A Method and System for Defect Location and Removal in Nonwoven Fabrics Based on a Self-calibration Model
By using a self-calibration model to analyze and compensate for texture offset in the nonwoven fabric defect detection system, the problem of identification and positioning deviation caused by dynamic disturbances during the nonwoven fabric production process is solved, and the accurate positioning and efficient removal of defects are achieved.
Patent Information
- Application Number
- CN202611054727.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-16
- Publication Date
- 2026-11-17
AI Technical Summary
Existing nonwoven fabric defect detection systems lack adaptive capabilities when faced with dynamic disturbances during the production process. This leads to a decrease in the defect identification model's ability to distinguish between normal textures and real defects, resulting in false alarms, missed alarms, and discrepancies between the location results and the actual removal positions.
A self-calibration model is used to locate edge references or marked reference points by acquiring fabric images, extract multi-dimensional texture features, and use the self-calibration model to perform texture offset analysis and compensation. By combining the texture correction results and positioning compensation, the physical location of abnormal texture areas is determined, and the removal method is identified.
It improves the accuracy of defect location and efficiency of removal in continuous nonwoven fabric production, reduces false alarms, missed alarms and removal position deviations, and enhances the system's adaptability and stability.
Smart Images

Figure CN123118450A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of nonwoven fabric defect identification technology, and in particular to a method and system for locating and removing nonwoven fabric defects based on a self-calibration model. Background Technology
[0002] For nonwoven fabric products, fabric quality, structural uniformity, and surface integrity are important factors affecting product grade and performance. To achieve online detection, location, and removal of defects in nonwoven fabrics, existing technologies typically include a nonwoven fabric conveying device, an image acquisition device, a light source illumination device, an image processing device, a control device, a position detection device, and a defect removal execution device. The system comprises several components: an image acquisition device (typically an industrial camera) for real-time imaging of the continuously conveyed fabric surface; an illumination device to provide a stable detection light field; a position detection device using an encoder, tachometer, or displacement sensor to acquire fabric speed and displacement information; an image processing device for denoising, enhancing, segmenting, and extracting features from the acquired fabric images; and a recognition model based on preset thresholds, template matching rules, or a trained model to identify defective areas such as holes, black spots, wrinkles, foreign objects, and uneven thickness, outputting the corresponding defect image coordinates. The control device then converts the defect image coordinates into physical position coordinates at the removal station based on the structural distance between the image acquisition location and the removal station, the fabric conveying speed, and the time delay. It then drives the cutter, blowing mechanism, marking mechanism, or rejection mechanism to process the target defect area, thus forming an online detection and removal system for nonwoven fabric defects.
[0003] However, most existing systems rely on relatively stable camera installation positions, lighting conditions, fabric conveying speeds, and coordinate conversion relationships, lacking the ability to adaptively correct for dynamic disturbances in the production process. When nonwoven fabrics are affected by factors such as tension fluctuations, equipment vibrations, and random changes in fiber texture during continuous operation, the fabric images are prone to brightness fluctuations, local blurring, and texture distortion. This reduces the ability of the defect identification model to distinguish between normal textures and real defects, leading to false alarms and missed detections. Simultaneously, factors such as cumulative displacement errors, coding tracking errors, and actuator response lags commonly exist during fabric conveying. These factors, combined with defect identification errors, can easily cause deviations between the defect location results and the actual removal location, resulting in missed removal, false removal, or over-removal. This makes it difficult to meet the requirements for accurate defect location and efficient removal in continuous nonwoven fabric production. Summary of the Invention
[0004] Therefore, embodiments of the present invention provide a method and system for locating and removing defects in nonwoven fabrics based on a self-calibration model. The technical solution is as follows: On the one hand, a method for locating and removing defects in nonwoven fabrics based on a self-calibration model is provided, which includes: S1. Acquire images of the fabric surface during the continuous conveying process of the nonwoven fabric, and perform edge benchmark positioning or mark reference point positioning to obtain the target fabric surface image for fabric texture offset analysis.
[0005] S2, extract the multidimensional texture features representing the fiber arrangement state of the target fabric image, and match them with the preset reference texture points to obtain the texture offset; the texture offset is used to characterize the degree of deviation of the current fabric texture from the reference texture in terms of orientation distribution, texture density and texture position distribution.
[0006] S3 inputs the texture offset into the preset self-calibration model. After bias learning analysis and compensation parameter solving, the corresponding texture correction result and positioning compensation result are output.
[0007] S4. Based on the texture correction results, determine the abnormal texture regions in the target fabric image, and combine the positioning compensation results to determine the physical coordinates of the abnormal texture regions at the cleaning station.
[0008] S5. Based on the regional distribution of physical location coordinates, determine the clearing method. The clearing method includes at least single fixed-point clearing and segmented continuous clearing.
[0009] On the other hand, a nonwoven fabric defect localization and removal system based on a self-calibration model is provided, which includes: The image acquisition and positioning module is used to acquire images of the fabric surface during the continuous conveying process of nonwoven fabric, and to perform edge benchmark positioning or mark reference point positioning to obtain the target fabric surface image for fabric texture offset analysis.
[0010] The texture offset extraction module is used to extract multidimensional texture features representing the fiber arrangement state of the target fabric image and match them with preset reference texture points to obtain the texture offset.
[0011] The self-calibration compensation module is used to input the texture offset into the preset self-calibration model. After deviation learning analysis and compensation parameter solving, it outputs the corresponding texture correction results and positioning compensation results.
[0012] The abnormal region localization module is used to determine the abnormal texture region in the target fabric image based on the texture correction result, and to determine the physical location coordinates of the abnormal texture region at the cleaning station in combination with the localization compensation result.
[0013] The clearing method determination module is used to determine the clearing method based on the regional distribution of physical location coordinates.
[0014] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: 1. This invention first acquires images of the nonwoven fabric surface during continuous conveying and obtains the target fabric surface image through edge reference positioning or marked reference point positioning, which helps reduce image positioning instability caused by factors such as fabric deviation and image offset. Subsequently, multi-dimensional texture features representing the fiber arrangement state of the fabric surface are extracted from the target fabric surface image and matched with preset reference texture points to obtain texture offset, effectively enhancing the ability to represent the difference between normal texture and real defects, thereby mitigating false alarms and missed alarms caused by brightness fluctuations, local blurring, and texture distortion. Further, the texture offset is input into a preset self-calibration model, and after deviation learning analysis and compensation parameter solving, texture correction results and positioning compensation results are output to adaptively correct texture distortion and position deviation formed during dynamic operation, thereby improving the stability of abnormal texture recognition results and the accuracy of defect position conversion. Then, based on the texture correction results, abnormal texture regions in the target fabric surface image are determined, and combined with the positioning compensation results, the physical position coordinates of the abnormal texture regions at the cleaning station are determined, thereby reducing the cleaning position deviation caused by cumulative displacement error, coding tracking error, and actuator response lag. Finally, the removal method is determined based on the regional distribution of physical location coordinates. Single-point removal or segmented continuous removal is selected as needed to avoid missed removal, accidental removal or over-removal, thereby improving the accuracy of defect location and efficient removal effect in continuous production of nonwoven fabrics.
[0015] 2. First, the acquired fabric image undergoes edge enhancement, breakpoint connection, and smoothing to obtain a stable edge curve. An edge baseline is then formed through edge line fitting. Combined with the positional and angular offsets between the baseline and the pre-defined baseline, translational and angular corrections are applied to the fabric image, thereby improving the consistency of the edge baseline during continuous transport. Simultaneously, pre-defined marked regions in the fabric image are identified, and region positioning is performed based on color, grayscale, or contour features. Candidate marked points are obtained through binarization segmentation and connected component analysis. Candidate marked points that meet the pre-defined positional continuity condition are determined as valid marked points, and their center coordinates are used as reference points for positional correction of the fabric image, further enhancing the spatial positioning stability of the target fabric image. Subsequently, the target fabric image is divided into several texture analysis sub-regions according to a preset grid division method. The texture energy features, texture direction features, and texture uniformity features corresponding to each sub-region are extracted and compared with the corresponding texture feature templates. Then, the feature deviation of each sub-region in multiple dimensions is calculated, and the sub-region offset is obtained through vector synthesis. Then, the offset of each sub-region is fitted with a surface to form a continuously distributed texture offset field, and the local extreme value distribution is extracted to obtain the texture offset. This can more comprehensively characterize the degree of distortion of the fabric texture at the local and global levels, improve the sensitivity of abnormal texture identification and the accuracy of texture offset analysis.
[0016] 3. First, the texture offset is input into a preset self-calibration model. Deviation learning analysis is performed on the input texture offset to obtain the texture deviation distribution result corresponding to the current fabric image. This characterizes the degree of local texture distortion and its distribution location in different texture analysis sub-regions, thus facilitating accurate identification of local distortions caused by tension fluctuations, equipment disturbances, and random texture changes during continuous fabric operation. Subsequently, based on the texture deviation distribution result, the texture compensation parameters and position compensation parameters corresponding to each texture analysis sub-region are calculated. Local texture distortion and image coordinates are then corrected to obtain texture correction results and positioning compensation results. This improves the fabric texture restoration effect and reduces position deviations caused by conveying errors, tension changes, and actuator response lag. Further, the corrected target fabric image is scanned point-by-point based on a preset sliding window. Texture direction features and grayscale change features corresponding to each detection position are extracted, and texture deviation parameters are calculated. Detection positions exceeding preset abnormal texture parameters are then aggregated to obtain abnormal texture regions, thereby improving the accuracy of identifying real defect areas.
[0017] 4. First, the coordinates of the center point and boundary range of the abnormal texture area are obtained. Based on the camera calibration parameters and preset coordinate mapping relationship, the coordinates are transformed from the image coordinate system to the fabric detection coordinate system to obtain the initial position coordinates, thus realizing the transformation of the abnormal area from image information to the actual spatial position of the fabric. Subsequently, combined with the relative installation distance between the detection station and the removal station, as well as the conveying speed, encoding displacement, and tension change parameters corresponding to the current detection moment, the position migration of the abnormal texture area during the conveying process is calculated. Based on the position compensation relationship, the initial position coordinates are corrected to obtain the physical position coordinates at the removal station, effectively reducing the positioning deviation caused by fabric conveying error, tension fluctuation, and actuator response lag. Further, based on the physical position coordinates corresponding to each abnormal texture area, the continuous distribution length of the abnormal texture area in the fabric conveying direction and the distribution range in the fabric width direction are determined. The normalized continuous distribution length and distribution range are combined to obtain a comprehensive distribution value, which determines whether to use a single fixed-point removal or segmented continuous removal method, so that the removal strategy can match the actual distribution state of different defect areas. Finally, after completing the corresponding removal method, the residual evaluation value is determined by the combined analysis results of the residual area parameter and residual texture deviation parameter in the current removal area, and the removal is judged based on the evaluation value. If it is not completed, the removal control is triggered again, and abnormal removal warning information is output by combining the number of repeated removals and residual changes, thereby forming a closed-loop verification after removal, further improving the removal reliability and production stability. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 A flowchart of a nonwoven fabric defect location and removal method based on a self-calibration model provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a nonwoven fabric defect location and removal system based on a self-calibration model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the nonwoven fabric defect location and removal system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the nonwoven fabric defect identification results provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the fabric image interface input and level conversion control circuit provided in an embodiment of the present invention. Figure 6 This is a closed-loop control block diagram for nonwoven fabric defect location and removal based on a self-calibration model provided in an embodiment of the present invention. Figure 7 This is a schematic diagram of abnormal texture quantity monitoring provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the training convergence result of the self-calibration model provided in an embodiment of the present invention. Detailed Implementation
[0020] The technical solution of the present invention will now be described with reference to the accompanying drawings.
[0021] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.
[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.
[0023] like Figure 3The diagram shows the structure of a nonwoven fabric defect location and removal system. This system mainly includes a fabric detection and position sensing unit, a main controller, an execution and output unit, and a power module. The fabric detection and position sensing unit on the left is used to collect real-time data on the nonwoven fabric's operating status. Specifically, it includes an industrial camera CAM1, an encoder ENC1, and a tension sensor TEN1. The industrial camera CAM1 acquires fabric images and generates fabric image signals. The encoder ENC1 detects the speed and displacement information of the fabric during its movement and outputs speed pulse signals. The tension sensor TEN1 detects the fabric tension and outputs a tension signal. The main controller is used for comprehensive analysis of the acquired data. It can be implemented using a microcontroller unit (MCU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), and internally runs core algorithms such as texture image analysis, self-calibration offset compensation, and removal strategy determination.
[0024] The main controller's input interfaces include an image input interface and a position input interface, while its output interfaces include a light source control interface, an execution control interface, and an alarm output interface. The execution and output unit on the right includes a light source control device LAMP1, a cleaning execution mechanism ACT1, and an alarm ALM1. LAMP1 is used to adjust the brightness of the light source or the detection illumination; ACT1 is used to perform defect removal actions such as cutting, blowing, or spraying liquid; and ALM1 is used to output audible and visual alarm signals in abnormal situations. The power module PWR1 below converts the AC input voltage to the DC operating voltage required by the system, providing power to the camera, sensor, controller, and execution mechanism. In the signal transmission relationship shown by the arrows in the diagram, the fabric image signal can be input to the main controller via CameraLink, MIPI, or GigE interfaces; the speed pulse signal can be input in A / B / Z phase pulse form; and the tension signal can be input in 0–10V or 4–20mA form. The main controller then outputs PWM / DO light source control signals, DO / PWM execution control signals, and DO alarm output signals to complete the closed-loop control of nonwoven fabric defect detection, location, and removal.
[0025] like Figure 4The schematic diagram of nonwoven fabric defect identification results shown is a fabric image obtained by an industrial camera after real-time acquisition of the nonwoven fabric surface during continuous conveying. The image shows the identification results of defect areas against different fabric texture backgrounds. The rectangular annotation boxes in each fabric image are used to characterize the identified target defect areas, and the text labels above the annotation boxes are used to characterize the corresponding defect types. The target defect areas can be stains, foreign matter attachment areas, or local texture abnormalities, which can intuitively reflect the distribution location, size, and identification category of defects in the fabric image.
[0026] like Figure 5 The schematic diagram of the fabric image interface input and level conversion control circuit shown is used to protect, shape, and level-match the signals input from the fabric image interface before transmitting them to the microcontroller control terminal. In the diagram, J1 is the fabric image interface, used to receive external fabric image-related input signals, one of which is a signal input terminal, and the other is a ground terminal; SIG_IN is the input signal network identifier. R1 is a series resistor with a resistance of 100Ω, used to limit the input current and suppress transient surges and high-frequency interference. D1 is a transient suppression diode, model SMBJ5.0A, with one end connected to the input node and the other end grounded, used to clamp and protect the input terminal from surges, voltage spikes, or electrostatic discharge. R2 is a resistor with a resistance of 10kΩ, used to bias or pull down the relevant control node to prevent false triggering due to floating input. U1 is a level conversion chip, model 74LVC245, used to achieve level matching and buffered driving between the interface signal and the controller signal.
[0027] A1-A4 are the A-side channel pins, B1-B4 are the B-side channel pins, DIR is the direction control terminal, used to set the signal transmission direction, labeled A→B in the diagram, indicating that the signal is transmitted from the interface side to the controller side, OE is the enable control terminal, shown as low-level enable in the diagram, VCC is the power supply terminal, and GND is the ground terminal. C1 is a decoupling capacitor with a capacitance of 0.1μF, connected between the +3.3V power supply and ground, used to filter power supply ripple and stabilize the chip power supply. R3 is the output series resistor with a resistance of 33Ω, used to limit the output current, reduce signal edges, and suppress reflections and electromagnetic interference. MCU1 is a microcontroller used as a downstream controller, its VCC terminal is connected to the +3.3V power supply, and its GND terminal is grounded. GPIO and PA0 are general purpose input / output pins used to receive the interface signals converted by U1. Overall, this circuit completes the input protection, level conversion, and reliable reception of the fabric image interface signal by the microcontroller.
[0028] Through the above description of the nonwoven fabric defect location and removal system structure, defect identification results, and fabric image interface input and level conversion control circuit, the implementation basis of this invention embodiment in terms of hardware configuration, signal acquisition, data transmission, and execution control can be clearly understood. Specifically, the system structure provides the functional carrier for method execution, the defect identification results verify the detection effect of abnormal areas on the fabric surface, and the interface circuit provides circuit support for stable image signal input and reliable controller processing. Under this premise, this invention embodiment provides, as follows: Figure 1 The flowchart shown is for a nonwoven fabric defect location and removal method based on a self-calibration model. The processing flow of this method may include the following steps: S1: Because the stable reference information for position correction is inconsistent in the fabric images acquired by the industrial camera CAM1 at different installation positions, with different framing ranges, and under different fabric operating conditions, this embodiment divides the positioning processing of the fabric images into two scenarios: Scenario 1 is a scenario where the edge of the fabric is fully presented in the image, and the edge contour is continuous and clear. In this scenario, the edge of the fabric itself can stably represent the actual position and tilt state of the current fabric, so edge reference positioning is adopted.
[0029] Scenario 2 is a scenario where the edge of the fabric is difficult to extract stably due to occlusion, curling, wrinkles, brightness fluctuations, or limited field of view, but there are identifiable marks on the fabric. In this scenario, the pre-set marks can more stably represent the current position of the fabric, so mark reference point positioning is used.
[0030] The main controller can first determine the edge integrity, edge continuity, and marker recognizability in the current fabric image, and then select the corresponding positioning processing method.
[0031] Furthermore, relevant benchmark information and judgment parameters can be obtained through calibration. Specifically, during the equipment installation and commissioning phase, the main controller controls the industrial camera CAM1 to acquire calibration images of the nonwoven fabric under normal conveying conditions. For edge benchmark positioning, the main controller fits the fabric edges in the calibration images to obtain benchmark edge lines. For marker reference point positioning, the main controller identifies the marked areas in the calibration images to obtain the coordinates of the benchmark reference points. At the same time, the main controller can also combine multiple frames of calibration images to statistically analyze the spacing between edge segments, direction changes, color range, grayscale range, contour size range of marked areas, and coordinate change range between adjacent frames under normal conditions. This yields connection distance thresholds, direction difference thresholds, color filtering ranges, grayscale filtering ranges, contour conditions, position change thresholds, and coordinate deviation thresholds for subsequent online positioning.
[0032] Scenario 1: Edge reference positioning.
[0033] When the current fabric image meets the condition for complete edge positioning, the main controller performs edge reference positioning. First, the main controller performs grayscale conversion and contrast enhancement on the fabric image acquired by the industrial camera CAM1 to improve the grayscale difference between the fabric edge and the background area. Based on the area where the fabric edge is located determined in the calibration stage, the main controller delineates an edge search area near the left and right edges or front and back edges of the fabric, and performs edge enhancement processing and edge extraction processing on the enhanced fabric image within this edge search area to obtain the edge contour data in the fabric image.
[0034] Next, the main controller performs breakpoint connection and smoothing on the edge contour data to obtain a stable edge curve. Breakpoint connection refers to splicing adjacent edge segments whose endpoint spacing is less than the connection distance threshold and whose endpoint direction difference is less than the direction difference threshold. Smoothing refers to performing moving average processing or curve smoothing on the spliced edge curve to reduce edge jitter caused by fiber burrs, local wrinkles and brightness fluctuations.
[0035] Then, the main controller fits the edge line to the stable edge curve to obtain the edge baseline, and calculates the position offset and angle offset between the edge baseline and the baseline edge line obtained in the calibration stage; wherein, the position offset is used to characterize the coordinate difference between the current edge baseline and the baseline edge line in the horizontal and / or vertical directions, and the angle offset is used to characterize the tilt angle difference between the current edge baseline and the baseline edge line.
[0036] Finally, the main controller performs translation correction on the fabric image based on the position offset and rotation correction on the fabric image based on the angle offset to obtain the target fabric image at the corresponding reference edge position.
[0037] Scenario 2: Marking reference points for positioning.
[0038] When the current fabric image does not meet the condition for complete edge positioning but fabric markings can be identified, the main controller performs mark reference point positioning. First, during the equipment installation and commissioning phase, marks can be set in the non-woven fabric edge area, at fixed intervals, or in a designated detection area. The main controller extracts the center coordinates of these marks based on the calibration image, using them as the reference point coordinates. During online operation, the main controller defines a search window in the current fabric image based on the reference point coordinates and the location of valid mark points from the previous sampling period. When the system has just started and has not yet obtained valid mark points from the previous sampling period, the area where the reference point coordinates are located is used as the center of the search window. Subsequently, the main controller identifies the marked areas in the fabric image within this search window and filters the areas based on their corresponding color, grayscale, or contour features to obtain the marked areas to be analyzed.
[0039] Next, the main controller performs binarization segmentation and connected component analysis on the region to be analyzed to obtain candidate marker points. Binarization segmentation refers to separating the marker pixels from the background pixels according to the grayscale threshold, while connected component analysis refers to aggregating adjacent target pixels into candidate regions and using the center coordinates of each candidate region as candidate marker points.
[0040] Then, the main controller performs validity screening on the candidate marker points and records the candidate marker points that meet the position continuity condition as valid marker points. The position continuity condition means that the coordinate change of the candidate marker point in two consecutive sampling periods or multiple consecutive frames of cloth image is less than the position change threshold, and the coordinate deviation between the candidate marker point and the position of the valid marker point or the reference point in the previous sampling period is less than the deviation threshold. This condition is used to exclude random noise points, reflective points, and misidentified points.
[0041] Finally, the main controller uses the center coordinates of the valid marker points as the marker reference points. When there are multiple valid marker points, the main controller selects the valid marker point with the smallest distance from the valid marker point position of the previous sampling period as the current marker reference point. When no position of the previous sampling period is available, the valid marker point with the smallest distance from the reference point position is selected as the current marker reference point. The main controller calculates the coordinate offset between the marker reference point and the reference reference point, and performs horizontal and / or vertical translation correction on the fabric image based on the coordinate offset to obtain the target fabric image at the corresponding reference position.
[0042] S2: First, during the equipment calibration stage, standard fabric images corresponding to qualified nonwoven fabric samples are acquired. The standard fabric images are partitioned according to the same grid division method as the images to be tested. Texture energy features, texture direction features, and texture uniformity features corresponding to each standard partition are extracted. The above features corresponding to each standard partition are combined to form a texture feature template. The texture feature template is used to characterize the standard texture state of the normal fabric surface at different positions.
[0043] During online detection, the main controller divides the target fabric image into several texture analysis sub-regions according to a preset grid division method. The preset grid division method refers to dividing the target fabric image into several rows and columns of rectangular regions according to the horizontal and vertical dimensions of the fabric image. Each texture analysis sub-region has corresponding region center coordinates.
[0044] Then, the main controller extracts the texture energy features, texture direction features, and texture uniformity features corresponding to each texture analysis sub-region, and compares them with the texture feature templates at the corresponding locations. Among them, the texture energy feature is used to characterize the strength of texture changes in the sub-region, and can be obtained by statistically analyzing the sum of gray-level gradient magnitudes, the mean of local filter response intensity, or the texture response energy value of each pixel in the texture analysis sub-region; the texture direction feature is used to characterize the dominant direction of fiber arrangement or texture extension in the sub-region, and can be obtained by statistically analyzing the gradient direction distribution of each pixel in the texture analysis sub-region and extracting the dominant direction angle; the texture uniformity feature is used to characterize the consistency of texture distribution in the sub-region, and can be obtained by the gray-level variance, gray-level co-occurrence matrix uniformity, or local texture consistency index in the texture analysis sub-region.
[0045] Furthermore, the main controller calculates the feature deviations of each texture analysis sub-region in three dimensions: texture energy, texture direction, and texture uniformity. The texture energy deviation can be obtained from the absolute or relative difference between the texture energy feature of the current texture analysis sub-region and the template texture energy feature in the corresponding texture feature template. The texture direction deviation can be obtained from the angle difference between the principal direction angle of the current texture analysis sub-region and the principal direction angle of the template. The texture uniformity deviation can be obtained from the absolute difference between the texture uniformity feature of the current texture analysis sub-region and the template texture uniformity feature. To ensure the comparability of the deviations in the three dimensions, the main controller first normalizes the texture energy deviation, texture direction deviation, and texture uniformity deviation to convert them all to a uniform numerical range.
[0046] Based on this, the main controller performs vector synthesis processing on the feature deviations of each texture analysis sub-region in three dimensions to obtain the sub-region offset of each texture analysis sub-region. The vector synthesis processing refers to taking the normalized texture energy deviation, texture direction deviation, and texture uniformity deviation as the three components of the three-dimensional deviation vector, and obtaining the sub-region offset of the corresponding texture analysis sub-region by calculating the magnitude of the three-dimensional deviation vector. The sub-region offset is used to characterize the overall deviation of the texture analysis sub-region relative to the corresponding template region.
[0047] Subsequently, the main controller performs surface fitting based on the region center coordinates and corresponding sub-region offsets of each texture analysis sub-region to obtain a continuously distributed texture offset field. Here, surface fitting refers to using the region center coordinates of each texture analysis sub-region as planar coordinate input and the corresponding sub-region offset as height value, and constructing a continuous offset distribution surface covering the entire target fabric image through bilinear interpolation, bicubic interpolation, polynomial fitting, or spline surface fitting. The texture offset field is used to characterize the continuous change of the degree of texture deviation at each position in the target fabric image.
[0048] Then, the main controller performs local extremum search on the texture offset field, extracting local maxima, local minima, and their spatial distribution. The local extremum distribution characterizes the location and clustering of regions with abrupt changes in texture deviation within the target fabric image. The main controller can further analyze the number of local maxima, peak values, distance between adjacent extrema, and the coverage of high-offset regions to characterize the concentration and extent of fabric texture anomalies.
[0049] Finally, the main controller determines the texture offset corresponding to the target fabric image based on the overall offset level and local extremum distribution results of the texture offset field. Specifically, the texture offset can be determined by one or more of the maximum offset value, average offset value, and local maximum distribution density in the texture offset field, which is used to characterize the comprehensive deviation of the overall texture state of the current target fabric image from the standard texture state.
[0050] S3: To adaptively correct texture distortion and positioning deviations caused by factors such as fabric vibration, tension fluctuations, lighting changes, conveying errors, and actuator response lag during continuous nonwoven fabric conveying, a self-calibration model for texture correction and position compensation is pre-constructed. The basic framework of the self-calibration model includes at least an input layer, a feature fusion layer, a deviation distribution output layer, and a compensation parameter output layer. Specifically, the input layer receives the texture offset, operating parameters, and historical deviation data corresponding to the target fabric image; the feature fusion layer establishes the correspondence between the texture offset state, the current operating state, and the historical deviation evolution; the deviation distribution output layer outputs the texture deviation distribution results corresponding to each texture analysis sub-region in the current fabric image; and the compensation parameter output layer outputs the texture compensation parameters and position compensation parameters corresponding to each texture analysis sub-region.
[0051] Furthermore, the self-calibration model can be pre-built and trained before the system goes online. Specifically, firstly, multiple sets of fabric images, conveying parameters, and operating condition parameters are collected under different working conditions such as normal conveying, tension fluctuations, speed changes, light changes, and response lag of the actuator. Combined with the standard sample alignment results, manual review results, and post-cleaning re-inspection results, a corresponding training sample set is constructed. Then, the target fabric images in each training sample are divided into several texture analysis sub-regions, and the texture offset corresponding to each texture analysis sub-region is extracted. The conveying speed, tension value, light intensity, historical deviation data, and actual position deviation results are used as model inputs, and the texture deviation distribution results, texture compensation parameters, and position compensation parameters are used as model output labels. The self-calibration model is iteratively trained and its parameters are updated.
[0052] like Figure 8In the schematic diagram of the training convergence results of the self-calibration model shown, the horizontal axis represents the number of iterations, the left vertical axis represents the loss value, and the right vertical axis represents the average accuracy. The blue curve indicates that the training loss gradually decreases with the increase of the number of iterations and eventually tends to stabilize, while the red curve indicates that the average accuracy gradually increases with the increase of the number of iterations and eventually stabilizes at a high level. This shows that the constructed self-calibration model has established a mapping relationship between texture offset state, operating conditions, and compensation parameters, and has reached a convergence state that can be used for online correction and positioning compensation.
[0053] During online operation, the texture offset is input into a preset self-calibration model. Based on the self-calibration model, deviation learning analysis is performed on the input texture offset to obtain the texture deviation distribution result corresponding to the current fabric image. The texture deviation distribution result is used to characterize the degree of local texture distortion and its distribution location corresponding to different texture analysis sub-regions in the target fabric image. Specifically, the main controller first inputs the texture offset obtained in the current sampling period along with the corresponding transport parameters, tension parameters, illumination parameters, and historical deviation data into the self-calibration model. The self-calibration model outputs the deviation distribution value of each texture analysis sub-region based on the correlation between the current texture offset state and the historical compensation state, thereby forming a texture deviation distribution result that corresponds one-to-one with the target fabric image.
[0054] Based on the texture deviation distribution results, the texture compensation parameters and position compensation parameters corresponding to each texture analysis sub-region are calculated. The texture compensation parameters are used to correct local texture distortion in the target fabric image and may include at least one or more of the following: 1. Directional compensation amount, used to correct the deviation of the main texture direction relative to the standard texture direction within the current texture analysis sub-region.
[0055] 2. Energy compensation amount, used to correct the deviation of the texture intensity in the current texture analysis sub-region from the standard texture intensity.
[0056] 3. Uniformity compensation amount, used to correct problems such as uneven gray-level distribution, abnormal local contrast, or inconsistent local texture thickness within the current texture analysis sub-region.
[0057] IV. Position fine-tuning amount, used to correct local misalignment of the texture structure within the current texture analysis sub-region relative to the standard texture template.
[0058] Position compensation parameters are used to correct the image coordinates in the target fabric image and compensate for positional deviations during subsequent erasure and positioning processes. These parameters may include at least one or more of the following: 1. Lateral coordinate compensation, used to correct the positional deviation of the defect area in the lateral direction of the image.
[0059] 2. Vertical coordinate compensation amount, used to correct the positional deviation of the defect area in the vertical direction of the image.
[0060] 3. Angle compensation amount, used to correct the overall or partial deviation of the fabric image relative to the reference direction.
[0061] IV. Timing lag compensation amount, used to convert the response lag time of the actuator into a position compensation distance along the fabric conveying direction.
[0062] 5. Proportional compensation amount, used to correct the mapping deviation between image coordinates and actual physical positions caused by tension changes or slight changes in image scale.
[0063] The local texture distortion in the target fabric image is corrected based on the texture compensation parameters to obtain the texture correction result, which is used to correct the texture distortion in the current fabric image.
[0064] Specifically, the main controller can perform the following correction operations for each texture analysis sub-region: When the orientation compensation is not zero, a local orientation rotation adjustment is performed on the texture analysis sub-region to bring the main texture direction in the sub-region closer to the standard texture direction. When the energy compensation is not zero, a local texture intensity adjustment is performed on the texture analysis sub-region by changing the local gray-level gradient response or local contrast to make the current texture intensity closer to the standard texture intensity. When the uniformity compensation is not zero, a local smoothing, enhancement, or equalization process is performed on the texture analysis sub-region to reduce distortion caused by abnormal local gray-level distribution and uneven texture thickness. When the position fine adjustment is not zero, a small translation adjustment is performed on the local texture structure in the texture analysis sub-region to make the texture position distribution in the sub-region more consistent with the standard texture template.
[0065] After each texture analysis sub-region has completed local correction, it is then stitched together according to the original grid position to form the corresponding texture correction result.
[0066] The image coordinates in the target fabric image are corrected based on the position compensation parameters to obtain the positioning compensation result. The positioning compensation result is used to compensate for the position deviation caused by fabric conveying errors, tension changes, and actuator response lag. If there is actuator response lag, the displacement compensation distance along the conveying direction is calculated based on the current fabric conveying speed and the timing lag compensation amount, and this displacement compensation distance is superimposed on the corrected image coordinates to obtain the final positioning compensation result.
[0067] In other words, the positioning compensation result not only corrects the static position deviation in the current image, but also corrects the dynamic displacement deviation between the image acquisition time and the actual cleaning time, so that the subsequent cleaning execution device can act more accurately on the target defect area.
[0068] In one embodiment, to improve the model's adaptability to new operating conditions during online operation, the self-calibration model can be iteratively updated based on the re-inspection results after removal. Specifically, after defect removal is completed, an industrial camera re-captures a re-inspection image of the removed area and determines whether the target defect in the re-inspection image has been completely removed and whether the removal position has shifted. If the re-inspection results indicate that residual defects still exist or the removal position deviation exceeds the preset tolerance range, the texture offset of the current sampling period, operating condition parameters, compensation parameters output by the model, and re-inspection error results are used together as incremental samples and fed back to the self-calibration model to retrain or fine-tune the model parameters online, thereby gradually improving the model's compensation accuracy under different conveying states, different tension states, and different execution lag states.
[0069] S4: The main controller uses each detection position in the corrected target fabric image as the center, and extracts a local texture image block according to a preset sliding window size. It then moves the sliding window point by point along the transverse and conveying directions of the fabric to continuously scan the entire target fabric image. The sliding window size can be preset based on the nonwoven fabric texture period, the minimum defect size, and the industrial camera resolution. For each local texture image block corresponding to a detection position, the main controller calculates its texture direction features and grayscale variation features. The texture direction features are obtained by statistically analyzing the directional distribution of each pixel within the local texture image block under the influence of transverse and longitudinal gradients, with the mean of the directional gradient representing the main texture extension direction at the detection position. The grayscale variation features are obtained by calculating the dispersion of pixel grayscale within the local texture image block, with the grayscale variance representing the brightness fluctuations and local texture variation at the detection position. Simultaneously, the main controller retrieves the reference texture data corresponding to the current detection position, obtaining the corresponding reference directional gradient mean and reference grayscale variance. The reference texture data can be pre-extracted and stored from qualified nonwoven fabric samples at the same coordinate position.
[0070] After obtaining the texture direction features and grayscale variation features of the current detection location, the main controller calculates the absolute value of the difference between the mean gradient of the current detection location and the mean gradient of the corresponding reference direction, as well as the ratio of the grayscale variance of the current detection location to the reference grayscale variance. These two values are then combined using preset weighting coefficients to obtain the texture deviation parameter. The preset weighting coefficients are pre-calibrated based on the influence of orientation distortion and grayscale fluctuations on defect identification results in historical samples. When changes in the nonwoven fabric fiber orientation are more sensitive to defect identification, the weight of the orientation deviation term can be increased; when local grayscale anomalies caused by stains, foreign objects, or indentations are more pronounced, the weight of the grayscale variation term can be increased. Therefore, the texture deviation parameter comprehensively reflects the degree of deviation of the current detection location from the standard texture state in terms of both texture direction and grayscale distribution, and a larger value indicates a higher likelihood of abnormal texture at that location.
[0071] After obtaining the texture deviation parameters at each detection location, the main controller records the detection locations where the texture deviation parameters are greater than a preset abnormal texture threshold as abnormal candidate points, and aggregates these candidate points according to the principle of adjacent location connectivity. This adjacent location connectivity principle can employ a four-neighbor connectivity method or an eight-neighbor connectivity method, or it can be based on whether the Euclidean distance between candidate points is less than a preset aggregation distance threshold. For the candidate regions formed after aggregation, the main controller further calculates their area, length, width, and the average value of their internal texture deviation parameters, and removes noise regions with an area smaller than a preset minimum area threshold or a length smaller than a preset minimum length threshold, thereby retaining the true abnormal texture regions.
[0072] After identifying the abnormal texture region, the main controller extracts the bounding box or minimum envelope contour of each abnormal texture region and calculates the coordinates of the region center point based on the horizontal and vertical coordinates of all pixels in the abnormal texture region. At the same time, the main controller calculates the left, right, upper and lower boundaries of the abnormal texture region in the image to determine the region boundary range.
[0073] Subsequently, during the equipment installation and commissioning phase, the industrial camera is pre-calibrated to obtain camera intrinsic parameters, extrinsic parameters, pixel size, camera mounting height, mounting angle, and lens distortion correction parameters, and to establish a coordinate mapping relationship between the image coordinate system and the fabric detection coordinate system. During online operation, the main controller, based on this coordinate mapping relationship, converts the coordinates of the center point and boundary range of the abnormal texture area from pixel coordinates to the actual position coordinates in the fabric detection coordinate system. The fabric detection coordinate system can be defined as a two-dimensional coordinate system with the fabric width direction as the horizontal axis and the fabric conveying direction as the vertical axis. The initial position coordinates obtained after the conversion reflect the actual fabric position of the abnormal texture area at the corresponding moment at the detection station.
[0074] Furthermore, the relative installation distance between the detection station and the cleaning station can be pre-measured after the equipment is installed and recorded as a fixed reference distance; the conveying speed can be obtained in real time by the encoder, the encoder displacement can be calculated by the cumulative pulse amount of the encoder from the detection time to the expected cleaning trigger time, and the tension change parameter can be determined by the difference between the current tension value detected by the tension sensor and the reference tension value. The main controller calculates the position migration amount of the abnormal texture area in the conveying direction when it moves from the detection station to the cleaning station based on the fixed reference distance, the encoder displacement, and the fabric elongation or shrinkage correction amount caused by tension changes; among them, the correction amount caused by tension changes can be determined based on the pre-established tension and position compensation relationship. When the tension increases, the fabric may stretch, causing the actual position of the abnormal area to reach the cleaning station to move relatively backward; when the tension decreases, the fabric may shrink, causing the actual position of the abnormal area to reach the cleaning station to move relatively forward.
[0075] The tension-position compensation relationship refers to the mapping relationship between the elastic deformation of the fabric caused by tension changes during conveying and its positional offset in the conveying direction. This relationship is typically established through offline calibration experiments: first, a baseline tension value is set, the conveying equipment is run under this tension, the fixed baseline distance between the detection station and the cleaning station is recorded, and the theoretical displacement is calibrated using an encoder; then, the fabric tension is sequentially changed to different preset values, and the deviation between the actual reference point position and the theoretical position at the cleaning station is measured under the same conveying distance; the tension change is used as input, and the positional offset as output, establishing a one-to-one corresponding compensation curve. Considering the differences in fabric material, thickness, and elastic modulus, independent tension-position compensation relationships need to be established for different batches or types of fabric to ensure compensation accuracy. In actual operation, the main controller calculates the corresponding position correction amount based on the real-time collected tension change value through this compensation relationship by looking up a table or interpolation. This correction is used to correct the position migration calculation of abnormal areas, thereby eliminating the impact of tension fluctuations on the cleaning positioning accuracy.
[0076] After obtaining the position migration amount, the main controller first superimposes the position migration amount onto the conveying direction coordinate value of the initial position coordinate to reflect the position change of the abnormal texture area as the fabric moves; then, it calculates the time delay displacement compensation amount based on the current conveying speed and the response delay of the actuator, and superimposes it onto the corrected conveying direction coordinate value; at the same time, based on the fabric elongation or retraction caused by tension changes, it further corrects the lateral coordinate value and / or longitudinal coordinate value of the target position coordinate to obtain the physical position coordinate of the abnormal texture area at the cleaning station.
[0077] Through the above processing, not only can image recognition and region extraction of abnormal texture areas be achieved, but the image coordinates under the detection station can also be accurately converted into the actual physical position coordinates under the cleaning station. This compensates for the positioning deviation caused by fabric feeding errors, tension fluctuations and actuator response lag, thereby improving the accuracy and stability of non-woven fabric defect positioning and cleaning.
[0078] S5: The main controller first reads the set of physical position coordinates corresponding to each abnormal texture region at the cleaning station, and then extracts the start and end position coordinates of the abnormal texture region in the fabric conveying direction, as well as the minimum and maximum boundary position coordinates in the fabric width direction, based on this set of physical position coordinates. The continuous distribution length represents the distance difference between the start and end position coordinates of the same abnormal texture region along the fabric conveying direction, characterizing the extent of the abnormal texture region's extension in the fabric travel direction; the distribution range represents the coverage interval between the minimum and maximum boundary position coordinates of the same abnormal texture region in the fabric width direction, characterizing the extent of the abnormal texture region's expansion in the lateral width.
[0079] To avoid the influence of different dimensions and value ranges on subsequent judgments, the main controller normalizes the continuous distribution length and distribution range separately. Specifically, it uses extreme value normalization to map the continuous distribution length to the [0,1] interval and the distribution range to the same interval, ensuring they are on the same comparable scale. Since the continuous distribution length and distribution range respectively characterize the extent of the abnormal texture region's extension in the fabric conveying direction and width direction, and are orthogonal dimensions with different physical meanings, direct multiplication would introduce dimensional conflicts and weight imbalances, making it difficult for the fusion result to accurately reflect the comprehensive distribution characteristics of the abnormal region. Therefore, after normalization, weight coefficients α and β are set to correspond to the contribution proportions of the continuous distribution length and distribution range, respectively, with α+β=1. The two normalized parameters are then weighted and summed to obtain the comprehensive distribution value. This index is dimensionless and ranges from [0,1], reasonably characterizing the comprehensive distribution severity of the abnormal texture region in the two-dimensional plane, providing a unified quantitative basis for subsequent removal priority determination and removal strategy selection.
[0080] The aforementioned weighting coefficients are set through a combination of offline calibration and online adaptive methods. During offline calibration, multiple sets of abnormal texture samples with typical distribution characteristics are selected, and different weight combinations are used to calculate the comprehensive distribution value. Correlation analysis is then performed with the severity levels judged manually, and the weight combination with the highest correlation coefficient is selected as the initial calibration value. Simultaneously, the removal success rate is used for verification and optimization. During the online operation phase, the system dynamically adjusts the weights based on real-time detection and removal feedback: when frequent anomalies in a certain dimension lead to poor removal results, the weighting coefficient of the corresponding dimension is automatically increased to achieve process adaptability optimization. The weighting coefficients are fixed in the control parameter table, and a manual correction interface is provided to facilitate fine-tuning by on-site personnel based on fabric characteristics, batch differences, or process changes.
[0081] When the overall distribution value is less than the preset distribution threshold, it indicates that the abnormal texture area has a small extension in the fabric conveying direction and a limited coverage area in the width direction. The main controller determines it as a locally dispersed defect area and controls the cleaning execution device to use a single-point cleaning method for the abnormal texture area. Specifically, the main controller generates a single-action control command based on the coordinates of the center position of the abnormal texture area, driving the cutter, blower head, or spray head to perform a one-time cleaning action at the corresponding target position, so as to achieve rapid removal of abnormal texture areas that are small in area and concentrated in location.
[0082] When the overall distribution value is not less than the preset distribution threshold, it indicates that the abnormal texture area has a long continuous distribution length along the fabric conveying direction, or a large expansion range in the fabric width direction. The main controller determines it as a continuously expanding defect area and controls the cleaning execution device to use a segmented continuous cleaning method for the abnormal texture area. Specifically, the main controller can generate multiple segmented cleaning positions along the extension path of the abnormal texture area in the conveying direction, and set the effective range of each segment action in combination with the coverage range of the abnormal texture area in the width direction, so that the cleaning execution device performs multiple cleaning actions in sequence according to a preset pitch, preset timing, or continuous trajectory, thereby covering the entire continuously expanding defect area.
[0083] After the corresponding removal method is used to complete the removal, the industrial camera re-acquires images of the corresponding area after removal. The main controller performs anomaly residual analysis on the currently removed area in the re-inspection image to obtain the residual area parameter and residual texture deviation parameter within the area. The residual area parameter characterizes the size of the remaining abnormal area after removal; the larger the abnormal area, the higher the degree of residual deviation. The residual texture deviation parameter characterizes the degree of residual deviation of the texture state of the removed area relative to the baseline texture state. The specific method for obtaining the residual texture deviation parameter is as follows: The main controller first extracts texture features from the currently removed area, using methods such as gray-level co-occurrence matrix or Gabor filter, to generate a feature vector that characterizes the texture distribution of the area. Then, this feature vector is compared with the pre-stored baseline texture feature vector, and the cosine distance or Euclidean distance between them is calculated. This distance value reflects the degree of deviation of the texture of the currently removed area relative to the baseline texture. Finally, the calculated distance value is compared with a preset texture deviation upper limit or mapped to the [0,1] interval using an extreme value normalization method to obtain the residual texture deviation parameter. The closer this parameter is to 0, the higher the degree of texture restoration; the closer it is to 1, the more obvious the residual abnormal texture features.
[0084] The normalized residual area parameter and residual texture deviation parameter are multiplied to obtain a residual removal evaluation value that characterizes the degree of abnormal texture residue within the current removal area. The normalization method is as follows: the residual area parameter is normalized with a preset maximum allowable residual area as the upper limit, and the ratio of the actual residual area to the maximum allowable area is used as the normalized value; the residual texture deviation parameter is normalized based on a preset upper limit of texture deviation, and the ratio of the actual texture deviation degree to the maximum allowable deviation degree is used as the normalized value. Both parameters are on the same dimension after normalization. The residual removal evaluation value obtained through the product operation is dimensionless, ranging from 0 to 1. The closer the value is to 0, the better the removal effect; the closer it is to 1, the higher the degree of residue.
[0085] If the residual removal evaluation value is not greater than the preset residual removal threshold, it indicates that the abnormal textures in the current removal area have been basically removed, and the main controller determines that the corresponding single-point removal or segmented continuous removal has been completed. Conversely, if the residual removal evaluation value is greater than the preset residual removal threshold, it indicates that there are still obvious abnormal residues in the current removal area, and the main controller determines that the corresponding single-point removal or segmented continuous removal has not been completed, and triggers the removal control again.
[0086] Furthermore, during the subsequent cleanup control process, the main controller continuously monitors the residual changes corresponding to the currently cleaned area. These residual changes represent the degree to which the residual evaluation value after the current round of cleanup decreases compared to the residual evaluation value after the previous round. If the residual evaluation value re-acquired after the subsequent cleanup control is still higher than the preset residual cleanup threshold, it indicates that the current abnormal texture area is difficult to remove effectively through conventional repeated cleanup methods. At this time, the main controller outputs an abnormal cleanup warning message.
[0087] The preset distribution threshold is determined by summing and averaging the historical comprehensive distribution values corresponding to historical scattered defect area samples and continuously expanding defect area samples. The preset residual removal threshold is determined by summing and averaging the historical residual removal evaluation values corresponding to historical samples that have been completely removed and samples that have not been completely removed. Abnormal removal early warning information can be sent to the alarm or host computer to prompt operators to further check the status of the removal actuator, the fabric running status, or the nature of the abnormal area, thereby avoiding missed removal, over-removal, or fabric damage caused by continuous residue, repeated removal failure, or actuator malfunction.
[0088] The above method can adaptively select the removal method according to the longitudinal extension and lateral coverage of the abnormal texture area, and verify the removal effect through re-inspection after removal, thereby improving the pertinence, completeness and stability of non-woven fabric defect location and removal.
[0089] To implement the aforementioned method for locating and removing defects in nonwoven fabrics based on a self-calibration model, this invention further provides a corresponding nonwoven fabric defect location and removal system based on a self-calibration model, such as... Figure 2 The diagram shown illustrates the structure of a nonwoven fabric defect localization and removal system based on a self-calibration model. This system may include: The image acquisition and positioning module is used to acquire images of the fabric surface during the continuous conveying process of nonwoven fabric, and to perform position correction by combining the edge information or marker information in the current image; when the fabric edge is clear and complete, edge reference positioning is performed; when the fabric edge is unstable but there are preset markers, marker reference point positioning is performed, thereby obtaining the target fabric image for fabric texture offset analysis.
[0090] The texture offset extraction module is used to partition the target fabric image according to a preset analysis method, extract texture direction features, texture energy features, and texture uniformity features that characterize the fiber arrangement state of the fabric, and match and compare them with the corresponding reference texture feature templates to calculate the deviation results corresponding to each analysis area, thereby obtaining the texture offset amount that reflects the current degree of fabric texture deviation.
[0091] The self-calibration compensation module is used to input the texture offset into the preset self-calibration model, and perform deviation learning analysis in combination with the transmission parameters, operating condition parameters and historical deviation data. It outputs the texture deviation distribution results, texture correction results and positioning compensation results corresponding to the current fabric state, so as to achieve adaptive correction of fabric texture distortion and position deviation.
[0092] The abnormal region localization module is used to identify and extract abnormal texture regions in the target fabric image based on the texture correction results, obtain the coordinates of the center point of the region and the boundary range of the region corresponding to the abnormal texture region, and combine the localization compensation results to convert them from image coordinates to physical location coordinates at the cleaning station, thereby determining the actual cleaning location of the abnormal texture region.
[0093] The removal method determination module is used to determine the continuous distribution length of the abnormal texture area in the fabric conveying direction and the distribution range in the fabric width direction based on the physical position coordinates of the abnormal texture area at the removal station, and to determine the removal method based on the comprehensive distribution of the two to determine whether to use a single fixed-point removal method or a segmented continuous removal method.
[0094] It should be added that, such as Figure 6 The diagram shows a closed-loop control block diagram for defect location and removal in nonwoven fabric based on a self-calibration model. This diagram illustrates the relationship between the main setpoint, main measurement, compensation adjustment, and process execution in the system of this invention using a closed-loop control approach. In the diagram, the main setpoint r(t) represents the main setpoint quantity for the fabric surface, i.e., the target fabric surface state without abnormalities; the main measurement y(t) represents the real-time feedback quantity of the fabric surface texture state; the two are compared at the summation node to form a deviation input to the main controller. The main controller can be implemented by the main control MCU, used to generate the removal strategy and output the secondary measurement value m(t). The self-calibration model outputs the compensation parameter θ(t) based on m(t), and the controller performs deviation learning and compensation solution on the compensation parameter to output the adjustment quantity c(t). Here, c(t) represents the correction or removal control quantity for the abnormal texture area, and e(t) represents the secondary controlled quantity, i.e., the residual deviation quantity of the abnormal texture state area of the fabric surface. The nonwoven fabric process consists of an actuator and a continuous nonwoven fabric conveying and removal execution process. The main measurement transmitter uses a camera plus positioning detection to complete the texture and position feedback.
[0095] like Figure 7The diagram shown illustrates the monitoring of abnormal texture counts. This diagram is a control chart (c) used to characterize the fluctuation of abnormal texture counts during nonwoven fabric inspection. The horizontal axis represents the sample number or time sequence, and the vertical axis represents the number of inconsistencies (c), in units of one detection window, indicating the number of abnormal textures identified within each detection window. The black line with black dots represents the changing trend of the actual number of abnormal textures (c) corresponding to each detection window, reflecting the fluctuation of fabric quality over time. The green horizontal line is the center line CLc, representing the average line or target texture quality level. The two red dashed lines represent the upper control limit UCLc and the lower control limit LCLc, respectively. UCLc = 10.33, which indicates the upper boundary of the allowable fluctuation in the number of abnormal textures, and LCLc = 0, which indicates that the lower boundary is zero. As can be seen from the figure, all sample points are located between the upper and lower control limits, indicating that although the number of abnormal textures on the fabric surface fluctuates, it is still under control overall.
[0096] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0097] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0100] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0101] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for locating and removing defects in nonwoven fabrics based on a self-calibration model, characterized in that, The method includes: S1. Acquire images of the fabric surface during the continuous conveying process of nonwoven fabric, and perform edge benchmark positioning or mark reference point positioning to obtain the target fabric surface image for fabric texture offset analysis. S2, extract the multidimensional texture features representing the fiber arrangement state of the target fabric image, and match them with the preset reference texture points to obtain the texture offset; The texture offset is used to characterize the degree of deviation of the current fabric texture from the reference texture in terms of orientation distribution, texture density, and texture position distribution. S3, the texture offset is input into the preset self-calibration model, and after deviation learning analysis and compensation parameter solution, the corresponding texture correction result and positioning compensation result are output; S4. Based on the texture correction result, determine the abnormal texture region in the target fabric image, and combine the positioning compensation result to determine the physical location coordinates of the abnormal texture region at the cleaning station. S5. Based on the regional distribution of the physical location coordinates, determine the clearing method. The clearing method includes at least single fixed-point clearing and segmented continuous clearing.
2. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 1, characterized in that, The edge reference positioning includes: Edge enhancement processing is performed on the acquired fabric images to extract edge contour data from the fabric images, and breakpoint connection and smoothing processing are performed to obtain stable edge curves; The edge baseline is obtained by fitting the stable edge curve with an edge line. Based on the positional and angular offsets between the edge baseline and the preset baseline edge line, the fabric image after edge enhancement is subjected to translational and angular corrections to obtain the target fabric image at the corresponding baseline edge position.
3. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 1, characterized in that, The positioning of the marker reference point includes: Identify the preset marked areas in the fabric image, and locate the areas based on the color features, grayscale features, or contour features corresponding to the preset marked areas to obtain the marked areas to be analyzed; Binarization segmentation and connected component analysis are performed on the region to be analyzed to obtain candidate marker points; Candidate marker points that meet the preset position continuity condition are recorded as valid marker points, and the center coordinates of the valid marker points are used as marker reference points; The fabric image is positionally corrected based on the marked reference points to obtain the target fabric image at the corresponding reference position.
4. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 2 or 3, characterized in that, The texture offset is obtained as follows: The target fabric image is divided into several texture analysis sub-regions according to a preset grid division method, and the texture energy features, texture direction features and texture uniformity features corresponding to each texture analysis sub-region are extracted respectively. Calculate the feature deviation of each texture analysis sub-region in three dimensions: texture energy, texture direction, and texture uniformity, and perform vector synthesis on the feature deviation in the three dimensions to obtain the sub-region offset of each texture analysis sub-region. Surface fitting is performed on the offset of each sub-region to obtain a continuously distributed texture offset field, and the local extreme value distribution of the texture offset field is extracted to obtain the texture offset corresponding to the target fabric image.
5. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 1, characterized in that, S3 includes: The texture offset is input into a preset self-calibration model, and the input texture offset is subjected to deviation learning analysis based on the self-calibration model to obtain the texture deviation distribution result corresponding to the current fabric image. The texture deviation distribution results are used to characterize the degree of local texture distortion and its distribution location corresponding to different texture analysis sub-regions in the target fabric image; Based on the texture deviation distribution results, solve for the texture compensation parameters and position compensation parameters corresponding to each texture analysis sub-region; Based on the texture compensation parameters, the local texture distortion in the target fabric image is corrected to obtain a texture correction result, which is used to correct the texture distortion in the current fabric image. The image coordinates in the target fabric image are corrected according to the position compensation parameters to obtain the positioning compensation result. The positioning compensation result is used to compensate for the position deviation caused by fabric conveying error, tension change and actuator response lag.
6. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 5, characterized in that, S4 includes: The target fabric image after correction is scanned point by point based on a preset sliding window, and the texture direction features and grayscale change features corresponding to each detection position are extracted. Calculate the texture deviation parameters corresponding to each detection location; The texture deviation parameter is obtained by weighting the absolute value of the difference between the mean directional gradient of the current detection position and the mean directional gradient of the corresponding reference position, and the ratio of the gray-level variance of the current detection position to the gray-level variance of the reference position. The detection locations corresponding to the texture deviation parameters that exceed the preset abnormal texture parameters are aggregated to obtain the abnormal texture regions in the target fabric image.
7. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 6, characterized in that, S4 further includes: Obtain the coordinates of the center point of the region corresponding to the abnormal texture region and the region boundary range; Based on the camera calibration parameters and the preset coordinate mapping relationship, the coordinates of the center point of the region and the boundary range of the region are transformed from the image coordinate system to the cloth detection coordinate system to obtain the initial position coordinates; The relative installation distance between the detection station and the cleaning station is obtained, as well as the conveying speed, coding displacement and tension change parameters of the nonwoven fabric at the current detection time. The position migration amount of the abnormal texture area from the detection station to the cleaning station is calculated. Based on the position compensation relationship between the position migration amount and the fabric conveying speed, tension change parameters and actuator response delay, the initial position coordinates are corrected to obtain the physical position coordinates of the abnormal texture area at the cleaning station.
8. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 1, characterized in that, S5 includes: Based on the physical location coordinates corresponding to each abnormal texture region, the continuous distribution length of the abnormal texture region in the fabric conveying direction and the distribution range in the fabric width direction are determined. The continuous distribution length represents the distance difference between the starting and ending coordinates of the same abnormal texture area along the fabric conveying direction; The distribution range represents the coverage interval between the minimum boundary position coordinates and the maximum boundary position coordinates of the same abnormal texture area in the fabric width direction; The normalized continuous distribution length and distribution range are combined and processed to obtain the comprehensive distribution value corresponding to the abnormal texture region.
9. The method for locating and removing defects in nonwoven fabrics based on a self-calibration model as described in claim 8, characterized in that, The S5 also includes: If the overall distribution value is less than the preset distribution threshold, the corresponding abnormal texture area is determined to be a localized scattered defect area, and the clearing execution device is controlled to use a single fixed-point clearing method for the abnormal texture area. Conversely, if the corresponding abnormal texture area is determined to be a continuously expanding defect area, the clearing execution device is controlled to use a segmented continuous clearing method for the abnormal texture area. After completing the removal using the corresponding removal method, the removal effect is verified, specifically as follows: The combined analysis results of the residual area parameter and the residual texture deviation parameter in the current clearing area are denoted as the clearing residual evaluation value, which characterizes the degree of abnormal texture residue in the current clearing area. If the residual removal evaluation value is not greater than the preset residual removal threshold, then it is determined that the corresponding single fixed-point removal or segmented continuous removal is completed. Conversely, if the corresponding single-point clearing or segmented continuous clearing is not completed, the clearing control is triggered again. At the same time, the residual changes in the current clearing area are monitored. If the clearing residual evaluation value obtained after the clearing control is again higher than the preset clearing residual threshold, an abnormal clearing warning message is output.
10. A nonwoven fabric defect location and removal system based on a self-calibration model, characterized in that, The system includes: The image acquisition and positioning module is used to acquire images of the fabric surface during the continuous conveying process of nonwoven fabric, and to perform edge benchmark positioning or mark reference point positioning to obtain the target fabric surface image for fabric texture offset analysis. The texture offset extraction module is used to extract multi-dimensional texture features that represent the fiber arrangement state of the target fabric image and match them with preset reference texture points to obtain the texture offset. The self-calibration compensation module is used to input the texture offset into the preset self-calibration model. After deviation learning analysis and compensation parameter solving, it outputs the corresponding texture correction results and positioning compensation results. The abnormal region localization module is used to determine the abnormal texture region in the target fabric image based on the texture correction result, and to determine the physical location coordinates of the abnormal texture region at the cleaning station in combination with the localization compensation result. The clearing method determination module is used to determine the clearing method based on the regional distribution of physical location coordinates.
