Machine vision-based polyester staple fiber spinning on-line detection system and method

Abstract

The application discloses a polyester staple fiber spinning online detection system and method based on machine vision and relates to the technical field of industrial automation detection.The system comprises a pneumatic optical constraint component, an imaging component and a processing unit.The pneumatic optical constraint component is integrated with a Venturi flow channel and provided with a backboard with an inclined stripe background;the imaging component acquires image data of the tow constrained by the Venturi flow channel.The processing unit performs calculation of the vertical gradient component and the gradient direction, generates a first feature map marking the defect area by calculating the difference between the gradient direction and the extension direction of the stripe background, weights the vertical gradient component by using the first feature map, performs projection integral operation in the direction perpendicular to the tow movement direction to offset the gradient component of the stripe background, calculates and outputs the tow width data according to the projection curve generated by the integral.The application realizes transparent defect detection and tow width measurement in a single image at the same time by means of the fluid dynamics constraint and the integral offset algorithm.

Description

Machine Vision-Based Online Inspection System and Method for Polyester Staple Fiber Spinning Technical Field

[0001] This invention relates to the field of industrial automation inspection technology, and in particular to an online inspection system and method for polyester staple fiber spinning based on machine vision. Background Technology

[0002] In the spinning and drawing process of polyester staple fiber, the fiber bundle exhibits a high-speed, flexible fluid state. This physical characteristic presents a complex technical challenge for online inspection, involving both optical imaging and feature extraction. On the one hand, the radial vibration generated by the high-speed movement of the fiber bundle makes it difficult to lock the imaging focal plane, while suspended fluff in the production environment can create noise interference if it enters the imaging depth of field. On the other hand, defects such as gel clumps in the fiber bundle are transparent or translucent, similar to the optical transmission characteristics of normal fibers, making them difficult to distinguish using conventional light intensity imaging methods. If a structured light stripe background is used to enhance the refractive characteristics of transparent defects, the gradient signal of the background stripes themselves will be superimposed on the signal at the edge of the fiber bundle, interfering with the accurate measurement of the physical width of the fiber bundle. Existing detection technologies cannot simultaneously address the problems of spatial imaging instability, environmental noise interference, and geometric measurement interference under structured light backgrounds within the same optical imaging system, resulting in a reduced signal-to-noise ratio of the detection data. Summary of the Invention

[0003] This invention provides an online detection system and method for polyester staple fiber spinning based on machine vision, aiming to solve the technical problems of difficulty in extracting optical features and limited measurement accuracy caused by radial vibration, environmental foreign objects and metamerism in high-speed flexible yarn bundles in open environments.

[0004] In view of the above problems, the present invention provides an online detection system for polyester staple fiber spinning based on machine vision, including a pneumatic optical constraint component, an imaging component and a processing unit arranged sequentially along the fiber bundle movement path;

[0005] The aero-optical confinement assembly includes a Venturi channel and a backplate located on the backlight side of the Venturi channel. The backplate is provided with a striped background whose extension direction is at an angle to the direction of filament movement.

[0006] The imaging component is configured to acquire single-frame image data of a filament bundle constrained by the Venturi channel;

[0007] The processing unit is configured to perform the following steps:

[0008] Calculate the vertical gradient component and gradient direction of the image data;

[0009] Calculate the difference between the gradient direction and the extension direction of the striped background, and mark the region where the difference is greater than a first threshold as the first feature map;

[0010] The vertical gradient components of the image data are weighted using the first feature map, and a projection integral operation is performed perpendicular to the direction of filament movement to generate a one-dimensional projection curve.

[0011] Based on the edge features of the one-dimensional projection curve, the filament width data is calculated and output.

[0012] Furthermore, when the processing unit performs projection integration, the height of the integration window is equal to an integer multiple of the projection period of the striped background in the vertical direction.

[0013] Furthermore, the central axis of the Venturi channel coincides with the focal plane of the imaging component; the depth of field of the imaging component covers the radial constraint range of the Venturi channel, but does not cover the external region of the airflow layer of the Venturi channel.

[0014] Furthermore, the processing unit is configured as follows:

[0015] Calculate the gradient direction of each pixel in the image;

[0016] Calculate the absolute value of the difference between the gradient direction and the included angle;

[0017] The pixels whose absolute value is greater than the first threshold are assigned a first value, and the remaining pixels are assigned a second value to generate the first feature map.

[0018] Furthermore, the processing unit calculates the intermediate variables according to the following formula. :

[0019] in, This is the normalized value of the vertical gradient component;

[0020] The signal-to-noise ratio enhancement index is greater than 2;

[0021] The formula for calculation is:

[0022] in The first feature map in coordinates The value at that location, Sensitivity is a coefficient. The second threshold;

[0023] The processing unit is configured to perform vertical summation on the intermediate variables to obtain the one-dimensional projection curve.

[0024] Furthermore, the included angle ranges from 30° to 60°; the value of the imaging component is less than 1.4.

[0025] A machine vision-based online detection method for polyester staple fiber spinning utilizes the Venturi channel to constrain the movement range of the fiber bundle.

[0026] Obtain image data of the filament bundle containing the striped background;

[0027] A first feature map is generated based on the gradient direction of the image data;

[0028] The vertical gradient components of the image data are weighted using the first feature map;

[0029] The weighted vertical gradient component is projected and integrated along the direction perpendicular to the filament movement to calculate the filament width.

[0030] The technical solution provided in this application has at least the following technical effects: By applying radial constraint to the filament bundle through the Venturi channel in the aero-optical constraint component, and in conjunction with the depth-of-field configuration of the imaging component, the vibration range of the filament bundle is limited at the physical level, and the interference of suspended foreign objects outside the airflow layer is filtered out by utilizing the defocus effect; By combining the tilted striped background with vertical projection integration, the gradient component of the background stripes is canceled at the mathematical level by utilizing the geometric matching relationship between the integration window and the stripe period, and the transparent defect features are extracted by utilizing the gradient direction difference and the width measurement is weighted and corrected, thereby realizing the width measurement and defect detection of the non-steady flexible filament bundle in a single imaging process.

[0031] This invention resolves the inherent contradiction in optical background requirements between transparent defect detection and high-precision width measurement. Typically, detecting transparent defects requires a high-contrast structured light (striped) background to highlight refractive features; however, the structured light background itself is high-frequency noise, severely interfering with the extraction of filament edges. This invention utilizes the mathematical orthogonality between the integral window and the background period (i.e., the period integral returns to zero) to achieve mathematical-level background cloaking in the width measurement path while retaining the striped background for defect detection. By achieving dual-path feature decoupling through a single optical input, a high signal-to-noise ratio is achieved. Attached Figure Description

[0032] Figure 1 is a schematic diagram of the overall structure of the online detection system for polyester staple fiber spinning based on machine vision provided in an embodiment of the present invention;

[0033] Figure 2 is a schematic flowchart of the online detection method for polyester staple fiber spinning based on machine vision provided in an embodiment of the present invention. Detailed Implementation

[0034] 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.

[0035] As shown in Figures 1 and 2, this embodiment provides an online detection system for polyester staple fiber spinning based on machine vision, and a corresponding online detection method. The system consists of a pneumatic optical constraint component, an imaging component, and a processing unit arranged sequentially along the fiber bundle movement path.

[0036] The aero-optical confinement assembly includes a Venturi channel and a backplate located on the backlight side of the Venturi channel. The backplate is provided with a striped background whose extension direction is at an angle to the direction of filament movement.

[0037] The imaging component is configured to acquire single-frame image data of a filament bundle constrained by the Venturi channel;

[0038] The processing unit is configured to perform the following steps:

[0039] Calculate the vertical gradient component and gradient direction of the image data;

[0040] Calculate the difference between the gradient direction and the extension direction of the striped background, and mark the region where the difference is greater than a first threshold as the first feature map;

[0041] The vertical gradient components of the image data are weighted using the first feature map, and a projection integral operation is performed perpendicular to the direction of filament movement to generate a one-dimensional projection curve.

[0042] Based on the edge features of the one-dimensional projection curve, the filament width data is calculated and output.

[0043] In practical implementation, the operation of the machine vision-based online detection system for polyester staple fiber spinning begins with introducing a high-speed moving filament bundle into the Venturi channel inside the aero-optical confinement assembly to create a physically stable imaging environment. To form an aerodynamic field around the filament bundle that suppresses radial vibrations, the Venturi channel is precisely machined along the filament bundle's direction of motion into a series of interconnected sections: a contraction section, a flow-stabilizing throat section, and a diffusion section. The contraction section employs a streamlined curved surface structure to guide boundary layer airflow. The ratio of the cross-sectional area at the inlet to the outlet of the contraction section is set within the range of 3:1 to 5:1 (this ratio is a preferred embodiment and can be adjusted according to fluid viscosity). This cross-sectional contraction ratio is designed to force the airflow entering the Venturi channel to undergo a rapid acceleration before reaching the flow-stabilizing throat section.

[0044] The stabilizing throat section, immediately following the contraction section, constitutes the core physical region for optical detection. The channel height of the stabilizing throat section is designed to be slightly greater than the natural thickness of the filament bundle, typically set between 2 and 5 millimeters. When a high-speed airflow passes through the stabilizing throat section with its reduced cross-sectional area, according to Bernoulli's principle in fluid mechanics, the increased airflow velocity causes the static pressure inside the stabilizing throat section to be significantly lower than the atmospheric pressure outside the Venturi channel. This results in a symmetrically distributed negative pressure adsorption field forming on the upper and lower inner wall surfaces of the stabilizing throat section.

[0045] This negative pressure adsorption field acts as a non-contact air clamp, applying a restoring force perpendicular to the direction of filament movement to the filament bundle passing through the steady-flow throat section. When the filament bundle attempts to deviate from the central axis of the Venturi channel due to fluctuations in stretching tension, the reduced gap on the side of the bundle near the inner wall of the Venturi channel leads to a further increase in airflow velocity and a further decrease in pressure on that side. The resulting pressure difference forces the filament bundle back to the geometric center plane of the Venturi channel. Through the hydrodynamic adaptive balancing effect of the Venturi channel, the radial vibration amplitude of the filament bundle in the optical axis direction is physically limited to within ±0.5 mm, a range that matches the preset clear imaging depth range of the imaging assembly. After flowing through the steady-flow throat section, the gas enters a diffuser section with an expansion angle, decelerates, and is discharged to prevent turbulence at the Venturi channel outlet from interfering with the subsequent operation of the filament bundle. Furthermore, the high-speed fluid environment within the Venturi channel, combined with the pressure difference design at the inlet and outlet, creates a continuously flowing high-speed air protective layer on the observation window surface of the imaging assembly. This air protection layer effectively blocks the adhesion of high-temperature oil volatiles and fine lint during the spinning process by utilizing the air curtain effect. It gives the pneumatic optical constraint component the ability to self-clean the optical window in high-temperature and high-humidity industrial environments, thereby maintaining the transparency of optical imaging for a long time without the need for additional mechanical wiping mechanisms.

[0046] After establishing a physically stable imaging environment, the online inspection system for polyester staple fiber spinning establishes a tomographic imaging optical path by precisely adjusting the spatial position of the imaging component. The imaging component is mounted on one side of the observation window of the aero-optical constraint component and is equipped with a large-aperture lens with an F-number less than 1.4 to obtain millimeter-level shallow depth-of-field optical characteristics. To ensure that the imaging sharpness matches the physical locking range, the optical focal plane of the imaging component must be precisely calibrated to strictly coincide with the physical central axis of the Venturi channel. Through this coaxial calibration operation, the sharp imaging depth-of-field range of the imaging component is precisely superimposed on the radial constraint range of the Venturi channel. The sharp imaging depth-of-field range is typically set to cover a spatial thickness of ±1.0 mm relative to the central axis of the Venturi channel.

[0047] In this optical configuration, the filament bundle, locked in place by the negative pressure of the Venturi channel and located in the central plane, can form a sharp, real image on the photosensitive element of the imaging component. Simultaneously, suspended fluff or oil mist particles floating outside the Venturi channel airflow layer in the production environment, due to their axial distance from the optical focal plane exceeding the clear imaging depth-of-field limit of the imaging component, will produce a severe defocus blur effect on the image plane according to the laws of geometric optics. This defocus blur effect transforms the originally high-frequency edge-feature fluff into low-contrast diffuse spots, causing the gradient energy of these environmental noises to be significantly attenuated during the physical imaging stage. This effectively filters them out at the source, eliminating them from subsequent edge feature extraction processes and achieving optical physical noise reduction without computational intervention. To further eliminate specular reflection spots on the filament bundle surface and enhance the contrast of transmission imaging, the optical imaging system employs an orthogonal polarization configuration, with a first linear polarizer and a second linear polarizer, with mutually perpendicular polarization directions, installed on the light source side and in front of the lens, respectively.

[0048] The background illumination environment on the opposite side of the imaging optical path is provided by a backplate integrated into the backlight side of the aero-optical constraint assembly. The backplate uses a high-brightness LED surface light source in conjunction with a mask printed with high-contrast black and white periodic stripes, which constitute the striped background for transmission imaging. To facilitate subsequent integral cancellation algorithms and avoid texture interference, the geometric layout of the striped background follows specific angular constraints: the extension direction of the black and white stripes in the striped background forms a fixed angle with the direction of movement of the filament bundle. The angle is set between 30 and 60 degrees, with 45 degrees being a typical preferred value. This obliquely intersecting geometric positional relationship ensures that the stripe boundaries in the background always cross the vertical physical edge of the filament bundle obliquely, avoiding moiré fringe interference or edge feature fading caused by the stripe extension direction being parallel to the filament bundle edge direction. The oblique striped background projected by the backplate passes through the filament bundle located in the Venturi channel and enters the imaging component. In a single frame image data, a composite light field information containing the filament bundle outline and background texture is formed, providing an optical carrier for the subsequent processing unit to perform refractive index-based defect separation and integral-based width measurement.

[0049] The single-frame image data acquired by the imaging component is transmitted to the processing unit to initiate a dual-path parallel decoupled computation process. The processing unit first performs a global scan and analysis of the geometric and optical features of the background texture in the defect development path, aiming to transform the imperceptible refraction phenomena of the transparent medium into computable digital features.

[0050] To quantify the grayscale variation trend of each pixel in the image data in the spatial domain, the processing unit performs convolution operations on the image data using a discrete difference operator. The Sobel or Scharr operator is used as the convolution kernel to calculate the grayscale gradient components in the horizontal and vertical directions of the image data, respectively. After obtaining the horizontal and vertical gradient components of each pixel, the processing unit calculates the arctangent function of the ratio of the vertical to the horizontal gradient components to determine the gradient direction value corresponding to each pixel coordinate. The set of gradient direction values ​​from all pixels forms a local gradient direction field describing the geometric topology of the image texture.

[0051] After establishing the local gradient direction field, the processing unit performs a deviation analysis between the calculated gradient direction value and the system's preset physical parameters. Since the striped background within the aero-optical constraint assembly is configured with a fixed extension direction angle, light transmitted through normal filaments will maintain a propagation path and texture direction consistent with the striped background. However, light transmitted through transparent gel blocks or crystal points will experience disordered deflection of the local texture direction due to microlens refraction. The processing unit performs point-by-point subtraction between the gradient direction value of each pixel and the preset extension direction angle, calculating the absolute value of the difference. This absolute value of the angle difference physically characterizes the severity of optical distortion at the current spatial location.

[0052] To separate the specific regions of refractive distortion from the image data, the processing unit introduces a preset angle threshold for logical discrimination. The typical range of the angle threshold is set to 10 to 15 degrees. When the absolute value of the angle difference at a specific pixel coordinate is greater than the preset angle threshold, the processing unit classifies the pixel as a refractive singularity caused by a defect and assigns it a first value (usually 1). Conversely, when the absolute value of the angle difference is less than or equal to the preset angle threshold, the processing unit classifies the pixel as normal background or a filament region and assigns it a second value (usually 0). Through this binary classification process, the processing unit generates a first feature map with the same resolution as the original image data. It should be understood that the gradient direction deviation-based analysis method described in this embodiment has broad defect universality. In addition to the refractive distortion caused by transparent slurry, defects such as fused filaments, fused filaments, stiff filaments, columnar filaments, and broken filaments, which are common in the production process, although their optical transmittance is similar to that of normal filament bundles, also manifest as local disturbances in the texture gradient direction in the tomographic phase image due to their abnormal physical morphology. Therefore, the above logic can accurately capture defect features including the aforementioned multiple types. Combined with the optical magnification of the imaging component, the system can detect a minimum defect size of 0.5 mm x 0.5 mm at a spinning speed of 1600 m / min. The first feature map serves as a digital mask accurately marking the spatial distribution of various defects, providing data guidance for weight suppression in subsequent width measurement calculations. The first feature map also serves as a digital mask accurately marking the spatial distribution of transparent defects, providing data guidance for weight suppression in subsequent width measurement calculations.

[0053] While generating the first feature map marking the defect locations, the processing unit initiates an energy density model construction program for the image data within the width measurement path. To eliminate interference from background stripes on the extraction of the filament edges, the processing unit discards the simple vertical gradient component in this path and instead extracts the signed vertical gradient component for each pixel in the image data. The processing unit retains the sign information of this vertical gradient component to reflect the grayscale transition direction of the background stripes from light to dark and from dark to light.

[0054] Subsequently, the processing unit calculates intermediate variables using a pre-set mathematical model. The calculation formula introduces a symbolic function. To maintain the original polarity of the gradient during power-law amplification:

[0055] In this formula, The normalized vertical gradient component; A signal-to-noise ratio enhancement index greater than 2 (typically 3) is used to enhance the saliency of edge signals. This represents the texture suppression factor, which the processing unit calculates based on a mathematical model that includes the Sigmoid function.

[0056] In the formula for calculating the texture suppression factor, This represents the first feature map at the corresponding coordinates. The value at that location. The sensitivity coefficient (preferably in the range of 5 to 15) is used to control the steepness of the gradient of the function; The second threshold (preferably ranging from 0.5 to 0.8) is used to define the defect determination boundary. For any region overlapping with the transparent slurry, the gradient energy is abruptly reduced to zero, thus physically eliminating the contribution of defects to the width in subsequent integration calculations.

[0057] Obtain intermediate variables after defect removal and signal enhancement processing Next, the processing unit performs vertical projection integration to extract the physical edge features of the filament bundle. Since the intermediate variables still retain the gradient information of the background stripes, directly searching for the peaks would lead to measurement errors. Therefore, the processing unit performs mathematical hidden surface removal of the background texture by setting an integration window with specific geometric constraints.

[0058] The first step in performing vertical projection integration is to determine the vertical span of the integration domain. Upon system startup, the processing unit performs spectral analysis or autocorrelation calculations on the background region, calibrating the projection period of the background fringes perpendicular to the filament movement direction in real time. This projection period depends on the physical grid pitch of the fringe background, the angle between the fringe background and the filament movement direction, and the magnification of the imaging components. The processing unit then calculates the integration window height for the vertical projection integration operation. Strictly set as projection period Integer multiples of, that is, satisfying the mathematical relation ,in The period is a positive integer. It can be obtained through system calibration or by performing autocorrelation / spectral analysis on the background image. Within a defined integration window height, the processing unit algebraically sums the intermediate variables containing positive and negative signs along the vertical direction to generate a one-dimensional projection curve. Since the background fringes exhibit a periodic alternating bright and dark structure, the vertical gradient components generated by the background fringes spatially exhibit a fluctuating pattern of alternating positive values ​​(bright areas transitioning to dark areas) and negative values ​​(dark areas transitioning to bright areas). Under the geometric condition that the integration window height is an integer multiple of the projection period, the positive and negative gradient energies generated by the background fringes undergo precise algebraic cancellation during vertical integration, causing the total contribution of the background fringes to mathematically zero in the integration result.

[0059] Meanwhile, the physical edges of the filament bundle exhibit a monotonically consistent gray-level step characteristic in the vertical direction, and the gradient signals generated by the edges of the filament bundle remain superimposed with the same sign within the integration window. After vertical projection integration, the two-dimensional image data containing background noise is compressed into a one-dimensional projection curve. In this one-dimensional projection curve, the high-frequency fluctuations caused by background stripes are mathematically smoothed out, while the signal peaks representing the edges of the filament bundle are coherently enhanced, thus achieving the separation of background texture from the data stream while preserving the characteristics of the filament bundle edges.

[0060] After generating a high signal-to-noise ratio one-dimensional projection curve, the processing unit initiates a centroid calculation procedure to obtain the precise location of the filament bundle edges. The processing unit uses a centroid calculation formula incorporating a regularization factor to multiply the coordinate values ​​of the one-dimensional projection curve by their corresponding absolute energy values, summing the results, and then dividing by the total sum of absolute energy values ​​to obtain the sub-pixel coordinates of the left and right edges of the filament bundle. This centroid calculation method utilizes all grayscale information within the edge transition region, enabling the final output edge coordinate accuracy to exceed the physical pixel resolution limitations of the imaging sensor. The processing unit combines this with pixel equivalent coefficients obtained beforehand using standard gauge blocks to output width data characterizing the actual physical size of the filament bundle. Thus, the system completes a closed-loop process from physical image stabilization, optical imaging, defect separation to precision measurement.

[0061] After completing the measurement process from physical imaging to feature decoupling, the online inspection system for polyester staple fiber spinning further integrates an adaptive protection mechanism for abnormal operating conditions and a quality closed-loop control logic to ensure numerical stability and production safety during continuous operation.

[0062] During the final calculation of the sub-pixel centroid, the processing unit employs a centroid calculation model incorporating a regularization factor to handle extreme conditions such as the imaging component outputting a completely black image or the filament completely detaching from the detection field of view. According to the mathematical definition of centroid calculation, the solution for the filament edge coordinates relies on the total sum of energy from the one-dimensional projection curves as the divisor. When the imaging component experiences a backlight failure, complete obstruction of the optical path by foreign objects, or filament breakage and disappearance, resulting in the absence of effective gradient energy in the image data, the total sum of energy from the one-dimensional projection curves will mathematically approach zero.

[0063] To prevent division-by-zero errors from causing the processing unit's computation process to crash, the processing unit introduces a very small positive number as a regularization factor (typically a value of...) into the denominator of the centroid calculation formula. Regardless of the fluctuations in the actual gradient energy of the image data, the existence of the regularization factor ensures that the denominator of the centroid calculation formula is never zero. When the system detects that the calculated total energy is lower than the effective signal threshold, the processing unit uses the regularization factor to maintain the mathematical convergence of the algorithm and simultaneously outputs a signal loss status code instead of an invalid width value, thereby ensuring the continuous availability of the detection system in harsh industrial environments at the software level.

[0064] In addition to outputting tow width data to monitor the stability of the drawing process, the processing unit simultaneously monitors the first feature map originating from path A to assess the product quality status. The processing unit performs statistical analysis on the first feature map corresponding to a single frame image, calculating the total number of pixels marked with a first value (representing a defect) in the first feature map, thereby quantifying the cumulative defect area within the current field of view. When the calculated cumulative defect area exceeds the area threshold preset by the production process, the processing unit determines that the tow currently passing through the pneumatic optical constraint assembly has severe slurry blockages or large-area crystal point defects. In response to this determination, the processing unit immediately generates a severe defect trigger signal. This severe defect trigger signal is transmitted to the external programmable logic controller (PLC) on the production line via an industrial communication interface. The PLC, based on this signal, activates the on-site audible and visual alarm device to alert the operator, or directly drives the tow cutting mechanism to perform a blocking operation when continuous severe defect trigger signals are detected, preventing unqualified polyester staple fiber tows from flowing into subsequent high-temperature heat setting or winding processes, thereby achieving automatic closed-loop control of production quality.

[0065] 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 online inspection system for polyester staple fiber spinning, characterized in that, The system includes an aero-optical constraint component, an imaging component, and a processing unit arranged sequentially along the filament movement path. The aero-optical constraint component includes a Venturi channel and a backplate located on the backlight side of the Venturi channel. The backplate has a striped background whose extension direction forms an angle with the filament movement direction. The imaging component is configured to acquire single-frame image data of the filament constrained by the Venturi channel. The processing unit is configured to perform the following steps: calculate the vertical gradient component and gradient direction of the image data; calculate the absolute value of the difference between the gradient direction and the extension direction of the striped background; assign a first value to pixels whose absolute value is greater than a first threshold, and assign a second value to the remaining pixels to generate a first feature map; weight the vertical gradient component of the image data using the first feature map, and perform a projection integral operation perpendicular to the filament movement direction to generate a one-dimensional projection curve; calculate and output filament width data based on the edge features of the one-dimensional projection curve.

2. The system as described in claim 1, characterized in that, Specifically, the step of determining the edge position based on the edge features of the one-dimensional projection curve involves: calculating the centroid of the one-dimensional projection curve and determining the edge position based on the centroid.

3. The system according to claim 1, characterized in that, When the processing unit performs projection integration, the height of the integration window is equal to an integer multiple of the projection period of the striped background in the vertical direction.

4. The system according to claim 1, characterized in that, The central axis of the Venturi channel coincides with the focal plane of the imaging component; the depth of field of the imaging component covers the radial constraint range of the Venturi channel, but does not cover the external region of the airflow layer of the Venturi channel.

5. The system according to claim 1, characterized in that, The processing unit is configured to: calculate the gradient direction of each pixel in the image; calculate the absolute value of the difference between the gradient direction and the included angle; assign a first value to pixels whose absolute value is greater than the first threshold, and assign a second value to the remaining pixels, thereby generating the first feature map.

6. The system according to claim 1, characterized in that, The processing unit calculates the intermediate variables according to the following formula. ： in, This is the normalized value of the vertical gradient component; The signal-to-noise ratio enhancement index is greater than 2; The formula for calculation is: in The first feature map in coordinates The value at that location, Sensitivity is a coefficient. The second threshold is used; the processing unit is configured to perform vertical summation on the intermediate variables to obtain the one-dimensional projection curve.

7. The system according to claim 1, characterized in that, The included angle ranges from 30° to 60°; the value of the imaging component is less than 1.

4.

8. A machine vision-based online detection method for polyester staple fiber spinning, executed using the system described in any one of claims 1 to 7, characterized in that, include: The Venturi channel is used to constrain the movement range of the filament bundle; image data of the filament bundle containing the striped background is acquired. A first feature map is generated based on the gradient direction of the image data; The vertical gradient components of the image data are weighted using the first feature map; the weighted vertical gradient components are then integrated by projection along the direction perpendicular to the filament movement to calculate the filament width.

Images