A method and system for monitoring the forging of a shaft part

Abstract

The present application relates to the technical field of image processing, and more particularly to a shaft part forging forming monitoring method and system. The method comprises: obtaining feature values of a pixel point in the surface image of a formed shaft part in multiple channels in HSV color space, and a gradient direction angle of the pixel point; determining local consistency of the pixel point; determining a weighted Euclidean distance between the pixel points; clustering all pixel points according to the weighted Euclidean distance to obtain a plurality of clustering clusters; determining gradient saliency and candidate corner point number of an image region where the clustering cluster is located; and obtaining candidate corner points by using a Harris corner point detection algorithm based on the candidate corner point number, and measuring whether the shaft part forging is qualified according to the Euclidean distance between the candidate corner points. The present application solves the problem of inaccurate recognition caused by large feature differences in different geometric regions by adaptively clustering the candidate corner point number of the image region where the cluster is located, thereby improving the accuracy of shaft part forging quality monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a shaft part forging forming monitoring method and system. BACKGROUND

[0002] In the manufacturing process of complex metal parts, shaft parts are usually produced by forging process, which puts strict requirements on dimensional accuracy. However, the problems of internal friction of machine tool and loosening of parts commonly existing in the forging process often lead to insufficient pressure or uneven pressure distribution in the hydraulic system, which directly affects the flowability and filling effect of metal materials, and further causes the inhomogeneity of the internal structure of the parts. In the subsequent cooling process, this inhomogeneity is further amplified, resulting in significant differences in strength and hardness of different regions of the parts, which ultimately leads to uneven deformation such as distortion or dimensional deviation after cooling forming, seriously affecting the quality and performance indicators of the parts. Therefore, it is crucial to monitor the shape of the cooled and formed shaft parts in order to timely discover and handle unqualified parts and ensure product quality.

[0003] The shaft part itself has many geometric regions, such as transition fillets and shaft shoulders, and the edges of these regions often bear complex stress and strain states during forging. The flowability and filling effect of metal materials in these regions are more easily hindered, leading to uneven distribution of materials in the edge regions, and these edge regions are more prone to deformation after cooling forming, which can cause significant changes in the local region gradient. However, the traditional Harris corner point detection algorithm relies on a global threshold to identify obvious gradient direction angle change points. However, due to the large difference in gradient of different geometric regions of shaft parts, a fixed global threshold cannot accurately adapt to the characteristics of all geometric regions, resulting in reduced accuracy of corner point recognition, which further affects the monitoring of shaft part forging forming. SUMMARY

[0004] To solve the problem that the traditional Harris corner point detection algorithm relies on a global threshold to identify obvious gradient direction angle change points, but due to the large difference in gradient of different geometric regions of shaft parts, a fixed global threshold cannot accurately adapt to the characteristics of all geometric regions, resulting in reduced accuracy of corner point recognition, which further affects the monitoring of shaft part forging forming, the present application provides a shaft part forging forming monitoring method and system.

[0005] In the first aspect, the present application provides a shaft part forging forming monitoring method, which adopts the following technical solution:

[0006] The application discloses a method for monitoring forging forming of shaft parts, and comprises the following steps: acquiring feature values of pixels in a surface image of a formed shaft part in multiple channels of an HSV color space and gradient direction angles of the pixels; regarding any pixel as a target pixel, determining local consistency of the target pixel according to feature values of the target pixel and pixels in a neighborhood in each channel of the HSV color space; in response to a cosine similarity between feature vectors corresponding to all channels of the target pixel and any other pixel being greater than 0, weighting an Euclidean distance between the two pixels according to the local consistency of the two pixels and the cosine similarity, to obtain a weighted Euclidean distance between the two pixels; clustering all the pixels according to the weighted Euclidean distance, to obtain a plurality of clustering clusters; determining gradient saliency of an image region where the clustering cluster is located according to differences between gradient direction angles of the pixels in the clustering cluster and a number of the pixels in the clustering cluster; determining a number of candidate corner points of the image region where the clustering cluster is located according to the gradient saliency, the number of the pixels and a preset global threshold; and acquiring the candidate corner points by using a Harris corner point detection algorithm based on the number of the candidate corner points, and measuring whether forging of the shaft part is qualified according to an Euclidean distance between the candidate corner points.

[0007] The application utilizes local consistency analysis, dynamically adjusts a detection standard according to features of different geometric regions, adapts to various gradient differences, and thus improves accuracy of corner point recognition; the clustering of the pixels by the weighted Euclidean distance can effectively distinguish different feature regions, reduces a problem of reduced clustering accuracy caused by a traditional clustering method, and increases sensitivity to subtle structures; the saliency of the clustering cluster is determined in combination with the gradient direction angles and the number of the pixels, which improves recognition ability of key features and provides a more reliable basis for monitoring of forging forming quality; the number of the candidate corner points of the image region where each clustering cluster is located is adaptively adjusted, effectively solving a problem of inaccurate recognition caused by large feature differences of different geometric regions, and improving accuracy of forging quality monitoring of the shaft parts.

[0008] Further, the feature value acquisition method comprises the following steps: converting an RGB color space of the surface image to an HSV color space, to obtain the feature values of the pixels in the multiple channels of the HSV color space.

[0009] Further, the gradient direction angle acquisition method comprises the following steps: performing grayscale processing on the surface image, and acquiring the gradient direction angles of the pixels on the surface image after the grayscale processing by using a Sobel operator.

[0010] Further, the local consistency satisfies:

[0011] ; in the formula, is a first gradient direction angle of the target pixel, local consistency of the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the number of channels, is a natural exponential function.

[0012] The application determines local consistency by comprehensively considering HSV multi-channel features and neighborhood information, accurately describes local color characteristics of shaft parts, and provides rich and reliable information for corner point recognition; the local consistency is calculated in an exponential form, the sensitivity to feature differences is enhanced, the numerical stability is ensured, and subsequent weighted Euclidean distance clustering operations are facilitated.

[0013] Further, the weighted Euclidean distance satisfies:

[0014] ; in the formula, is the weighted Euclidean distance between the i-th pixel point and the j-th pixel point, is the Euclidean distance between the i-th pixel point and the j-th pixel point, is the local consistency of the i-th pixel point, is the local consistency of the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the i-th pixel point, is the cosine similarity between the i-th pixel point and the j-th pixel point, is the i-th pixel point, is an absolute value symbol.

[0015] The application comprehensively measures the relationship between pixels by fusing the Euclidean distance, local consistency and cosine similarity of the feature vector of the pixel points, and provides more accurate distance measurement for clustering; when the cosine similarity of the feature vector is greater than 0, the weight is adaptively adjusted, the distance is dynamically adjusted according to the difference in local consistency, and the characteristics of different regions of the shaft part are adapted.

[0016] Further, the clustering adopts an iterative self-organizing clustering algorithm.

[0017] Further, the gradient saliency satisfies:

[0018] ; in the formula, is the gradient saliency of the image region where the m-th cluster is located, is the gradient saliency of the image region where the m-th cluster is located, is the gradient direction angle of the n-th pixel point in the m-th cluster, is the gradient direction angle of the n-th pixel point in the m-th cluster, is the gradient direction angle of the n-th pixel point in the m-th cluster, is the gradient direction angle of the n-th pixel point in the m-th cluster, is the gradient direction angle of the n-th pixel point in the m-th cluster, is the gradient direction angle of the n-th pixel point in the m-th cluster, is the number of pixel points in the m-th cluster, is the number of pixel points in the m-th cluster, is a standard normalization function.

[0019] The gradient saliency can fully reflect the visual feature complexity of the region by calculating the gradient direction angle difference of all pixel points in the same cluster, so that the algorithm not only focuses on the characteristics of a single pixel, but also focuses on the feature changes of the entire region, thereby providing more comprehensive feature description. The calculation of the gradient saliency emphasizes the gradient direction angle changes between pixels in the same cluster, which can effectively identify the complexity of the structure in the region. The increase in saliency usually means that this region may contain more edge or detail information, thereby providing a more important basis for subsequent corner point identification.

[0020] Further, the number of candidate corner points satisfies:

[0021] ; in the formula, is the number of candidate corner points of the image region where the m-th cluster is located, is the number of candidate corner points of the image region where the m-th cluster is located, is the gradient saliency of the image region where the m-th cluster is located, is the gradient saliency of the image region where the m-th cluster is located, is the number of clusters, is a preset global threshold, is the number of pixel points in the surface image, is a rounding function.

[0022] The number of candidate corner points can be dynamically adjusted according to the relative complexity of the region features, and this adaptability ensures that more corner points are detected in feature-rich regions, improving the sensitivity and accuracy of monitoring. Combined with the preset global threshold, the number of candidate corner points can be controlled while ensuring that the detection result meets certain quality standards, so that the number of corner points not only depends on local features, but also considers global detection targets, improving the stability and reliability of quality monitoring.

[0023] Further, the measuring whether the shaft part forging is qualified according to the Euclidean distance between the candidate corner points comprises: in response to the difference between the maximum value of the Euclidean distance between all candidate corner points in the surface image of the formed shaft part and the maximum value of the Euclidean distance between all candidate corner points in the surface image of the standard shaft part being greater than a preset abnormal threshold, determining that the forging of the current formed shaft part is unqualified, and issuing a warning prompt, thereby completing the shaft part forging forming monitoring.

[0024] In a second aspect, the present application provides a shaft part forging forming monitoring system, which adopts the following technical scheme:

[0025] A shaft part forging forming monitoring system comprises a processor and a memory, and the memory stores computer program instructions.

[0026] By adopting the above technical scheme, the above-mentioned shaft part forging forming monitoring method is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is made according to the memory and the processor, and the use is convenient.

[0027] The present application has the following technical effects:

[0028] The HSV color space multi-channel feature values of the shaft part surface image pixel points are acquired, and gradient direction angle analysis is combined to determine local consistency. When the target pixel point and the neighborhood point in the HSV channel feature values are determined, the color information is comprehensively considered, the color difference interference is reduced, the corner point recognition is more accurate, and when the cosine similarity of the feature vectors of the target pixel point and the remaining pixel points is greater than 0, the local consistency and the cosine similarity are combined to weight the Euclidean distance to obtain a weighted Euclidean distance. This weighting method can comprehensively consider the color feature similarity and the local consistency between the pixel points, so that the clustering result is more reasonable, the pixel points with similar colors and local structures in the shaft part image are more accurately divided into the same clustering cluster, and a more reliable basis is provided for subsequent corner point recognition. When the pixel point clustering obtains a clustering cluster and the gradient saliency is determined, the gradient direction angle difference and the number of the pixel points in the cluster are considered, the gradient and the pixel distribution analysis are performed according to different geometric regions, the uniform global threshold is not relied on, the characteristics of the large gradient difference of the shaft part region are matched, the number of candidate corner points is determined in combination with the gradient saliency, the total number of pixel points and the preset global threshold, this dynamic mode fuses local and global pixel information, accurately identifies potential corner points in different regions, and avoids missed detection and false detection caused by a fixed threshold. The accurate identification of the corner points can capture the shape change of the part forming, compare the distance with the qualified standard, timely detect shape deviation and other abnormalities, provide a reliable basis for quality control, and improve the monitoring efficiency of the shaft part forging forming. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a method flow chart in a shaft part forging forming monitoring method. DETAILED DESCRIPTION

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

[0031] The embodiments of the present application disclose a shaft part forging forming monitoring method, referring to Figure 1 , comprising steps S1-S6:

[0032] S1: Obtain the feature values of the pixel points in the surface image of the formed shaft part in multiple channels in the HSV color space, and the gradient direction angle of the pixel points.

[0033] It should be noted that by using an industrial camera to collect images in a uniform light environment, according to the size and shape of the surface of the formed shaft part, a suitable shooting angle and distance are selected to ensure that the surface details can be fully captured. The surface of the shaft part has various geometric features (transition fillets, shaft shoulders, etc.), which may indeed be disturbed by various factors in the RGB color space, for example, under the same lighting conditions, different geometric features may produce different shadow or light reflection effects, resulting in inaccurate or difficult-to-distinguish color features in the RGB color space of different geometric regions. Considering that in the HSV space, the hue, saturation and brightness three parameters can more intuitively reflect the color information in the image, and have a certain robustness to light changes. Therefore, the surface image is converted from the RGB color space to the HSV color space to more accurately extract the color features.

[0034] Specifically, the feature value acquisition method is:

[0035] The RGB color space of the surface image is converted to the HSV color space to obtain the feature values of the pixel points in the surface image in multiple channels in the HSV color space.

[0036] Specifically, the gradient direction angle acquisition method is:

[0037] The surface image is subjected to grayscale processing, and the gradient direction angle of the pixel points is obtained on the grayscale-processed surface image by using a Sobel operator.

[0038] S2: Determine the local consistency of the pixel points.

[0039] It should be noted that different geometric regions of shaft parts often exhibit significant variations in visual features such as color and brightness. These variations can be effectively reflected in the HSV color space. Pixels with similar local consistency often belong to the same geometric region because they have similar color characteristics and spatial distribution. Conversely, pixels with significant differences in local consistency may belong to different geometric regions. Therefore, by analyzing the local consistency of each pixel, we can obtain the spatial distribution consistency of these features for each pixel.

[0040] Any pixel is recorded as the target pixel, and the local consistency of the target pixel is determined based on the characteristic values ​​of each channel of the target pixel and the pixels in the neighborhood in the HSV color space.

[0041] Specifically, the local consistency satisfies:

[0042] ;

[0043] Where, For the The local consistency of pixels, For the The pixel at the The eigenvalues ​​of the channels, For the The pixel in the neighborhood The pixel at the The eigenvalues ​​of the channels, For the The number of pixels in the neighborhood of a pixel, is the number of channels, is the natural exponential function.

[0044] Among them, local consistency reflects the consistency of the HSV color space features of the current pixel and the pixels in its neighborhood. The larger the value, the closer the color features of the pixel and the surrounding pixels are in different channels. Indicates the The pixel point and the pixel point in the neighborhood are The sum of the absolute differences of the eigenvalues ​​of each channel. The smaller the value, the smaller the difference between the pixel and the pixels in the neighborhood on this channel, which means that the color feature changes of the pixel and the pixels in the neighborhood on this channel are more consistent; the local consistency of the color features on each channel is summed to reflect the overall consistency of the pixel relative to its neighborhood in the entire HSV color space, that is, The smaller the value, the higher the local consistency of the current pixel.

[0045] S3: Determine the weighted Euclidean distance between pixels.

[0046] It's important to note that traditional clustering methods often only consider pixel color and coordinate information for clustering. However, at the intersection of different regions, color differences are minimal and locations are close, ignoring the consistency of local pixel features. This can lead to misclassification of pixels in these regions, which are crucial for corner identification. Therefore, by introducing local consistency (the similarity of the local color features of two pixels) and cosine similarity (the similarity of the color features of two pixels) to measure the cluster distance between pixels, we can more accurately reflect the true relationship between pixels.

[0047] In response to the cosine similarity between the feature vectors corresponding to all channels of the target pixel and any other pixel points being greater than 0, the Euclidean distance between the two pixel points is weighted according to the local consistency of the two pixel points and the cosine similarity to obtain a weighted Euclidean distance between the two pixel points.

[0048] Specifically, the weighted Euclidean distance satisfies:

[0049] ;

[0050] Where, For the Pixels and The weighted Euclidean distance between pixels, For the Pixels and The Euclidean distance between pixels, For the The local consistency of pixels, No. The local consistency of pixels, For the The feature vector corresponding to each pixel in all channels is: For the The feature vector corresponding to each pixel in all channels is: For the Pixels and The cosine similarity between the feature vectors corresponding to all channels of the pixel points, is the absolute value symbol.

[0051] wherein the weighted Euclidean distance reflects the spatial distance and color feature similarity between the pixel points, the smaller the value, the closer the spatial distribution between the two pixel points and the more similar the color features, meaning that the two pixel points are likely to be a geometric region (different geometric regions are prone to different shadow or light reflection effects due to different light, which means that the color features of different geometric regions change significantly); the more similar the feature vectors between the two pixel points, i.e. , the more similar the color features of the two pixel points in each channel; in this case, the smaller the difference in local consistency after normalization of the two pixel points, the more similar the local color features of the two pixel points, meaning that the two pixel points have a greater probability of being distributed in the same geometric region; therefore, the current Euclidean distance needs to be reduced so that the two pixel points can be clustered. Conversely, the greater the difference in local consistency between the two pixel points, the more likely the two pixel points are located at the junction of different regions, in which case the current Euclidean distance needs to be increased to avoid clustering.

[0052] The more dissimilar the feature vectors between the two pixel points, i.e. , meaning that the two pixel points do not belong to the same region, and the original distance can be maintained.

[0053] S4: clustering all pixel points according to the weighted Euclidean distance to obtain a plurality of clustering clusters.

[0054] Specifically, the clustering adopts an iterative self-organizing clustering algorithm.

[0055] The weighted Euclidean distance between the pixel points is used to replace the distance in the traditional iterative self-organizing clustering algorithm for clustering, until the clustering clusters no longer change and the clustering stops, obtaining the final clustering result, i.e. a plurality of clustering clusters.

[0056] S5: determining the gradient saliency and the number of candidate corner points of the image region where the clustering cluster is located.

[0057] It should be noted that pixel points with similar color features and spatial distribution are clustered into a class, thereby distinguishing different geometric regions that the part may have. Different geometric regions often bear different complex stress and strain states during forging, which leads to differences in gradient saliency of different regions. The traditional Harris corner detection algorithm relies on a global threshold to identify obvious gradient direction angle change points, but ignores the fact that the gradient of actual geometric mutation regions has great differences, and a fixed global threshold cannot accurately adapt to the characteristics of all mutation regions, resulting in inaccurate corner recognition. Therefore, the gradient direction angle change degree of each region pixel point needs to be analyzed to evaluate the gradient saliency of the region, so as to adaptively adjust the suitable threshold for each region.

[0058] According to the difference between gradient direction angles of pixel points in the clustering cluster and the number of pixel points in the clustering cluster, gradient saliency of an image region where the clustering cluster is located is determined.

[0059] Specifically, the gradient saliency satisfies:

[0060] ;

[0061] In the formula, G is the gradient saliency of the image region where the i-th clustering cluster is located, N is the number of pixel points in the i-th clustering cluster, and θi is the gradient direction angle of the j-th pixel point in the i-th clustering cluster. is a standard normalization function.

[0062] According to the gradient saliency, the number of all pixel points, and a preset global threshold, the number of candidate corner points of the image region where the clustering cluster is located is determined.

[0063] Specifically, the number of candidate corner points satisfies:

[0064] ;

[0065] In the formula, C is the number of candidate corner points of the image region where the i-th clustering cluster is located, G is the gradient saliency of the image region where the i-th clustering cluster is located, N is the number of pixel points in the i-th clustering cluster, K is the number of clustering clusters, and T is the preset global threshold. is a rounding function.

[0066] The global threshold can be set by the implementer according to specific implementation conditions, for example, 1%; the traditional Harris corner point detection algorithm uses a formula of a corner point response function to obtain a response value of each pixel point according to eigenvalues of a second-order matrix, and generally, a global threshold is set to take the first 10% of the response values as candidate corner points.

[0067] In the gradient saliency calculation formula, θi represents the gradient direction angle of the j-th pixel point in the i-th clustering cluster. ​​​​​​​​​​​​​​​​​​​The mean value of the angle difference between the gradient directions of two pixel points in the image region of each cluster (the pixel points in the cluster and the cluster are completely consistent), the greater the value, the more obvious the gradient change of the current image region, that is, the image region contains more details and edge information, which may mean that the image region has more potential corner points; in the candidate corner point number calculation formula represents the significance weight of the image region of the first cluster, if the gradient significance of the image region of the cluster is greater, the threshold value needs to be reduced to identify more potential corner points, therefore obtains the number of candidate corner points that each image region should be allocated, since represents the significance weight of the image region of the first cluster, therefore the sum of the significance weights of all clusters is 1, and finally the total candidate corner points are ensured unchanged by the rounding function.

[0068] S6: based on the number of candidate corner points, using the Harris corner point detection algorithm to obtain candidate corner points, and according to the Euclidean distance between the candidate corner points, whether the shaft type part forging is qualified is measured.

[0069] It should be noted that for each image region of each cluster, the response values of each pixel point are sorted from large to small, and the first pixel points with the largest response value after sorting are selected as candidate corner points, and here is the number of candidate corner points of the image region of the cluster.

[0070] Specifically, the measurement of whether the shaft type part forging is qualified according to the Euclidean distance between the candidate corner points comprises:

[0071] In response to the difference between the maximum value of the Euclidean distance between all candidate corner points in the surface image of the formed shaft type part and the maximum value of the Euclidean distance between all candidate corner points in the surface image of the standard shaft type part being greater than a preset abnormal threshold value, it is determined that the forging of the current formed shaft type part is unqualified, and a warning prompt is issued, and the shaft type part forging forming monitoring is completed.

[0072] The implementer can set the abnormal threshold value according to the specific implementation, for example, 0.1cm.

[0073] The embodiment of the application also discloses a shaft type part forging forming monitoring system, comprising a processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, a shaft type part forging forming monitoring method according to the application is realized.

[0074] The system also comprises other components well known to those skilled in the art, such as a communication bus and a communication interface, which are arranged and function as known in the art and thus will not be described in more detail here.

[0075] The above are preferred embodiments of the present application, which do not limit the protection scope of the present application, and thus: any equivalent changes made on the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for monitoring the forging of shaft parts, characterized in that: include: Obtain the characteristic values ​​of multiple channels of the pixel points in the surface image of the formed shaft part in the HSV color space, as well as the gradient direction angle of the pixel points; Any pixel is marked as the target pixel, and the local consistency of the target pixel is determined based on the characteristic values ​​of each channel of the target pixel and the pixels in the neighborhood in the HSV color space; In response to the cosine similarity between the feature vectors corresponding to all channels of the target pixel and any other pixel being greater than 0, the Euclidean distance between the two pixels is weighted according to the local consistency of the two pixels and the cosine similarity to obtain a weighted Euclidean distance between the two pixels, satisfying: Where, For the Pixels and The weighted Euclidean distance between pixels, For the Pixels and The Euclidean distance between pixels, For the The local consistency of pixels, No. The local consistency of pixels, For the The feature vector corresponding to each pixel in all channels is: For the The feature vector corresponding to each pixel in all channels is: For the Pixels and The cosine similarity between the feature vectors corresponding to all channels of the pixel points, is the absolute value symbol; Clustering all pixels according to the weighted Euclidean distance to obtain a number of clusters; According to the difference in gradient direction angles between pixels in the cluster and the number of pixels in the cluster, the gradient significance of the image region where the cluster is located is determined, including: Where, For the The gradient significance of the image region where the clusters are located, For the The first The gradient direction angle of each pixel, For the The first The gradient direction angle of each pixel, For the The number of pixels in a cluster, is the standard normalization function; Determining the number of candidate corner points in the image region where the cluster is located based on the gradient significance, the number of all pixels, and a preset global threshold, including: Where, For the The number of candidate corner points in the image region where the cluster is located, is the number of clusters, is the preset global threshold, is the number of pixels in the surface image, is the rounding function; Based on the number of candidate corner points, the Harris corner point detection algorithm is used to obtain candidate corner points, and the euclidean distance between the candidate corner points is used to measure whether the forging of the shaft parts is qualified.

2. A method for monitoring the forging of shaft parts according to claim 1, characterized in that: The method for obtaining the characteristic value is: The RGB color space of the surface image is converted to the HSV color space, and the characteristic values ​​of multiple channels of the pixel points in the surface image in the HSV color space are obtained.

3. The method for monitoring the forging of shaft parts according to claim 1, characterized in that: The method for obtaining the gradient direction angle is: The surface image is gray-scaled, and the Sobel operator is used to obtain the gradient direction angle of the pixel point on the gray-scaled surface image.

4. A method for monitoring the forging of shaft parts according to claim 1, characterized in that: The local consistency satisfies: ； Where, For the The local consistency of pixels, For the The pixel at the The eigenvalues ​​of the channels, For the The pixel in the neighborhood The pixel at the The eigenvalues ​​of the channels, For the The number of pixels in the neighborhood of a pixel, is the number of channels, is the natural exponential function.

5. The method for monitoring the forging of shaft parts according to claim 1, characterized in that: The clustering adopts an iterative self-organizing clustering algorithm.

6. A method for monitoring the forging of shaft parts according to claim 1, characterized in that: The method of measuring whether the forging of shaft parts is qualified based on the Euclidean distance between the candidate corner points includes: In response to the fact that the difference between the maximum value of the Euclidean distance between all candidate corner points in the surface image of the formed shaft part and the maximum value of the Euclidean distance between all candidate corner points in the surface image of the standard shaft part is greater than a preset abnormality threshold, the forging of the current formed shaft part is determined to be unqualified, and an early warning prompt is issued to complete the forging monitoring of the shaft part.

7. A monitoring system for forging shaft parts, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a method for monitoring the forging of shaft parts according to any one of claims 1 to 6 is implemented.

Images