal-difference fringe image based on the optimal spectrogram; and obtaining the linear expansion coefficient of the metal at any temperature based on the equal-difference fringe image so as to obtain linear expansion coefficients at different temperatures, thereby realizing prediction of the linear expansion coefficient of the metal at the to-be-measured temperature. According to the scheme, the prediction accuracy of the linear expansion coefficient of the metal can be improved.

Description

Technical Field

[0001] The present invention relates to the field of measurement technology, and more specifically, to a method for predicting expansion rate in high-purity metal based on image processing. Background Art

[0002] In modern industrial production, high-purity metals are widely used in various high-tech fields, such as aerospace, semiconductor manufacturing, etc., due to their excellent physical and chemical properties. In the manufacturing process of high-purity metals, accurate prediction of their expansion rate is the key to ensure product quality and stable performance.

[0003] The prior art discloses a method for measuring the linear expansion coefficient of metal materials based on image analysis. The method is based on the moiré fringe theory, and derives the functional relationship between the linear expansion coefficient of metal materials and the slope of the arithmetic difference fringe of the moiré fringe and the temperature change of the metal material due to thermal expansion. With the help of image analysis technology, the three-layer operation of image optimization, image difference frequency information processing and image data calculation is performed on the collected grating change image. The slope of the arithmetic difference fringe of the moiré fringe and the temperature change of the metal material due to thermal expansion are used as basic parameters, and finally the strain amount and linear expansion coefficient of the metal copper material within a certain temperature change range are obtained.

[0004] For example, the patent application document with application publication number CN102879418A and invention name “A method for measuring the linear expansion coefficient of metal materials” discloses that the acquired moiré fringes are subjected to Fourier transform, filtering and extreme value method refinement to obtain refined arithmetic stripes, and then the slope and inclination of the arithmetic stripes are calculated by the least squares method to calculate the expansion coefficient.

[0005] The above method has the advantages of non-contact, high precision and real-time performance. However, it should be noted that the above measurement method often uses a low-pass filter to select low-frequency information. When the low-pass filter is not set properly, the obtained low-frequency information may have more interference information or too much low-frequency information loss after restoration, resulting in a low accuracy rate of the final expansion coefficient detection result, which in turn makes the subsequent expansion coefficient prediction inaccurate. Summary of the invention

[0006] The purpose of the present invention is to propose a method for predicting the expansion coefficient of high-purity metals based on image processing, so as to solve the problem that the detection accuracy of the expansion coefficient of metals in the prior art is low, resulting in inaccurate subsequent prediction of the expansion coefficient; to this end, the present invention provides a solution in the following aspect.

[0007] The present invention provides a method for predicting expansion rate in high-purity metal based on image processing, comprising the following steps: Obtain grayscale images of the grating image of the metal surface at different temperatures, and obtain a frequency spectrum of the grayscale image corresponding to any temperature; Determine the optimal filter radius of the filter of the spectrum graph; filter the spectrum graph according to the optimal filter radius to obtain an optimal spectrum graph; obtain an equidistant fringe image based on the optimal spectrum graph; The linear expansion coefficient of the metal at any temperature is obtained based on the arithmetic difference fringe image, and the linear expansion coefficient of the metal at the temperature to be measured is predicted based on the obtained linear expansion coefficients at different temperatures; Wherein, the optimal filtering radius is: Acquire a peak point sequence in the spectrum graph; construct a triangulated network model based on the position coordinates of each peak point in the peak point sequence; the triangulated network model includes multiple triangulated networks; Calculate the priority value of each filter radius, which is positively correlated with the fitness value and importance of each triangulated network; the fitness value represents the possibility of the corresponding triangulated network belonging to the equidistant stripes, and the importance is the overlap of any triangulated network with the circular area; the center of the circular area is the center point of the spectrum graph, and the radius is any filter radius; Based on each filter radius and the corresponding priority value, a corresponding fitting function is generated, and the filter radius corresponding to the maximum value of the derivative of the fitting function is taken as the optimal filter radius.

[0008] In the above scheme, by using the triangulated network model, the relationship between the area enclosed by each peak point in the spectrum diagram corresponding to any temperature and the equidistant fringes is analyzed, and different filtering radii can be adaptively analyzed, so as to determine the optimal filtering radius of the spectrum diagram, and the spectrum diagram is filtered by the optimal filtering radius to obtain the optimal spectrum diagram after filtering, and then the equidistant fringe image is obtained, and finally the expansion coefficient is calculated; that is, the scheme of the present invention has a corresponding optimal filtering radius for the spectrum diagrams at different temperatures, and has a stable filtering effect when filtering different spectrum diagrams, preventing the corresponding equidistant fringes from having more interference information or lacking too much information, so that when the equidistant fringes are used to obtain the pure metal expansion coefficient, it has stability, improves the accuracy of the pure metal expansion coefficient detection, and can more accurately predict the expansion coefficient of the metal at the temperature to be tested.

[0009] In one embodiment, the process of acquiring the grating image is as follows: The metal to be tested is used as a screen, and a moiré fringe experiment is performed using two gratings with the same grating parameters. During the moiré fringe experiment, the angle between the two gratings is a set angle, and a moiré fringe image is formed on the surface of the metal to be tested, and the moiré fringe image is used as a grating image.

[0010] In one embodiment, the fitness value for: ; in, is the ratio of the relative angle of the jth triangulation to 90°, where the relative angle is the angle between the straight line formed by the center of gravity of the jth triangulation and the center point of the spectrum graph and the straight line formed along the angle β direction, is the ratio of the Euclidean distance from the centroid of the jth triangulated network to the center of the spectrum graph to the diagonal length of the grayscale image, is the ratio of the mean spectral intensity of all internal points in the jth triangulated network to the spectral intensity of the center point of the spectrum graph, is an exponential function with the natural constant e as base, , θ is the set angle.

[0011] The above scheme can determine the possibility that the triangulated network belongs to equidistant stripes by obtaining the appropriate value of each triangulated network.

[0012] In one embodiment, the process of obtaining the importance is as follows: Randomly select any filter radius, take the center point of the spectrum graph as the center of the circle, and obtain the circular area corresponding to any filter radius; The number of overlapping data between the circular area and the j-th triangulated network is counted, and the ratio of the number of overlapping data to the number of all internal points in the j-th triangulated network is used as the importance of any filtering radius.

[0013] The above scheme can measure the overlap between the circular area under any filter radius and each triangulated network by obtaining the importance.

[0014] In one embodiment, the priority value for: ; Where J is the total number of triangulated networks, is the importance of the sth filter radius, is the appropriate value of the jth triangulation.

[0015] The above scheme can adaptively analyze different filter radii by obtaining priority values.

[0016] In one embodiment, before constructing the triangulated network model, the step of screening the peak point sequence to obtain a new peak point sequence is further included, specifically: Obtain the spectrum intensity of each peak point in the peak point sequence; Based on the position coordinates of each peak point and the corresponding spectrum intensity, the peak point sequence is divided into an interference sequence, a weak feature sequence, and a strong feature sequence; wherein the interference sequence is a category with the smallest spectrum intensity mean, the weak feature sequence is a category with the spectrum intensity mean in the middle, and the strong feature sequence is a category with the largest spectrum intensity mean; Eliminate the remaining peak points except the central peak point in the interference sequence and the strong feature sequence to obtain the weak feature sequence and the central peak point; The weak feature sequence and the central peak point are combined into a new peak point sequence.

[0017] The above solution reduces the interference problem of other peak points and improves accuracy by removing all other peak points in the strong feature sequence.

[0018] In one embodiment, the specific process of realizing the prediction of the linear expansion coefficient of the metal at the temperature to be measured is: performing temperature and linear expansion coefficient data fitting on the linear expansion coefficient of the metal at different temperatures to obtain a fitting curve, substituting the temperature to be measured into the fitting curve to obtain a predicted linear expansion coefficient.

[0019] In one embodiment, after the grayscale image of each grating image is acquired, the grayscale image is further denoised to obtain a denoised grayscale image.

[0020] The beneficial effects of the present invention are: The solution of the present invention has a corresponding optimal filtering radius for the frequency spectrum at different temperatures. When filtering different frequency spectrums, it has a stable filtering effect, so that when using arithmetic stripes to obtain the expansion coefficient of pure metal, it has stability, that is, it improves the accuracy of the expansion coefficient detection of pure metal, and thus can more accurately predict the expansion coefficient of the metal at the temperature to be tested. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 is a flowchart of the steps of a method for predicting expansion rate in high-purity metal based on image processing in this embodiment; Figure 2 Schematic diagram of local moiré fringes; Figure 3 Schematic diagram of moiré fringes with arrangement direction; Figure 4 for Figure 3 Spectrum diagram of . DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0023] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0024] The scenario targeted by the present invention is: for the process of measuring the linear expansion coefficient of metal materials based on the characteristics of gratings and moiré fringes. Since when the linear expansion coefficient of metal is detected at a certain temperature, when the spectrum diagram of the grating image is filtered, it does not consider how to select the low-pass filter. Therefore, when the spectrum diagram is filtered, there may be errors, resulting in inaccurate linear expansion coefficients calculated subsequently, thereby affecting the prediction of the linear expansion coefficient at the temperature to be measured.

[0025] Based on the above problems, the present invention provides a method for predicting the expansion rate of high-purity metals based on image processing. The method mainly focuses on the selection of low-pass filters, and does not improve other processing in the grating image (such as Fourier transform of the grating image, inverse Fourier transform of the filtered spectrum, refinement of the equidistant fringe image, and least squares processing).

[0026] The expansion rate in the present invention is actually the linear expansion coefficient of the metal, which is used to indicate the change in length value caused by temperature change. The linear expansion coefficient is a type of thermal expansion coefficient; the thermal expansion coefficient also includes the surface expansion coefficient and the volume expansion coefficient. Different metals have different expansion rates, that is, the amount of expansion per unit length at different temperatures.

[0027] Specifically, Figure 1 As shown, a method for predicting expansion rate in high-purity metal based on image processing in this embodiment includes the following steps: In step S1, grayscale images of the grating image of the metal surface at different temperatures are acquired to obtain a frequency spectrum of the grayscale image corresponding to any temperature.

[0028] In this embodiment, grating images on the metal surface to be tested at different temperatures are collected, and each grating image is grayed to obtain a gray image corresponding to each temperature; the different temperatures are at least two temperatures set at a set temperature interval.

[0029] Since noise is often a highlight, this embodiment also performs denoising on each grayscale image. Specifically, minimum filtering can be used to perform denoising to obtain a denoised grayscale image.

[0030] In this embodiment, the specific process of obtaining the grating image is: using the metal to be tested as the screen, and using two gratings with the same grating parameters to perform a moiré fringe experiment; during the moiré fringe experiment, the angle between the two gratings with the same grating parameters is a set angle, the set angle is θ, and a moiré fringe image is formed on the surface of the metal to be tested, and the moiré fringe image is used as the grating image.

[0031] In the above-mentioned moiré fringe experiment, an existing grating device is used. The angle between the grating lines of two gratings with the same grating parameters is θ (the angle θ is generally very small). The metal to be tested is used as a screen to form moiré fringes on the metal to be tested. At the same time, a high-definition camera is installed near the grating device to collect the moiré fringes on the surface of the metal to be tested and obtain a grating image. Among them, since the moiré fringes are black and white stripes formed by the superposition of two cosine gratings, the grating image is black and white stripes.

[0032] For example, when the angle between the grating lines of two gratings with the same grating parameters is θ, alternating light and dark stripes appear, such as Figure 2 As shown, I is a dark stripe, and the I line position is opaque, and II is a bright stripe, and the II line position has the largest amount of light transmittance. This alternating light and dark stripe is a moiré stripe. At this time, the light energy distribution after the light passes through the two gratings is an ideal triangular wave; however, in actual applications, light has diffraction, and the illumination light source has width. The grating pitch, slit width, and grating line width of the two gratings are not completely equal. Therefore, the actual light energy distribution is an approximate sine wave.

[0033] In this embodiment, the direction of line I or line II is the arrangement direction of the moiré fringes, and the arrangement direction of the moiré fringes is perpendicular to the bisector of the angle between the two grating lines; Figure 2 The angle between the arrangement direction of the moiré fringes and the horizontal direction is , angle The angle is half of the set angle θ. It should be noted that since the angle θ is generally very small, the angle Generally small.

[0034] The two gratings in this embodiment can be attached to the metal to be measured, or there can be a gap between them, and the gap can be set according to actual conditions.

[0035] The different temperatures mentioned above can be multiple set temperatures, which are based on the set initial temperature and are set according to the set temperature intervals; for example, when the initial temperature is 70°C, the set temperature interval is 10°C, and the number of the multiple set temperatures is 7, the multiple set temperatures are specifically 70°C, 80°C, 90°C, 100°C, 110°C, 120°C, and 130°C. Of course, the initial temperature and the set temperature interval mentioned above can also be set according to actual conditions, and are not limited to the above specific set values, such as the initial temperature can also be 25°C, and the set temperature interval can also be 5°C or 15°C; and the number setting is not limited to the above specific number, and can also be 8, 9 or more.

[0036] In one embodiment, a Fourier transform is also performed on the grayscale image corresponding to any temperature to obtain a corresponding spectrum diagram. It should be noted that the spectrum diagram in this embodiment is a centralized spectrum diagram; and the change of the grayscale image has a vertical relationship with the change of the spectrum diagram, that is, the direction in the spectrum diagram is vertical to the spatial direction. Therefore, the arrangement direction of the equidistant fringe image in the grayscale image is perpendicular to the direction in the spectrum diagram. At this time, on the spectrum image, the center point is taken as the origin, and the angle β (β is 90 degrees and the angle between the arrangement direction of the moire fringe and the horizontal direction) is The greater the possibility that the straight lines formed by the straight lines in the direction of (the difference of) belong to arithmetic stripes.

[0037] In step S2, the optimal filter radius of the filter of the spectrum graph is determined; the spectrum graph is filtered according to the optimal filter radius to obtain an optimal spectrum graph; and an equidistant fringe image is obtained based on the optimal spectrum graph.

[0038] At the same time, since equidistant fringes are the main features of the moire fringe image, they belong to low-frequency information, and the closer the data is to the center in the spectrum graph, the more it belongs to low-frequency information; and the grayscale fluctuation of equidistant fringes in the moire fringe image is large, so the corresponding spectrum intensity in the spectrum graph is greater. Therefore, it is necessary to filter the spectrum graph, remove high-frequency information, and only retain low-frequency information to obtain an equidistant fringe image.

[0039] Specifically, in this embodiment, a circular low-pass filter is selected for filtering, wherein the center of the spectrum graph is used as the center of the circular low-pass filter, and the spectrum graph is filtered by setting the radius of the circular low-pass filter to remove high-frequency information and retain only low-frequency information. However, when setting the radius of the low-pass filter, if the filter radius is not set properly, there is a problem of poor filtering effect, which results in a low accuracy rate in detecting the expansion rate of the metal to be tested.

[0040] Therefore, it is necessary to select the best filter radius from multiple different filter radii. Specifically, the process of obtaining the best filter radius is as follows: Step S21, obtaining a peak point sequence in the spectrum diagram; constructing a triangulated network model based on the position coordinates of each peak point in the peak point sequence; the triangulated network model includes multiple triangulated networks; calculating the fitness value of any triangulated network; for any filtering radius, calculating the importance of any triangulated network and the any filtering radius; taking the product of the fitness value and the importance as the priority of the any filtering radius, and taking the sum of the priorities of all triangulated networks as the priority value of the any filtering radius.

[0041] Specifically, the coordinates of all peak points in the peak point sequence are obtained to construct a triangulated network model, wherein the triangulated network model includes multiple triangulated networks, each triangulated network includes vertices, edges and internal points. Since the triangulated network model is an existing technology, it will not be described in detail here.

[0042] The triangulated network model in this embodiment is a Delaunay triangulated network model.

[0043] The internal points mentioned above are actually the data that constitute the triangulated network profile. Since the profile of the triangulated network is an area, the internal points are the data other than the vertices and edges in the triangulated network. All the internal points constitute the profile data sequence. For example, the profile data sequence of the jth triangulated network is .

[0044] It should be noted that the reason for not considering the vertex and edge information of the triangulated network is that the vertices and edges may be common information of multiple different triangulated networks, and thus the ownership of the vertices and edges is difficult to determine. Therefore, in order to avoid the problem of difficulty in determining the ownership, only the profile data sequence composed of the internal points of the triangulated network is considered.

[0045] Furthermore, before constructing the triangulated network model, the step of screening the peak point sequence to obtain a new peak point sequence is also included, specifically: First, the spectrum intensity of each peak point in the peak point sequence is obtained. Specifically, in this embodiment, a peak point detection algorithm is used to detect peak points on the spectrum graph to obtain a peak point sequence, wherein the information of each peak point in the peak point sequence includes position coordinates and spectrum intensity. Since the peak point detection algorithm is a prior art, it will not be described in detail here.

[0046] Secondly, based on the position coordinates of each peak point and the corresponding spectrum intensity, the peak point sequence is divided into interference sequence, weak feature sequence, and strong feature sequence; among which, the interference sequence is the category with the smallest spectrum intensity mean, the weak feature sequence is the category with the spectrum intensity mean in the middle, and the strong feature sequence is the category with the largest spectrum intensity mean.

[0047] In this embodiment, the peak point sequence is classified, and specifically a k-means clustering algorithm can be used, where k is set to 3; in the classification process, the cluster distance is obtained by calculating the Euclidean distance of information of any two peak points; of course, the hyperparameter f can also be set to adjust the calculated Euclidean distance, specifically f=0.3.

[0048] It should be noted that, since the spectrum intensity information of each peak point in the peak point sequence may contain some obvious features, some weakly obvious features and interference noise data, it needs to be divided into three categories.

[0049] Furthermore, since the information of each peak point in the peak point sequence is three-dimensional data (abscissa, ordinate and spectrum intensity) with inconsistent dimensions, the peak point sequence needs to be standardized before classification.

[0050] Then, the interference sequence and the remaining peak points in the strong feature sequence except the central peak point are removed to obtain the weak feature sequence and the central peak point. Since the purpose of this embodiment is to obtain low-frequency information, the interference sequence can be removed by using the above classification idea, and the remaining data in the strong feature sequence except the central peak point can also be removed to finally obtain the weak feature sequence and the central peak point.

[0051] The reason for removing the remaining data except the central peak point in the strong feature sequence in the above embodiment is that the central peak point in the strong feature sequence has the highest spectrum intensity, that is, the center point of the spectrum diagram; the texture of the arithmetic fringe image is an obvious stripe, and the arrangement direction of the stripe is perpendicular to the direction of the corresponding spectrum information in the spectrum diagram. Since the arithmetic fringe image is obvious stripe information, its spectrum intensity is relatively high. Therefore, the spectrum information of the obvious stripe information passes through the center point of the spectrum diagram. Therefore, the center point of the spectrum diagram needs to be retained; and all other peak points except the central peak point in the strong feature sequence may have interference problems, resulting in inaccuracy, so all other peak points in the strong feature sequence are removed.

[0052] For example, Figure 3 and Figure 4 As shown, Figure 4 for Figure 3 The spectrum of Figure 3 The arrangement direction of the equidistant fringe images in is along Figure 3 The direction of the black line in Figure 3 The angle between the black line in and the horizontal direction is ), then in the spectrum diagram, the direction of the spectrum information corresponding to the equidistant fringe image is along Figure 4 The direction of the black straight line; at this time, Figure 4The angle between the black straight line in and the horizontal direction is β. It should be noted that Figure 3 and Figure 4 There are two possible directions for the black straight line in the figure, which can be determined based on actual conditions.

[0053] Finally, the weak feature sequence and the central peak point are combined into a new peak point sequence. This embodiment can use the position coordinates of each peak point in the new peak point sequence to construct a triangulated network model, which can improve the accuracy of the constructed triangulated network model and reduce noise interference.

[0054] In this embodiment, the appropriate value of the jth triangulated network for: ;in, is the ratio of the relative angle of the jth triangulation to 90°, where the relative angle is the angle between the straight line formed by the center of gravity of the jth triangulation and the center point of the spectrum graph and the straight line formed along the angle β direction, where, The smaller it is, the more the frequency domain distribution in the frequency domain conforms to the characteristic distribution of arithmetic stripes, and the smaller it is, the better. is the ratio of the Euclidean distance from the centroid of the jth triangulated network to the center of the spectrum graph to the diagonal length of the grayscale image, The smaller it is, the closer it is to the center of the frequency domain, and the internal points of the j-th triangulation network are low-frequency information, and the arithmetic fringe image is low-frequency information. is the ratio of the mean spectral intensity of all internal points in the jth triangulated network to the spectral intensity of the center point of the spectrum graph, where: The lower it is, the higher the possibility that the internal points in the j-th triangulation network belong to interference data, which is not conducive to generating arithmetic stripes; the appropriate value represents the possibility that the corresponding triangulation network belongs to arithmetic stripes.

[0055] in, is an exponential function with the natural constant e as the base, where the angle β is 90° and the angle The difference, that is , θ is the setting angle, angle Half of the set angle.

[0056] The diagonal length of the grayscale image can be obtained according to the size of the grayscale image. For example, when the size of the grayscale image is n×n, the diagonal length is ; When the size of the grayscale image is n×m, the length of the diagonal is .

[0057] Furthermore, using Function pair and Perform negative correlation mapping, The larger the value is, the more likely the filter radius will be selected.

[0058] In this embodiment, since the constructed triangulated network model adopts the position coordinates of the peak point, and the peak point itself is in the spectrum diagram, the triangulated network corresponding to the peak point is also in the spectrum diagram. Therefore, the Euclidean distance value from the center of gravity of the triangulated network to the center point of the spectrum diagram can be directly calculated.

[0059] It should be noted that the spectrum intensity of the internal points in the triangulation network determines the characteristics of the internal points. Therefore, the spectrum intensity can be used to determine whether there is interference. The spectrum intensity of all internal points within is generally low, so the profile data series All internal points are inconspicuous features. For restoring the equidistant fringe image, the profile data sequence All internal points within may be interference factors.

[0060] In this embodiment, the process of obtaining the importance is as follows: Randomly select any filter radius, take the center point of the spectrum graph as the center of the circle, and obtain the circular area corresponding to any filter radius; count the number of overlapping data between the circular area and the j-th triangulated network, calculate the ratio of the number of overlapping data to the number of all internal points of the j-th triangulated network, and use the ratio as the importance between the filter radius and the triangulated network.

[0061] The number of overlapping data mentioned above is obtained by counting the position coordinates of the data within the sth filter radius and the position coordinates of the internal points in the jth triangulation network. The data with repeated position coordinates are overlapping data. The larger the number of overlapping data, the larger the overlapping part of the circular area under the sth filter radius and the jth triangulation network.

[0062] In this embodiment, the priority values ​​corresponding to each filtering radius are specifically as follows: for any filtering radius, the product of the appropriateness value of any triangulated network and the corresponding importance is used as the priority of any filtering radius, and the sum of the priorities of all triangulated networks is used as the priority value of any filtering radius.

[0063] Specifically, taking the sth filter radius as an example, the priority value corresponding to the sth filter radius is : ; Where J is the total number of triangulated networks, is the importance of the sth filter radius, is the appropriate value of the jth triangulation.

[0064] In this embodiment, since the circular area under the sth filter radius and the jth triangulated network only partially overlap, the repetition ratio is used to compare The weighted values ​​are added up for all triangulated meshes covered by the sth filter radius, and the weighted sum is used as the priority value corresponding to the sth filter radius.

[0065] Step S22: Generate a corresponding fitting function based on each filter radius and the corresponding priority value, and take the filter radius corresponding to the maximum value of the derivative of the fitting function as the optimal filter radius.

[0066] After obtaining the priority values ​​under different filter radii, the least squares method is used to fit different filter radii and their corresponding priority values ​​to obtain the corresponding fitting function. The fitting function is differentiated to obtain the filter radius corresponding to the maximum derivative, which is the optimal filter radius. However, since there are only integers and no decimals when selecting pixels during filtering, the optimal filter radius needs to be rounded to obtain the final optimal filter radius.

[0067] The fitting in the above embodiment may also adopt other fitting methods, such as gradient descent method.

[0068] It should be noted that the reason for the above fitting is that, since the moderation value and the importance are both always greater than or equal to 0, and the moderation value of each triangulated mesh in the triangulated mesh model is unchanged, for any triangulated mesh, the product of its moderation value and importance increases with the increase of the filter radius (the importance increases with the increase of the filter radius), then the priority value of all triangulated meshes must be a data that increases with the increase of the filter radius; therefore, in this embodiment, it is necessary to fit the priority value to obtain the position of the maximum slope of the fitting function (wherein the maximum slope is obtained by derivatizing the fitting function). Since the position of the maximum slope represents the point with the largest change in priority value, the filter radius corresponding to the point with the largest change in priority value at this time has a better priority value.

[0069] The purpose of using the Delaunay triangulation model in this embodiment is that the peak point sequence may still have a lot of interference information, and not all peak points in the peak point sequence are pixels that have a strong correlation with the arithmetic stripes. If the filter radius of the low-pass filter is too small, it will cause too much loss of information and fail to obtain relatively obvious arithmetic stripes. Therefore, it is necessary to use the relationship between the Delaunay triangulation model and the filter radius to screen the filter radius to determine whether the filter radius is appropriate.

[0070] After obtaining the optimal filter radius, a circular filter is created with the optimal filter radius as a low-pass filter for the spectrum graph, so that only the data covered by the filter is retained after filtering to obtain the filtered spectrum graph data. The filtered spectrum graph is then inverse Fourier transformed to restore the spectrum graph to the spatial domain to obtain an equidistant fringe image.

[0071] At this point, the optimal filtering radius of the spectrum graph corresponding to each temperature can be obtained in this embodiment to filter the spectrum graph at each temperature. Since each spectrum graph corresponds to an optimal filtering radius, each optimal spectrum graph after filtering can be obtained, thereby improving the accuracy of the arithmetic fringe image of each optimal spectrum graph obtained subsequently.

[0072] In step S3, the linear expansion coefficient of the metal at any temperature is obtained based on the arithmetic difference fringe image, and then the linear expansion coefficients at different temperatures are obtained, thereby realizing the prediction of the linear expansion coefficient of the metal at the temperature to be measured.

[0073] In this embodiment, the equidistant fringe image is subjected to thinning processing and least squares processing to obtain the linear expansion coefficient of the metal to be tested at any temperature.

[0074] Specifically, the grayscale image corresponding to the equidistant stripes is first binarized, and the binary image is thinned using a thinning algorithm. Since the equidistant stripes have multiple stripe lines, the stripe lines corresponding to all the stripes are obtained, and each stripe line is fitted using the least squares method to obtain the angle value between the fitting curve and the X-axis, and the difference between the angle value and the angle θ between the two grating lines is obtained. Thus, the slope and difference of the stripes can be obtained, and the strain of the metal material can be calculated; then the maximum value of the strain in all the stripes is obtained, and the linear expansion coefficient of the metal material can be calculated based on the maximum value of the strain. It should be noted that since the process of measuring the linear expansion coefficient of metal materials based on the characteristics of gratings and moiré fringes is a prior art, it will not be described in detail here.

[0075] At this point, the linear expansion coefficients of metals at multiple different temperatures can be obtained.

[0076] In this embodiment, the linear expansion coefficients of metals at multiple different temperatures are obtained, and the linear expansion coefficient at the temperature to be tested is predicted based on an existing prediction model to complete the expansion rate prediction in high-purity metals.

[0077] Among them, the prediction model can be data fitting, ARIMA algorithm or random forest algorithm, etc. Since the specific prediction process of the above prediction model is an existing technology, it will not be described in detail here.

[0078] Exemplarily, the linear expansion coefficient of the metal at different temperatures is subjected to data fitting of temperature and linear expansion coefficient to obtain a fitting curve, and the temperature to be measured is substituted into the fitting curve to obtain a predicted linear expansion coefficient.

[0079] In the description of this specification, “plurality” means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0080] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for predicting expansion rate in high-purity metals based on image processing, characterized in that: The following steps are involved: Obtain grayscale images of the grating image of the metal surface at different temperatures, and obtain a frequency spectrum of the grayscale image corresponding to any temperature; Determine the optimal filter radius of the filter of the spectrum graph; filter the spectrum graph according to the optimal filter radius to obtain an optimal spectrum graph; obtain an equidistant fringe image based on the optimal spectrum graph; The linear expansion coefficient of the metal at any temperature is obtained based on the arithmetic difference fringe image, and the linear expansion coefficient of the metal at the temperature to be measured is predicted based on the obtained linear expansion coefficients at different temperatures; Wherein, the optimal filtering radius is: Acquire a peak point sequence in a spectrum graph; construct a triangulated network model based on the position coordinates of each peak point in the peak point sequence; the triangulated network model includes multiple triangulated networks; calculate the priority value of each filter radius, the priority value is positively correlated with the fitness value and importance of each triangulated network; the fitness value represents the possibility of belonging to equidistant stripes in the corresponding triangulated network, and the importance is the overlap of any triangulated network with a circular area; the center of the circular area is the center point of the spectrum graph, and the radius is any filter radius; Based on each filter radius and the corresponding priority value, a corresponding fitting function is generated, and the filter radius corresponding to the maximum value of the derivative of the fitting function is taken as the optimal filter radius.

2. The method for predicting expansion rate in high-purity metal based on image processing according to claim 1, characterized in that: The acquisition process of the grating image is as follows: The metal to be tested is used as the screen, and a moiré fringe experiment is performed using two gratings with the same grating parameters. During the moiré fringe experiment, the angle between the grating lines of the two gratings is a set angle, and a moiré fringe image is formed on the surface of the metal to be tested, and the moiré fringe image is used as the grating image.

3. The method for predicting expansion rate in high-purity metal based on image processing according to claim 2, characterized in that: The appropriate value for: ； in, is the ratio of the relative angle of the jth triangulation to 90°, where the relative angle is the angle between the straight line formed by the center of gravity of the jth triangulation and the center point of the spectrum graph and the straight line formed along the angle β direction, is the ratio of the Euclidean distance from the centroid of the jth triangulated network to the center of the spectrum graph to the diagonal length of the grayscale image, is the ratio of the mean spectral intensity of all internal points in the jth triangulated network to the spectral intensity of the center point of the spectrum graph, is an exponential function with the natural constant e as base, , θ is the set angle.

4. The method for predicting expansion rate in high-purity metal based on image processing according to claim 1, characterized in that: The process of obtaining the importance is as follows: Randomly select any filter radius, take the center point of the spectrum graph as the center of the circle, and obtain the circular area corresponding to any filter radius; The number of overlapping data between the circular area and the j-th triangulated network is counted, and the ratio of the number of overlapping data to the number of all internal points in the j-th triangulated network is used as the importance of any filtering radius.

5. The method for predicting expansion rate in high-purity metal based on image processing according to claim 3 or 4, characterized in that: The priority value for: ； Among them, J is the total number of triangulated networks, is the importance of the sth filter radius, is the appropriate value of the jth triangulation.

6. The method for predicting expansion rate in high-purity metal based on image processing according to claim 1, characterized in that: Before constructing the triangulated network model, the steps of screening the peak point sequence to obtain a new peak point sequence are also included, specifically: Obtain the spectrum intensity of each peak point in the peak point sequence; Based on the position coordinates of each peak point and the corresponding spectrum intensity, the peak point sequence is divided into an interference sequence, a weak feature sequence, and a strong feature sequence; wherein the interference sequence is a category with the smallest spectrum intensity mean, the weak feature sequence is a category with the spectrum intensity mean in the middle, and the strong feature sequence is a category with the largest spectrum intensity mean; Eliminate the remaining peak points except the central peak point in the interference sequence and the strong feature sequence to obtain the weak feature sequence and the central peak point; The weak feature sequence and the central peak point are combined into a new peak point sequence.

7. The method for predicting expansion rate in high-purity metal based on image processing according to claim 1, characterized in that: The specific process of realizing the prediction of the linear expansion coefficient of the metal at the temperature to be measured is: performing temperature and linear expansion coefficient data fitting on the linear expansion coefficient of the metal at different temperatures to obtain a fitting curve, substituting the temperature to be measured into the fitting curve to obtain the predicted linear expansion coefficient.

8. The method for predicting expansion rate in high-purity metal based on image processing according to claim 1, characterized in that: After the grayscale image of each grating image is acquired, the grayscale image is further denoised to obtain a denoised grayscale image.

Images