Method and system for optimizing routing parameters of suture line correction process

Abstract

The invention discloses a routing parameter optimization method and system for a suture line correction procedure, and relates to the technical field of visual data analysis. According to the scheme, light interference judgment is carried out on a suture line area image, whether multi-light interference reduction regulation is carried out or not is decided, preliminary detection is carried out on the edge of a suture line, and the accuracy of the suture line correction procedure is improved. Whether suture line edge detection optimization is carried out or not is determined; time sequence sample set standardization is carried out on the suture line edge points, whether time sequence sample set multi-stage combination filtering optimization is carried out or not is determined, and if the standardization result is not qualified, time sequence sample set multi-stage combination filtering optimization is carried out; on the contrary, PID control routing parameter self-adaptive optimization is carried out on the suture line routing, whether PID calibration frequency increasing operation is carried out or not is determined, and the problems that in the prior art, due to the fact that the precision of a control instruction sent based on a PID controller is reduced, an output continuous offset signal generates accumulated deviation, and the accuracy of the control instruction is reduced are solved. And finally, the accuracy of suture line routing correction is difficult to break through the bottleneck.

Description

Technical Field

[0001] This invention relates to the field of visual data analysis technology, and in particular to a method and system for optimizing the stitching parameters in a suture correction process. Background Technology

[0002] Since the Industrial Revolution, stitching technology has undergone profound changes, from hand sewing to mechanization, and then to automation and intelligence. With the invention and popularization of sewing machines, sewing efficiency has leaped forward. However, the pursuit of consistent stitching, strength, and aesthetics has always been the core driving force behind the evolution of manufacturing processes. This evolution is particularly evident in automotive parts manufacturing. In the automotive industry, the stitching of interior automotive parts is far more than a simple means of connection; it is a core detail concerning function, safety, and quality. From seat leather that needs to withstand long-term friction and deformation to materials requiring specific breathability and elasticity... From the fabric used in car armrest covers to the airbags that must maintain structural integrity under instantaneous impact to protect lives, the stitching on these components not only needs to withstand extreme tension, pressure, and fatigue, but also needs to adapt to complex usage environments such as temperature differences, light exposure, and chemical contact. Therefore, the automotive industry has put forward almost stringent standardization requirements for the stitching parameters (such as tension, stitch pitch, and trajectory accuracy) of the stitching, which are far greater than those for general textiles. It is precisely because of these stringent standardization requirements that it is especially important to ensure that every inch of stitching conforms to the preset safe stitching trajectory during the actual sewing process.

[0003] Therefore, stitch correction in the stitch correction process becomes a core link in ensuring the quality of stitching in automotive interior parts. The stitch correction process refers to the real-time correction of stitch deviations in the automated sewing process of fabric automotive parts (such as seat fabric, door panel covering fabric, headliner, etc.) through a combination of visual inspection and motion control. For example, the stitch correction process for the fabric stitching used in the armrest cover of a car can be achieved through existing fabric stitch correction mechanisms. First, when the pre-attached fabric armrest cover is placed on the armrest cover support platform, a push rod pushes a clamp to the rear end of the armrest cover, so that the clamp is in contact with the side of the fabric. Another push rod pushes its corresponding clamp to both sides of the armrest cover, so that the clamp is in contact with the side of the fabric, thereby aligning the fabric stitching with the edges of the two clamps, thus correcting the fabric stitching. However, traditional stitching parameter setting methods often rely on empiricism or simple trial and error, lacking systematicity and precision. Faced with the massive and diverse demand for sewing parts on modern automobile production lines, this method is difficult to adapt to different situations quickly and accurately, and is prone to sewing defects such as skipped stitches, broken stitches, and uneven stitches.

[0004] Based on the aforementioned industry pain points, introducing a stitching parameter optimization strategy has become an inevitable path to solve the problem of insufficient precision in the current stitching correction process for automotive interior parts. By constructing a parameter optimization model with material characteristics, product structure, and equipment performance as input variables, and combining it with real-time stitching deviation data from visual inspection feedback, dynamic and precise control of key parameters such as stitch length, thread tension, and sewing speed can be achieved. This effectively breaks through the technical bottleneck of traditional empiricism, improves the stability and adaptability of the stitching correction process, and thus ensures the consistency of stitching quality for automotive interior parts.

[0005] To achieve precise stitch correction for fabric seams used in automotive armrest covers, existing technologies first employ a series of techniques during the seam correction process. These include: continuously capturing images of the stitching trajectory using an industrial camera (such as a CCD (Charge-Coupled Device) vision camera); performing edge detection based on a local flow processing algorithm; and then acquiring the real-time offset of the stitching relative to a reference path. Simultaneously, a tension sensor collects the tension of the stitching during the dynamic process. The acquired offset and tension data are then pre-processed through filtering and noise reduction to form a standardized signal for real-time analysis, creating a time-series sample set.

[0006] The time-series sample set is input into the convolutional neural network model to accurately identify the edge of the fabric stitching used in the car armrest cover. Based on the regression algorithm, the lateral and longitudinal offset of the stitching path of the fabric used in the car armrest cover from the reference path is calculated and output as a continuous offset signal. At the same time, the tension sensing signal is converted into a standardized waveform for real-time monitoring of the tension status and compared with the ideal tension range preset based on the material database.

[0007] Based on the comparison results obtained above, an offset signal is output and input into the PID controller. The PID (Proportional-Integral-Derivative) controller obtains the adjustment amount required for stitch line correction according to the offset signal and sends the control command to the fabric stitch line correction mechanism (such as a fine-tuning guide driven by a linear motor or piezoelectric ceramic) to correct the stitching trajectory in real time, so that the fabric stitch line is aligned with the edge of the clamping plate of the fabric stitch line correction mechanism, thereby improving the accuracy of stitch line correction for the fabric used in the armrest cover of the car.

[0008] Because the stitching commonly used in automotive interiors (such as contrasting stitching and transparent stitching) has little grayscale difference with fabrics, such as the fabric used for car armrest covers, and the stitching correction process may involve natural light, factory ceiling lights, and indicator lights or lighting from other equipment, these light sources have different colors, intensities, and angles, creating complex mixed reflections on the stitching and fabric surfaces. This mixed reflection produces uneven light, and since fabric surfaces usually have a velvety or textured structure, under uneven light, raised areas of the fabric texture will form highlight areas, and recessed areas will form shadow areas. The grayscale gradients, amplitudes, and directions of these highlights and shadows highly overlap with the gradients of the stitching edges. This causes the fabric stitching correction mechanism, based on a local flow processing algorithm, to continuously generate pseudo-edges that are highly similar to the actual stitching edges when detecting the edges of various stitching types. In this scenario... In practice, local flow processing algorithms struggle to accurately distinguish between actual seams and fabric, introducing noise and misjudgments during edge detection. This results in inaccurate real-time offset measurements. Furthermore, the quality of the time-series sample set, composed of these noisy standardized signals, is affected. When these noisy time-series sample sets are input into a convolutional neural network model, the noise in the features leads to the model extracting invalid features. This introduces inherent errors in the regression-based calculation of lateral and longitudinal offsets, resulting in insufficient accuracy in identifying the seam edges of the fabric on the car armrest cover. Consequently, the output continuous offset signal accumulates deviations, ultimately causing a decrease in the accuracy of control commands sent by the PID controller in existing technologies. This leads to the cumulative deviations in the output continuous offset signal, making it difficult to overcome the bottleneck in seam correction accuracy. Summary of the Invention

[0009] This invention provides a method and system for optimizing suture routing parameters in a suture correction process. The technical solution provided by this application is as follows:

[0010] Firstly, a method for optimizing the stitching parameters in the suture correction process is provided, and the specific implementation of this method is as follows:

[0011] For the acquired suture region image, light interference is assessed, and the corresponding assessment results are obtained to determine whether to implement multi-light interference reduction control. For the suture edges corresponding to the qualified assessment results, preliminary detection is performed to determine whether to implement suture edge detection optimization. For the suture edge points corresponding to the qualified detection results, time-series sample set standardization is performed to determine whether to implement time-series sample set multi-level combined filtering optimization. If the standardization result is unqualified, time-series sample set multi-level combined filtering optimization is implemented. If the standardization result is qualified, PID control routing parameter adaptive optimization is performed on the suture routing corresponding to the qualified standardization result to determine whether to implement PID calibration frequency increase operation.

[0012] On the other hand, a suture correction process routing parameter optimization system is provided. This system specifically includes: a suture region image light interference determination module, a suture edge preliminary detection module, a time-series sample set standardization module, and a PID control routing parameter adaptive optimization module. The suture region image light interference determination module performs light interference determination on the acquired suture region image, obtains the corresponding determination result, and then decides whether to implement multi-light interference reduction control. The suture edge preliminary detection module performs preliminary detection on the suture edges corresponding to the qualified determination results, and then decides whether to implement suture edge detection optimization. If the standardization result is unqualified, then time-series sample set multi-level combined filtering optimization is implemented. The time-series sample set standardization module performs time-series sample set standardization on the suture edge points corresponding to the qualified detection results if the standardization result is qualified, and then decides whether to implement time-series sample set multi-level combined filtering optimization. The PID control routing parameter adaptive optimization module performs PID control routing parameter adaptive optimization on the suture routing corresponding to the qualified standardization results, and then decides whether to implement PID calibration frequency increase operation.

[0013] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0014] 1. Light interference assessment is performed on the acquired suture area image to obtain corresponding assessment results. This helps to accurately identify the degree of light and color temperature differences within the suture correction area, enhancing the light uniformity and detail clarity of the suture area image. It achieves targeted suppression of multi-source interference, thereby reducing the probability of false edges. Preliminary detection is performed on the suture edges to obtain corresponding detection results, effectively evaluating the accuracy of the local flow processing algorithm in identifying suture edges. This improves the accuracy and continuity of suture edge recognition, strengthens the ability to distinguish between real suture edges and fabric texture false edges, thus improving the quality of subsequent time-series sample sets. Time-series sample set standardization is performed on suture edge points to obtain corresponding standardization results, reducing the interference of residual noise components on offset calculation, improving the anti-interference ability and signal purity of the time-series sample set. Adaptive optimization of PID control routing parameters is performed on the suture routing to obtain corresponding optimization results. This helps to enhance the real-time response and adjustment capability of the PID controller to suture offset, achieving precise optimization of routing parameters, thereby improving the accuracy of suture routing correction.

[0015] 2. Optimization through seam edge detection helps to accurately distinguish between actual seam edges and false edges formed by fabric texture, highlights, and shadows. Compared to existing technologies, the grayscale gradients, amplitudes, and directions of highlights and shadows highly overlap with the gradients of seam edges. This causes the fabric seam correction mechanism, based on the local flow processing algorithm, to continuously generate false edges that are highly similar to actual seam edges when detecting the edges of various seams. This makes it difficult for the local flow processing algorithm to accurately distinguish between actual seams and fabric, thus introducing noise and misjudgments during edge detection, resulting in inaccurate real-time offset measurements. This solution can effectively filter out noise and misjudgments caused by false edges, ensuring the integrity and authenticity of seam edges, thereby improving the accuracy of real-time offset data.

[0016] 3. Multi-level combined filtering optimization of time-series sample sets is used to improve the anti-interference ability and signal purity of the time-series sample sets, and reduce the interference of residual noise on the feature extraction of convolutional neural network models. Compared with existing technologies, the quality of time-series sample sets composed of standardized signals containing noise is affected. When these noisy time-series sample sets are input into the convolutional neural network model, the features of the time-series sample sets contain noise. This solution uses the quantization of the noise suppression ratio of the time-series signal to ensure the qualification of the filtering effect, thereby improving the overall suture correction accuracy.

[0017] 4. Feature recognition using CNN models helps measure the effective recognition accuracy of convolutional neural network models for seam features and the stability of output offsets. Compared to existing technologies, the extraction of invalid features by convolutional neural network models leads to errors in the calculation of lateral and longitudinal offsets based on regression algorithms, resulting in inaccurate recognition of the seam edges of fabric used in car armrest covers. This further causes cumulative deviations in the output continuous offset signals. This solution helps reduce invalid feature extraction, improve the accuracy of offset calculation, eliminate seam edge recognition errors, and avoid cumulative deviations in the output continuous offset signals. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are 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 This is a flowchart of a method for optimizing the stitching parameters in a suture correction process according to an embodiment of the present invention;

[0020] Figure 2 This is a framework diagram of a method for optimizing the stitching parameters in a suture correction process provided by an embodiment of the present invention;

[0021] Figure 3 This is an architecture diagram of the suture area image light interference determination method of the suture thread correction process provided in an embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of the suture parameter optimization system for a suture correction process provided in an embodiment of the present invention;

[0023] Figure 5 This is a flowchart of a convolutional neural network model for processing suture data. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0025] Before providing a detailed explanation of the embodiments of this application, the application scenarios of these embodiments will be described first.

[0026] Example 1: This embodiment of the invention provides a method for optimizing the suture routing parameters in a suture correction process. For example... Figure 1 The flowchart shown represents a method for optimizing suture routing parameters in a suture correction process. The process can include the following steps:

[0027] First, the image of the suture region is subjected to light interference. The illumination difference degree of the corrected region is compared with the illumination difference threshold of the corrected region. If the illumination difference degree of the corrected region is greater than or equal to the illumination difference threshold of the corrected region, the corresponding suture region image is marked as a qualified suture region image, and preliminary suture edge detection is performed. If the illumination difference degree of the corrected region is greater than the illumination difference threshold of the corrected region, multi-light interference reduction adjustment is performed, and preliminary suture edge detection is performed again. It is determined whether the success rate of suture edge detection is greater than or equal to the preset suture edge detection success rate. If not, the suture edge detection is optimized. After this, the temporal sample set is standardized. If so, the temporal sample set is standardized, and it is determined whether the temporal signal-to-noise ratio is greater than the temporal signal-to-noise ratio threshold. If so, CNN (Convolutional Neural Network) is applied. The process involves several steps: First, it performs feature recognition on the CNN (Convolutional Neural Network) model. Second, it optimizes the multi-level combined filtering of the time-series sample set, checking if the noise suppression ratio of the re-acquired time-series signal after multi-level combined filtering optimization is greater than the time-series signal noise suppression ratio threshold. If not, it sends an abnormal prompt for multi-level combined filtering optimization of the time-series sample set to the designated personnel. Third, it checks if the suture feature region positioning accuracy is greater than or equal to the preset suture feature region positioning accuracy, and if the suture correction error is less than or equal to the preset suture correction error. If yes, it performs adaptive optimization of the PID control routing parameters. Otherwise, it sends an abnormal prompt for the CNN model feature recognition operation to the designated personnel. Fourth, it performs adaptive optimization of the PID control routing parameters, checking if the routing correction deviation is equal to 0. If not, it increases the PID calibration frequency. If yes, it sends a prompt indicating that the suture correction process is qualified to the designated personnel.

[0028] It should be added that, before designing the method for optimizing the suture parameters in the suture correction process in this application, a database is established to store various types of setting data. The database includes, but is not limited to, the color temperature difference threshold of the correction area, the illuminance difference threshold of the correction area, the preset number of suture edge points, and the time-series signal noise suppression ratio threshold.

[0029] The database's data sources include pre-set personnel combining industry technical specifications, extensive historical data on suture correction processes, comparative experimental data on different fabric materials (including texture features) and suture types (such as contrasting colors and transparent threads), and effective operating parameters accumulated in actual production scenarios. The data structure employs a structured design, categorized by parameter function into light detection, edge detection, time-series processing, model recognition, and PID optimization categories. Each category includes associated parameter names, data types, value ranges, default values, and update timestamps, enabling categorized management and rapid association of parameters. Storage utilizes an industrial-grade distributed relational database, supporting high-concurrency read / write, multi-terminal synchronous access, and scheduled automatic backups. It features data encryption protection mechanisms to ensure data security and reserves flexible data update interfaces. This allows for dynamic supplementation and iterative optimization of various set thresholds and parameters based on subsequent production process upgrades, new fabric materials, or new suture application scenarios, adapting to the needs of suture parameter optimization in different production scenarios.

[0030] like Figure 2 The diagram shown is a framework diagram of a method for optimizing the suture routing parameters in a suture correction process according to an embodiment of the present invention.

[0031] Light interference determination in suture area images: Before performing the suture correction process, light interference determination is performed on the acquired suture area images to assess the interference of mixed reflected light on the suture area images. The corresponding determination results are obtained, and then it is decided whether to implement multi-light interference reduction control to reduce uneven light interference. The suture area image refers to the image corresponding to the fabric suture area captured by the industrial camera. Performing light interference determination in suture area images helps to enhance the targeted suppression of multi-light source interference, thereby reducing the probability of false edges.

[0032] Preliminary detection of suture edges: For suture edges corresponding to qualified judgment results, preliminary detection is carried out to assess the probability of false edges in the suture area image, and then it is decided whether to implement suture edge detection optimization to improve the accuracy of suture edge point recognition. The suture edge refers to the boundary contour formed between the fabric suture and the fabric base on both sides. Preliminary detection of suture edges helps to strengthen the ability to distinguish between real suture edges and false edges of fabric texture, thereby improving the quality of subsequent time series sample sets.

[0033] Temporal sample set standardization: Temporal sample set standardization is carried out for the seam edge points corresponding to qualified detection results to reduce the calculation offset error. Then, it is decided whether to implement multi-level combined filtering optimization of temporal sample set to improve the anti-interference ability of temporal sample set. Seam edge points refer to pixels located on the transition boundary between the seam and the fabric substrate on both sides, which are characterized by abrupt changes in grayscale value, texture features or color attributes. Temporal sample set standardization helps to reduce the interference of residual noise components on offset calculation and improve the anti-interference ability and signal purity of temporal sample set.

[0034] Adaptive optimization of PID control routing parameters: Adaptive optimization of PID control routing parameters is carried out for the suture routing corresponding to qualified standardized results to improve the accuracy of routing parameter optimization. Then, it is decided whether to implement the operation of increasing the PID calibration frequency to improve the real-time performance of PID adjustment of routing parameters. Adaptive optimization of PID control routing parameters helps to achieve precise optimization of routing parameters, thereby improving the accuracy of suture routing correction.

[0035] Furthermore, the specific process for determining light interference in the acquired suture area image is as follows:

[0036] Images corresponding to the suture correction area are acquired based on a preset frame rate (typically 30 frames / second) and represented as suture area images. The ratio of the minimum to the maximum irradiance within the suture correction area is represented as the irradiance difference, used to characterize the degree of light difference within the suture correction area. The minimum irradiance represents the minimum irradiance value among all pixels within the suture correction area; the maximum irradiance represents the maximum irradiance value among all pixels within the suture correction area. Irradiance is monitored by an industrial-grade irradiance sensor. The suture correction area refers to the area within the field of view of the fabric suture line captured by the industrial camera, including the fabric suture line, the fabric surface, and the pre-defined edge of the clamping plate.

[0037] The illuminance difference in the correction area is compared with the illuminance difference threshold in the correction area set in advance by the preset personnel. If the illuminance difference in the correction area is greater than or equal to the illuminance difference threshold in the correction area, the corresponding suture area image is marked as a qualified suture area image, and preliminary detection of the suture edge is carried out. Otherwise, after implementing multi-light interference reduction control, preliminary detection of the suture edge is carried out.

[0038] It should be added that, such as Figure 3The diagram shown is an architecture diagram of the suture region image light interference determination method of the suture correction process according to an embodiment of the present invention. First, light interference determination of the suture region image is carried out to obtain the illumination difference degree of the correction region. The illumination difference degree of the correction region is compared with the illumination difference threshold of the correction region. If the illumination difference degree of the correction region is greater than or equal to the illumination difference threshold of the correction region, the corresponding suture region image is marked as a qualified suture region image, and preliminary detection of suture edges is carried out. If the illumination difference degree of the correction region is greater than the illumination difference threshold of the correction region, multi-light interference reduction control is carried out to determine whether the number of pixels of the candidate pixel block is within the preset upper and lower fluctuation range of the number of pixels of the pixel block to be processed. If so, a first fusion weight is assigned to the corresponding candidate pixel block. If not, a second fusion weight is assigned to the corresponding candidate pixel block. Weighted average weighted fusion is performed on the pixel blocks in the uniformly lit suture region image to obtain a qualified suture image, and preliminary detection of suture edges is carried out.

[0039] In this embodiment, this process can quantify the degree of light difference within the correction area, screen out qualified suture area images whose lighting conditions meet the edge detection requirements, and ensure that subsequent preliminary suture edge detection can be carried out based on high-quality images, avoiding false edge interference caused by uneven and inconsistent lighting.

[0040] Furthermore, the specific process of reducing interference from multiple light sources is as follows: For the suture region image, the brightness component and the reflection component are separated based on a preset multi-scale image processing algorithm (such as the Retinex algorithm); the brightness component refers to the light intensity in the suture region image, and the reflection component refers to the light intensity reflected in the suture region image.

[0041] Histogram equalization enhances the contrast of the reflection component, normalizes the luminance component, and then a preset multi-scale image processing algorithm is used to merge the processed luminance component and the reflection component to obtain a preliminary uniform suture image.

[0042] It should be added that the suture line area image is divided into a preset number of sub-regions, and a gray-level histogram array is calculated independently for each sub-region. Histogram equalization with contrast limitation is then performed to prevent excessive noise amplification. Next, for the boundary pixels of each sub-region, bilinear interpolation is used to eliminate artifacts at the boundaries of the sub-regions, ensuring the global gray-level uniformity and smooth transition between regions of the enhanced initially uniform suture line image.

[0043] In addition, the specific process of normalizing and correcting the luminance component is as follows: the median of the pixel intensity of each sub-region is obtained as the luminance reference value of that sub-region. The sub-region containing the center of the suture line region is marked as the reference sub-region, and its luminance reference value is marked as the reference luminance reference value. The sub-regions corresponding to the luminance reference values ​​within the preset upper and lower fluctuation range of the reference luminance reference value are retained. Conversely, the luminance reference values ​​corresponding to the maximum values ​​within the preset upper and lower fluctuation range of the reference luminance reference value are compressed to the reference luminance reference value, and the luminance reference values ​​corresponding to the minimum values ​​within the preset upper and lower fluctuation range of the reference luminance reference value are stretched to the reference luminance reference value.

[0044] For example, the specific process of calculating the grayscale histogram array is as follows: Assume that a sub-region has a size of 10×10 (a total of 100 pixels), of which there are 15 pixels with a grayscale value of 50, 20 pixels with a grayscale value of 100, and the remaining pixels are distributed in other grayscale levels. Then, in the grayscale histogram array of this sub-region, (H(50)=15, H(100)=20), the sum of all grayscale histogram arrays H(k) is equal to the total number of pixels in this sub-region. k represents the pixel grayscale value.

[0045] The initial uniform suture image is filtered using a homomorphic filtering function to obtain a filtered image of the uniform suture region. Specifically, this involves filtering the initial uniform suture image using a homomorphic filtering function (such as a Gaussian function), as follows:

[0046] The initial uniform suture image f(x, y) is defined as consisting of two parts: luminance component i(x, y) and reflectance component r(x, y): , where x and y represent the rows and columns of pixels in the initial uniform suture image.

[0047] The multiplication is converted into addition through logarithmic transformation, which facilitates frequency domain separation. .

[0048] By performing a Fourier transform on both sides of the equation, we can then process the data in the frequency domain. .

[0049] Where, F(u, v), F i (u, v), F r (u, v) are the Fourier transforms of lnf(x, y), lni(x, y), and lnr(x, y), respectively, where u and v represent the brightness variation frequencies in the horizontal and vertical directions of the preliminary uniform stitch line image after the frequency domain Fourier transform.

[0050] Set a filter H(u, v) to filter F(u, v) to obtain the frequency domain of the logarithmic domain of the preliminary uniform suture image: .

[0051] It should be added that the Gaussian function H(u, v) has the following form: Where D(u, v) is the distance from the coordinate (u, v) to the preset origin in the frequency domain, and D0 is a pre-set auxiliary adjustment parameter: the smaller D0 is, the more strict the division between high and low frequencies.

[0052] Applying the inverse Fourier transform to S(u, v) yields s(x, y), as follows: Where M and N represent the number of rows and columns of pixels in the initial uniform suture image, the filtered initial uniform suture image is: g .

[0053] A weighted average is performed on the pixel blocks in the image of the uniformly lit seam line region to obtain a qualified seam line image, thereby eliminating small-scale false edges of the fabric. Specifically, the image of the uniformly lit seam line region is divided into non-overlapping pixel blocks (usually 8x8 or 16x16), the pixel block containing the center point of the image of the uniformly lit seam line region is selected as the pixel block to be processed, and the remaining pixel blocks are marked as candidate pixel blocks; the number of pixels in the pixel block to be processed and the candidate pixel blocks are obtained and monitored by an industrial camera.

[0054] The system determines whether the number of pixels in a candidate pixel block is within a preset fluctuation range of the number of pixels in the pixel block to be processed, which is pre-set by a pre-defined person. If it is, a first fusion weight is assigned to the corresponding candidate pixel block; otherwise, a second fusion weight is assigned to the corresponding candidate pixel block. The gray values ​​of all candidate pixel blocks are weighted and fused (i.e., weighted average) based on the first and second fusion weights to obtain a qualified seam line image. Both the first and second fusion weights are pre-set by a pre-defined person based on historical experience, and the first fusion weight is greater than the second fusion weight.

[0055] Weighted fusion specifically refers to multiplying the gray value of each candidate pixel block by the weight corresponding to that candidate pixel block, then summing all the multiplication results to obtain the corresponding pixel block gray value, and marking the image corresponding to the pixel block gray value as a qualified suture line image.

[0056] In this embodiment, this process can reduce the influence of light interference and fabric texture pseudo edges. By processing, the image of the suture area is made to achieve light uniformity, enhance the contrast of suture details, eliminate small-scale pseudo edges, and output a qualified suture image, ensuring the accuracy and reliability of subsequent offset-related data.

[0057] Furthermore, the specific process for preliminary inspection of the suture edges is as follows:

[0058] The number of suture edge points in qualified suture images is extracted based on the local flow processing algorithm and represented as the number of successfully identified suture edge points. Existing machine vision software (such as Halcon, VisionPro, and OpenCV) is used for monitoring. The ratio of the number of successfully identified suture edge points to the preset number of suture edge points set in advance by preset personnel based on historical experience is marked as the suture edge detection success rate, which is used to quantify the accuracy of edge detection.

[0059] Determine whether the success rate of suture edge detection is greater than or equal to the preset success rate of suture edge detection. If it is greater than or equal to, mark the set of suture data in qualified suture images as a time series sample set and perform time series sample set standardization. The preset success rate of suture edge detection is represented by the average success rate of suture edge detection over a historical time period.

[0060] Suture line data refers to structured data extracted from qualified suture line images that is directly related to suture line features. It includes the spatial coordinates of successfully identified suture line edge points, frame sequence number, and time point of successful suture line edge point identification. If the data is less than 1, then suture line edge detection optimization is performed.

[0061] The specific process for optimizing suture edge detection is as follows:

[0062] Based on the Canny edge detection method, edge points of suture lines in qualified suture line images are screened. The specific process is as follows: The average value of the brightness difference between adjacent pixels in the horizontal direction and the average value of the brightness difference between adjacent pixels in the vertical direction are obtained in the qualified suture line image and represented as horizontal brightness difference and vertical brightness difference, respectively. Both are monitored by a high-resolution scanner. The square root of the sum of the squares of the horizontal and vertical brightness differences is represented as the gradient magnitude. If the gradient magnitude is higher than the high gradient threshold, the edge point is marked as a strong edge point and directly retained. If the gradient magnitude is between the high gradient threshold and the low gradient threshold, the corresponding edge point is marked as a weak edge point. If there is one or more strong edge points in the preset neighborhood of a weak edge point, it indicates that it is part of the real edge and is retained. Otherwise, it is regarded as noise or irrelevant details and directly removed to ensure the continuity of the suture line edge. Edge points with gradient magnitudes lower than the low gradient threshold are directly removed. The high gradient threshold is greater than the low gradient threshold. The high gradient threshold and the low gradient threshold are preset by preset personnel.

[0063] In this embodiment, this process can effectively suppress the interference of different grades of fabric texture noise on the image, output a qualified suture image with low noise and clear suture outline, and ensure the accuracy and effectiveness of the data required for optimizing suture thread parameters.

[0064] Furthermore, the specific process of time series sample set standardization is as follows: obtain the time series signal noise suppression ratio; the time series signal noise suppression ratio refers to the ratio of the amplitude of the signal in the time series sample set to the amplitude of the residual noise component, which is used to quantify the noise suppression level of the time series sample set.

[0065] The amplitude of the time-series sample set signal is represented by the average amplitude of the time-series sample set signal acquired within a preset time period. The amplitude of the residual noise component is represented by the absolute value of the difference between the amplitude of the time-series sample set signal at the beginning of the preset time period and the amplitude of the time-series sample set signal at the end of the preset time period. The amplitude of the time-series sample set signal is monitored by a spectrum analyzer.

[0066] The time-series sample set signal refers to the signal corresponding to the time-series sample set. If the noise suppression ratio of the time-series signal is greater than the time-series signal noise suppression ratio threshold set in advance by the preset personnel, then the CNN model feature recognition operation is performed; otherwise, multi-level combination filtering optimization of the time-series sample set is implemented.

[0067] In this embodiment, this process can quantitatively evaluate the accuracy of suture edge detection, screen out edge detection results that meet the accuracy requirements to promote the standardization of time-series sample sets, ensure the continuity and accuracy of suture edges, reduce the deviation of subsequent CNN model feature extraction and offset calculation, and improve the accuracy of the basic data required for routing parameter optimization from the perspective of edge detection accuracy.

[0068] Furthermore, the specific process of multi-level combined filtering optimization of time-series sample sets is as follows:

[0069] Wavelet decomposition is performed on the time-series sample set signal to obtain wavelet coefficients at different frequency scales. Low-frequency wavelet coefficients with a wavelet coefficient signal frequency less than or equal to the offset signal frequency are retained, while low-frequency wavelet coefficients with a wavelet coefficient signal frequency greater than the offset signal frequency are removed. The removed wavelet coefficients are then subjected to inverse transform to obtain the time-series sample set signal with pseudo-edge noise removed.

[0070] The frequency of wavelet coefficient signals is represented by the ratio of the center frequency of the mother wavelet monitored by a spectrum analyzer to a scaling factor set in advance by pre-set personnel based on historical experience. The specific process of inverse transformation of the removed wavelet coefficients is as follows: using the preset wavelet basis as a reference, the retained low-frequency wavelet coefficients are multiplied by the corresponding preset scale, and then all the product results are summed over the entire time axis to finally obtain a continuous time-domain signal, which is represented as a time-series sample set signal.

[0071] The offset signal frequency refers to the critical frequency value used to distinguish between the effective low-frequency signal and the low-frequency component of pseudo-edge noise in wavelet coefficients. For example, the effective low frequency of the suture line time series signal is concentrated in 0~50Hz, while the mixed component of pseudo-edge noise is mainly concentrated in 50~100Hz. It is determined whether the noise suppression ratio of the time series signal re-acquired after multi-level combined filtering optimization of the time series sample set is greater than the time series signal noise suppression ratio threshold. If so, the CNN model feature recognition operation is performed; otherwise, an abnormal prompt of multi-level combined filtering optimization of the time series sample set is sent to the preset personnel.

[0072] In this embodiment, this process can quantitatively evaluate the noise suppression level of the time series sample set, and select sample sets that meet the noise suppression standard to ensure that the time series sample set input to subsequent stages has low noise and high effectiveness, avoiding invalid feature extraction or offset calculation deviation of the CNN model due to noise in the time series sample set.

[0073] Furthermore, the specific process of CNN model feature recognition is as follows: the overlapping area between the suture feature region obtained by the CNN model and the actual suture feature region is marked as the suture feature localization area.

[0074] It should be added that the specific process of obtaining the suture feature region by the CNN model is as follows: The original image of a qualified suture (such as a suture image taken by an industrial camera) is standardized, including size normalization (uniformly scaling to the CNN input size, such as 224×224) and pixel value normalization (mapping pixel values ​​to the 0-1 range or standardizing them to mean 0 and variance 1). Multiple learnable convolutional kernels (such as 3×3 and 5×5) are used to perform sliding convolution operations on the preprocessed image to extract features at different levels: shallow convolutional kernels capture basic features such as edges and textures; deep convolutional kernels fuse basic features to form high-level semantic features related to the suture (such as the direction of the suture, the spacing between stitches, and the outline of the stitches). The feature map output by the convolution is downsampled (such as max pooling and average pooling) to reduce the feature dimension while retaining key features. Combined with the target detection or semantic segmentation branch, the bounding box coordinates or pixel-level mask of the suture feature region are output, thus obtaining the suture feature region.

[0075] The ratio of the suture feature area to the preset suture feature area set in advance by the preset personnel based on historical experience is expressed as the suture feature region localization accuracy, which is used to quantify the effective feature recognition accuracy of the CNN model; the standard deviation of the suture offset output by the CNN model is expressed as the suture correction error, which is used to quantify the stability of the CNN model output.

[0076] The suture offset refers to the vertical distance between the center point of the suture and the feature center point of the theoretical reference position, which is monitored by a high-resolution industrial camera; the center point of the suture refers to the geometric center point of the suture, while the feature center point of the theoretical reference position refers to the reference center point of the suture determined based on the suture sample.

[0077] For example, in the sewing process of the fabric used for the top cover of a car armrest, the image of the sewing line area is semantically segmented based on the CNN model to obtain a binary mask containing only the sewing line. The skeleton of the mask is extracted to determine the extension direction of the sewing line as it bends clockwise along the curvature of the top cover of the car armrest. Then, a center point is selected every 4 pixels along the skeleton and represented as the center point of the sewing line.

[0078] While determining whether the accuracy rate of suture feature region positioning is greater than or equal to the preset accuracy rate, it is also determined whether the suture correction error is less than or equal to the preset suture correction error. If the accuracy rate of suture feature region positioning is greater than or equal to the preset accuracy rate, and the suture correction error is less than or equal to the preset correction error, then adaptive optimization of PID control routing parameters is performed. Otherwise, an abnormal operation prompt for CNN model feature recognition is sent to the preset personnel. The preset accuracy rate of suture feature region positioning is represented by the average value of the accuracy rate of suture feature region positioning over a historical time period, and the preset correction error is represented by the average value of the correction error over a historical time period.

[0079] In this embodiment, this process can remove false edge noise from the time-series sample set, retain the effective low-frequency signal related to the offset, improve the noise suppression level of the time-series sample set, and ensure that the noise suppression ratio of the optimized time-series sample set meets the standard to promote the feature recognition of the CNN model.

[0080] Furthermore, the specific process of adaptive optimization of PID control routing parameters is as follows: The difference in linear distance between the actual position of the suture line monitored by the industrial camera and the preset position set by the preset personnel based on historical experience is expressed as the routing correction deviation, which is used to reflect the optimization level of the suture routing parameters; it is determined whether the routing correction deviation is equal to 0; if the routing correction deviation is equal to 0, it means that the routing parameter optimization is qualified, and a qualified suture correction process prompt is sent to the preset personnel; if the routing correction deviation is not equal to 0, the PID calibration frequency is increased.

[0081] The specific process for increasing the PID calibration frequency is as follows:

[0082] A prompt is sent to the preset personnel to gradually increase the PID calibration frequency based on the original PID calibration frequency and the preset calibration frequency ratio. At the same time, the routing correction deviation is monitored in real time until the routing correction deviation is equal to 0. The PID calibration frequency increase operation is stopped, and a suture correction process qualified prompt is sent to the preset personnel. The preset calibration frequency ratio is represented by the ratio of the current suture offset to the suture offset output by the CNN model.

[0083] In this embodiment, this process can quantitatively evaluate the CNN model's recognition accuracy of suture features and the stability of its output offset, ensuring that PID control can accurately adjust the suture parameters based on high-quality data, thereby enhancing the accuracy of suture correction.

[0084] It should be added that, such as Figure 4 The diagram shows a schematic of a suture correction process parameter optimization system provided in an embodiment of the present invention. The suture area image light interference judgment module is used to perform light interference judgment on the acquired suture area image, obtain the corresponding judgment result, and implement multi-light interference reduction control for unqualified judgment results. The suture edge preliminary detection module is used to perform preliminary detection on the suture edge corresponding to the qualified judgment result, obtain the corresponding detection result, and implement suture edge detection optimization for unqualified detection results. The time-series sample set standardization module is used to perform time-series sample set standardization on the suture edge points corresponding to the qualified detection result, obtain the corresponding standardization result, and implement time-series sample set multi-level combined filtering optimization for unqualified standardization results. The PID control path parameter adaptive optimization module is used to perform PID control path parameter adaptive optimization on the suture path corresponding to the qualified standardization result, obtain the corresponding optimization result, and if the optimization result is unqualified, the PID calibration frequency is increased; if the optimization result is qualified, a suture correction process qualified prompt is sent to a preset personnel.

[0085] It should be added that, such as Figure 5 The diagram shows the flowchart of the convolutional neural network model for suture data processing in this invention. It mainly extracts local features of suture data through a one-dimensional convolutional layer, then performs dimensionality reduction of local features and preservation of key information through a one-dimensional max pooling layer, then adjusts the data dimension through a dimension transpose layer to adapt to the subsequent network, then enters the LSTM (Long Short-Term Memory) time-series modeling layer to perform in-depth modeling of the sequence features in the time dimension, and finally outputs the final suture offset by a fully connected regression head.

[0086] Example 2: Building upon Example 1, to deeply suppress small-scale noise such as fuzz and texture while strictly preserving the continuous edge structure of the actual seam, bilateral filtering from Example 2 can be used in production scenarios with relatively low fabric texture noise (such as short-pile or plain weave fabrics, texture protrusion height ≤ 0.1mm, and grayscale contrast ≤ 15). Compared to non-local mean filtering, which requires traversing all pixel blocks within the search window to calculate similarity, bilateral filtering uses dual weights of fabric texture spacing and texture grayscale contrast for rapid calculation, reducing processing time to less than 30ms. While ensuring texture noise suppression, it significantly improves the image preprocessing efficiency of the seam correction area, avoiding correction delays caused by excessive algorithm time. The specific process is as follows:

[0087] The image of the uniformly lit seam area is filtered based on a preset bilateral filter. The specific process is as follows: The fabric texture spacing and grayscale contrast of the pure texture area without seams on the fabric are extracted based on image analysis software such as OpenCV and Halcon. The fabric texture spacing refers to the average distance between texture protrusions, and the texture grayscale contrast refers to the average value of the grayscale ratio of the texture to the fabric background color. Both are monitored by an industrial high-resolution camera.

[0088] If either of the following two situations applies:

[0089] In the first case, the fabric texture spacing is greater than the maximum value within the preset fabric texture spacing reference range, and the texture grayscale contrast is greater than the minimum value within the preset fabric texture contrast reference range; in the second case, the texture grayscale contrast is greater than the maximum value within the preset fabric texture contrast reference range, and the fabric texture spacing is greater than the minimum value within the preset fabric texture spacing reference range; then the corresponding fabric is marked as high-noise fabric, and filtered using preset high-level bilateral filtering core parameters.

[0090] If the fabric texture spacing is within the preset fabric texture spacing reference range and the texture grayscale contrast is within the preset fabric texture contrast reference range, then the corresponding fabric will be marked as medium noise fabric and filtered using the preset medium bilateral filter core parameters.

[0091] If either of the following two phenomena occurs: First, the fabric texture spacing is less than the minimum value within the preset fabric texture spacing reference range, and the texture grayscale contrast is less than the maximum value within the preset fabric texture contrast reference range; Second, the texture grayscale contrast is less than the minimum value within the preset fabric texture contrast reference range, and the fabric texture spacing is less than the maximum value within the preset fabric texture spacing reference range; then the corresponding fabric will be marked as low-noise fabric, and filtered using preset low-level bilateral filtering core parameters. The filtered image of the uniformly lit seam area will be marked as a qualified seam image, and preliminary detection of the seam edge will be performed.

[0092] It should be added that the preset bilateral filter core parameters include: preset low-level bilateral filter core parameters, preset medium-level bilateral filter core parameters, and preset high-level bilateral filter core parameters, all of which are determined by the bilateral filter properties. The preset fabric texture spacing reference range and preset fabric texture contrast reference range are both set in advance by the preset personnel.

[0093] In summary, in the embodiments of the application, performing light interference judgment on the acquired suture area image to obtain the corresponding judgment result helps to accurately identify the degree of light difference and color temperature difference within the suture correction area, enhances the light uniformity and detail clarity of the suture area image, and achieves targeted suppression of multi-source interference, thereby reducing the probability of false edge generation. Preliminary detection of the suture edge to obtain the corresponding detection result can effectively evaluate the accuracy of the local flow processing algorithm in identifying the suture edge, improves the accuracy and continuity of suture edge recognition, strengthens the ability to distinguish between real suture edge and fabric texture false edge, thereby improving the quality of subsequent time-series sample sets. Time-series sample set standardization is performed on the suture edge points to obtain the corresponding standardization result, which is used to reduce the interference of residual noise components on offset calculation, improve the anti-interference ability and signal purity of the time-series sample set. Adaptive optimization of PID control routing parameters for the suture routing to obtain the corresponding optimization result helps to enhance the real-time response and adjustment capability of the PID controller to suture offset, achieves precise optimization of routing parameters, and thus improves the accuracy of suture routing correction.

[0094] The above-disclosed embodiments are merely some examples of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for optimizing suture routing parameters in a suture correction process, characterized in that, The method includes: Light interference is determined for the acquired suture area image, the corresponding determination result is obtained, and then it is decided whether to implement multi-light interference reduction control. A preliminary inspection is conducted on the suture edges corresponding to the qualified judgment results, and then a decision is made on whether to implement suture edge inspection optimization. For the suture edge points corresponding to qualified test results, time-series sample set standardization is carried out, and then it is decided whether to implement time-series sample set multi-level combined filtering optimization. If the standardization result is not qualified, time-series sample set multi-level combined filtering optimization is implemented. If the standardization result is satisfactory, then adaptive optimization of the PID control routing parameters will be carried out for the suture routing corresponding to the satisfactory standardization result, and then it will be decided whether to implement the PID calibration frequency increase operation.

2. The method for optimizing suture parameters in a suture correction process as described in claim 1, characterized in that, The specific process for determining light interference in the acquired suture area image is as follows: The image corresponding to the suture correction area is acquired based on a preset frame rate and represented as the suture area image. The ratio of the minimum to the maximum irradiance within the suture correction area is expressed as the irradiance difference in the correction area. Compare the illuminance difference in the corrected area with the illuminance difference threshold in the corrected area; If the illuminance difference in the correction area is greater than or equal to the illuminance difference threshold in the correction area, the corresponding suture area image is marked as a qualified suture area image, and preliminary detection of the suture edge is carried out. Otherwise, after implementing multi-light interference reduction control, preliminary detection of the suture edge is carried out.

3. The method for optimizing suture routing parameters in a suture correction process as described in claim 2, characterized in that, The specific process of reducing and controlling multi-ray interference is as follows: For the suture area image, the brightness component and the reflection component are separated based on a preset multi-scale image processing algorithm; Based on histogram equalization to enhance the contrast of the reflection component, the brightness component is normalized and corrected, and then the processed brightness component and reflection component are merged based on a preset multi-scale image processing algorithm to obtain a preliminary uniform suture image. The initial uniform suture image is filtered using a homomorphic filtering function to obtain the filtered image of the uniform suture region. The image of a uniform suture region is weighted and averaged based on nonlocal mean filtering to obtain a qualified suture image. Specifically, the image of a uniform suture region is divided into non-overlapping pixel blocks. The pixel block containing the center point of the image of the uniform suture region is selected as the pixel block to be processed, and the remaining pixel blocks are marked as candidate pixel blocks. Obtain the number of pixels in the pixel block to be processed and the candidate pixel block; Determine whether the number of pixels in the candidate pixel block is within a preset fluctuation range of the number of pixels in the pixel block to be processed, which is pre-set by the preset personnel. If the first fusion weight is assigned to the corresponding candidate pixel block, the second fusion weight is assigned to the corresponding candidate pixel block. The gray values ​​of all candidate pixel blocks are weighted and fused based on the first and second fusion weights to obtain a qualified seam line image.

4. The method for optimizing suture routing parameters in a suture correction process as described in claim 2, characterized in that, The specific process for the preliminary inspection of the suture edge is as follows: The number of suture edge points in qualified suture images is extracted based on the local flow processing algorithm and represented as the number of successfully identified suture edge points. The ratio of the number of successfully identified suture edge points to the preset number of suture edge points is marked as the suture edge detection success rate; Determine whether the success rate of suture edge detection is greater than or equal to the preset success rate of suture edge detection; If it is greater than or equal to, the set containing suture data from qualified suture images will be labeled as a time series sample set, and time series sample set standardization will be performed. If it is less than, then perform suture edge detection optimization; The specific process for optimizing the suture edge detection is as follows: Based on the Canny edge detection method, the edge points of the suture line in the qualified suture line image are screened. The specific process is as follows: the average value of the brightness difference between adjacent pixels in the horizontal direction and the average value of the brightness difference between adjacent pixels in the vertical direction are obtained in the qualified suture line image, and expressed as horizontal brightness difference and vertical brightness difference, respectively. The result of taking the square root of the sum of the squares of the horizontal and vertical brightness differences is represented as the gradient magnitude. If the gradient magnitude is higher than the high gradient threshold, the edge point is marked as a strong edge point and is retained directly; If the gradient magnitude is between the high gradient threshold and the low gradient threshold, the corresponding edge point is marked as a weak edge point. If there is one or more strong edge points in the preset neighborhood of the weak edge point, it is retained; otherwise, it is directly removed. Edge points with gradient magnitudes below the low gradient threshold are directly removed.

5. The method for optimizing suture routing parameters in a suture correction process as described in claim 4, characterized in that, The specific process of standardizing the time-series sample set is as follows: Obtain the noise suppression ratio of the timing signal; The time-series signal noise suppression ratio refers to the ratio of the amplitude of the signal to the amplitude of the residual noise component for a time-series sample set. If the time-series signal noise suppression ratio is greater than the time-series signal noise suppression ratio threshold, then the CNN model feature recognition operation is performed; otherwise, multi-level combination filtering optimization of the time-series sample set is implemented.

6. The method for optimizing suture routing parameters in a suture correction process as described in claim 5, characterized in that, The specific process of multi-level combined filtering optimization of the time-series sample set is as follows: Wavelet decomposition was performed on the time-series sample set signal to obtain the wavelet coefficients corresponding to different frequency scales; Low-frequency wavelet coefficients whose wavelet coefficient signal frequency is less than or equal to the offset signal frequency are retained, while low-frequency wavelet coefficients whose wavelet coefficient signal frequency is greater than the offset signal frequency are removed. The removed wavelet coefficients are then subjected to inverse transform to obtain the time-series sample set signal with pseudo-edge noise removed. The specific process of performing inverse transformation on the removed wavelet coefficients is as follows: using the preset wavelet basis as a reference, the retained low-frequency wavelet coefficients are multiplied by the corresponding preset scale, and then all the product results are summed over the entire time axis to finally obtain a continuous time-domain signal, which is represented as a time-series sample set signal. Determine whether the noise suppression ratio of the time-series signal re-acquired after multi-level combined filtering optimization of the time-series sample set is greater than the time-series signal noise suppression ratio threshold; If yes, then perform CNN model feature recognition; otherwise, send an anomaly alert for multi-level combined filtering optimization of time-series sample sets to the designated personnel.

7. The method for optimizing suture routing parameters in a suture correction process as described in claim 6, characterized in that, The specific process of the feature recognition operation of the CNN model is as follows: The overlapping area between the suture feature region obtained through the CNN model and the actual suture feature region is marked as the suture feature localization area; The ratio of the suture feature area to the preset suture feature area is expressed as the suture feature area positioning accuracy. The standard deviation of the suture offset output by the CNN model is expressed as the suture correction error; While determining whether the accuracy rate of suture feature area positioning is greater than or equal to the preset accuracy rate of suture feature area positioning, it is also determined whether the suture correction error is less than or equal to the preset suture correction error. If the accuracy rate of suture feature region positioning is greater than or equal to the preset accuracy rate of suture feature region positioning, and the suture correction error is less than or equal to the preset suture correction error, then adaptive optimization of PID control routing parameters will be carried out; otherwise, an abnormal operation prompt for CNN model feature recognition will be sent to the preset personnel.

8. The method for optimizing suture routing parameters in a suture correction process as described in claim 7, characterized in that, The specific process of adaptive optimization of the PID control routing parameters is as follows: The difference between the actual position and the preset position of the suture is expressed as the suture correction deviation. Determine if the trace correction deviation is equal to 0; If the stitching correction deviation is 0, a notification that the stitching correction process is qualified will be sent to the designated personnel. If the trace correction deviation is not equal to 0, then perform a PID calibration frequency increase operation; The specific process of increasing the PID calibration frequency is as follows: A prompt is sent to the preset personnel to gradually increase the PID calibration frequency based on the original PID calibration frequency ratio, while monitoring the stitch correction deviation in real time until the stitch correction deviation equals 0. At this point, the PID calibration frequency increase operation is stopped, and a prompt indicating that the stitch correction process is qualified is sent to the preset personnel. The preset calibration frequency ratio is represented by the ratio of the current suture offset to the suture offset output by the CNN model.

9. The method for optimizing suture routing parameters in a suture correction process as described in claim 2, characterized in that, The specific process of reducing and controlling multi-ray interference is as follows: Based on a preset bilateral filter, the image of the uniform light seam area is filtered. The specific process is as follows: extract the fabric texture spacing and grayscale contrast of the pure texture area on the fabric without seams. If either of the following two situations applies: In the first case, the fabric texture spacing is greater than the maximum value within the preset fabric texture spacing reference range, while the texture grayscale contrast is greater than the minimum value within the preset fabric texture spacing reference range. In the second case, the texture grayscale contrast is greater than the maximum value within the preset fabric texture spacing reference range, while the fabric texture spacing is greater than the minimum value within the preset fabric texture spacing reference range. The corresponding fabric is then marked as high-noise fabric and filtered using preset high-noise bilateral filter core parameters; If the fabric texture spacing is within the preset fabric texture spacing reference range and the texture grayscale contrast is within the preset fabric texture spacing reference range, then the corresponding fabric will be marked as medium noise fabric and filtered using the preset medium bilateral filter core parameters. If either of the following two phenomena occurs: The first phenomenon is that while the fabric texture spacing is less than the minimum value within the preset fabric texture spacing reference range, the texture grayscale contrast is less than the maximum value within the preset fabric texture spacing reference range. The second phenomenon is that the texture grayscale contrast is less than the minimum value within the preset fabric texture spacing reference range, while the fabric texture spacing is less than the maximum value within the preset fabric texture spacing reference range. The corresponding fabric is then marked as low-noise fabric and filtered using preset low-noise bilateral filter core parameters; The filtered light uniformly suture area image is marked as a qualified suture image, and preliminary detection of the suture edge is carried out.

10. A suture parameter optimization system for a suture correction process, used to implement the suture parameter optimization method for a suture correction process as described in any one of claims 1-9, characterized in that, The system includes: a suture region image light interference determination module, a suture edge preliminary detection module, a time-series sample set standardization module, and a PID control routing parameter adaptive optimization module; The suture region image light interference determination module is used to determine light interference in the acquired suture region image, obtain the corresponding determination result, and then decide whether to implement multi-light interference reduction control. The preliminary detection module for suture edges is used to perform preliminary detection on the suture edges corresponding to qualified judgment results, and then decide whether to implement suture edge detection optimization. If the standardized result is unqualified, multi-level combination filtering optimization of time-series sample set is implemented. The time series sample set standardization module is used to perform time series sample set standardization on the suture edge points corresponding to the qualified detection results if the standardization result is qualified, and then decide whether to implement time series sample set multi-level combined filtering optimization. The PID control routing parameter adaptive optimization module is used to perform adaptive optimization of PID control routing parameters for suture routing corresponding to qualified standardized results, and then decide whether to implement PID calibration frequency increase operation.

Images