ETFE film cutting control method and system based on image recognition

Abstract

The invention relates to the technical field of image analysis, in particular to an ETFE film cutting control method and system based on image recognition, and the method comprises the following steps: enhancing an edge through multi-scale Gaussian filtering, extracting a gray level change rate pixel by a Sobel operator, generating a path segment, outputting a coordinate, calculating a gradient direction, constructing a direction continuous sequence, and recording an angle deviation; a dynamic window judges abnormal fluctuation to trigger calibration, a fluctuation segment vector is replaced to update a motion track, and a B spline is fitted to generate a track output cutting instruction. According to the method, multi-scale Gaussian filtering is combined with Sobel operator enhanced edge extraction continuous pixels, the fracture problem caused by a fixed threshold value is solved, vector space projection constructs a direction sequence to quantify angle deviation, dynamic window range judges fluctuation to trigger calibration, and a vector replacement strategy corrects an abnormal section control point direction. A smooth track is generated through B spline interpolation, mechanical vibration is reduced, and the flexible material cutting precision and the equipment cooperation efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and particularly to an ETFE film cutting control method and system based on image recognition. Background Art

[0002] Image analysis technology performs structured processing on static or dynamic images through computer algorithms to extract information such as edges, shapes, and textures. It encompasses technologies such as image segmentation, object recognition, registration, and enhancement, and is widely applied in fields such as industrial inspection, medical imaging, remote sensing mapping, security monitoring, and autonomous driving, emphasizing image space and semantic parsing to support high-level decision-making.

[0003] Among them, the ETFE film cutting control method based on image recognition uses technologies such as edge detection and image segmentation to achieve film material feature extraction and boundary recognition, and combines geometric matching and path planning algorithms to generate cutting paths. The image acquisition device captures the film material image, forms path instructions after analysis, and drives the cutting execution mechanism to achieve automatic control. The existing technology uses fixed-threshold edge detection, which is difficult to adapt to the interference of film material reflection or wrinkles, and is prone to false detection and requires manual repair. The path planning ignores the pixel gradient continuity and global direction consistency, resulting in path conflicts and instability; lacks a real-time calibration mechanism and cannot cope with deformation errors; linear interpolation does not adapt to the flexibility of the film material, is prone to trajectory burrs, and affects the equipment life and accuracy. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose an ETFE film cutting control method and system based on image recognition.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: The ETFE film cutting control method based on image recognition includes the following steps: S1: Perform edge enhancement on the ETFE film image through multi-scale Gaussian filtering, extract pixel points whose gray change rate exceeds the dynamic threshold based on the Sobel gradient operator, generate candidate cutting path segments, and extract the starting and ending pixel coordinates of each path segment; S2: For adjacent pixel points of the candidate cutting path segments, call the vector space projection algorithm, calculate the gradient direction vectors in the horizontal and vertical directions of each point, construct a path segment direction continuity sequence, and store the angle deviation value between it and the preset cutting direction; S3: Based on the path segment direction continuity sequence, use the dynamic window range determination method to perform window sliding calculation on the angle deviation values of 5 consecutive pixel points. If the range within the window exceeds the set threshold, mark this path segment as an abnormal fluctuation segment and trigger a calibration instruction; S4: For the abnormal fluctuation segment, extract the mean of the gradient vectors of its adjacent path segments, replace the direction vectors of all control points in the fluctuation segment with the mean vector through the vector replacement strategy, generate calibrated path data, and synchronously update the motion trajectory parameters of the cutting equipment.

[0006] As a further solution of the present invention, the candidate clipping path segment specifically includes the coordinates of the starting and ending points of the path segment, and the grayscale change rate parameters; the path segment directional continuity sequence includes the horizontal gradient vector, the longitudinal gradient vector, and the angle deviation value; the abnormal fluctuation segment specifically refers to the abnormal path segment mark and the calibration trigger threshold; the calibrated path data includes the replaced control point vector and the motion trajectory parameters.

[0007] As a further solution of the present invention, the step of obtaining the candidate clipping path segment is specifically as follows: S101: acquiring original data of the ETFE film image, performing convolution operation on the image using Gaussian kernels of multiple scales, linearly superimposing multiple groups of convolution results according to preset weight coefficients, and generating a multi-scale edge-enhanced image; S102: Based on the multi-scale edge enhanced image, calling the Sobel operator to calculate the pixel gradient component, synthesizing the gradient amplitude matrix, setting a dynamic threshold according to the statistical characteristics of the gradient amplitude distribution, screening the pixel points whose gradient amplitude exceeds the threshold, and generating a gradient over-limit pixel set; S103: For the gradient excess pixel set, perform a morphological closing operation to connect adjacent pixel points, extract the circumscribed rectangular boundary of the closed area, remove small isolated areas, retain the remaining path segments, and generate candidate clipping path segments.

[0008] As a further solution of the present invention, the step of obtaining the path segment direction continuity sequence is specifically: S201: calling the horizontal and vertical convolution kernel matrices, performing point-by-point convolution operations on adjacent pixel points of the candidate clipping path segment, recording the horizontal gradient component and the vertical gradient component, and generating a gradient direction vector set; S202: Calculating the direction angles of multiple pixel points based on the gradient direction vector set, defining the direction angle differences of adjacent pixel points and storing them in a queue in order, and constructing a direction angle difference queue; S203: Call the direction angle difference queue, using the formula: ; Calculate the continuity score of the path segment, compare it with the preset threshold to screen the path segment, and generate the path segment direction continuity sequence; in, represents the path segment direction continuity score, is the difference in direction angles between adjacent pixels, , is the gradient modulus of adjacent pixels, is the Euclidean distance between adjacent pixel points, is the total number of adjacent pixel point pairs within the path segment.

[0009] As a further aspect of the present invention, the steps for obtaining the abnormal fluctuation segment are specifically as follows: S301: Obtain the angular deviation values of 5 consecutive pixel points in the path segment direction continuity sequence, traverse the path segment in a window sliding manner, calculate the range of the angular deviation values within each window, and generate a dynamic window range value; S302: Invoke the dynamic window range value, and based on the difference in the range change trend between adjacent windows and the path segment curvature change rate, use the formula: ; Calculate to obtain a range fluctuation determination coefficient, and generate a range fluctuation determination coefficient through weighted superposition; Among them, represents the range fluctuation determination coefficient, represents the maximum angular deviation within the window, represents the minimum angular deviation within the window, represents the mean value of the range change trend difference between adjacent windows, represents the path segment curvature change rate, represents the environmental interference factor; S303: Invoke the range fluctuation determination coefficient, compare it with a preset path fluctuation determination threshold, and if the coefficient exceeds the threshold, mark the path segment covered by the corresponding window as an abnormal fluctuation segment.

[0010] As a further aspect of the present invention, the steps for obtaining the calibrated path data are specifically as follows: S401: Obtain the data of the adjacent path segments on both sides of the abnormal fluctuation segment, extract the gradient vectors of the control points within the path segment, calculate the arithmetic mean of the adjacent segment gradient vectors, and generate an adjacent gradient mean vector; S402: Based on the adjacent gradient mean vector, traverse the direction vectors of the control points within the abnormal fluctuation segment, compare the multi-component values of the direction vectors with the mean vector, replace the original components with the mean components, update the control point coordinates and direction parameters, and generate a set of replacement vector control points; S403: Invoke the coordinates of the unaffected path segments in the original path, integrate the set of replacement vector control points, reconstruct the connection relationship and curvature parameters of adjacent control points, and generate the calibrated path data.

[0011] As a further aspect of the present invention, the method further includes: S5: According to the calibrated path data, call the B-spline curve interpolation algorithm to perform smooth fitting at the connection of path segments, generate the ETFE film cutting trajectory, drive the multi-axis linkage parameters of the cutting head based on the coordinate differences of adjacent path points in the trajectory, and output cutting instructions; The ETFE film cutting trajectory is specifically the B-spline curve parameters and multi-axis linkage parameters.

[0012] As a further solution of the present invention, the steps for obtaining the ETFE film cutting trajectory are specifically as follows: S501: Call the calibrated path data, extract the control point coordinates and curvature parameters at the connection of path segments, calculate the distance and angle difference between adjacent control points, set the knot vector and weight coefficient, and generate a path node parameter set; S502: Based on the path node parameter set, divide the interpolation interval and solve the cubic polynomial coordinate components of each interval, adjust the high-order derivative terms of the polynomial coefficient matrix to meet the curvature continuity condition, and generate a fitting trajectory coordinate set; S503: Traverse the fitting trajectory coordinate set, calculate the curvature change rate of adjacent interpolation points, screen out the jump points exceeding the continuity threshold, re-interpolate and compensate, and output a path sequence to generate the ETFE film cutting trajectory.

[0013] An ETFE film cutting control system based on image recognition, the ETFE film cutting control system based on image recognition is used to execute the above-mentioned ETFE film cutting control method based on image recognition, and the system includes: An edge enhancement module, which is used to perform filtering and noise reduction on the ETFE film image through a multi-scale Gaussian filtering algorithm, call the Sobel gradient operator to calculate the horizontal and vertical gray change rates, screen out the set of continuous pixel points whose gray change rates exceed the dynamic threshold, generate candidate cutting path segments and output their start and end point coordinates, and transfer the candidate cutting path segments to the direction continuity analysis module; A direction continuity analysis module, which is used for adjacent pixel points of the candidate cutting path segment, constructs the gradient direction vectors of each pixel point by using the vector space projection algorithm, calculates the direction angle between adjacent vectors, generates a path segment direction continuity sequence, stores the angle deviation value between the angle and the preset direction, and transfers the path segment direction continuity sequence to the abnormal fluctuation determination module; An abnormal fluctuation determination module, which is used to call the dynamic window range determination method to perform a sliding window calculation on the angle deviation values of 5 consecutive pixel points based on the path segment direction continuity sequence, judge whether the range within the window exceeds the set threshold, mark the path segment with an excessive range as an abnormal fluctuation segment and generate a calibration instruction, and transfer the abnormal fluctuation segment to the path calibration module; A path calibration module, which is used to extract the mean value of the gradient direction vectors of the adjacent path segments of the abnormal fluctuation segment, replace the direction vectors of all control points in the fluctuation segment with the mean vectors through a vector replacement strategy, generate calibrated path data and update the motion trajectory parameters of the cutting device, and transfer the calibrated path data to the trajectory generation module; A trajectory generation module, which is used to perform smooth fitting on the connection of path segments by calling the B-spline curve interpolation algorithm based on the calibrated path data, calculate the coordinate differences of adjacent path points to generate an ETFE film cutting trajectory, and drive the multi-axis linkage parameters of the cutting head to output a cutting instruction.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, continuous pixels are enhanced by multi-scale Gaussian filtering combined with the Sobel operator to solve the fracture problem caused by a fixed threshold. The vector space projection is used to construct a direction sequence to quantify the angular deviation. The dynamic window range difference is used to determine the trigger of calibration. The vector replacement strategy is used to correct the direction of control points in the abnormal segment. The B-spline interpolation is used to generate a smooth trajectory to reduce mechanical vibration, and improve the cutting accuracy of flexible materials and the cooperation efficiency of the equipment. Description of the Drawings

[0015] Figure 1 It is a schematic diagram of the working process of the present invention. Detailed Embodiments

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] Embodiment 1

[0018] Please refer to Figure 1 , the present invention provides a technical solution: an ETFE film cutting control method based on image recognition, including the following steps: S1: Perform edge enhancement on the ETFE film image through multi-scale Gaussian filtering, extract the pixel points whose gray-scale change rate exceeds the dynamic threshold based on the Sobel gradient operator, generate candidate cutting path segments, and extract the starting and ending pixel coordinates of each path segment; S2: For adjacent pixel points of the candidate cutting path segments, call the vector space projection algorithm, calculate the gradient direction vectors of each point in the horizontal and vertical directions, construct a path segment direction continuity sequence, and store its angular deviation value from the preset cutting direction (that is, calculate the angular deviation between the direction vector of each pixel point in the sequence and the preset cutting direction, and store it for subsequent determination); S3: Based on the path segment direction continuity sequence, use the dynamic window range determination method to perform window sliding calculations on the angle deviation values of 5 consecutive pixels. If the range within the window exceeds the set threshold (that is, the difference between the maximum angle deviation and the minimum angle deviation within the window exceeds the threshold, and this difference is the "range fluctuation value", defined as: Δθ = max(θᵢ) - min(θᵢ), and when Δθ > T, it is regarded as abnormal), mark this path segment as an abnormal fluctuation segment and trigger a calibration instruction; S4: For the abnormal fluctuation segment, extract the average gradient vector of its adjacent path segments, and replace the direction vectors of all control points within the fluctuation segment with the average vector through the vector replacement strategy to generate calibrated path data, and synchronously update the motion trajectory parameters of the cutting device; S5: According to the calibrated path data, call the B-spline curve interpolation algorithm to perform smooth fitting on the connection of path segments to generate the ETFE film cutting trajectory, and drive the multi-axis linkage parameters of the cutting head based on the coordinate differences between adjacent path points in the trajectory to output a cutting instruction.

[0019] The candidate cutting path segment specifically includes the start and end point coordinates of the path segment and the gray change rate parameter. The path segment direction continuity sequence includes the horizontal gradient vector, the vertical gradient vector, and the angle deviation value. The abnormal fluctuation segment specifically refers to the abnormal path segment mark and the calibration trigger threshold. The calibrated path data includes the replaced control point vectors and the motion trajectory parameters. The ETFE film cutting trajectory specifically includes the B-spline curve parameters and the multi-axis linkage parameters.

[0020] The specific steps for obtaining the candidate cutting path segment are as follows: S101: Obtain the original image data of the ETFE film. This data is a grayscale image file with 1920x1080 pixels, and the file name is ETFE_raw.png. The average gray value of the main area of the film material in the image is stable at about 150, the average gray value of the background area is about 50, and the edge area shows a gray gradient due to the influence of light and the curved surface.

[0021] Perform convolution operations on the original image using Gaussian kernels of multiple scales. Three different scales of Gaussian kernels are selected, and their parameter settings are based on the response analysis of the edge features of the ETFE film image at different observation scales, aiming to capture both fine edge details and smooth overall contours simultaneously. The specific parameters are shown in Table 1.

[0022] Table 1 Gaussian kernel parameter table

[0023] As shown in Table 1, the specific sizes and standard deviations of the three Gaussian kernels used in this embodiment are listed.

[0024] For the original image data matrix Perform the first convolution using the Gaussian kernel of scale 1 in Table 1 (3x3, ). Align the center of the kernel with each pixel point in the image in turn, such as pixel point and its 3x3 neighborhood, calculate the weighted sum of the kernel coefficients and the corresponding neighborhood pixel values, and obtain the convolution result of this point at the first scale . Perform this operation on the entire image to generate the first set of convolution result matrices .

[0025] Next, perform the second convolution on the original image data matrix using the Gaussian kernel of scale 2 in Table 1 (5x5, ). Similarly, traverse all pixel points, calculate the weighted sum, and obtain the convolution result of pixel point at the second scale , and generate the second set of convolution result matrices .

[0026] Then, perform the third convolution on the original image data matrix using the Gaussian kernel of scale 3 in Table 1 (7x7, ). Calculate the convolution result of pixel point at the third scale , and generate the third set of convolution result matrices .

[0027] Linearly superimpose these three sets of convolution results according to the preset weight coefficients. The setting of the weight coefficients aims to balance the contributions of different scale information. According to empirical analysis, the medium scale (5x5 kernel) contributes the most to the extraction of the main contour of the ETFE film, while the small scale (3x3 kernel) helps to retain details, and the large scale (7x7 kernel) helps to suppress noise and connect discontinuous edges. Therefore, set the weight coefficients . Specifically, by evaluating the effects of a set of test images, it is found that when , the edge enhancement effect is the most balanced and satisfies . Calculate the final enhancement value of each pixel point. For pixel point , its enhanced pixel value is calculated as . Perform this linear superposition calculation on all pixel points to generate the final multi-scale edge enhanced image .

[0028] S102: Based on the multi-scale edge enhanced image , call the Sobel operator to calculate the pixel gradient components. Use the standard Sobel horizontal direction gradient kernel and the vertical direction gradient kernel .

[0029] For each pixel in , perform a convolution operation. First, extract the pixel values in its 3x3 neighborhood and perform a weighted sum with the kernel coefficients to obtain the horizontal gradient component of this point. Then, use the kernel to perform a convolution operation on the 3x3 neighborhood of the same point to obtain the vertical gradient component of this point. Repeat this process for all pixels to generate the complete horizontal gradient component matrix and the vertical gradient component matrix .

[0030] Based on the calculated gradient components and , synthesize the gradient magnitude of each pixel. The calculation formula is . Consider a pixel , and its gradient components obtained through convolution calculation are and , then its gradient magnitude is . The unit of the gradient magnitude here is gray intensity unit per pixel. Perform this calculation for all pixels to generate the gradient magnitude matrix .

[0031] Set a dynamic threshold according to the statistical characteristics of the gradient magnitude distribution of all pixels in the gradient magnitude matrix . First, calculate the average value and the standard deviation of all gradient magnitudes in the matrix . For the in this embodiment, the calculated is gray / pixel, gray / pixel. The dynamic threshold is set in the way to adapt to the contrast and noise level of the image and identify the pixels that are significantly higher than the average edge intensity. The value of the coefficient determines the strictness of the screening, and its setting is based on the test of multiple sample images containing clear edges and noise regions. When , too many noise points may be introduced; when , some weak edges may be lost. By testing, selecting can effectively suppress noise while retaining most of the real edges. Therefore, calculate the dynamic threshold gray / pixel.

[0032] Apply the gradient magnitude matrix The gradient magnitude of each pixel point in is compared with the calculated dynamic threshold . If , then it is determined that the pixel point is a pixel with gradient exceeding the limit. For the point calculated above, its gradient magnitude , since , this point is screened out. Record the coordinates of all pixel points that meet the condition, and generate a set of pixels with gradient exceeding the limit .

[0033] S103: For the set of pixels with gradient exceeding the limit , this set appears as a series of discrete points or short pixel chains on the image.

[0034] Perform a morphological closing operation to connect adjacent pixel points. Define a 3x3 pixel square structuring element . First, perform a morphological dilation operation on the binary image (the value of the pixel with gradient exceeding the limit is 1, and the rest are 0) represented by : Slide across the image. If the center of or any pixel it covers touches a pixel with a value of 1, then set the values of all pixel points covered by this to 1. This operation will expand the area of the pixels with gradient exceeding the limit and connect adjacent pixels. Then, perform a morphological erosion operation on the dilated result: Slide the same across the image. Only when all the pixel values within the area completely covered by are 1, the value of its central pixel point is retained as 1, otherwise it is set to 0. This operation removes small noise points and thin connecting bridges. The combination of dilation followed by erosion is the morphological closing operation, which can effectively fill small gaps and connect adjacent edge segments.

[0035] Perform connected component analysis on the binary image processed by the closing operation to identify all independent regions composed of connected pixels. Calculate the bounding rectangle of each independent closed region, that is, determine the smallest rectangular range that can completely enclose the region and whose sides are parallel to the image coordinate axes, and record its upper left coordinate and lower right coordinate .

[0036] Eliminate small isolated areas. Calculate the area of each bounding rectangle, with the unit being square pixels. Set an area threshold 。The setting of this threshold is based on the statistical analysis of the sizes of typical noise spots and the minimum size of effective edge segments in the ETFE film image. It is observed that the area of isolated regions formed by sensor noise or minute impurities is usually less than 50 square pixels, while the area of connected regions formed by meaningful edge segments is much larger than this value. Therefore, set square pixels. Compare the area calculated for each region with the threshold . If , then determine that the region is a small-area isolated region and remove all the pixel points it contains from the set. A calculation example is that the upper left corner of the circumscribed rectangle of a connected region is (50, 50), and the lower right corner is (54, 56), and its area square pixels. Since

[0037] , the region is removed. Retain the set of pixel points within all regions with an area . The set of these retained pixel points that are morphologically connected and have a large enough area constitutes the path segments. Generate the candidate cropping path segment set .

[0038] The specific steps for obtaining the path segment direction continuity sequence are as follows: S201: Invoke each path segment in the candidate cropping path segment set . Select one of the path segments , which contains a series of ordered pixel points .

[0039] To calculate the local direction information of each pixel point, the gradient information needs to be utilized again. Here, invoke the horizontal gradient component matrix and the vertical gradient component matrix calculated in S102. For each pixel point in the path segment ( ), directly read the gradient component values corresponding to this coordinate from the matrices and , that is, and .

[0040] Combine the gradient components of each pixel point into a two-dimensional gradient direction vector . Taking the pixel point in S102 as an example, if it belongs to the point in the path segment , then its gradient direction vector is . For the path segment Perform this operation on all pixel points, record the gradient direction vectors of all points, and generate the gradient direction vector set of this path segment. Perform this process for all path segments in to obtain the gradient direction vector sets of all candidate path segments respectively.

[0041] S202: Based on the gradient direction vector set of a certain path segment where ... ...

[0042] Calculate the gradient direction angle of each pixel point in the path segment. Use the two-parameter arctangent function atan2 for calculation, and the formula is ... This function can correctly determine the quadrant where the vector is located according to the signs of and and return a radian value within the range of ... For the vector in S201, its direction angle is radians (about 52.26 degrees). Calculate the direction angle ( ) of all pixel points in the path segment to obtain the direction angle sequence ...

[0043] Calculate the difference in direction angles between adjacent pixel points. For the th pair of adjacent pixel points in the path segment (i.e., and , ), calculate the difference in their direction angles ... To handle the periodicity of angles (e.g., jumping from near to near ), the calculated difference needs to be normalized to the range of ... The implementation method is: if , then ; if , then ... A specific calculation: if radians and radians, the direct subtraction is radians, and this value is less than , so an adjustment is made: radians. If radians and radians, then radians, and this value is within , so no adjustment is needed.

[0044] Calculate all of the difference in the direction angles of adjacent pixel points . Store these calculated and normalized differences in the direction angles in a queue data structure in the order in which they occur in the path segment (from to ). Construct the queue of the differences in the direction angles of this path segment . .

[0045] S203: Call the queue of the differences in the direction angles of the path segment , and prepare to calculate the continuity score of this path segment using the formula . The formula is: . , Obtaining and explanation of each parameter in the formula: : The difference in the direction angles (in radians) of the th pair of adjacent pixel points ( and ), and its value is obtained in order from the queue . Indicates taking the absolute value of this difference. : Are respectively the gradient magnitudes (i.e., gradient amplitudes, in gray levels / pixel) of the adjacent pixel points and . Their values are obtained from the gradient magnitude matrix calculated in S102 according to the coordinates of the points and and . : The Euclidean distance between the adjacent pixel points and . On the pixel grid, if two points are horizontally or vertically adjacent, pixels; if diagonally adjacent, pixels. : The total number of pairs of adjacent pixel points within the path segment, equal to the number of pixels in the path segment minus 1. : Find the maximum value among the squares of all the differences in the direction angles in the queue . Its square root is equal to , that is, the maximum absolute difference in the direction angles (in radians) in the entire path segment.

[0046] The calculation process of the formula is illustrated with an example path segment . This path segment contains 4 pixel points ​, so there are adjacent points. The relevant parameter values are shown in Table 2 below, where the gradient magnitude is obtained from S102, and the direction angle difference is obtained from the queue constructed in S202 , and the adjacent distance is calculated according to the pixel coordinates.

[0047] Table 2 Path Segment Example Parameter Table

[0048] As shown in Table 2, the specific parameter values used to calculate the path segment continuity score are listed.

[0049] Substitute into the formula for calculation: The first part (summation term): : ; : ; : ; Summation result = ; The second part (maximum absolute difference term): Calculate radians; Or calculate according to the formula radians.

[0050] Final continuity score ; This score is a comprehensive measure with a mixed unit (derived from a combination of radians, grayscale / pixel, and pixels), mainly used for relative comparison. The smaller the value, the smoother and more continuous the direction change of the path segment.

[0051] Set a preset threshold for path segment continuity . The setting of this threshold is based on scoring calculations and manual evaluations of path segments extracted from a large number (such as 100) of ETFE film sample images. The path segments are classified into three categories according to the visual effect: "continuously smooth" (scores generally lower than 25), "slightly fluctuating" (scores between 25 - 35), and "broken / abrupt turn" (scores higher than 35). To select high-quality initial path segments, set the threshold . This is an engineering setting value based on statistical analysis and application requirements, used to distinguish path segments with sufficiently good direction continuity.

[0052] Compare the calculated path segment continuity score with a preset threshold . Since , it indicates that the direction continuity of this path segment meets the requirements and is determined to be qualified. Therefore, retain this path segment. Repeat the process of S201 - S203 for all candidate cropping path segments in , and screen out all path segments that meet the conditions to generate a path segment direction continuity sequence .

[0053] The steps for obtaining the abnormal fluctuation segment are specifically as follows: S301: Obtain a path segment from the path segment direction continuity sequence , denoted as . This path segment contains a series of pixel points, and its corresponding queue of direction angle differences is known, where (unit: radian) is the angle deviation value of the th pair of adjacent pixel points.

[0054] Set an analysis window with a fixed size, and the window size is defined as containing 5 consecutive pixel points. Such a window covers 4 consecutive angle deviation values (because the angle deviation is between adjacent points). Traverse the path segment in the way of window sliding.

[0055] Initially, the first window covers the first to 5 pixel points of the path segment start, so the analyzed angle deviation values are . Calculate the range of these 4 angle deviation values within this window (denoted as ), that is, find the maximum value and the minimum value among these 4 values, and calculate the range . Let the deviation value sequence within the window be radians, then radians,[[]] radians, and the range of this window radians.

[0056] Record the range value radians of the first window. Then, slide the window forward one pixel position along the path segment, and the second window covers to 5 pixel points, and the involved angle deviation values are Calculate the range of the angular deviation values within this new window Repeat this sliding and calculation process, with the windows being For each position on the path segment that can form a complete 5-point window, calculate and record the corresponding range of angular deviation values. Generate a sequence of range values for the dynamic windows .

[0057] S302: Call the sequence of range values for the dynamic windows For each range value in the sequence (corresponding to the th window), use the following formula to calculate and obtain the coefficient for judging the range fluctuation of this window :[[]] ; Obtaining and explanation of each parameter in the formula (taking the calculation of the th window's as an example): : They are respectively the maximum and minimum values of the angular deviation values within the th window (unit: radian). These two values have been obtained when calculating the range in S301. : The average value of the difference in the changing trend of the ranges of the adjacent windows corresponding to the th window (unit: radian). This value reflects the local smoothness of the range change of the current window. The calculation method is: Consider the current window and the ranges and of one window before and after it (if or , only consider one side). Calculate the absolute values of the changes in the adjacent ranges and . Take the average value of these absolute values of the changes. If and , then . : The absolute value of the average curvature change rate of the path segment covered by the th window. The curvature can be approximately calculated from the angular deviation and the pixel pitch , that is (unit: radian / pixel).

[0058] The curvature change rate is the average value of the differences in the curvatures of adjacent points calculated within the window. Therefore Note that the unit here is (radian / pixel) / pixel = radian / pixel². : Environmental interference factor, which is a dimensionless adjustment coefficient. Its setting is based on the evaluation of the overall quality of the input image, especially the noise level. It is quantified by calculating the signal-to-noise ratio (SNR) of the image. Setting rule: If SNR > 30dB (judged as a low-noise image), ; if 20dB < SNR 30dB (judged as a medium-noise image), ; if SNR 20dB (judged as a high-noise image), . This factor is used to adjust the sensitivity of the algorithm to noise fluctuations.

[0059] Continuing with the example of S301, calculate the coefficient of the first window . Given rad, rad, rad, rad. It is necessary to calculate . This requires knowing . Assume the deviation values of window are [-0.10, 0.08, 0.02, 0.06] rad, then , the range rad. Since it is the first window, only consider the difference from : . To avoid a zero denominator, a small positive lower limit will be set in practical applications, or the average value of a wider range (such as the previous and next 2 windows) will be used. Here, the correction rule is: If the calculated is less than a small value (such as ), then it is set to this small value. In this example, set the lower limit to , then rad. It is necessary to calculate . The points involved in window are , and the angular deviation is . The corresponding pixel distance is assumed to be pixels.

[0060] Calculate the approximate curvature: , , , (unit: rad / pixel).

[0061] Calculate the absolute value of the curvature change: ; ; .

[0062] Calculate the average curvature change rate rad / pixel²; Set the environmental interference factor . By analyzing the input image ETFE_raw.png, its SNR is calculated to be approximately 25 dB, belonging to the medium noise level. According to the set rules, take .

[0063] Substitute the obtained values into the formula for calculation : ; ; ; ; , this value is very large, mainly because is close to zero. Let's reconsider 's calculation, using a more stable method, such as calculating the average of the absolute values of the range changes of the two windows before and after the calculation window , and setting a reasonable minimum denominator value. Or, taking the calculation window as an example (assuming is not a boundary), its range . Its previous window range , and the next window range .

[0064] Calculate rad; Use the rad / pixel² calculated above (assuming this value is also applicable to the window ) and .

[0065] ; ; ; ; , this The coefficient is a dimensionless comprehensive score used to determine whether there are abnormal fluctuations within a window. For each valid window position on the path segment, the corresponding value is calculated.

[0066] S303: Call the range fluctuation determination coefficient of each window position , for example .

[0067] Set a preset path fluctuation determination threshold . The setting of this threshold is based on calculating the coefficient for a large number of path segment samples that contain known normal bends (such as designed arcs) and known abnormal fluctuations (such as material wrinkles, sawteeth caused by image acquisition jitter). Statistical findings show that the values of normal bend segments are usually distributed between 5 and 10, while the values of abnormal fluctuation segments are significantly higher, with most exceeding 15. To effectively identify abnormal fluctuations and avoid misjudging normal large curvature changes, set the threshold . This is an engineering setting value based on empirical data and discrimination.

[0068] Compare the calculated range fluctuation determination coefficient of a specific window with the preset threshold . Perform the judgment: If , then determine that the path segment covered by this window (that is, the part from the starting pixel point to the ending pixel point of the window) has abnormal fluctuations. In our calculation example, . Because , the path segment covered by this window is not marked as an abnormal fluctuation segment. If there is another window , and its calculated coefficient is . Because , the path segment covered by this window will be marked as an abnormal fluctuation segment.

[0069] Perform this comparison and judgment on all calculated values on the path segment. Mark the path segments covered by all windows determined to have a coefficient exceeding the threshold . Since the windows are sliding, there may be cases where multiple adjacent or overlapping windows are marked. Merge the set of pixel points covered by all marked and possibly overlapping windows to form one or more continuous abnormal fluctuation segments. Finally, output the set of these abnormal fluctuation segments.

[0070] The specific steps for obtaining the calibrated path data are as follows: S401: Obtain a marked abnormal fluctuation segment . Meanwhile, from the path segment direction continuity sequence , identify the normal path segment that is not marked as abnormal and is adjacent in path order before it, and the normal path segment that is not marked as abnormal and is adjacent after it. .

[0071] From the data of these two adjacent normal path segments and , extract the gradient vectors of all control points (pixel points) inside them. These vectors are obtained in S201. Let contain control points, and its gradient vector set is . Let contain control points, and its gradient vector set is .

[0072] Calculate the arithmetic mean of all gradient vectors in these two adjacent normal segments and to obtain an adjacent gradient mean vector representing the stable gradient direction of this area . The calculation method is: ; Taking a specific scenario as an example, contains points, and its gradient vector is . contains points, and its gradient vector is . The two segments have a total of points. Calculate the average x component: ; Calculate the average y component: ; The finally generated adjacent gradient mean vector is .

[0073] S402: Based on the adjacent gradient mean vector . Traverse all control points (pixel points) inside the marked abnormal fluctuation segment . Let the th control point inside the abnormal segment be , and its original gradient direction vector is (from S201).

[0074] For each control point within the abnormal fluctuation segment perform a direction vector replacement operation: directly replace its original gradient direction vector with the adjacent gradient mean vector calculated in S401 . That is, the updated vector . The basis for this operation is that it is considered that the true direction of the abnormal segment should smoothly transition to the directions of the stable segments on both sides. Therefore, the average gradient directions on both sides are used to correct the direction information of all points within the abnormal segment.

[0075] After replacing the direction vector, the direction parameters of this control point must be updated synchronously, mainly the direction angle . Use the replaced vector to calculate the new direction angle : radians (about 50.9 degrees). For all control points within the abnormal segment , their direction angles will be uniformly updated to this calculated radians. In this step, the coordinates of the control point remain unchanged, and only its associated direction parameters are updated.

[0076] Collect all the control points within all abnormal fluctuation segments after direction vector replacement and direction parameter update to generate a replacement vector control point set .

[0077] S403: Call the data of the unaffected path segments in the original path, that is, the generated by S203, excluding all abnormal fluctuation segments marked by S303 and denote the remaining part as . This part of the data contains the coordinates and original parameters of the path segments considered to be reliable. At the same time, call the replacement vector control point set that has been directionally calibrated and generated by S402 (it represents the calibrated version of the original abnormal fluctuation segment

[0078] Integrate these two parts of data and . According to their order in the original complete path, insert into the corresponding position in , that is, replace the original with the calibrated to form a preliminary integrated complete path point sequence containing all points.

[0079] On this basis, reconstruct the connection relationship and curvature parameters between adjacent control points on the path. Especially at the interface between the normal segment and the calibration segment (i.e., between the end point of and the start point of , and between the end point of and the start point of ), it is necessary to recalculate the distance , the difference in direction angles . And according to the new, integrated point sequence and the updated direction parameters (the original parameters from and the calibrated parameters from ), recalculate the curvature parameters of each point on the entire path. At the connection points, to ensure a smooth transition of the path, it is necessary to check and ensure the continuity of the curvature. If there are obvious jumps, local smoothing processing is required, such as by fine-tuning the parameters of the control points near the connection points. Finally, generate the calibrated path data that includes the coordinates of all control points, the updated direction parameters, and the recalculated connection relationship and curvature parameters.

[0080] The steps for obtaining the cutting trajectory of the ETFE film are specifically as follows: S501: Call the calibrated path data . This data is a series of ordered control points , and each point has coordinates (unit: pixel) and the updated direction or curvature parameters (unit: radian / pixel).

[0081] Extract the key information required for subsequent curve fitting. Select all the control points on the path as the nodes for fitting. Extract the coordinates and curvature parameters of each node. Calculate the Euclidean distance and between adjacent control points (unit: pixel). Calculate the angle difference between adjacent tangent vectors, or the angle change amount (unit: radian) defined by three consecutive control points .

[0082] Set the knot vector for B-spline or NURBS curve fitting. The setting of the knot vector affects the shape and parameterization process of the curve. Use the chord length parameterization method to set the knot vector, which can better reflect the spatial distance between the control points. Calculate the cumulative chord length: , for . Then normalize the entire node sequence to the interval [0, 1] by dividing by the total chord length as follows: . For cubic B-splines (d = 3), the complete knot vector also needs to have repeated knots added at both ends, in the form of . Consider a path that contains 4 control points: pixels.

[0083] Calculate the distances: pixels; pixels; pixels.

[0084] Calculate the cumulative chord lengths: . The total length .

[0085] Normalize the internal knots: ; .

[0086] If using cubic B-splines, the knot vector .

[0087] Set the weight coefficients . For standard B-spline curve fitting, the weight coefficients of all control points are all set to 1.

[0088] Integrate the extracted control point coordinates , curvature , the calculated distances , the angle differences , as well as the generated knot vector and the weight coefficients to generate a set of path node parameters for curve fitting .

[0089] S502: Based on the set of path node parameters , specifically the control point coordinates , the knot vector and the weights , perform fitting using a cubic B-spline curve.

[0090] Divide the parameter domain (usually [0, 1]) into multiple parameter sub-intervals according to the knot vector . Within each parameter sub-interval , the points on the curve ​Given by the weighted sum of control points and the corresponding cubic B-spline basis functions as follows: , where the basis functions are defined by the knot vector through the Cox-deBoor recurrence formula. Calculate the specific cubic polynomial expressions for the coordinate components and in each subinterval (i.e., determine the coefficients and ).

[0091] To ensure that the generated cutting trajectory is smooth enough to meet the cutting process requirements, it is necessary to adjust the curve parameters to meet continuity, i.e., curvature continuity. The standard cubic B-spline curve automatically satisfies continuity at non-repeated knots. If the knot vector design or control point arrangement results in non-satisfaction of continuity at some knots (which is not common in the standard construction unless there are special designs of the knot vector or the control points are collinear and other degenerate cases), then it is necessary to adjust the positions of the control points or increase the number of control points to ensure that at all connection points, the first derivative (tangent) and the second derivative (related to curvature) of the curve are continuous.

[0092] Perform B-spline curve calculations for all parameter intervals. By densely sampling in the range of the parameter from 0 to 1, a series of coordinate points that define the entire smooth curve are generated. The selection of the sampling step determines the density of the trajectory points. According to the required cutting accuracy, set . For each sampled parameter , calculate the corresponding curve point . Generate the fitted trajectory coordinate set , where . The coordinates of these points are still in pixels.

[0093] S503: Traverse the fitted trajectory coordinate set , where are the trajectory points arranged in order (unit: pixel).

[0094] Calculate the curvature change rate between adjacent interpolation points on the trajectory and perform unit conversion to match the physical cutting requirements. First, establish the conversion rule between pixel coordinates and physical coordinates (millimeters, mm). Through camera calibration or reference to the known object size, determine the physical scale of the image. Set the conversion rule: 1 millimeter corresponds to 10 pixels in the image. That is, the scale factor pixels_per_mm = 10 pixels / mm. Therefore, 1 pixel = 0.1 millimeter.

[0095] Calculate each sampled point on the trajectory Curvature . Since the analytical expression of the B-spline curve is known, its curvature can be accurately calculated through its first and second derivatives: ; The calculated curvature has the unit of 1 / pixel. According to the conversion rule, it is converted to the physical unit of 1 / mm:

[0096] Calculate the physical distance between adjacent trajectory points and : ; Calculate the physical curvature change rate between adjacent points (unit: (1 / mm) / mm = 1 / mm²): ; Set a curvature continuity threshold . The setting of this threshold is based on the specific properties of the ETFE film material and the dynamic performance of the cutting tool used (such as a laser cutting head). Excessive curvature changes may lead to unstable cutting speed, poor cutting edge quality, or material tearing. Through experimental tests, for laser cutting of an ETFE film with a thickness of 0.2 mm, it is found that when the curvature change rate exceeds 0.8 mm⁻², the cutting quality begins to become unstable. Therefore, set mm⁻².

[0097] Compare the calculated curvature change rate of each point with the threshold . Screen out all points that satisfy , and these points are marked as "curvature jump points".

[0098] For each screened curvature jump point , perform re-interpolation compensation on the local path segment near it. The method is to increase the control point density of the B-spline near this point, or adjust the positions of the nearby control points, so that the curvature change of the re-fitted local curve at this point is smoother, ensuring that its value is lower than the threshold mm⁻².

[0099] After compensating for all detected curvature jump points, traverse and check the curvature change rate of the entire trajectory again to ensure that all points . The final output path point sequence, whose coordinates need to be converted to physical units (millimeters). For each point (pixels), the output coordinate is ​(mm). This coordinate sequence after smoothing and unit conversion is the final ETFE film cutting trajectory. . The following Table 3 shows a small segment of example data of the final trajectory.

[0100] Table 3 Example Table of ETFE Film Cutting Trajectory Segment

[0101] As shown in Table 3, a small segment of data in the finally generated ETFE film cutting trajectory is listed, including physical coordinates, the curvature of each point, and the curvature change rate between adjacent points. All curvature change rates are lower than the set threshold of 0.8 mm⁻².

[0102] The ETFE film cutting control system based on image recognition is used to execute the above-mentioned ETFE film cutting control method based on image recognition. The system includes: Edge enhancement module, which is used to perform filtering and noise reduction on the ETFE film image through a multi-scale Gaussian filtering algorithm, call the Sobel gradient operator to calculate the horizontal and vertical gray change rates, screen the set of continuous pixel points whose gray change rates exceed the dynamic threshold, generate candidate cutting path segments and output their start and end point coordinates, and transfer the candidate cutting path segments to the direction continuity analysis module; Direction continuity analysis module, which is used for adjacent pixel points of the candidate cutting path segment, constructs the gradient direction vectors of each pixel point by using the vector space projection algorithm, calculates the direction angle between adjacent vectors, generates the path segment direction continuity sequence, stores the angle deviation value between the angle and the preset direction, and transfers the path segment direction continuity sequence to the abnormal fluctuation determination module; Abnormal fluctuation determination module, which is used to perform a sliding window calculation on the angle deviation values of 5 consecutive pixel points based on the path segment direction continuity sequence by calling the dynamic window range determination method, judge whether the range within the window exceeds the set threshold, mark the path segment with an excessive range as an abnormal fluctuation segment and generate a calibration instruction, and transfer the abnormal fluctuation segment to the path calibration module; Path calibration module, which is used to extract the mean value of the gradient direction vectors of adjacent path segments of the abnormal fluctuation segment, replace the direction vectors of all control points within the fluctuation segment with the mean vector through the vector replacement strategy, generate the calibrated path data and update the motion trajectory parameters of the cutting device, and transfer the calibrated path data to the trajectory generation module; Trajectory generation module, which is used to perform smooth fitting on the connection of path segments based on the calibrated path data by calling the B-spline curve interpolation algorithm, calculate the coordinate differences between adjacent path points to generate the ETFE film cutting trajectory, and drive the multi-axis linkage parameters of the cutting head to output cutting instructions.

[0103] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An ETFE film cutting control method based on image recognition, characterized in that: The following steps are involved: S1: Perform edge enhancement on the ETFE film image through multi-scale Gaussian filtering, extract pixel points whose grayscale change rate exceeds the dynamic threshold based on the Sobel gradient operator, generate candidate clipping path segments, and extract the starting and ending pixel coordinates of each path segment; S2: for the adjacent pixel points of the candidate clipping path segment, call the vector space projection algorithm, calculate the gradient direction vector of each point in the horizontal and vertical directions, construct a path segment direction continuity sequence, and store its angle deviation value from the preset clipping direction; S3: Based on the continuity sequence of the path segment direction, a dynamic window range determination method is used to perform window sliding calculation on the angle deviation values ​​of five consecutive pixel points. If the range in the window exceeds a set threshold, the path segment is marked as an abnormal fluctuation segment, and a calibration instruction is triggered; S4: For the abnormal fluctuation segment, extract the mean of the gradient vectors of its adjacent path segments, replace the direction vectors of all control points in the fluctuation segment with the mean vector through the vector replacement strategy, generate calibrated path data, and synchronously update the motion trajectory parameters of the cutting equipment.

2. The ETFE film cutting control method based on image recognition according to claim 1, characterized in that: The candidate clipping path segment specifically includes the coordinates of the starting and ending points of the path segment and the grayscale change rate parameters. The path segment directional continuity sequence includes the horizontal gradient vector, the vertical gradient vector and the angle deviation value. The abnormal fluctuation segment specifically refers to the abnormal path segment mark and the calibration trigger threshold. The calibrated path data includes the replaced control point vector and the motion trajectory parameters.

3. The ETFE film cutting control method based on image recognition according to claim 2 is characterized in that: The steps of obtaining the candidate clipping path segments are specifically as follows: S101: acquiring original data of the ETFE film image, performing convolution operation on the image using Gaussian kernels of multiple scales, linearly superimposing multiple groups of convolution results according to preset weight coefficients, and generating a multi-scale edge-enhanced image; S102: Based on the multi-scale edge enhanced image, calling the Sobel operator to calculate the pixel gradient component, synthesizing the gradient amplitude matrix, setting a dynamic threshold according to the statistical characteristics of the gradient amplitude distribution, screening the pixel points whose gradient amplitude exceeds the threshold, and generating a gradient over-limit pixel set; S103: For the gradient excess pixel set, perform a morphological closing operation to connect adjacent pixel points, extract the circumscribed rectangular boundary of the closed area, remove small isolated areas, retain the remaining path segments, and generate candidate clipping path segments.

4. The ETFE film cutting control method based on image recognition according to claim 3 is characterized in that: The steps for obtaining the path segment direction continuity sequence are specifically as follows: S201: calling the horizontal and vertical convolution kernel matrices, performing point-by-point convolution operations on adjacent pixel points of the candidate clipping path segment, recording the horizontal gradient component and the vertical gradient component, and generating a gradient direction vector set; S202: Calculating the direction angles of multiple pixel points based on the gradient direction vector set, defining the direction angle differences of adjacent pixel points and storing them in a queue in order, and constructing a direction angle difference queue; S203: Call the direction angle difference queue, using the formula: ； Calculate the continuity score of the path segment, compare it with the preset threshold to screen the path segment, and generate the path segment direction continuity sequence; in, represents the path segment direction continuity score, is the difference in direction angles between adjacent pixels, , is the gradient modulus length of adjacent pixels, is the Euclidean distance between adjacent pixels, is the total number of adjacent pixel pairs in the path segment.

5. The ETFE film cutting control method based on image recognition according to claim 4 is characterized in that: The steps for obtaining the abnormal fluctuation segment are specifically as follows: S301: Obtaining angle deviation values ​​of five consecutive pixel points in the path segment direction continuity sequence, traversing the path segment in a window sliding manner, calculating the range of the angle deviation value in each window, and generating a dynamic window range value; S302: calling the dynamic window range value, based on the range change trend difference of adjacent windows and the curvature change rate of the path segment, using the formula: ； Obtain the range fluctuation determination coefficient by calculation, and generate the range fluctuation determination coefficient by weighted superposition; in, represents the coefficient of determination of the range fluctuation, Represents the maximum angle deviation in the window, Represents the minimum angle deviation in the window, Represents the mean of the trend difference of the range change of adjacent windows, represents the rate of change of curvature of the path segment, represents environmental disturbance factors; S303: calling the extreme difference fluctuation determination coefficient and comparing it with the preset path fluctuation determination threshold. If the coefficient exceeds the threshold, marking the path segment covered by the corresponding window as an abnormal fluctuation segment.

6. The ETFE film cutting control method based on image recognition according to claim 5, characterized in that: The steps for obtaining the calibrated path data are specifically as follows: S401: Acquire the data of the adjacent path segments on both sides of the abnormal fluctuation segment, extract the gradient vectors of the control points in the path segment, calculate the arithmetic mean of the gradient vectors of the adjacent segments, and generate an adjacent gradient mean vector; S402: based on the adjacent gradient mean vector, traverse the direction vectors of the control points in the abnormal fluctuation segment, compare the multi-component values ​​of the direction vector and the mean vector, replace the original components with the mean components, update the coordinates and direction parameters of the control points, and generate a replacement vector control point set; S403: calling the coordinates of the unaffected path segments in the original path, integrating the replacement vector control point set, reconstructing the connection relationship and curvature parameters of adjacent control points, and generating calibrated path data.

7. The ETFE film cutting control method based on image recognition according to claim 6 is characterized in that: The method further comprises: S5: according to the calibrated path data, calling the B-spline curve interpolation algorithm to perform smooth fitting on the connection of the path segments, generating the ETFE film cutting trajectory, driving the multi-axis linkage parameters of the cutting head based on the coordinate difference of adjacent path points in the trajectory, and outputting the cutting instruction; The ETFE film cutting trajectory is specifically a B-spline curve parameter and a multi-axis linkage parameter.

8. The ETFE film cutting control method based on image recognition according to claim 7 is characterized in that: The steps for obtaining the ETFE film cutting trajectory are specifically as follows: S501: calling the calibrated path data, extracting the coordinates and curvature parameters of the control points at the connection of the path segments, calculating the distance and angle difference between adjacent control points, setting the node vector and weight coefficient, and generating a path node parameter set; S502: Based on the path node parameter set, divide the interpolation interval and solve the cubic polynomial coordinate components of each interval, adjust the high-order derivative terms of the polynomial coefficient matrix to meet the curvature continuity condition, and generate a fitting trajectory coordinate set; S503: traverse the fitting trajectory coordinate set, calculate the curvature change rate of adjacent interpolation points, screen the jump points exceeding the continuity threshold, re-interpolate and compensate and output the path sequence to generate the ETFE film cutting trajectory.

9. The ETFE film cutting control system based on image recognition is characterized in that: The system is used to implement the ETFE film cutting control method based on image recognition according to any one of claims 1 to 8, and the system comprises: The edge enhancement module is used to perform filtering and noise reduction on the ETFE film image through a multi-scale Gaussian filtering algorithm, call the Sobel gradient operator to calculate the horizontal and vertical grayscale change rates, screen the continuous pixel point set whose grayscale change rate exceeds the dynamic threshold, generate candidate clipping path segments and output their starting and ending point coordinates, and pass the candidate clipping path segments to the directional continuity analysis module; A directional continuity analysis module is used to construct the gradient direction vector of each pixel point using a vector space projection algorithm for the adjacent pixel points of the candidate clipping path segment, calculate the direction angle between adjacent vectors, generate a path segment direction continuity sequence, store the angle deviation value between the angle and the preset direction, and transmit the path segment direction continuity sequence to the abnormal fluctuation determination module; An abnormal fluctuation determination module is used to call a dynamic window range determination method to perform sliding window calculation on the angle deviation values ​​of five consecutive pixel points based on the path segment direction continuity sequence, determine whether the range in the window exceeds a set threshold, mark the path segment with an excessive range as an abnormal fluctuation segment and generate a calibration instruction, and transmit the abnormal fluctuation segment to a path calibration module; A path calibration module is used to extract the mean of the gradient direction vectors of the path segments adjacent to the abnormal fluctuation segment, replace the direction vectors of all control points in the fluctuation segment with the mean vector through a vector replacement strategy, generate calibrated path data and update the motion trajectory parameters of the cutting device, and pass the calibrated path data to the trajectory generation module; The trajectory generation module is used to call the B-spline curve interpolation algorithm to perform smooth fitting on the connection of the path segments based on the calibrated path data, calculate the coordinate difference of adjacent path points to generate the ETFE film cutting trajectory, and drive the multi-axis linkage parameters of the cutting head to output cutting instructions.

Images