Circular curve burr detection method and system based on multi-scale phase consistency and contour model, and medium
By employing a multi-scale phase consistency and contour model approach, the challenges of illumination and contrast in circular curve burr detection have been addressed, achieving high-precision, low-miss-rate automatic burr detection, suitable for quality control of components such as bearings and gears.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HANGZHOU HUICUI INTELLIGENT TECH CO LTD
- Filing Date
- 2026-04-21
- Publication Date
- 2026-05-19
AI Technical Summary
Traditional manual visual inspection methods are inefficient and susceptible to subjective factors. Machine vision technology faces challenges in the detection of burrs on circular curves, such as light sensitivity, low contrast, morphological diversity, and complex background interference, making it difficult to achieve high-precision and efficient burr detection.
A multi-scale phase consistency and contour model is adopted. Image features are extracted through Log-Gabor filter banks to construct an active contour model. Combining contour curvature and local phase consistency energy features, a linear weighted scoring function is used for fusion and discrimination to mark spur regions.
It can stably extract contour edges in scenarios with uneven lighting and contrast differences, significantly reduce the false negative rate, achieve high-precision burr detection, and has a high degree of automation and intelligence.
Smart Images
Figure CN122066705A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of curve burr detection technology, and more specifically, to a method, system, and medium for detecting burrs on circular curves based on multi-scale phase consistency and contour models. Background Technology
[0002] In modern precision manufacturing, components with circular or arc-shaped contours, such as bearings, gears, cams, and sealing rings, are widely used. The contour quality of these parts, especially the smoothness of their edges, directly determines their service performance, lifespan, and reliability. Burrs, a common machining defect, refer to tiny, protruding excess material generated at the edges of workpieces due to plastic deformation. On circular curves, burrs manifest as local deviations from the ideal smooth arc or curve. These tiny burrs can cause stress concentration, accelerated wear, abnormal noise, and even system failure in high-speed operating mechanisms. Therefore, high-precision and high-efficiency burr inspection of circular contours in the final stage of production is a key process to ensure product quality.
[0003] Traditional manual visual inspection methods suffer from inherent drawbacks such as low efficiency, high labor intensity, susceptibility to subjective factors of inspectors, and poor consistency, making them unsuitable for the large-scale, high-standard production demands of modern industry. Machine vision technology, with its advantages of non-contact operation, high speed, high precision, and good repeatability, has become the preferred solution for industrial automated inspection. It acquires workpiece images using industrial cameras and processes and analyzes these images using computer algorithms to automatically identify and locate defects.
[0004] However, applying machine vision technology to the detection of burrs on circular curves faces many challenges:
[0005] Light sensitivity: The reflective properties of the workpiece surface and the non-uniformity of ambient light can seriously affect image quality, leading to inaccurate edge extraction.
[0006] Low contrast: Burrs are made of the same material as the workpiece body, and their grayscale contrast in the image is often very low, making them difficult to distinguish from background noise.
[0007] Morphological diversity: The shape, size, and direction of burrs are random, and there is no fixed template to match.
[0008] Subpixel-level accuracy requirement: High-precision workpieces typically allow burr sizes in the tens of micrometers range, requiring the detection algorithm to have subpixel or even higher edge positioning capabilities.
[0009] Complex background interference: The workpiece image may contain non-burr features such as scratches, oil stains, and textures that cause interference.
[0010] In conclusion, developing a robust, accurate, and fast automatic detection method for burrs on circular curves that can overcome the above challenges has significant industrial application value. Summary of the Invention
[0011] The purpose of this application is to provide a method, system, and medium for detecting burrs on circular curves based on multi-scale phase consistency and contour models. By using phase consistency as the core image feature, the detection algorithm has natural invariance to uneven illumination, brightness changes, and contrast differences. In scenarios where existing technologies fail due to poor image quality, this invention can still stably and completely extract contour edges, significantly reducing the false negative rate.
[0012] This application also provides a method for detecting burrs on circular curves based on multi-scale phase consistency and contour models, including:
[0013] The input image is acquired, a multi-scale Log-Gabor filter bank is constructed, and the input image is convolved to obtain the response vector;
[0014] The phase consistency measure of each pixel is calculated based on the response vector, and the initial edge point set is extracted by the sub-pixel localization method.
[0015] The initial circular contour is fitted based on the edge point set, and an active contour model is constructed. The initial circular contour is then subjected to energy minimization evolution based on the active contour model to fit the true contour of the burr.
[0016] A two-dimensional feature vector is constructed based on the curvature features of the contour points after the evolution of the true contour of the burr and the local average phase consistency energy features.
[0017] The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function, the burr region is marked, and the burr detection result is output.
[0018] Optionally, in the circular curve burr detection method based on multi-scale phase consistency and contour model described in the embodiments of this application, a multi-scale Log-Gabor filter bank is constructed, specifically including:
[0019] At multiple scales and direction Construct a Log-Gabor filter bank;
[0020] In the frequency domain The transfer function of the Log-Gabor filter is:
[0021]
[0022] in: It is the first Center frequency at each scale;
[0023] Indicates the bandwidth of the filter. It is a constant;
[0024] It is the first The center angle in each direction Control angle bandwidth;
[0025] Even-symmetric filters in the spatial domain are obtained based on inverse Fourier transform. And odd symmetric filters Filters based on even symmetry in the spatial domain And odd symmetric filters Construct a Log-Gabor filter bank.
[0026] Optionally, in the circular curve burr detection method based on multi-scale phase consistency and contour model described in the embodiments of this application, the input image is convolved to obtain a response vector, specifically including:
[0027] For the input image At each scale and direction The input image is convolved with the filter to obtain the response vector.
[0028]
[0029] Then point amplitude and phase for:
[0030]
[0031] Phase consistency measurement The calculation formula is:
[0032]
[0033] in:
[0034] It is the direction All scales The phase-weighted average;
[0035] The term calculates the weighted sum of local energy in all directions; the maximum value is taken when the phases of all scales are aligned.
[0036] It is a threshold used to suppress noise;
[0037] It is a constant, and Greater than zero;
[0038] The value is in the range [0, 1], and the larger the value, the more likely it is to be a point. The higher the probability that it is a feature point;
[0039] This represents an even-symmetric filter, also known as a cosine filter. It is a symmetric function in the spatial domain and is sensitive to changes in edge intensity (i.e., a smooth transition of gray values) in an image.
[0040] This represents an odd-symmetric filter, also known as a sine filter. It is an antisymmetric function in the spatial domain and is sensitive to changes in the direction of edges in an image (i.e., step changes in grayscale values).
[0041] The even-symmetric response represents the result of convolving the image with an even-symmetric filter at scale s and direction o. It reflects the cosine component of the image at that scale and direction, typically corresponding to the intensity variation of edges.
[0042] The odd-symmetric response represents the result of convolving the image with an odd-symmetric filter at scale s and direction o. It reflects the sinusoidal component of the image at that scale and direction, typically corresponding to the directional changes of the edges.
[0043] Optionally, in the circular curve burr detection method based on multi-scale phase consistency and contour model described in the embodiments of this application, the edge point extraction method is as follows:
[0044] based on Build Energy map, find local maxima along the row and column directions of the image respectively;
[0045] For candidate points Utilizing the neighborhood Values are used to locate sub-pixel coordinates through quadratic function fitting. ;
[0046] In the x-direction:
[0047] set up , , ,and The subpixel offset is then:
[0048]
[0049] Similarly, we can obtain Then the coordinates of the sub-pixel edge point are ;
[0050] An initial set of edge points is obtained based on the sub-pixel edge point coordinates. ;
[0051] Used to calculate sub-pixel offset in the x-direction This represents the PC value of the candidate point itself. Used to calculate sub-pixel offset in the x-direction Indicates the phase consistency response value. This represents the j-th sub-pixel edge point. This represents the sub-pixel coordinates of the j-th edge point.
[0052] Optionally, in the circular curve burr detection method based on multi-scale phase consistency and contour model described in the embodiments of this application, the method further includes feature extraction, specifically as follows: The method involves fusing and discriminating the two-dimensional feature vector based on a linear weighted scoring function.
[0053] Extract each point based on the true contour of the burr. Two key features:
[0054] Contour curvature Curvature is a geometric quantity that describes the degree of curvature of a curve. It is very sensitive to local bumps (burrs). Discrete curvature is estimated using adjacent points.
[0055]
[0056] Wherein, the first derivative and second derivative The formula for approximation using the central difference method is as follows:
[0057]
[0058] Similarly, burr regions will exhibit local maxima of curvature and are thus identified as burr regions;
[0059] Local average phase coherence energy :
[0060] It induces a strong phase-consistent response across multiple scales and directions, and calculations are performed using points... A small neighborhood centered on the outline, taking values to the left and right. Average within each point value:
[0061]
[0062] Will Areas with values greater than or equal to a set threshold are identified as burr areas.
[0063] Optionally, in the circular curve burr detection method based on multi-scale phase consistency and contour model described in this application embodiment, the two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function to mark the burr region and output the burr detection result, specifically including:
[0064] Constructing two-dimensional feature vectors Each contour point is described, and a linear weighted scoring function is used for fusion and discrimination. The formula for the linear weighted scoring function is as follows:
[0065]
[0066] max(k) represents the curvature value at the point on the entire circle where the curvature is the greatest;
[0067] min(k) represents the curvature value of the point with the minimum curvature on the entire circle;
[0068] in and It is a weighting coefficient, and The denominator is the normalized feature;
[0069] Set a comprehensive score threshold ;
[0070] like Then point Marked as a suspected defect point;
[0071] Clustering of consecutive or clustered suspicious points forms a complete defect region. ;
[0072] The average score and length of points within the defect area are calculated as a measure of the severity of the defect, thus obtaining the burr detection results.
[0073] Secondly, embodiments of this application provide a circular curve burr detection system based on multi-scale phase consistency and contour model. The system includes a memory and a processor. The memory includes a program for a circular curve burr detection method based on multi-scale phase consistency and contour model. When the program for the circular curve burr detection method based on multi-scale phase consistency and contour model is executed by the processor, it performs the following steps:
[0074] The input image is acquired, a multi-scale Log-Gabor filter bank is constructed, and the input image is convolved to obtain the response vector;
[0075] The phase consistency measure of each pixel is calculated based on the response vector, and the initial edge point set is extracted by the sub-pixel localization method.
[0076] The initial circular contour is fitted based on the edge point set, and an active contour model is constructed. The initial circular contour is then subjected to energy minimization evolution based on the active contour model to fit the true contour of the burr.
[0077] A two-dimensional feature vector is constructed based on the curvature features of the contour points after the evolution of the true contour of the burr and the local average phase consistency energy features.
[0078] The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function, the burr region is marked, and the burr detection result is output.
[0079] Optionally, in the circular curve burr detection system based on multi-scale phase consistency and contour model described in the embodiments of this application, a multi-scale Log-Gabor filter bank is constructed, specifically including:
[0080] At multiple scales and direction Construct a Log-Gabor filter bank;
[0081] In the frequency domain The transfer function of the Log-Gabor filter is:
[0082]
[0083] in: It is the first Center frequency at each scale;
[0084] Indicates the bandwidth of the filter. It is a constant;
[0085] It is the first The center angle in each direction Control angle bandwidth;
[0086] Even-symmetric filters in the spatial domain are obtained based on inverse Fourier transform. And odd symmetric filters Filters based on even symmetry in the spatial domain And odd symmetric filters Construct a Log-Gabor filter bank.
[0087] Optionally, in the circular curve burr detection system based on multi-scale phase consistency and contour model described in this application embodiment, convolution calculation is performed on the input image to obtain the response vector, specifically including:
[0088] For the input image At each scale and direction The input image is convolved with the filter to obtain the response vector.
[0089]
[0090] Then point amplitude and phase for:
[0091]
[0092] Phase consistency measurement The calculation formula is:
[0093]
[0094] in:
[0095] It is the direction All scales The phase-weighted average;
[0096] The term calculates the weighted sum of local energy in all directions; the maximum value is taken when the phases of all scales are aligned.
[0097] It is a threshold used to suppress noise;
[0098] It is a constant, and Greater than zero;
[0099] The value is in the range [0, 1], and the larger the value, the more likely it is to be a point. The higher the probability that it is a feature point.
[0100] Thirdly, embodiments of this application also provide a computer-readable storage medium, which includes a circular curve burr detection method program based on a multi-scale phase consistency and contour model. When the circular curve burr detection method program based on a multi-scale phase consistency and contour model is executed by a processor, it implements the steps of the circular curve burr detection method based on a multi-scale phase consistency and contour model as described in any of the above claims.
[0101] As can be seen from the above, the circular curve burr detection method, system, and medium provided in this application embodiment based on multi-scale phase consistency and contour model acquires an input image, constructs a multi-scale Log-Gabor filter bank, performs convolution calculation on the input image to obtain a response vector; calculates the phase consistency metric of each pixel based on the response vector, and extracts an initial edge point set through a sub-pixel localization method; fits an initial circular contour based on the edge point set to construct an active contour model; performs energy minimization evolution on the initial circular contour based on the active contour model to fit the true contour of the burr; extracts the curvature features and local average phase consistency energy features of the evolved contour points based on the true contour of the burr to construct a two-dimensional feature vector; fuses and discriminates the two-dimensional feature vector based on a linear weighted scoring function, marks the burr region, and outputs the burr detection result; by using phase consistency as the core image feature, the detection algorithm has natural invariance to uneven illumination, brightness changes, and contrast differences. In scenarios where existing technologies fail due to poor image quality, this invention can still stably and completely extract the contour edge, significantly reducing the false negative rate. Attached Figure Description
[0102] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0103] Figure 1 A flowchart of a circular curve burr detection method based on multi-scale phase consistency and contour model provided in an embodiment of this application;
[0104] Figure 2 This is a schematic diagram of the dynamic contour evolution process of the circular curve burr detection method based on multi-scale phase consistency and contour model provided in the embodiments of this application, wherein (a) the initial circle (red) is covered on the phase consistency energy map (background); (b) the intermediate evolution state; and (c) the final convergence state (blue), which accurately fits the real contour containing burrs.
[0105] Figure 3 This is a schematic diagram illustrating the feature fusion and discrimination of a circular curve burr detection method based on multi-scale phase consistency and contour model provided in the embodiments of this application. Detailed Implementation
[0106] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0107] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0108] Please refer to Figures 1-3 As shown, the circular curve burr detection method based on multi-scale phase consistency and contour model is used in terminal devices. This method includes the following steps:
[0109] S101: Obtain the input image, construct a multi-scale Log-Gabor filter bank, perform convolution calculation on the input image, and obtain the response vector;
[0110] S102, calculate the phase consistency measure of each pixel based on the response vector, and extract the initial edge point set through the sub-pixel localization method;
[0111] S103: Fit the initial circular contour based on the edge point set, construct the active contour model, and perform energy minimization evolution on the initial circular contour based on the active contour model to fit the true contour of the burr.
[0112] S104, based on the curvature features of the contour points after the evolution of the true contour of the burr, and the local average phase consistency energy features, a two-dimensional feature vector is constructed;
[0113] S105 performs fusion and discrimination on two-dimensional feature vectors based on a linear weighted scoring function, marks burr regions, and outputs burr detection results.
[0114] It should be noted that, firstly, the phase consistency feature, which is insensitive to illumination and contrast, is used to robustly extract edge points with sub-pixel precision; then, a dynamic active contour model is constructed, which adaptively converges to the true contour of the workpiece (regardless of whether it contains burrs) under the combined action of internal constraints and external image forces; finally, by analyzing the curvature change and local phase consistency energy anomalies of the converged contour model, burr defects are accurately located and identified.
[0115] In addition to the classic parametric active contour model (Snake model), the geometric active contour model (level set method) can also be used. The level set method represents the contour by evolving a zero level set of a high-dimensional function, which can naturally handle changes in topology. Although it is more computationally intensive, the level set method is more robust for extremely irregular shapes or cases that may have fracture edges. Its external energy field can also be driven by a phase consistency map.
[0116] According to an embodiment of the present invention, constructing a multi-scale Log-Gabor filter bank specifically includes:
[0117] At multiple scales and direction Construct a Log-Gabor filter bank;
[0118] In the frequency domain The transfer function of the Log-Gabor filter is:
[0119]
[0120] in: It is the first Center frequency at each scale;
[0121] Indicates the bandwidth of the filter. It is a constant;
[0122] It is the first The center angle in each direction Control angle bandwidth;
[0123] Even-symmetric filters in the spatial domain are obtained based on inverse Fourier transform. And odd symmetric filters Filters based on even symmetry in the spatial domain And odd symmetric filters Construct a Log-Gabor filter bank.
[0124] According to an embodiment of the present invention, convolution calculation is performed on the input image to obtain a response vector, specifically including:
[0125] For the input image At each scale and direction The input image is convolved with the filter to obtain the response vector.
[0126]
[0127] Then point amplitude and phase for:
[0128]
[0129] Phase consistency measurement The calculation formula is:
[0130]
[0131] in:
[0132] It is the direction All scales The phase-weighted average;
[0133] The term calculates the weighted sum of local energy in all directions; the maximum value is taken when the phases of all scales are aligned.
[0134] It is a threshold used to suppress noise;
[0135] It is a constant, and Greater than zero;
[0136] The value is in the range [0, 1], and the larger the value, the more likely it is to be a point. The higher the probability that it is a feature point;
[0137] This represents an even-symmetric filter, also known as a cosine filter. It is a symmetric function in the spatial domain and is sensitive to changes in edge intensity (i.e., a smooth transition of gray values) in an image.
[0138] This represents an odd-symmetric filter, also known as a sine filter. It is an antisymmetric function in the spatial domain and is sensitive to changes in the direction of edges in an image (i.e., step changes in grayscale values).
[0139] The even-symmetric response represents the result of convolving the image with an even-symmetric filter at scale s and direction o. It reflects the cosine component of the image at that scale and direction, typically corresponding to the intensity variation of edges.
[0140] The odd-symmetric response represents the result of convolving the image with an odd-symmetric filter at scale s and direction o. It reflects the sinusoidal component of the image at that scale and direction, typically corresponding to the directional changes of the edges.
[0141] According to an embodiment of the present invention, the edge point extraction method is as follows:
[0142] based on Build Energy map, find local maxima along the row and column directions of the image respectively;
[0143] For candidate points Utilizing the neighborhood Values are used to locate sub-pixel coordinates through quadratic function fitting. ;
[0144] In the x-direction:
[0145] set up , , ,and The subpixel offset is then:
[0146]
[0147] Similarly, we can obtain Then the coordinates of the sub-pixel edge point are ;
[0148] An initial set of edge points is obtained based on the sub-pixel edge point coordinates. ;
[0149] Used to calculate sub-pixel offset in the x-direction This represents the PC value of the candidate point itself. Used to calculate sub-pixel offset in the x-direction Indicates the phase consistency response value. This represents the j-th sub-pixel edge point. This represents the sub-pixel coordinates of the j-th edge point.
[0150] According to an embodiment of the present invention, the fusion and discrimination of two-dimensional feature vectors based on a linear weighted scoring function further includes feature extraction, as detailed below:
[0151] Extract each point based on the true contour of the burr. Two key features:
[0152] Contour curvature Curvature is a geometric quantity that describes the degree of curvature of a curve. It is very sensitive to local bumps (burrs). Discrete curvature is estimated using adjacent points.
[0153]
[0154] Wherein, the first derivative and second derivative The formula for approximation using the central difference method is as follows:
[0155] ;
[0156] In the formula, This represents the first derivative in the x-direction. Indicates the first on the outline The x-coordinate of the point following this point. Indicates the first on the outline The x-coordinate of the point preceding the point. This represents the second derivative in the x-direction;
[0157] Similarly, burr regions will exhibit local maxima of curvature and are thus identified as burr regions;
[0158] Local average phase coherence energy :
[0159] It induces a strong phase-consistent response across multiple scales and directions, and calculations are performed using points... A small neighborhood centered on the outline, taking values to the left and right. Average within each point value:
[0160]
[0161] Will Areas with values greater than or equal to a set threshold are identified as burr areas.
[0162] It should be noted that phase consistency analysis can use texture features based on Local Binary Pattern (LBP) to construct an energy map. By calculating the LBP value of each point in the image and constructing an LBP variance map or gradient map that highlights edges, it serves as the external energy of the active contour. Although LBP is sensitive to texture and may introduce more noise, it may be more computationally efficient and effective in specific textured contexts.
[0163] According to an embodiment of the present invention, a two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function, spur regions are marked, and spur detection results are output, specifically including:
[0164] Constructing two-dimensional feature vectors Each contour point is described, and a linear weighted scoring function is used for fusion and discrimination. The formula for the linear weighted scoring function is as follows:
[0165]
[0166] in and It is a weighting coefficient, and The denominator is the normalized feature;
[0167] Set a comprehensive score threshold ;
[0168] like Then point Marked as a suspected defect point;
[0169] Clustering of consecutive or clustered suspicious points forms a complete defect region. ;
[0170] The average score and length of points within the defect area are calculated as a measure of the severity of the defect, thus obtaining the burr detection results.
[0171] It should be noted that, in addition to using a linearly weighted scoring function, more advanced machine learning classifiers can be employed. For example, the feature vector of each contour point can be... The data is input into a pre-trained classifier (such as a support vector machine (SVM), decision tree, or simple neural network), which then determines whether the point belongs to a spur. This method requires training with a large number of samples but may achieve better discrimination performance than the fixed threshold method, especially when the spur features are complex and diverse.
[0172] According to embodiments of the present invention, the method further includes employing an improved Snake model (active contour model), specifically including:
[0173] Model energy function definition:
[0174] The activity profile is a parametric curve , Its total energy From internal energy and external energy composition:
[0175]
[0176] The goal of model evolution is to find that makes Minimized curve .
[0177] Internal energy : Controls the smoothness and continuity of the contour.
[0178]
[0179] The first term is elastic energy, which is the stretching of the penalty curve, determined by the weights. Control. The second term is bending energy, a penalty for bending the curve, determined by weights. Control. To allow the profile to bend sharply at the burrs, It becomes an adaptive parameter that depends on local image features. In phase-consistency energy... Reduce the size of high areas (which may be burrs). This allows for greater curvature; in flat regions, it increases... Keep the outline smooth.
[0180]
[0181] in It is the basic bending weight. It is a parameter that controls sensitivity.
[0182] external energy : Guides the contour to move towards image features. Uses the calculated phase-consistency energy map The negative value is used as the external energy field:
[0183]
[0184] in These are weighting coefficients. This way, the contour will be attracted. The edge position with high value.
[0185] Model initialization and evolution:
[0186] The obtained edge point set An initial circle is fitted using the least squares method. Discretize this initial circle as Control points This serves as the initial state of the activity outline.
[0187] The energy minimization problem is solved by addressing the Euler-Lagrange equations, typically using the gradient descent method (or force balance method). The discretized evolution equations are as follows:
[0188]
[0189] In the formula, This represents the current coordinates of the i-th control point of the active contour. and express Direct neighbors on the contour sequence and express Indirect neighbor points on the contour sequence; Represents the elastic (stretch) weight at point i; This represents the curvature weight at point i;
[0190] in This refers to the number of iteration steps. The position of each control point is updated iteratively until the profile converges (i.e., the average movement distance of the control points is less than a threshold). (This process continues until the maximum number of iterations is reached.) Finally, the converged set of contour points is obtained. , This represents the final coordinates of the i-th control point of the contour after evolution convergence. It accurately depicts the actual contour of the workpiece, and the burr area is represented by a local protrusion on the contour.
[0191] In summary, the present invention has the following beneficial effects:
[0192] This invention achieves a perfect balance between high positioning accuracy and strong noise resistance: the phase consistency feature itself is insensitive to noise because it requires phase alignment at multiple scales, a condition that random noise cannot meet. Simultaneously, sub-pixel positioning is performed directly on the phase consistency energy map, avoiding edge shifting caused by Gaussian blurring in traditional methods. Therefore, this invention can achieve sub-pixel-level edge positioning accuracy while suppressing noise, making it possible to detect minute glitch patterns.
[0193] This invention fundamentally eliminates the "contamination" effect of model fitting: it innovatively uses a dynamic active contour model. This model does not force the contour to converge to a preset geometry (such as a circle), but rather adaptively fits the real contour, which may contain defects, under the balance of internal constraints and external image forces. Since no "fitting" is performed, there is no problem of baseline model deviation caused by burr points participating in the calculation. This allows the method to accurately depict the true shape of burrs while completely avoiding the phenomenon of false alarms on the opposite side found in existing technologies.
[0194] Multi-feature fusion discrimination significantly improves recognition accuracy and specificity: This invention abandons the simplistic approach of relying solely on geometric deviations, creatively combining contour curvature (geometric feature) and local phase consistency energy (image feature). This fusion strategy greatly enhances the algorithm's discriminative ability: it can effectively distinguish genuine burrs (which simultaneously cause geometric protrusions and abrupt changes in image features) from simple noise, scratches (which may only have high PC energy but gentle curvature), or poor fitting (which may only have curvature peaks but low PC energy). This results in an extremely low false detection rate, making the detection results more reliable.
[0195] High degree of automation and intelligence: From phase consistency detection to active contour evolution, and then to feature fusion and discrimination, the entire process parameters are fixed, exhibiting strong adaptability and reducing reliance on operator experience. This is particularly evident in the adaptive bending weighting. The design enables the model to intelligently balance smooth contours and capture details, demonstrating the advanced intelligence of the algorithm.
[0196] Secondly, embodiments of this application provide a circular curve burr detection system based on multi-scale phase consistency and contour model. The system includes a memory and a processor. The memory includes a program for a circular curve burr detection method based on multi-scale phase consistency and contour model. When the program for the circular curve burr detection method based on multi-scale phase consistency and contour model is executed by the processor, it performs the following steps:
[0197] The input image is acquired, a multi-scale Log-Gabor filter bank is constructed, and the input image is convolved to obtain the response vector;
[0198] The phase consistency measure of each pixel is calculated based on the response vector, and the initial edge point set is extracted by the sub-pixel localization method.
[0199] An initial circular contour is fitted based on the edge point set, an active contour model is constructed, and the initial circular contour is evolved by minimizing energy based on the active contour model to fit the true contour of the burr.
[0200] A two-dimensional feature vector is constructed based on the curvature features of the contour points after the evolution of the true contour of the burr and the local average phase consistency energy features.
[0201] The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function, the spur region is marked, and the spur detection result is output.
[0202] According to an embodiment of the present invention, constructing a multi-scale Log-Gabor filter bank specifically includes:
[0203] At multiple scales and direction Construct a Log-Gabor filter bank;
[0204] In the frequency domain The transfer function of the Log-Gabor filter is:
[0205]
[0206] in: It is the first Center frequency at each scale;
[0207] Indicates the bandwidth of the filter. It is a constant;
[0208] It is the first The center angle in each direction Control angle bandwidth;
[0209] Even-symmetric filters in the spatial domain are obtained based on inverse Fourier transform. And odd symmetric filters Filters based on even symmetry in the spatial domain And odd symmetric filters Construct a Log-Gabor filter bank.
[0210] According to an embodiment of the present invention, convolution calculation is performed on the input image to obtain a response vector, specifically including:
[0211] For the input image At each scale and direction The input image is convolved with the filter to obtain the response vector.
[0212]
[0213] Then point amplitude and phase for:
[0214]
[0215] Phase consistency measurement The calculation formula is:
[0216]
[0217] in:
[0218] It is the direction All scales The phase-weighted average;
[0219] The term calculates the weighted sum of local energy in all directions; the maximum value is taken when the phases of all scales are aligned.
[0220] It is a threshold used to suppress noise;
[0221] It is a constant, and Greater than zero;
[0222] The value is in the range [0, 1], and the larger the value, the more likely it is to be a point. The higher the probability that it is a feature point.
[0223] A third aspect of the present invention provides a computer-readable storage medium including a circular curve burr detection method program based on a multi-scale phase consistency and contour model. When the circular curve burr detection method program based on a multi-scale phase consistency and contour model is executed by a processor, it implements the steps of the circular curve burr detection method based on a multi-scale phase consistency and contour model as described above.
[0224] This invention discloses a method, system, and medium for detecting burrs on circular curves based on multi-scale phase consistency and contour models. The method involves acquiring an input image, constructing a multi-scale Log-Gabor filter bank, and performing convolution calculations on the input image to obtain a response vector. Based on the response vector, a phase consistency metric is calculated for each pixel, and an initial set of edge points is extracted using a sub-pixel localization method. An initial circular contour is fitted based on the edge point set to construct an active contour model. The initial circular contour is then subjected to energy minimization evolution based on the active contour model to fit the true contour of the burr. The curvature features and local average phase consistency energy features of the evolved contour points are extracted based on the true contour of the burr to construct a two-dimensional feature vector. The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function to mark the burr region and output the burr detection result. By using phase consistency as the core image feature, the detection algorithm possesses inherent invariance to uneven illumination, brightness variations, and contrast differences. Even in scenarios where existing technologies fail due to poor image quality, this invention can still stably and completely extract the contour edges, significantly reducing the false negative rate.
[0225] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.
[0226] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.
[0227] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.
[0228] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0229] Alternatively, if the integrated units of the present invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.
Claims
1. A method for detecting burrs on circular curves based on multi-scale phase consistency and contour models, characterized in that, include: The input image is acquired, a multi-scale Log-Gabor filter bank is constructed, and the input image is convolved to obtain the response vector; The phase consistency measure of each pixel is calculated based on the response vector, and the initial edge point set is extracted by the sub-pixel localization method. The initial circular contour is fitted based on the edge point set, and an active contour model is constructed. The initial circular contour is then subjected to energy minimization evolution based on the active contour model to fit the true contour of the burr. A two-dimensional feature vector is constructed based on the curvature features of the contour points after the evolution of the true contour of the burr and the local average phase consistency energy features. The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function, the burr region is marked, and the burr detection result is output.
2. The circular curve burr detection method based on multi-scale phase consistency and contour model according to claim 1, characterized in that, Constructing a multi-scale Log-Gabor filter bank specifically includes: At multiple scales and direction Construct a Log-Gabor filter bank; In the frequency domain The transfer function of the Log-Gabor filter is: ； in: It is the first Center frequency at each scale; Indicates the bandwidth of the filter. It is a constant; It is the first The center angle in each direction Control angle bandwidth; Even-symmetric filters in the spatial domain are obtained based on inverse Fourier transform. And odd symmetric filters Filters based on even symmetry in the spatial domain And odd symmetric filters Construct a Log-Gabor filter bank.
3. The circular curve burr detection method based on multi-scale phase consistency and contour model according to claim 2, characterized in that, The input image is convolved to obtain the response vector, which specifically includes: For the input image At each scale and direction The input image is convolved with the filter to obtain the response vector. ； Then point amplitude and phase for: ； Phase consistency measurement The calculation formula is: ； in: It is the direction All scales The phase-weighted average; The term calculates the weighted sum of local energy in all directions; the maximum value is taken when the phases of all scales are aligned. It is a threshold used to suppress noise; It is a constant, and Greater than zero; The value is in the range [0, 1]. Indicates an even-symmetric response; This indicates an odd-symmetric response.
4. The circular curve burr detection method based on multi-scale phase consistency and contour model according to claim 3, characterized in that, The edge point extraction method is as follows: based on Build Energy map, find local maxima along the row and column directions of the image respectively; For candidate points Utilizing the neighborhood Values are used to locate sub-pixel coordinates through quadratic function fitting. ; The sub-pixel offsets in the x and y directions are calculated based on the sub-pixel coordinates to obtain the sub-pixel edge point coordinates; An initial set of edge points is obtained based on the sub-pixel edge point coordinates. ; This represents the j-th sub-pixel edge point. This represents the sub-pixel coordinates of the j-th edge point.
5. The circular curve burr detection method based on multi-scale phase consistency and contour model according to claim 4, characterized in that, The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function, and feature extraction is also included, as follows: Extract each point based on the true contour of the burr. Two key features: Based on contour curvature Geometric quantities describing the curvature of a curve; for local bulges and burrs, the discrete curvature is estimated using adjacent points: ； In the formula, This represents the first derivative in the x-direction. This represents the second derivative in the x-direction; This represents the first derivative in the y-direction. This represents the second derivative in the y-direction; A region exhibiting a local maximum of curvature is identified as a burr region. Local average phase coherence energy : Calculation by point Take the neighborhood of the center, and take the left and right sides along the outline. Average within each point value: ； Will Areas with values greater than or equal to a set threshold are identified as burr areas.
6. The circular curve burr detection method based on multi-scale phase consistency and contour model according to claim 5, characterized in that, The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function to mark spur regions and output spur detection results, specifically including: Constructing two-dimensional feature vectors Each contour point is described, and a linear weighted scoring function is used for fusion and discrimination. The formula for the linear weighted scoring function is as follows: ； Where max(k) represents the curvature value of the point with the largest curvature on the entire circle; min(k) represents the curvature value of the point with the smallest curvature on the entire circle; and It is a weighting coefficient, and The denominator is the normalized feature; Set a comprehensive score threshold ; like Then point Marked as a suspected defect point; Clustering of consecutive or clustered suspicious points forms a complete defect region. ; The average score and length of points within the defect area are calculated as a measure of the severity of the defect, thus obtaining the burr detection results.
7. A circular curve burr detection system based on multi-scale phase consistency and contour model, characterized in that, The system includes a memory and a processor. The memory contains a program for a circular curve burr detection method based on multi-scale phase consistency and contour model. When the program for the circular curve burr detection method based on multi-scale phase consistency and contour model is executed by the processor, it performs the following steps: The input image is acquired, a multi-scale Log-Gabor filter bank is constructed, and the input image is convolved to obtain the response vector; The phase consistency measure of each pixel is calculated based on the response vector, and the initial edge point set is extracted by the sub-pixel localization method. The initial circular contour is fitted based on the edge point set, and an active contour model is constructed. The initial circular contour is then subjected to energy minimization evolution based on the active contour model to fit the true contour of the burr. A two-dimensional feature vector is constructed based on the curvature features of the contour points after the evolution of the true contour of the burr and the local average phase consistency energy features. The two-dimensional feature vector is fused and discriminated based on a linear weighted scoring function, the burr region is marked, and the burr detection result is output.
8. The circular curve burr detection system based on multi-scale phase consistency and contour model according to claim 7, characterized in that, Constructing a multi-scale Log-Gabor filter bank specifically includes: At multiple scales and direction Construct a Log-Gabor filter bank; In the frequency domain The transfer function of the Log-Gabor filter is: ； in: It is the first Center frequency at each scale; Indicates the bandwidth of the filter. It is a constant; It is the first The center angle in each direction Control angle bandwidth; Even-symmetric filters in the spatial domain are obtained based on inverse Fourier transform. And odd symmetric filters Filters based on even symmetry in the spatial domain And odd symmetric filters Construct a Log-Gabor filter bank.
9. The circular curve burr detection system based on multi-scale phase consistency and contour model according to claim 8, characterized in that, The input image is convolved to obtain the response vector, which specifically includes: For the input image At each scale and direction The input image is convolved with the filter to obtain the response vector. ； Then point amplitude and phase for: ； Phase consistency measurement The calculation formula is: ； in: It is the direction All scales The phase-weighted average; The term calculates the weighted sum of local energy in all directions; the maximum value is taken when the phases of all scales are aligned. It is a threshold used to suppress noise; It is a constant, and Greater than zero; The value is in the range [0, 1].
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a circular curve burr detection method program based on a multi-scale phase consistency and contour model. When the circular curve burr detection method program based on a multi-scale phase consistency and contour model is executed by a processor, it implements the steps of the circular curve burr detection method based on a multi-scale phase consistency and contour model as described in any one of claims 1 to 6.