Two-for-one twister yarn quality online detection system and method based on machine vision

Abstract

The invention discloses a two-for-one twister yarn quality online detection system and method based on machine vision, and relates to the technical field of textile monitoring machinery, and the system comprises a light field acquisition unit, a geometric reconstruction unit and a vector filtering unit. The light field acquisition unit captures a trapezoidal image sequence covering a yarn ballooning motion range by using a non-parallel imaging light field meeting the Gremer's law; the geometric reconstruction unit maps the trapezoidal image sequence into an orthogonal image sequence with a unified spatial resolution coefficient through an inverse homography matrix; the vector filtering unit calculates the probability of flaws based on the motion vector field using a multiplicative probability model including a reference velocity and a directional deviation angle. The problems of imaging defocus and environment flying interference caused by balloon swing in the production of high count yarns of the two-for-one twister are solved, and online closed-loop monitoring of the yarn quality is realized.

Description

Technical Field

[0001] This invention relates to the field of textile monitoring machinery technology, and in particular to an online detection system and method for yarn quality of a doubling machine based on machine vision. Background Technology

[0002] During the production of high-count yarn (120 count and above) using a doubling machine, the spiral air ring formed by the yarn under the high-speed rotation of the spindle is in a highly unstable dynamic state. This instability manifests as high-frequency radial jump and Z-axis defocusing oscillation in the spatial dimension of the yarn, while in the temporal dimension, it is accompanied by a large amount of random drifting environmental fly waste interference.

[0003] Currently, machine vision-based inspection methods face a technical challenge: feature vanishing under spatiotemporal coupling interference. Because the radial oscillation amplitude of the yarn's air loop far exceeds the depth of field of conventional imaging systems, the original signal captured by the image sensor suffers from grayscale gradient vanishing due to physical defocusing. This causes the edge features originally used to distinguish defects to become diffused in the spatial dimension. In this context, randomly drifting fly yarn signals and defect signals moving at high speed with the yarn create semantic aliasing in the visual space. Traditional detection logic, lacking stable physical gradient support and a unified dynamic measurement benchmark, cannot extract statistically significant defect features from spatiotemporally intertwined noise, leading to detection blind spots and frequent false alarms in high-count yarn production environments. Summary of the Invention

[0004] This invention provides an online yarn quality detection system and method based on machine vision, which solves the problem of difficulty in capturing target features in the detection of high-count yarns in double twisting machines due to the non-intermittent movement of the air ring and interference from environmental fly waste.

[0005] In view of the above problems, the present invention provides an online yarn quality inspection system based on machine vision, comprising a light field acquisition unit, a geometric reconstruction unit, and a vector filtering unit connected sequentially along the data processing link: The light field acquisition unit includes an image sensor and an optical lens. An angle is set between the main plane of the optical lens and the photosensitive plane of the image sensor. The main plane, the photosensitive plane and the tangent of the movement trajectory of the yarn air ring to be measured intersect on the same straight line in three-dimensional space, which is used to output the first image sequence. The geometric reconstruction unit stores an inverse homography matrix and is configured to perform coordinate space transformation on the first image sequence using the inverse homography matrix to generate a second image sequence with spatial resolution coefficients. The vector filtering unit is configured to calculate the motion vector field of the second image sequence under a preset coordinate axis, and calculate the defect probability based on the velocity magnitude, directional deviation angle and neighborhood consistency density of each pixel in the motion vector field.

[0006] Preferably, the light field acquisition unit further includes a signal interface connected to the rotating component of the twisting machine; the signal interface is configured to trigger the image sensor to perform exposure when acquiring the rotation phase signal; wherein the included angle ranges from 8 degrees to 12 degrees.

[0007] Preferably, the geometric reconstruction unit is configured to perform perspective transformation logic on the first image sequence so that the yarn axis in the second image sequence is parallel to the preset coordinate axis in the image coordinate system.

[0008] Preferably, the vector filtering unit is configured to calculate the defect probability using the following formula. : in, The defect probability is mentioned above. The velocity modulus; For reference speed; It is a non-zero constant; The deviation angle of the direction; Angle threshold; The first coefficient; The neighborhood consistency density; This is the second coefficient.

[0009] Preferably, the reference speed is determined based on the product of the real-time rotational speed of the twisting machine, the diameter of the air ring, the image acquisition frame rate, and the spatial resolution coefficient; the reference speed is directly proportional to the real-time rotational speed of the twisting machine and the diameter of the air ring, and inversely proportional to the image acquisition frame rate and the spatial resolution coefficient; the order of magnitude of the non-zero constant is... .

[0010] Preferably, the calculation logic for the defect probability satisfies the following limitations: If the directional deviation angle is greater than the angle threshold, the defect probability tends to zero. If the neighborhood consistency density decreases, the defect probability decreases.

[0011] The machine vision-based online yarn quality inspection method for doubling machines, executed through the aforementioned system, includes the following steps: Step 1: Acquire the first image sequence using an optical imaging system that satisfies the collinearity constraint on three planes; Step 2: Reconstruct the first image sequence into a second image sequence using inverse transformation logic, so that the second image sequence has a uniform pixel physical scale transformation relationship across the entire field of view; Step 3: Extract the instantaneous motion vector features from the second image sequence, and obtain the orientation deviation angle and neighborhood consistency density of each pixel under the preset coordinate axis; Step 4: Substitute the instantaneous motion vector features into a multiplicative decision model that includes a reference velocity and a non-zero constant, and output the defect probability distribution matrix.

[0012] Furthermore, the method is applied to production conditions where the yarn count is greater than 120; in step four, if a maximum point is detected in the defect probability distribution matrix, a quality anomaly signal is output.

[0013] The technical solution provided in this application has at least the following technical effects: This invention constructs an accurate mapping from non-accidental physical motion to deterministic probabilistic discriminant solutions, realizes sub-pixel-level logical segmentation of physical defects and environmental interference, and finally outputs a defect probability matrix composed of time tags, phase tags and defect probability values, providing a deterministic physical meaning for the quality status judgment result representation of the twisting production process.

[0014] By using a theoretical layout in the light field acquisition unit that satisfies the collinearity constraint of three planes, the focal plane of the system is rotated to coincide with the tangent plane of the linear gas circle's motion. Using Sham adjustment, the inference jump range of the gas circle is covered without sacrificing the amount of light received, thus repairing the boundary that disappeared due to the jump from the physical source and providing a stable kinematic decoupling premise for the subsequent vector filtering unit. In conjunction with the inverse homography matrix transformation of the geometric reconstruction unit, the trapezoidal distortion image is restored to an orthogonal Euclidean space with a linear proportional relationship, eliminating the geometric consistency error introduced by attention error. Finally, using a probabilistic integral model that includes the multiplication of reference velocity and direction departure angle, the truncation characteristics of nonlinear gating are used to forcibly separate the disordered fly flower signal and the defect body signal that is combined with controlled motion in terms of shape and morphology. Attached Figure Description

[0015] Figure 1 This is a structural diagram of the online yarn quality detection system based on machine vision provided in an embodiment of the present invention; Figure 2 The flowchart illustrates the online yarn quality detection system based on machine vision provided in this embodiment of the invention. Detailed Implementation

[0016] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0017] Example 1: This example discloses an online yarn quality inspection system for a doubling machine based on machine vision. See [link to documentation]. Figure 1 The system includes a light field acquisition unit, a geometric reconstruction unit, and a vector filtering unit connected sequentially along the data processing sequence.

[0018] The light field acquisition unit includes an image sensor and a lens. The principal plane of the optical lens and the sensor's sensing plane are positioned at an angle, and the principal plane, the sensing plane, and the tangent plane to the trajectory of the measuring line's atmosphere intersect on the same straight line in three-dimensional space, used to output the first image sequence. The geometric reconstruction unit stores an inverse homography matrix and is configured to use this matrix to perform coordinate space transformation on the first image sequence, generating a second image sequence with spatial resolution coefficients. The vector filtering unit is configured to calculate the motion vector field of the second image sequence under preset coordinate axes and, based on the magnitude, orientation removal angle, and neighborhood consistency density of the pixels in the motion vector field, calculate the defect probability.

[0019] The system's operational logic is executed using a machine vision-based online quality detection method for the doubling machine sequence. See also... Figure 2 The method specifically includes: acquiring a first sequence using an optical image inference system that satisfies the collinearity constraint of three planes; reconstructing the first image sequence into a second image sequence using inverse transformation logic to obtain a unified pixel physical pixel conversion relationship in the entire field of view of the second image sequence; then extracting instantaneous motion features from the second image sequence to obtain the orientation column angle and neighborhood consistency density of each pixel under the reference coordinate axis; and substituting the instantaneous motion features into a multiplicative probability decision model that includes a reference velocity and a non-zero constant to output a defect probability distribution matrix.

[0020] The operation process of the online yarn quality detection system for doubling machines begins with the initialization and construction of the physical imaging environment by the light field acquisition unit.

[0021] The physical connection between the signal interface and the rotating component of the twisting machine forms the hardware basis for phase locking. The photoelectric encoder is fixed to the drive shaft of the rotating component and rotates synchronously with it. The electrical pulse signals generated during rotation are transmitted in real-time via the signal interface to the logic control module of the optical field acquisition unit. The logic control module performs counting processing on the received electrical pulse signals to map the rotational phase of the yarn's air pocket in three-dimensional space. When the count value reaches a preset phase threshold, a trigger command is synchronously sent to the stroboscopic illumination module and the image sensor. The stroboscopic illumination module generates pulsed illumination upon triggering, and the image sensor performs charge accumulation during the illumination cycle. Because the trigger delay is limited to within 10 microseconds, the positional offset of the yarn's air pocket in the first image sequence is suppressed to the sub-pixel level. The phase-locked trigger mechanism eliminates the influence of rotational speed fluctuations on the imaging position, ensuring that the output first image sequence has stable temporal phase characteristics.

[0022] The parameters of the tilting optical module are set by rigidly defining the relative positions of the imaging lens, image sensor, and the object under test. The principal plane of the imaging lens and the photosensitive plane of the image sensor are set at an angle of 8 to 12 degrees. Under three-dimensional spatial geometric constraints, the principal plane of the imaging lens, the photosensitive plane of the image sensor, and the tangent plane of the trajectory of the yarn's vapor ring under test intersect on the same straight line. This three-plane collinear spatial layout alters the focal plane orientation of the light field acquisition unit, causing the focal plane to coincide with the tangent plane of the yarn's vapor ring's trajectory. This physical constraint ensures that the effective depth-of-field envelope can completely cover the physical amplitude of the yarn's vapor ring during radial movement. When the yarn's vapor ring under test experiences a radial movement of ±5 mm, the scattered light signal can still converge on the photosensitive plane of the image sensor. This physical layout, which does not rely on a mechanical focusing module, ensures the acquisition of original optical features with steep edge gradients across the entire field of view.

[0023] The generation of the first image sequence originates from the digital conversion of optical signals by the image sensor. The reflected light signal, converged by the imaging lens, is received by the photosensitive pixels of the image sensor, generating a digital current signal proportional to the light intensity. The first image sequence output by the image sensor has a trapezoidal perspective shape caused by the tilted optical axis. The underlying pixel data of the first image sequence retains high-frequency edge gradients, and the grayscale gradient values ​​of the edge pixels are maintained within a preset grayscale range. This first image sequence, possessing contrast characteristics, is written in real-time to the buffer unit of the online yarn quality detection system for the twisting machine. As the underlying input for the subsequent geometric reconstruction unit to perform coordinate space transformation, the spatial gradient information contained in the first image sequence provides data support for the kinematic decoupling of the subsequent vector filtering unit.

[0024] The processing flow of the online yarn quality detection system for doubling machines shifts from the physical perception stage to the geometric regularization stage. The geometric reconstruction unit transforms the first image sequence with perspective distortion into a second image sequence with linear metric characteristics by establishing mathematical mapping relationships.

[0025] The implementation of the coordinate space transformation logic relies on a pre-determined inverse homography matrix. During the system deployment initialization phase, a standard checkerboard calibration board is placed on the tangential plane of the trajectory of the yarn's airfoil. After the light field acquisition unit captures an image of the calibration board, it extracts the pixel coordinates of the calibration board's feature points in the first image sequence and combines this with the actual coordinates of the feature points in physical space to calculate the homography matrix used to characterize perspective relationships. Subsequently, the homography matrix was examined. Perform algebraic inversion to obtain the inverse homography matrix. Inverse homography matrix It contains the geometric transformation parameters required to restore a trapezoidal field of view to an orthogonal field of view. Inverse homography matrix. It is embedded in the hardware storage module of the geometric reconstruction unit as a mathematical benchmark for real-time image reconstruction.

[0026] The real-time pixel redrawing process is executed by the field-programmable gate array (FPGA) module within the geometric reconstruction unit. The FPGA module receives the first image sequence and performs matrix mapping operations on each pixel in the sequence. The operation follows the mathematical expression below: In the above expression, Represents the original pixel coordinates in the first image sequence. These represent the reconstructed coordinates after the transformation. Because the reconstructed coordinates... The calculation result is usually not an integer. The field-programmable gate array (FPGA) module synchronously calls the bilinear interpolation algorithm for image reconstruction. The bilinear interpolation algorithm obtains the grayscale data of the four pixels adjacent to the reconstructed coordinates in the first image sequence and performs bidirectional linear weighting according to the spatial distance to determine the grayscale distribution of the corresponding pixels in the second image sequence. Through this coupled operation of coordinate mapping and grayscale compensation, the trapezoidal distortion in the first image sequence is undone, generating a second image sequence that is geometrically orthogonal.

[0027] The technical result of the geometric reconstruction process is a second image sequence with physical consistency. This second image sequence eliminates the non-uniform scaling effect introduced by tilted imaging, ensuring a linear relationship between pixel displacement and physical motion distance across the entire field of view. In the second image sequence, the spatial resolution coefficient in the vertical direction... It is set to a constant value. Spatial resolution coefficient. The physical height dimension represented by a single pixel length was clearly defined, thus establishing a globally unified measurement standard within the digital image space. In the second image sequence, the axis of the yarn was forcibly constrained to be parallel to the image coordinate axes. The second image sequence and its associated spatial resolution coefficients... It is transmitted to the vector filtering unit as the input source for subsequent processing.

[0028] After completing geometric reconstruction, the online yarn quality detection system for the doubling machine initiates a kinematic feature decoupling procedure for the second image sequence via the vector filtering unit.

[0029] The vector filtering unit extracts the dynamic trajectory information of target pixels by performing temporal correlation analysis on the second image sequence output by the geometric reconstruction unit. During execution, the vector filtering unit selects temporally adjacent frame images from the cache module as input. For two consecutive frames, the vector filtering unit calls the Farneback dense optical flow algorithm. The Farneback dense optical flow algorithm uses a multi-scale Gaussian pyramid structure to hierarchically approximate image features and establishes a quadratic polynomial expansion model in the local pixel neighborhood to fit the displacement bias of pixels over the inter-frame time span. The calculation process generates an instantaneous motion vector field with a resolution completely consistent with the second image sequence. Each element in the instantaneous motion vector field corresponds to an instantaneous motion vector in the image coordinate system. Instantaneous motion vector Includes horizontal displacement components Displacement components in the vertical direction Through this pixel-level dense computation, the relative motion state of the moving yarn and the environmental background is transformed into a digital vector field representation.

[0030] After the instantaneous motion vector field is generated, the vector filtering unit performs targeted data decoupling to obtain the dynamic characteristics that support subsequent mass determination. First, the velocity modulus is extracted. (Velocity modulus) The value is determined by the horizontal displacement component. With vertical displacement components The square root of the sum of squares is used to obtain the velocity, which represents the speed of the composite motion of the pixel in physical space. Then, the orientation deviation angle is extracted. Orientation deviation angle Defined as instantaneous motion vector The angular displacement between the direction of the axis and the preset coordinate axis. The preset coordinate axis direction is consistent with the yarn principal axis direction after geometric normalization in the second image sequence. Direction deviation angle. The calculation utilizes inverse trigonometric functions to quantify the degree to which the pixel motion trajectory deviates from the preset spiral path. Finally, neighborhood consistency density is extracted. The calculation involves a statistical analysis of the vector distribution within a 3x3 or 5x5 local pixel window surrounding the target pixel. This is achieved by calculating the direction variance of all instantaneous motion vectors within the local pixel window, and the neighborhood consistency density. Defined as the reciprocal (or negative exponential mapping) of the directional variance. That is, the smaller the variance, the larger the neighborhood consistency density, thus quantifying the coherence of motion behavior within a local region. Velocity modulus. , Direction deviation angle Consistent density with neighborhood Together, they form the basic data matrix for the subsequent execution of the multiplicative probability decision model.

[0031] The online yarn quality detection system for doubling machines transitions from the feature extraction stage to the final decision-making stage. The numerical processing module built into the vector filtering unit receives the velocity modulus, direction deviation angle, and neighborhood consistency density, and then calls the preset multiplicative probability decision model for calculation.

[0032] The generation of the reference speed relies on a real-time mapping of the physical operating state of the twisting machine. The real-time rotational speed signal of the twisting machine is transmitted to the vector filtering unit via a signal interface. Reference Speed The computational lattice follows the following mathematical relationship: In the formula, This is the real-time rotational speed of the doubling machine (unit: revolutions per minute). The diameter of the rotating cylinder (unit: millimeters). The constant spatial resolution coefficient (unit: mm / pixel) is determined by the geometric reconstruction unit. The image acquisition frame rate (unit: frames / second) for the light field acquisition unit, where 60 is a time unit conversion constant. Reference speed. The calculation is updated in real time according to the fluctuations in the spindle speed of the doubling twister. This dynamic calculation method establishes a linear mapping benchmark between pixel displacement velocity and physical linear velocity, ensuring that the multiplicative probability decision model can obtain accurate velocity normalization reference values ​​during the acceleration and deceleration of the doubling twister.

[0033] Defect probability The numerical values ​​are obtained by performing a nonlinear weighting of multidimensional features. The vector filtering unit calculates the velocity modulus. , Direction deviation angle Neighborhood Consistency Density and reference speed Substitute into the multiplicative probability decision formula: In the formula, the first coefficient Second coefficient and angle threshold The preset parameter matrix is ​​stored in a hardware register. First coefficient The second coefficient determines the sensitivity of orientation recognition. The saturation velocity and angle threshold of the rigid body determination logic were controlled. Tolerance boundaries for yarn spiral motion were defined. By performing a trinomial multiplication operation, the outlier value in a single dimension will affect the defect probability. The final output produces a product inhibition effect.

[0034] Under typical production conditions with a yarn count of 120 and a twisting machine spindle speed of 12,000 RPM, the parameter configuration in the multiplicative probability decision model is as follows: Angle threshold. The value range is from 25 degrees to 35 degrees, and the first coefficient The value range is from 0.4 to 0.6, the second coefficient The value range is from 1.8 to 2.2. Under the above parameter configuration, the online yarn quality detection system for the doubling machine has an extremely high accuracy rate in identifying yarn defects with a diameter of not less than 0.1 mm. This specific parameter quantification configuration, combined with the reference speed... The dynamic compensation ensures that the system can maintain the defect probability even when the oscillation frequency of the high-count yarn air vortex exceeds the preset threshold. The discrimination sensitivity.

[0035] Non-zero constants The settings are designed to eliminate the risk of singularities during the calculation process. In situations where the twisting machine stops abruptly or experiences a breakage, the velocity modulus decreases as the image sequence tends to remain still. The value of approaches zero. Non-zero constants. The order of magnitude is precisely limited to When the velocity modulus Compared with reference speed When the synchronous value drops to an extremely low value, the non-zero constant This ensures that the denominator of the first term in the formula is not zero. This makes the defect probability... The calculation results can smoothly converge to zero, avoiding division-by-zero overflow or logic lock failures in the computer processing unit, and ensuring the system's continuous operation capability in extreme production environments.

[0036] Separation of semantic signals is achieved through the synergy of direction-gated logic and rigid-body belief logic. The second term of the formula constitutes a nonlinear gating system based on the Sigmoid function. When the direction deviates by an angle... The value exceeds the angle threshold. At this point, the output value of the second term will drop drastically. This logic causes the random motion signal generated by the flying petals to be rejected due to excessive directional deviation. Meanwhile, the third term in the formula establishes the neighborhood consistency density. With the probability of defects A positive correlation mapping exists between them. Yarn defects exhibit a high degree of consistency in their direction of movement within their neighborhood, while defocused debris or scattered dust points show a chaotic vector distribution in local areas. Neighborhood consistency density. The reduction of this factor simultaneously drives a decrease in the probability of defects. Through this asymmetric weight modulation, the detection system achieves the preservation of rigid body defect signals and the shielding of outlier noise signals.

[0037] In the production of high-count yarns (counts greater than 120), the yarn's air pocket will enter a critical state of collapse. The severe shape fluctuations during this collapse cause radial drift in the yarn's physical trajectory. Because the optical field acquisition unit provides a full-depth-of-focus visual field covering the radial bounce envelope, and the vector filtering unit utilizes the non-linear truncation characteristic of the directional deviation angle, the online yarn quality detection system for the doubling machine can maintain detection accuracy even when the yarn's air pocket trajectory is unstable. This semantic recognition logic based on directional consistency ensures that the online yarn quality detection system for the doubling machine only locks onto valid signals moving along the yarn's axial direction.

[0038] The triggering of the quality anomaly signal relies on the numerical retrieval of the defect probability distribution matrix. The defect probability distribution matrix output by the vector filtering unit is transmitted to the judgment logic module to perform a global maximum value retrieval operation. The pixel values ​​in the defect probability distribution matrix represent the confidence level of the presence of a defect at the corresponding spatial location. By scanning the defect probability distribution matrix, the online yarn quality detection system of the twisting machine can identify areas where probability values ​​are concentrated. If the maximum value point in the defect probability distribution matrix exceeds a preset judgment threshold, the online yarn quality detection system of the twisting machine generates a quality anomaly signal. The quality anomaly signal contains the time information corresponding to the defect and the spindle phase information. The output of the quality anomaly signal triggers the alarm module or shutdown circuit of the twisting machine, realizing online closed-loop control of yarn production quality.

[0039] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine vision based on-line yarn quality detection system for a two-for-one twister, characterized in that, This includes a light field acquisition unit, a geometric reconstruction unit, and a vector filtering unit connected sequentially along the data processing link: The light field acquisition unit includes an image sensor and an optical lens. An angle is set between the main plane of the optical lens and the photosensitive plane of the image sensor. The main plane, the photosensitive plane and the tangent of the movement trajectory of the yarn air ring to be measured intersect on the same straight line in three-dimensional space, which is used to output the first image sequence. The geometric reconstruction unit stores an inverse homography matrix and is configured to perform coordinate space transformation on the first image sequence using the inverse homography matrix to generate a second image sequence with spatial resolution coefficients. The vector filtering unit is configured to calculate the motion vector field of the second image sequence under a preset coordinate axis, and calculate the defect probability based on the velocity magnitude, directional deviation angle and neighborhood consistency density of each pixel in the motion vector field.

2. The system according to claim 1, characterized in that, The light field acquisition unit also includes a signal interface connected to the rotating component of the twisting machine; the signal interface is configured to trigger the image sensor to perform exposure when acquiring the rotation phase signal; wherein the included angle ranges from 8 degrees to 12 degrees.

3. The system according to claim 1, characterized in that, The geometric reconstruction unit is configured to perform perspective transformation logic on the first image sequence so that the yarn axis in the second image sequence is parallel to the preset coordinate axis in the image coordinate system.

4. The system according to claim 1, characterized in that, The vector filtering unit is configured to calculate the defect probability using the following formula. : in, The defect probability is mentioned above. The velocity modulus; For reference speed; It is a non-zero constant; The deviation angle of the direction; Angle threshold; The first coefficient; The neighborhood consistency density; This is the second coefficient.

5. The system according to claim 4, characterized in that, The reference speed is determined by multiplying the real-time rotational speed of the twisting machine, the diameter of the air ring, the image acquisition frame rate, and the spatial resolution coefficient. The reference speed is directly proportional to the real-time rotational speed of the twisting machine and the diameter of the air ring, and inversely proportional to the image acquisition frame rate and the spatial resolution coefficient. The order of magnitude of the non-zero constant is... .

6. The system according to claim 4, characterized in that, The logic for calculating the defect probability satisfies the following constraints: If the directional deviation angle is greater than the angle threshold, the defect probability tends to zero. If the neighborhood consistency density decreases, the defect probability decreases.

7. A machine vision-based online yarn quality detection method for a doubling machine, wherein the method is executed by the machine vision-based online yarn quality detection system for a doubling machine as described in any one of claims 1 to 6, characterized in that, Includes the following steps: Step 1: Acquire the first image sequence using an optical imaging system that satisfies the collinearity constraint on three planes; Step 2: Reconstruct the first image sequence into a second image sequence using inverse transformation logic, so that the second image sequence has a uniform pixel physical scale transformation relationship across the entire field of view; Step 3: Extract the instantaneous motion vector features from the second image sequence, and obtain the orientation deviation angle and neighborhood consistency density of each pixel under the preset coordinate axis; Step 4: Substitute the instantaneous motion vector features into a multiplicative decision model that includes a reference velocity and a non-zero constant, and output the defect probability distribution matrix.

8. The method according to claim 7, characterized in that, The method is applied to production conditions where the yarn count is greater than 120; in step four, if a maximum point is detected in the defect probability distribution matrix, a quality anomaly signal is output.

Images