Bent plate image detection method for bending machine processing

Abstract

The invention discloses a bent plate image detection method used for bending machine processing, and relates to the field of image detection, and the method comprises the steps: obtaining the image data of a to-be-processed plate, and extracting an edge contour point sequence; calculating a curvature gradient sequence, identifying segmentation boundary points according to the curvature gradient sequence and a preset threshold value, and dividing segmentation areas; performing linear fitting on each segmented region to obtain a unit normal vector sequence; calculating normal difference values of adjacent segmented areas to form a normal difference value sequence; judging an error level according to the normal difference value sequence and a preset angle interval; marking a suspicious section according to the error level and a suspicious section judgment rule; and identifying and correcting the section needing to be corrected. According to the method, accurate detection of the bent plate image is realized through curvature gradient analysis and normal vector calculation, an abnormal region in the image can be automatically identified and corrected, and meanwhile, the problem of overall judgment failure caused by a primary detection error can be effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of image detection, specifically to a method for detecting images of bent sheet metal used in bending machine processing. Background Technology

[0002] With the rapid development of industrial automation and intelligent manufacturing, bending machines, as crucial equipment for metal sheet processing, have a decisive impact on product quality due to their processing accuracy and efficiency. During bending machine processing, sheet metal image detection technology has become a key link in ensuring processing quality. By accurately identifying and analyzing the edge contours of the sheet metal, abnormalities in the processing can be effectively detected, improving the product qualification rate.

[0003] In current bending inspection practices, visual sensors are typically used to capture images of the bending area of ​​the sheet metal. Edge detection and fitting methods are then used to identify bending line features to assess the bending angle and error. However, these methods largely rely on straight-line fitting of the overall contour, lacking detailed analysis of local variations in complex contours. They cannot accurately identify minute bending anomalies or slight deformation areas, especially when the sheet metal has complex multiple folds, subtle variations, or localized error clusters. Traditional methods are prone to misjudgment or missed detection. Furthermore, existing technologies often lack systematic error level judgment and feedback correction mechanisms, making it difficult to establish a stable, closed-loop error processing flow at the image level, thus limiting their application in high-precision bending processes.

[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for detecting images of bent sheet metal used in bending machine processing.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention discloses a method for detecting images of bent sheet metal used in bending machine processing, comprising the following steps:

[0008] Acquire image data of the sheet material to be processed, and extract edge contour point sequences from the image data to form a contour point set;

[0009] Calculate the curvature gradient sequence based on the set of contour points; identify segment boundary points and divide the set of contour points into several segmented regions based on the curvature gradient sequence and a preset curvature gradient jump threshold;

[0010] Linear fitting is performed on the points in each segmented region to obtain the unit normal vector of each segmented region, forming a normal vector sequence;

[0011] Based on the normal vector sequence, calculate the normal difference between unit normal vectors in adjacent segment regions to form a normal difference sequence;

[0012] Based on the normal difference sequence and the preset angle range, the error level of each segment region is determined, and the error level includes a first level, a second level and a third level.

[0013] Based on the error level and the preset suspicious segment judgment rules, suspicious segments are marked in the segmentation area;

[0014] Based on the error level and the suspicious segment, the segment that needs correction is identified, and the segment that needs correction is corrected.

[0015] In a second aspect, the present invention discloses an image detection system for bent sheet metal used in bending machine processing, comprising:

[0016] The data acquisition module is used to acquire image data of the sheet material to be processed, and extract edge contour point sequences from the image data to form a contour point set;

[0017] The region segmentation module is used to calculate the curvature gradient sequence based on the set of contour points; and to identify segment boundary points and divide the set of contour points into several segmented regions based on the curvature gradient sequence and a preset curvature gradient jump threshold.

[0018] The normal fitting module is used to perform linear fitting on the points in each segmented region to obtain the unit normal vector of each segmented region, thus forming a normal vector sequence.

[0019] The error analysis module is used to calculate the normal difference between unit normal vectors in adjacent segment regions based on the normal vector sequence, forming a normal difference sequence; and to determine the error level of each segment region based on the normal difference sequence and a preset angle interval, wherein the error level includes a first level, a second level, and a third level.

[0020] The feedback correction module is used to mark suspicious segments in the segmented area according to the error level and the preset suspicious segment judgment rules; identify segments that need to be corrected according to the error level and the suspicious segments, and correct the segments that need to be corrected.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] By employing a boundary recognition method based on a combination of curvature gradient and normal difference, precise identification of minute anomalies in complex contours is achieved, demonstrating higher contour recognition accuracy compared to single edge detection methods. An angle threshold is used to subdivide error levels, providing clear criteria for subsequent control strategies and correction operations, thus achieving quantitative classification of error levels. Local secondary sampling and correction are performed using updated edge information, forming an intelligent feedback mechanism that effectively avoids the problem of overall judgment failure due to a single detection error. It supports trend detection and response correction for typical error combinations, possessing certain trend recognition and adaptive capabilities, and is suitable for various types of bending deformation scenarios. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is an overall block diagram of the method in Embodiment 1 of the present invention;

[0025] Figure 2 This is a flowchart of the correction operation in Embodiment 1 of the present invention;

[0026] Figure 3 This is an overall block diagram of the system in Embodiment 2 of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Application Overview: In existing technologies, sheet metal bending inspection mainly relies on visual sensors to acquire images of the sheet metal and then assess the bending angle by fitting the overall contour straight line. This method lacks sensitivity to minor local deformations in complex contours, making it difficult to identify multi-bend areas or areas where errors accumulate, easily leading to false positives and missed detections. When the sheet metal has continuous bends or local material springback, traditional inspection methods cannot effectively distinguish between normal processing fluctuations and abnormal deformation areas, resulting in limited quality control accuracy.

[0029] To address the aforementioned issues, the research and development process revealed that existing technologies lack quantitative analysis methods for local curvature changes in the profile, making it impossible to establish a correlation mechanism between error levels and correction feedback. By studying the geometric characteristics of the sheet metal bending profile, a correspondence was found between curvature abrupt change points and bending boundaries, and the difference in normal vectors between adjacent regions can reflect the degree of processing error. Based on this, a proposed approach is to divide the profile into multiple segmented regions, identify boundary points through curvature gradients, and construct an error level system by combining normal difference values, ultimately forming a closed-loop correction process.

[0030] Example 1:

[0031] like Figure 1-2 As shown, the method for image detection of bent sheet metal used in bending machine processing includes the following steps:

[0032] Acquire image data of the sheet material to be processed, and extract edge contour point sequences from the image data to form a contour point set;

[0033] Calculate the curvature gradient sequence based on the set of contour points; identify segment boundary points and divide the set of contour points into several segmented regions based on the curvature gradient sequence and a preset curvature gradient jump threshold;

[0034] Linear fitting is performed on the points in each segmented region to obtain the unit normal vector of each segmented region, forming a normal vector sequence;

[0035] Based on the normal vector sequence, calculate the normal difference between unit normal vectors in adjacent segment regions to form a normal difference sequence;

[0036] Based on the normal difference sequence and the preset angle range, the error level of each segment region is determined, and the error level includes a first level, a second level and a third level.

[0037] Based on the error level and the preset suspicious segment judgment rules, suspicious segments are marked in the segmentation area;

[0038] Based on the error level and the suspicious segment, the segment that needs correction is identified, and the segment that needs correction is corrected.

[0039] This application further proposes that the image data be a sheet metal image containing the complete bending area, and the edge contour point sequence is a set of two-dimensional coordinate points arranged in the order of the image rows or columns.

[0040] The complete bending area image refers to an image covering all bending areas of the sheet metal, acquired through image acquisition equipment. This can be achieved using an industrial camera with a multi-angle supplementary lighting system to ensure that the bending line, transition area, and adjacent planar areas are completely captured. This feature is used to avoid contour analysis errors caused by missing parts of the image. The edge contour point sequence is a set of two-dimensional coordinate points arranged sequentially along the row or column direction of the image. This can be generated by using an edge detection algorithm to scan pixels along the image rows or columns, extracting edge point coordinates, and arranging them in the scan order. This feature provides an ordered data foundation for subsequent curvature gradient calculations by maintaining the spatial continuity of the point sequence.

[0041] Specifically, during the image acquisition stage, the sheet metal is placed on the bending machine's worktable. By adjusting the camera angle and light source layout, complete imaging of the bending area is ensured. For example, when the sheet metal has multiple bending surfaces, the image needs to cover all bending lines and their adjacent areas. In the edge extraction stage, the Canny edge detection algorithm is used to scan the image row by row or column by column, converting detected edge pixels into two-dimensional coordinate points and generating a point sequence according to the scanning order. Thus, the contour point set not only contains bending line features but also retains the spatial distribution pattern of points in the original image, enabling subsequent curvature gradient calculations to accurately reflect the local change trend of the contour.

[0042] Through the above technical solution, this application can effectively solve the problem of misjudgment of bending lines caused by incomplete image coverage or disordered point sequences in traditional detection, and improve the reliability of contour segmentation and error recognition. The complete bending region image provides a comprehensive data foundation for subsequent analysis, while the ordered edge point sequence ensures the logical consistency of curvature gradient calculation, thereby achieving more stable detection results in complex bending scenarios.

[0043] This application further proposes a process for calculating the curvature gradient sequence, which includes: calculating the curvature value of a local window consisting of three consecutive points in the contour point set to form a curvature sequence; and calculating the difference between adjacent curvature values ​​in the curvature sequence to form a curvature gradient sequence.

[0044] The local window refers to a set of three consecutively arranged two-dimensional coordinate points. This can be achieved by using a sliding window to traverse the contour point set. The geometric relationships of the local point set capture the contour curvature variation features, providing foundational data for subsequent segment boundary identification. The curvature gradient sequence refers to the sequence of differences between adjacent curvature values. This can be achieved by performing a first-order difference operation on the curvature sequence. This sequence quantifies the severity of contour curvature changes and is used to detect abrupt changes in the contour.

[0045] Specifically, in the sheet metal edge profile analysis process, three consecutive points are selected as computational units, and curvature values ​​characterizing the degree of local bending are obtained through geometric operations. For example, for a set of consecutive points, the angle or tangent change rate formed by them can be calculated as curvature values. After all profile points have been traversed, a complete curvature sequence is formed. Subsequently, by calculating the difference between adjacent curvature values, a curvature gradient sequence is generated. The high-amplitude regions in this sequence correspond to the bending boundaries or abnormal deformation locations of the profile, thereby achieving precise segmentation of the profile.

[0046] Through the above technical solution, this application effectively solves the problem that traditional detection methods are not sensitive to local curvature changes. By quantitatively analyzing the curvature gradient sequence, it improves the accuracy of segment boundary identification, provides a reliable basis for segmented region division for subsequent error level judgment, and avoids misjudgment or missed detection of error level caused by segmentation errors.

[0047] This application further proposes that the preset curvature gradient jump threshold is set empirically based on historical sample data; the process of identifying segment boundary points is to identify points in the curvature gradient sequence whose values ​​exceed the preset curvature gradient jump threshold.

[0048] The curvature gradient jump threshold is a critical value used to determine when a significant change in the profile curvature occurs. Specifically, it can be determined by statistically analyzing the curvature gradient distribution range of different sheet metal types in historical samples and selecting the corresponding quantile as an empirical threshold. This threshold can dynamically adapt to differences in profile changes caused by different sheet metal materials or processing conditions. The process of identifying segment boundary points involves using points in the curvature gradient sequence that exceed the threshold as the basis for region segmentation. Specifically, a sliding window can be used to traverse the gradient sequence, selecting consecutive peak points that exceed the threshold as candidate boundary points.

[0049] Specifically, the preset curvature gradient jump threshold is established by analyzing the contour features of typical sheet metal in historical processing data to create a curvature gradient distribution model corresponding to different bending shapes. For example, for right-angle bent sheet metal, the curvature gradient exhibits a steep jump at the corner, and the threshold can be set to the 80th quantile of the gradient value in this region; for arc bent sheet metal, the threshold is adjusted to the 50th quantile based on the magnitude of the curvature gradient change. In real-time detection, when the curvature gradient value of the contour point sequence exceeds this threshold, it indicates the presence of potential feature points at the bending boundary. By filtering these feature points, the contour can be divided into independent regions with different curvature characteristics, providing an accurate segmentation basis for subsequent normal analysis.

[0050] Through the above technical solution, this application can effectively improve the recognition accuracy of contour segment boundaries and avoid over-segmentation or under-segmentation problems caused by fixed thresholds. In complex bending sheet metal inspection scenarios, this solution can accurately distinguish the boundary features of different bending areas, providing a reliable segmentation basis for subsequent error level judgment, while reducing the risk of misjudgment caused by contour noise.

[0051] This application further proposes that linear fitting be performed using the least squares method, performing a straight line fitting once for the point set of each segmented region, and calculating the unit normal vector through the direction vector of the fitted line, wherein the unit normal vector is perpendicular to the direction of the fitted line.

[0052] The least squares linear fitting method determines the optimal line parameters by minimizing the sum of the squared perpendicular distances from each point to the fitted line. This can be achieved using matrix operations or iterative optimization algorithms. It is used to eliminate random noise interference in the local contour point distribution and extract the dominant directional trend. The unit normal vector is a vector perpendicular to the fitted line with a magnitude of 1. It is calculated by taking the negative reciprocal of the slope of the fitted line and normalizing it. This vector characterizes the geometric direction features of the segmented region and provides a directional difference metric for subsequent error level assessment.

[0053] Specifically, after the segmented regions are divided, the contour points within each region are input into the least-squares fitting module. The slope and intercept parameters of the best-fit line are obtained by solving a system of linear equations. Based on these line parameters, the direction vector is calculated and converted into a unit normal vector, ensuring that the normal vectors of different segmented regions have uniform dimensions. Since the normal vector is perpendicular to the direction of the fitted line, the difference in normal vectors between adjacent segmented regions directly reflects the degree of abrupt change in the contour direction, providing a quantitative basis for subsequent error level classification.

[0054] Traditional methods typically employ linear fitting of the overall contour or simple segmentation, failing to accurately capture subtle directional changes in local contours and easily overlooking small-scale bending anomalies. This proposed solution, however, utilizes least-squares fitting and normal vector calculation across segmented regions to analyze the continuity and abrupt changes in contour direction segment by segment, significantly improving the sensitivity for identifying local errors.

[0055] Through the above technical solution, this application achieves accurate quantitative characterization of the local orientation of complex contours, effectively distinguishes normal bending segments from abnormal deformation areas, provides a reliable data basis for subsequent suspicious segment marking and correction segment identification, and avoids misjudgment or missed detection caused by local fitting deviation.

[0056] This application further proposes an error level judgment process including: calculating the angle between any two adjacent unit normal vectors as the normal difference value; if the normal difference value is less than a first preset value, it is marked as the first level; if it is between the first preset value and the second preset value, it is marked as the second level; if it is greater than the second preset value, it is marked as the third level.

[0057] The normal difference refers to the angle between the unit normal vectors of adjacent segment regions, which can be calculated using the vector dot product formula. It is used to quantify the degree of geometrical difference between adjacent segment regions. The first preset value is a pre-set low error threshold, such as 5 degrees. When the normal difference is below this threshold, it is considered to have no significant error. The second preset value is a pre-set high error threshold, such as 15 degrees. When the normal difference exceeds this threshold, it is considered to have a serious geometric deviation. The error level is a classification label based on the comparison between the normal difference and the preset threshold, used to characterize the severity of geometric errors in different segment regions.

[0058] Specifically, after calculating the unit normal vector of the segmented region, the normal vector pairs of adjacent segmented regions are traversed sequentially, and the normal difference is obtained by calculating the angle between the vectors. This difference is input into a preset angle range judgment logic: if the difference is less than a first preset value, it indicates that the geometric direction of the adjacent regions is highly consistent, and it is marked as level one; if the difference is between the first and second preset values, it indicates that there is a moderate degree of geometric deviation, and it is marked as level two; if the difference exceeds the second preset value, it is judged that there is a significant geometric anomaly, and it is marked as level three. This forms an error level sequence covering all adjacent segmented regions, providing a data foundation for subsequent identification of suspicious segments.

[0059] Through the above technical solution, this application achieves refined hierarchical identification of geometric errors in the bending area of ​​sheet metal, effectively distinguishing between areas with minor deviations, moderate errors, and severe defects. Based on the hierarchical results, different levels of correction strategies can be triggered specifically, avoiding the false detection or missed detection problems caused by single threshold judgment in traditional methods, and significantly improving the fault tolerance and correction efficiency of the detection system.

[0060] This application further proposes a preset rule for judging suspicious segments, including one or a combination of the following conditions: there are two or more consecutive segment regions with an error level of the second or third level; the error level of a single segment region is the third level and the number of points it contains is less than a preset point threshold.

[0061] The presence of two or more consecutive segmented regions with error levels of level two or three refers to adjacent segmented regions exhibiting consecutively high anomaly levels. This can be achieved by traversing the normal difference sequence and counting the number of consecutive anomaly levels. This condition is used to identify regional anomalies caused by error accumulation or propagation. A single segmented region with an error level of level three and containing fewer points than a preset threshold refers to an isolated high-level error region containing fewer contour points than an empirically set value. This can be achieved by counting the number of points in the segmented region and comparing it with a threshold. This condition is used to eliminate misjudgments caused by local noise or random errors.

[0062] Specifically, after the error level is determined, the suspicious segment marking process is triggered. When two or more consecutive segmented regions are detected with error levels reaching the second or third level, the region is marked as a suspicious segment. This indicates that the error may be continuous or diffuse. Another situation is when a single segmented region is determined to be at the third level, but the number of contour points it contains is less than a preset point threshold, for example, the point threshold can be set to 5-10 points. In this case, the region is marked as a suspicious segment. This indicates that the anomaly may be caused by local interference or sampling defects. The combined use of these two conditions can cover anomaly patterns in different scenarios.

[0063] Through the above technical solution, this application can accurately identify bending areas with potential risks, avoiding over- or under-detection problems caused by single-condition judgment. By combining conditions for screening, it can capture suspicious segments formed by continuous error propagation and eliminate isolated noise segments with insufficient points, thereby providing reliable input data for subsequent correction operations and improving the robustness of the overall detection system.

[0064] This application further proposes that the conditions for identifying segments requiring correction must meet one of the following types: Type A is two or more consecutive segmented regions with an error level of the third grade, and at least one of the segmented regions is marked as a suspicious segment; Type B is segmented regions with error levels of the second and third grades appearing alternately, and the interval between adjacent segmented regions of different grades does not exceed two segmented regions; Type C is three consecutive segmented regions with an error level of the first grade, and all of them are marked as suspicious segments.

[0065] Among them, a segmented region with an error level of third grade refers to a region where the difference in normal vectors between adjacent segmented regions exceeds a second preset value. This can be achieved by calculating the vector angle and comparing it with a preset threshold, used to identify abnormal regions with significant angular deviations. A suspicious segment refers to a segmented region that meets preset suspicious segment judgment rules, such as multiple consecutive regions with error levels of second or third grade. This can be achieved using logical condition judgment rules, used to screen potentially abnormal regions requiring intervention. A segment requiring correction refers to a set of segmented regions that meet any one of types A, B, or C. This can be achieved using a multi-condition matching algorithm, used to locate contour regions that require correction operations.

[0066] Specifically, during the detection process, when two or more segmented regions with an error level of level 3 appear consecutively and are marked as suspicious segments, correction condition type A is triggered, indicating that there is a concentrated high-error anomaly in that region. When level 2 and level 3 segmented regions appear alternately with an interval of no more than two regions, correction condition type B is triggered, indicating that the error shows a spreading or propagating trend. When all three level 1 segmented regions are marked as suspicious segments, correction condition type C is triggered, indicating that there may be hidden local anomalies in the low-error region. By combining multiple types of conditions, different anomaly patterns can be covered, such as concentrated anomalies, gradual anomalies, and latent anomalies, thereby achieving comprehensive identification of complex contour errors.

[0067] Traditional methods rely solely on a single error threshold to identify anomalous regions, failing to distinguish between different anomalous patterns and their correlations, which can easily lead to a mismatch between correction strategies and anomalous characteristics. This proposed solution, by defining multiple types of correction conditions and combining error level distribution characteristics with the status of suspicious segment markers, can accurately identify isolated anomalies, related anomalies, and latent anomalies, making correction operations more targeted.

[0068] Through the above technical solution, this application effectively solves the problem of miscorrection or omission caused by the single judgment of abnormal patterns in the prior art. By combining multiple types of conditions, it improves the recognition coverage of complex contour abnormal areas, ensures that the correction operation is accurately matched with the abnormal features, and thus improves the reliability and correction effectiveness of bending contour detection.

[0069] This application further proposes a process for correcting the segment to be corrected, including local resampling of the original image region corresponding to the segment to be corrected to obtain a new set of edge contour points; re-dividing the new set of edge contour points into segmented regions and determining the error level; and performing a preset error propagation trend analysis and local correction based on the updated error level sequence. The error propagation trend analysis includes detecting whether there is a first-level combination of first error level → third error level → first error level in the updated error level sequence, and whether there is a second-level combination of first error level → second error level → third error level. The local correction operation includes detecting the first error level when the error level is detected. During level combination, normal average correction is performed on the middle third error level segment region, taking the arithmetic mean of the unit normal vectors of the adjacent segment regions as the correction normal. When a second level combination is detected, the reconstruction instruction is recursively pushed to the first second error level segment region in the sequence, and marked as a segment to be reconstructed. The correction coverage of the segment region after the correction segment is updated is calculated. The correction coverage is the ratio of the number of corrected points in the segment region to the total number of points in the segment region. If the correction coverage exceeds the preset coverage threshold, the segment region is marked as a low confidence segment. If more than a preset number of low confidence segments are detected globally or the number of corrections exceeds the preset number, an image resampling signal is triggered.

[0070] Local subsampling refers to acquiring higher-resolution images of the original image region corresponding to the segment requiring correction. This can be achieved by adjusting the sampling interval or focusing parameters of the image sensor to obtain finer edge contour data. The first-level combination in error propagation trend analysis refers to a continuous pattern of first, third, and first-level errors in the error level sequence. This can be achieved by using a sliding window to traverse the error sequence for pattern matching, used to identify local anomalous abrupt changes. Normal average correction involves arithmetically averaging the normal vectors of adjacent normal segment regions. This can be achieved by averaging the vector coordinate components separately and then normalizing, used to eliminate the influence of isolated anomalous segments on the overall contour. The correction coverage threshold can be an empirical value, such as 80%, used to evaluate the reliability of the correction operation. The low-confidence segment marking condition can be that when the correction coverage exceeds the threshold, it indicates the presence of uncorrected anomalous points in the region, requiring further processing.

[0071] Specifically, when a segment requiring correction is detected, the corresponding region is first located in the original image. Local secondary sampling is then performed by adjusting image acquisition parameters, for example, shortening the sampling interval from 0.5 mm to 0.2 mm. Curvature gradient analysis is then performed on the newly acquired high-precision edge point set to re-divide the segmented regions and calculate the normal difference. The updated error level sequence is input into the trend analysis module, which uses a sliding window to detect the presence of first-level or second-level combinations. For example, when a first, third, and first-level sequence is detected, the system automatically performs normal averaging correction on the middle third-level segment, taking the average of the normal vectors of the preceding and following first-level segments to cover the original value. If a second-level combination is detected, a reconstruction command is sent to the first second-level segment, triggering the re-division of the segment's boundary points. After correction, the number of corrected points in each segment is counted. When the correction coverage reaches a preset threshold, for example, if 80 out of 100 points in a segment are corrected, it is marked as a low-confidence segment. The system continuously monitors the number of low-confidence segments globally. For example, when the number of low-confidence segments exceeds 30% of the total number of segments or the number of corrections reaches 5, it triggers the image resampling signal to reacquire the entire image.

[0072] Compared to existing technologies, current methods typically only perform single-line fitting corrections on outlier regions, lacking strategies for tracking error propagation paths and dynamic correction. This proposed solution, by introducing secondary sampling and trend analysis, can identify error propagation patterns and select targeted correction strategies. For example, it uses normal averaging for isolated outlier segments and segmented reconstruction for continuously propagating errors. Simultaneously, by monitoring correction coverage and low-confidence segments, it achieves quantitative evaluation of the correction effect and system self-checking, avoiding the error accumulation problem caused by incomplete local corrections in traditional methods.

[0073] Through the above technical solutions, this application can improve the correction accuracy by using dynamic local sampling and multi-mode correction strategies when detecting contour anomalies, effectively suppressing the propagation of errors between segmented regions. For example, when there is local deformation at the bend of the sheet metal, the system can accurately locate the abnormal area and perform normal correction, avoiding the efficiency reduction caused by overall resampling in traditional methods. At the same time, the coverage monitoring mechanism ensures the reliability of the correction operation, reduces the false judgment rate in complex contour scenarios, and improves the robustness of the detection system.

[0074] Example 2:

[0075] like Figure 3 As shown, a bending sheet metal image detection system for bending machine processing includes:

[0076] The data acquisition module is used to acquire image data of the sheet material to be processed, and extract edge contour point sequences from the image data to form a contour point set;

[0077] The region segmentation module is used to calculate the curvature gradient sequence based on the set of contour points; and to identify segment boundary points and divide the set of contour points into several segmented regions based on the curvature gradient sequence and a preset curvature gradient jump threshold.

[0078] The normal fitting module is used to perform linear fitting on the points in each segmented region to obtain the unit normal vector of each segmented region, thus forming a normal vector sequence.

[0079] The error analysis module is used to calculate the normal difference between unit normal vectors in adjacent segment regions based on the normal vector sequence, forming a normal difference sequence; and to determine the error level of each segment region based on the normal difference sequence and a preset angle interval, wherein the error level includes a first level, a second level, and a third level.

[0080] The feedback correction module is used to mark suspicious segments in the segmented area according to the error level and the preset suspicious segment judgment rules; identify segments that need to be corrected according to the error level and the suspicious segments, and correct the segments that need to be corrected.

[0081] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0082] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0083] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for detecting images of bent sheet metal processed by a bending machine, characterized in that... This includes the following steps: Acquire image data of the sheet material to be processed, and extract edge contour point sequences from the image data to form a contour point set; Calculate the curvature gradient sequence based on the set of contour points; identify segment boundary points and divide the set of contour points into several segmented regions based on the curvature gradient sequence and a preset curvature gradient jump threshold; Linear fitting is performed on the points in each segmented region to obtain the unit normal vector of each segmented region, forming a normal vector sequence; Based on the normal vector sequence, calculate the normal difference between unit normal vectors in adjacent segment regions to form a normal difference sequence; Based on the normal difference sequence and the preset angle range, the error level of each segment region is determined, and the error level includes a first level, a second level and a third level. Based on the error level and the preset suspicious segment judgment rules, suspicious segments are marked in the segmentation area; Based on the error level and the suspicious segment, the segment that needs correction is identified, and the segment that needs correction is corrected.

2. The method for detecting images of bent sheet metal for bending machine processing according to claim 1, characterized in that: The image data is an image of a sheet metal containing the complete bending area; The edge contour point sequence is a set of two-dimensional coordinate points arranged in the order of the rows or columns of the image.

3. The method for image detection of bent sheet metal for bending machine processing according to claim 2, characterized in that: The process of calculating the curvature gradient sequence includes: The curvature value is calculated for a local window consisting of three consecutive points in the set of contour points, forming a curvature sequence; The difference between adjacent curvature values ​​in the curvature sequence is calculated to form the curvature gradient sequence.

4. The method for detecting images of bent sheet metal for bending machine processing according to claim 3, characterized in that: The preset curvature gradient jump threshold is set empirically based on historical sample data; The process of identifying segment boundary points involves identifying points in the curvature gradient sequence whose values ​​exceed the preset curvature gradient jump threshold.

5. The method for detecting images of bent sheet metal for bending machine processing according to claim 4, characterized in that: The linear fitting is performed using the least squares method. A straight line is fitted once for the point set of each segmented region, and the unit normal vector is calculated through the direction vector of the fitted line. The unit normal vector is perpendicular to the direction of the fitted line.

6. The method for detecting images of bent sheet metal for bending machine processing according to claim 5, characterized in that: The error level determination process includes: calculating the angle between any two adjacent unit normal vectors as the normal difference; if the normal difference is less than a first preset value, it is marked as the first level; if it is between the first preset value and the second preset value, it is marked as the second level; if it is greater than the second preset value, it is marked as the third level.

7. The method for detecting images of bent sheet metal for bending machine processing according to claim 6, characterized in that: The preset suspicious segment judgment rules include one or a combination of the following conditions: There are two or more consecutive segmented regions with error levels of level two or three; The error level of a single segment region is level three and it contains fewer points than the preset point threshold.

8. The method for detecting images of bent sheet metal for bending machine processing according to claim 7, characterized in that: The conditions for identifying the segment requiring correction must meet one of the following types: Type A: Two or more consecutive segmented regions with an error level of Level 3, and at least one of these segmented regions is marked as a suspicious segment; Type B: Segmented regions with error levels of the second and third levels appear alternately, and the interval between adjacent segmented regions of different levels does not exceed two segmented regions; Type C: Three consecutive segmented regions with error levels of Level 1, all of which are marked as suspicious segments.

9. The method for detecting images of bent sheet metal for bending machine processing according to claim 8, characterized in that: The process of correcting the segment to be corrected includes: performing local secondary sampling on the original image region corresponding to the segment to be corrected to obtain a new set of edge contour points; re-dividing the segmented region and determining the error level for the new set of edge contour points; and performing preset error propagation trend analysis and local correction based on the updated error level sequence. The error propagation trend analysis includes: Check if the updated error level sequence contains a first-level combination of first error level → third error level → first error level; Check if there is a second-level combination of first error level → second error level → third error level in the updated error level sequence; The local correction operation includes: When the first-level combination is detected, the normal average correction is performed on the middle third-error-level segment region: the arithmetic mean of the unit normal vectors of the adjacent segment regions is taken as the correction normal. When a second-level combination is detected, the reconstruction instruction is pushed forward to the first second-error-level segment region in the sequence and marked as a segment that needs to be reconstructed. Calculate the correction coverage of the segmented region after the segment to be corrected is updated. The correction coverage is the ratio of the number of corrected points in the segmented region to the total number of points in the segmented region. If the corrected coverage exceeds a preset coverage threshold, the segmented area is marked as a low-confidence segment; if more than a preset number of low-confidence segments are detected globally or the number of corrections exceeds a preset number, an image resampling signal is triggered.

10. An image detection system for bent sheet metal used in bending machine processing, characterized in that: The method for detecting images of bent sheet metal used in bending machine processing as described in any one of claims 1 to 9 includes: The data acquisition module is used to acquire image data of the sheet material to be processed, and extract edge contour point sequences from the image data to form a contour point set; The region segmentation module is used to calculate the curvature gradient sequence based on the set of contour points; and to identify segment boundary points and divide the set of contour points into several segmented regions based on the curvature gradient sequence and a preset curvature gradient jump threshold. The normal fitting module is used to perform linear fitting on the points in each segmented region to obtain the unit normal vector of each segmented region, thus forming a normal vector sequence. The error analysis module is used to calculate the normal difference between unit normal vectors in adjacent segment regions based on the normal vector sequence, forming a normal difference sequence; and to determine the error level of each segment region based on the normal difference sequence and a preset angle interval, wherein the error level includes a first level, a second level, and a third level. The feedback correction module is used to mark suspicious segments in the segmented area according to the error level and the preset suspicious segment judgment rules; identify segments that need to be corrected according to the error level and the suspicious segments, and correct the segments that need to be corrected.

Images