Large shaft forging blank center forging method and system based on machine vision

Abstract

The application belongs to the technical field of image data processing, and particularly relates to a large shaft forging blank center forging and pressing method and system based on machine vision, which comprises the following steps: pre-processing a high-temperature steel ingot image; adopting a logarithm-Gabor filter set to convolve the image, and obtaining a complex response and local direction energy; for each pixel point, obtaining its structural saliency based on the sum of the real part, the imaginary part and the local direction energy of the complex response; obtaining its direction certainty based on the local direction energy and the direction vector; obtaining the final voting weight by combining the structural saliency and the direction certainty; and based on the weight, accumulating the votes and calculating the steel ingot geometric axis, so that the anvil of the forging and pressing machine is aligned with the steel ingot geometric axis. The application effectively suppresses strong interference such as water vapor and oxide skin by combining the authenticity and directionality of the pixel points for weighted voting, and realizes stable and accurate positioning of the steel ingot axis under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology. More specifically, this invention relates to a machine vision-based method and system for center forging of large shaft-type forging blanks. Background Technology

[0002] In the manufacturing of large shaft forgings, center forging is a key process for improving the core quality of forgings and eliminating internal porosity and voids. This process is crucial for improving the density and mechanical properties of forgings and is an essential step in the production of critical load-bearing components such as heavy generator rotors and turbine shafts. The forging effect directly depends on whether the forging anvil can accurately act on the core of the high-temperature steel ingot, i.e., the geometric axis of the steel ingot. Any centering deviation will cause the forging pressure to deviate from the center, resulting in poor core compaction and thus affecting the forging effect.

[0003] To ensure that the forging press accurately strikes the core of the forging, existing technologies have begun to incorporate machine vision systems for automatic centering. These systems acquire images of high-temperature steel ingots by installing industrial cameras on the upper part of the forging press, and use image processing algorithms such as edge detection and contour extraction to identify the geometric contour of the steel ingot. By fitting the extracted contour data, the system calculates the geometric axis of the steel ingot, and the position information of this axis is then transmitted to the control system of the forging press to guide the anvil to achieve non-contact automatic centering.

[0004] However, the aforementioned machine vision-based positioning method has fundamental flaws when facing the actual working conditions of central forging. The steel ingot is in a high-temperature state, and its surface has strong thermal radiation, accompanied by randomly distributed oxide scale and a large amount of water vapor generated by surface cooling. These factors together constitute complex dynamic optical interference. The edge detection or contour extraction algorithms that existing technologies rely on will completely fail under the above strong interference. The algorithms will misjudge the edges of water vapor, the edges of oxide scale patches, and the uneven bright areas caused by thermal radiation as the real edges of the steel ingot, resulting in the extraction of a large amount of incorrect edge information.

[0005] The root cause of this defect lies in the fact that the edge detection algorithms relied upon by existing technologies are based on the gray-level gradient of images. These algorithms assume that the local maxima of the gray-level gradient are the edges of objects. However, in the strong interference environment of forging, the "pseudo-gradients" generated by water vapor edges or uneven thermal radiation often have a much larger amplitude than the relatively blurred edge gradient formed by the steel ingot body due to high temperature. The algorithm cannot distinguish the source of the gradient and will inevitably prioritize locking the interference source with the strongest gradient amplitude. This is an inherent limitation in its principle. Ultimately, the geometric axis position calculated based on these unreliable edges will frequently jump and have low accuracy, failing to meet the precise positioning requirements of center forging. Summary of the Invention

[0006] To address the technical problem that the aforementioned visual positioning method based on grayscale gradients cannot distinguish between real edges and pseudo-gradients generated by interference under strong interference such as high temperature, water vapor, and oxide scale, resulting in low accuracy of the calculated geometric axis position of the steel ingot and difficulty in meeting the precise positioning requirements of center forging, this invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides a machine vision-based method for center forging of large shaft-type forging blanks, comprising: preprocessing an original grayscale image of a high-temperature steel ingot at a forging station; performing convolution operations on the preprocessed image using a log-Gabor filter bank to generate a complex response matrix, the complex response matrix containing the complex response of each pixel at different scales and directions; calculating the amplitude of each complex response to obtain the local directional energy of each pixel at the corresponding scale and direction; and for each pixel, based on the real part and imaginary part corresponding to the pixel in the complex response matrix, and the local directional energy corresponding to the pixel at the corresponding scale and direction. The structural saliency of a pixel is obtained by using local directional energy. For each pixel, the local directional energy in each direction is summed to obtain the total directional energy of the pixel in each direction. The X and Y components of the pixel's direction vector are also obtained. The directional determinism of the pixel is obtained based on the X and Y components of the pixel's direction vector and the total directional energy. The final voting weight of each pixel is obtained based on its structural saliency and directional determinism. If the final voting weight of a pixel is greater than a voting threshold, an accumulated vote is performed on the pixel. The geometric axis of the steel ingot is calculated based on the voting results.

[0008] This invention identifies the real structure by calculating the structural saliency of pixels and evaluates edge sharpness by calculating their directional determinism. The two are combined as the final voting weights and then Hough transform is performed. This method can effectively suppress the pseudo gradients generated in the image by strong interference sources such as high temperature water vapor and oxide scale, so that the real steel ingot edge dominates the voting, thus obtaining a stable and accurate steel ingot geometric axis even under complex dynamic optical interference.

[0009] Preferably, the structural saliency of the pixel satisfies the expression: ;in, For pixels Structural salience; For scale indexing; For direction index; For pixels In scale and direction The real part of the complex response on; For pixels In scale and direction The imaginary part of the complex response on; For pixels In scale and direction Local directional energy; It is a preset noise threshold; It is to prevent positive numbers with a denominator of zero; This represents the square root operation; Represents the squaring operation; This indicates a summation operation.

[0010] This invention utilizes the phase consistency principle to calculate structural saliency. By comparing the vector sum magnitude and scalar sum of the multi-scale response, pixels representing real structures such as steel ingot edges receive high scores due to their high phase consistency, while pixels representing noise such as water vapor receive low scores because their vector sums cancel each other out due to random phase. This achieves effective differentiation between real structures and non-structural noise at the pixel level.

[0011] Preferably, the orientation of the pixel satisfies the expression: ;in, It is a pixel. directional certainty; It is a pixel. The X component of the direction vector; It is a pixel. The Y component of the direction vector; For direction index; It is a pixel. In direction Total energy in the upward direction; It refers to the number of directions; It is to prevent positive numbers with a denominator of zero; This represents the square root operation; Represents the squaring operation; This indicates a summation operation.

[0012] This invention calculates directional determinism by analyzing the energy distribution of pixels in different directions. The energy of a clear edge is concentrated in a single dominant direction, making its vector and magnitude close to the scalar sum and thus scoring high. In contrast, the energy distribution of noisy or structurally intersecting pixels is dispersed, causing the vector components to cancel each other out and scoring low. This allows the invention to assess whether the direction of the pixel as an edge is reliable.

[0013] Preferably, the final voting weight of the pixel satisfies the expression: ;in, For pixels The final voting weight; For pixels Structural salience; It is a pixel. directional certainty; It represents the multiplication operation.

[0014] This invention multiplies structural saliency and orientation determinism to obtain the final voting weight. This multiplication relationship ensures that only pixels that belong to the real structure and have a clear orientation can obtain high weights. Pixels with low scores in either index, such as low structural saliency pixels that may be water vapor pixels or low orientation determinism pixels that may be intersection pixels, will have their final weights suppressed, thereby improving the signal-to-noise ratio of subsequent Hough transform voting.

[0015] Preferably, the preprocessing of the original grayscale image of the high-temperature steel ingot at the forging station includes: taking a top-view image of the high-temperature steel ingot at the forging station using one or more industrial cameras installed on the upper part of the forging press, wherein the cameras are heat-resistant or equipped with cooling and protection devices, and preprocessing the high-temperature steel ingot image using a filtering algorithm.

[0016] Preferably, the log-Gabor filter bank includes filters of multiple scales and multiple directions; the preprocessed image is convolved with each filter in the filter bank; the real and imaginary parts of the complex response generated by the convolution operation are saved; the number of scales in the log-Gabor filter bank is set to 4, and the number of directions is set to 6.

[0017] Preferably, obtaining the X and Y components of the direction vector of a pixel includes: multiplying the total directional energy of the pixel in each direction by the cosine value of the angle corresponding to that direction (doubled); summing these products obtained in all directions, and using the summed result as the X component of the direction vector; multiplying the total directional energy of the pixel in each direction by the sine value of the angle corresponding to that direction (doubled); summing these products obtained in all directions, and using the summed result as the Y component of the direction vector.

[0018] Preferably, the step of accumulating votes for a pixel in response to its final voting weight being greater than a voting threshold includes: initializing a Hough accumulator; traversing each pixel in the image; and allowing the pixel to perform a voting operation when its final voting weight is greater than the voting threshold. The voting operation involves obtaining the pixel's normal direction and accumulating the Hough spatial unit corresponding to that normal direction only according to the final voting weight value.

[0019] This invention employs a weighted and directional voting mechanism, where the vote value is determined by the final voting weight rather than simply by adding 1, and voting is only conducted on the dominant normal direction of the pixel. This mechanism allows high-weighted votes from clear edges of the ingot to overwhelm low-weighted votes from noisy or blurred structures, while avoiding the dispersion of effective votes in the Hough space, making the cumulative peaks representing edges more prominent and easier to search.

[0020] Preferably, the step of calculating the geometric axis of the steel ingot based on the voting results includes: after voting is completed, searching for the two peak pixels with the highest accumulated value in the Hough accumulator, which represent the two parallel edges of the steel ingot; taking the average value of the parameters of these two parallel lines to calculate the final geometric axis of the steel ingot.

[0021] Secondly, the present invention provides a machine vision-based center forging system for large shaft forging blanks, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned machine vision-based center forging method for large shaft forging blanks is implemented.

[0022] By adopting the above technical solution, the above-mentioned machine vision-based method for center forging of large shaft forging blanks is generated into a computer program and stored in a memory for loading and execution by a processor. This allows for the creation of terminal devices based on the memory and processor, making them convenient to use.

[0023] The beneficial effects of this invention are as follows:

[0024] This invention utilizes a log-Gabor filter bank to perform multi-scale directional energy decomposition on images and introduces the phase consistency principle to calculate structural saliency. Based on the characteristic that the real structure has a phase consistency across multiple scales while the noise phase is random, this method can effectively distinguish between the edges of the steel ingot body and pseudo gradients generated by water vapor, thermal radiation, etc., thus solving the problem that traditional edge detection algorithms extract a large number of erroneous edges under strong interference.

[0025] Based on the identification of real structural pixels, this invention further calculates directional determinism by analyzing the distribution of their energy in various directions, in order to evaluate whether the pixel has a single, clear edge direction. By multiplying the structural saliency representing realism with the directional determinism representing clarity, the final voting weight is obtained, ensuring that only pixels that are both real and have clear directions can participate in subsequent calculations with high weight.

[0026] This invention employs a weighted and directional Hough transform for axis calculation. The high-weighted votes cast by pixels from the clear edge of the steel ingot can overwhelm all low-weighted votes from noise or blurred structures, making the two peaks representing the edge of the steel ingot in the Hough space more prominent. This enables accurate calculation of the geometric axis of the steel ingot under actual working conditions of high temperature and high interference, providing reliable position information for precise centering of the forging process. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the machine vision-based center forging method for large shaft forging blanks in this invention.

[0028] Figure 2 This is a schematic illustration of the final voting weight distribution histogram in this invention;

[0029] Figure 3 This is a schematic diagram illustrating the geometric axis of the steel ingot in this invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0032] This invention discloses a machine vision-based method for center forging of large shaft-type forging blanks, referring to... Figure 1 This includes steps S1-S5:

[0033] S1: Preprocess the original grayscale image of the high-temperature steel ingot at the forging station.

[0034] It should be noted that this invention aims to solve the problem of inaccurate axis positioning of large forgings caused by high temperature and high interference environments such as water vapor and oxide scale. To this end, this invention first acquires the original grayscale image of the high-temperature steel ingot at the forging station, and optionally performs preprocessing through median filtering and other means to remove occasional pulse noise, so as to provide high-quality image input with clear structural edges for subsequent steps.

[0035] Specifically, one or more industrial cameras mounted on the upper part of the forging press are used to capture images of the high-temperature steel ingots at the forging station. Since the steel ingots are at high temperatures, the camera selection should consider high-temperature resistance or be equipped with appropriate cooling and protection devices. The acquired images are raw grayscale images. ,in The coordinates of the pixel in the image coordinate system.

[0036] Understandably, in real industrial environments, images may be subject to occasional impulse noise interference. As a preferred solution, the acquired original grayscale image can be processed... Median filtering is used for preprocessing. Median filtering can effectively remove isolated bright or dark pixels, i.e., impulse noise, while maintaining the sharpness of real structures in the image, such as the edge of a steel ingot, providing higher quality input for subsequent feature analysis.

[0037] S2: The preprocessed image is convolved using a log-Gabor filter bank to generate a complex response matrix; the amplitude of each complex response is calculated to obtain the local directional energy of each pixel at the corresponding scale and direction.

[0038] It should be noted that, in order to overcome the defect of traditional gradient operators such as the Sobel operator being prone to generating spurious gradients in strong noise environments, this invention employs a log-Gabor filter bank to perform feature decomposition on the preprocessed image. The log-Gabor filter has a Gaussian response on the logarithmic frequency scale and does not contain a DC component, making it particularly suitable for analyzing edges and textures with wide bandwidth characteristics in natural images, effectively decomposing the image into feature spaces of different scales and orientations.

[0039] Specifically, a log-Gabor filter bank is obtained, which contains filters of multiple scales and multiple directions. In this embodiment, the number of scales is... Set to 4, number of directions Set it to 6. A good balance can be achieved between computational efficiency and the ability to capture structures of varying coarseness; This means that the directional resolution is ( The preprocessed image With each filter in the filter bank Perform convolution operations, where For scale indexing, Using the direction index, the convolution process produces a complex response matrix. .

[0040] Furthermore, the real part of the complex response is preserved. and the virtual part Calculate the amplitude of the complex response, denoted as the local directional energy. When the direction and frequency of local image structures such as edges are related to the filter When the parameters are matched, the energy of the convolution response It will achieve a larger value, and vice versa.

[0041] S3: For each pixel, the structural saliency of the pixel is obtained based on the real part and imaginary part of the complex response matrix corresponding to the pixel, and the local directional energy corresponding to the pixel.

[0042] It should be noted that the purpose of this invention is to distinguish between real structures in an image, such as the edges of steel ingots and oxide scale textures, and unstructured noise, such as water vapor and thermal radiation. This is because the Fourier components of real structures have highly consistent phases across multiple scales, while the phases of noise are random and uncorrelated at different scales. Therefore, this invention utilizes the principle of phase consistency to obtain a structurally salient image.

[0043] Specifically, at each pixel At this point, structural saliency satisfies the expression:

[0044]

[0045] in, For pixels Structural salience; For scale indexing; For direction index; For pixels In scale and direction The real part of the complex response on; For pixels In scale and direction The imaginary part of the complex response on; For pixels In scale and direction Local directional energy; It is a preset noise threshold; It is to prevent positive numbers with a denominator of zero; This represents the square root operation; Represents the squaring operation; This indicates a summation operation.

[0046] More specifically, a batch of pure noise sample images, excluding steel ingots, are collected at the forging station. These images should only contain interference sources such as background, water vapor, and thermal radiation. An algorithm is run on these pure noise images to specifically calculate the amplitude of the Gabor response vector sum for each pixel. The amplitude values ​​of all noise pixels are then aggregated, and their statistical distribution characteristics are analyzed and plotted as a histogram. From this noise amplitude distribution, a 99th percentile value is selected as the preset noise threshold. .

[0047] In the formula, the molecules in This represents the magnitude of the vector sum of all complex responses. If the components are in phase, this value will be large; if the phases are random, it will become very small due to mutual cancellation; the denominator... It is the scalar sum of all response amplitudes, representing the total energy of that pixel. For pixels in real structures such as the edge of a steel ingot, the response phases across multiple scales are highly consistent, causing the magnitude of the vector sum to be close to the scalar sum, i.e., the numerator is close to the denominator. Therefore... The value is close to 1; for noise pixels such as water vapor, their phase is random, and the magnitude of the vector sum is much smaller than the scalar sum, leading to... The value is close to 0.

[0048] S4: For each pixel, sum the local directional energies in each direction to obtain the total directional energy of the pixel in each direction; and obtain the X and Y components of the direction vector of the pixel; based on the X and Y components of the direction vector of the pixel and the total directional energy, obtain the directional determinism of the pixel.

[0049] It should be noted that while this invention has identified which pixels are real, a real pixel, such as an edge, must still have a definite direction. The purpose of this invention is to further analyze the direction of the pixel and whether the direction is reliable, based on structural saliency. A clear edge should have a single dominant direction, while the energy of a noisy pixel or a pixel at a structural intersection may be distributed in multiple directions.

[0050] Specifically, each pixel In every direction The local directional energies are added together to obtain the pixel. In direction Total energy in the upward direction By direction direction angle Double the power, iterate through all directions, multiply the total directional energy of the pixel in each direction by the cosine of the doubled angle in that direction, and then sum all these products calculated for all directions. The sum is the X component of the direction vector. Similarly, iterate through all directions, multiply the total directional energy of the pixel in each direction by the sine of the angle after doubling in that direction, and then sum all these products calculated for all directions. The sum is the Y component of the direction vector. .

[0051] Furthermore, the orientation determinism of a pixel satisfies the expression:

[0052]

[0053] It is a pixel. directional certainty; It is a pixel. The X component of the direction vector; It is a pixel. The Y component of the direction vector; For direction index; It is a pixel. In direction Total energy in the upward direction; It refers to the number of directions; It prevents positive numbers with a denominator of zero, such as... ; This represents the square root operation; Represents the squaring operation; This indicates a summation operation.

[0054] If the energy of a pixel is highly concentrated in one direction, then the magnitude of the sum of the directional vectors will be close to the scalar sum of the energy in all directions, meaning the numerator is close to the denominator. The value is close to 1; conversely, if the energy is uniformly distributed in all directions, such as in noise pixels, the vector components will cancel each other out, causing the molecule to approach 0, thus... The value is also close to 0.

[0055] S5: Based on the structural saliency and orientation determinism of each pixel, obtain the final voting weight of that pixel; in response to the final voting weight of a pixel being greater than the voting threshold, accumulate votes for that pixel; calculate the geometric axis of the steel ingot based on the voting results.

[0056] It should be noted that the purpose of this invention is to comprehensively utilize the structural saliency and directional determinism of pixels to detect two parallel straight lines representing the outline of a steel ingot from these sparse, noisy but high-confidence edge pixels. This invention multiplies two weights to obtain the final voting weight, ensuring that only pixels that are both real and directionally definite participate in the Hough transform voting with high weight. The two edge straight lines are determined by finding the two highest accumulated peak values ​​in the Hough space, and their average value is used to calculate the final geometric axis of the steel ingot.

[0057] Specifically, initialize a two-dimensional Hough accumulator. ,in This represents the distance from the line to the original pixel. The normal angle is represented by the accumulator, with all cells initialized to 0; calculations are performed for each pixel. Final voting weight Satisfies the expression:

[0058]

[0059] A pixel, if its A high value indicates that the pixel belongs to a real structure; The high height also indicates a clear direction. Only then will it be very high. If any value is very low, for example, A low value indicates that the pixel might be water vapor. If a pixel has a low weight, it may be a structural intersection pixel, and its final weight will be suppressed, thus it will not play a dominant role in the voting.

[0060] For example, Figure 2 This is the histogram of the final voting weight distribution.

[0061] Furthermore, after obtaining the final voting weights, this invention initializes a two-dimensional Hough accumulator; it iterates through each pixel in the image and sets a voting threshold. Only when the final voting weight of a pixel is greater than the threshold is it considered a high-confidence structural pixel and allowed to perform the voting operation; during voting, the pixel obtains its normal direction from the local dominant direction and only votes on the Hough space unit corresponding to that normal direction; the voting update rule is to accumulate the final voting weight value of the pixel, rather than simply adding 1 as in the traditional method. This weighted and directional voting mechanism ensures that the high-weight votes cast by pixels from the clear edges of the ingot can overwhelm all low-weight votes from noisy or blurred structures; after voting is completed, the two peak pixels with the highest accumulated values ​​are searched in the Hough accumulator. These two peaks represent the two parallel edges of the ingot; the system takes the average value of the parameters of these two parallel lines to calculate the final geometric axis of the ingot, so that the anvil of the forging press is aligned with the geometric axis of the high-temperature ingot.

[0062] For example, Figure 3 This is a schematic diagram of the geometric axis of a steel ingot. The blue and red line segments are the two parallel edges of the steel ingot; the green line segment is the geometric axis of the steel ingot.

[0063] Thus, a machine vision-based method for center forging of large shaft forging blanks has been realized.

[0064] This invention also discloses a machine vision-based center forging system for large shaft forging blanks, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the machine vision-based center forging method for large shaft forging blanks according to this invention.

[0065] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for center forging of large shaft-type forging blanks based on machine vision, characterized in that, include: Preprocess the original grayscale image of the high-temperature steel ingot at the forging station; For each pixel, a convolution operation is performed using a log-Gabor filter bank to generate complex responses of the pixel at different scales and directions; the magnitude of the complex response is used as the local directional energy of the pixel at the corresponding scale and direction; at the same scale and direction, the structural saliency of the pixel is obtained based on the real and imaginary parts of the complex response matrix corresponding to the pixel, as well as the local directional energy corresponding to the pixel. For any direction, the local directional energies of the pixel at different scales are added together to obtain the total directional energy of the pixel in that direction; the X and Y components of the pixel's direction vector are obtained based on the sum of the projections of the total directional energy of all directions onto the X and Y axes, respectively; the directional determinism of the pixel is obtained based on the sum of the total directional energy of all directions of the pixel and the X and Y components of the direction vector. The final voting weight of a pixel is obtained based on its structural saliency and orientation determinism. In response to a final voting weight exceeding a voting threshold, the pixel undergoes an accumulation voting process, including: initializing the Hough accumulator; traversing each pixel in the image; when a pixel's final voting weight exceeds a voting threshold, the pixel is allowed to perform a voting operation; the voting operation involves obtaining the pixel's normal direction and accumulating only the Hough space cells corresponding to that normal direction according to the final voting weight value; and calculating the ingot's geometric axis based on the voting results, including: After voting is completed, the two peak pixels with the highest accumulated value are searched in the Hough accumulator. These two peaks represent the two parallel edges of the steel ingot. The average value of the parameters of these two parallel lines is taken to calculate the final geometric axis of the steel ingot, so that the anvil of the forging press is aligned with the geometric axis of the steel ingot.

2. The method for center forging of large shaft forging blanks based on machine vision according to claim 1, characterized in that, The structural saliency satisfies the expression: ； in, For pixels Structural salience; For scale indexing; For direction index; For pixels In scale and direction The real part of the complex response on; For pixels In scale and direction The imaginary part of the complex response on; For pixels In scale and direction Local directional energy; It is a preset noise threshold; It is to prevent positive numbers with a denominator of zero; This represents the square root operation; Represents the squaring operation; This indicates a summation operation.

3. The method for center forging of large shaft forging blanks based on machine vision according to claim 1, characterized in that, The orientation of the pixel satisfies the expression: ； in, It is a pixel. directional certainty; It is a pixel. The X component of the direction vector; It is a pixel. The Y component of the direction vector; For direction index; It is a pixel. In direction Total energy in the upward direction; It refers to the number of directions; It is to prevent positive numbers with a denominator of zero; This represents the square root operation; Represents the squaring operation; This indicates a summation operation.

4. The method for center forging of large shaft forging blanks based on machine vision according to claim 1, characterized in that, The final voting weight of the pixel satisfies the expression: ； in, For pixels The final voting weight; For pixels Structural salience; It is a pixel. The certainty of direction.

5. The method for center forging of large shaft forging blanks based on machine vision according to claim 1, characterized in that, The preprocessing of the original grayscale image of the high-temperature steel ingot at the forging station includes: A top-view image of a high-temperature steel ingot at the forging station is captured by one or more industrial cameras installed on the top of the forging press. The cameras are heat-resistant or equipped with cooling and protection devices. The high-temperature steel ingot image is preprocessed by a filtering algorithm.

6. The method for center forging of large shaft forging blanks based on machine vision according to claim 1, characterized in that, The step of performing convolution operations on the preprocessed image using a log-Gabor filter bank includes: The log-Gabor filter bank contains filters of multiple scales and multiple directions; the preprocessed image is convolved with each filter in the filter bank; the real and imaginary parts of the complex response generated by the convolution operation are saved.

7. The method for center forging of large shaft forging blanks based on machine vision according to claim 1, characterized in that, The process of obtaining the X and Y components of the direction vector of a pixel includes: Multiply the total directional energy of a pixel in each direction by the cosine value of the angle corresponding to that direction (doubled); then sum these products in all directions and use the sum as the X component of the direction vector; multiply the total directional energy of a pixel in each direction by the sine value of the angle corresponding to that direction (doubled); then sum these products in all directions and use the sum as the Y component of the direction vector.

8. A machine vision-based center forging system for large shaft-type forging blanks, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a machine vision-based center forging method for large shaft forging blanks according to any one of claims 1-7.

Images