Image detection-based double-outside square bag zipper bag making detection method and system

Abstract

The application discloses a double-outside bag zipper bag manufacturing detection method and system based on image detection, and relates to the technical field of image detection. The method comprises the following steps: synchronously collecting images on the surfaces of double bags of a double-outside bag zipper bag, and generating a double-channel image dataset; performing symmetry analysis based on the double-channel image dataset, extracting a structural feature vector of the double bags, performing defect detection according to the structural feature vector, and generating a defect label set; introducing a bag manufacturing process parameter, combining the defect label set to perform dynamic matching, generating a sorting instruction, executing the sorting instruction to perform bag manufacturing defect detection, and determining defect distribution data; and based on the defect distribution data, correlating and optimizing the bag manufacturing process parameter to obtain a bag manufacturing process optimization parameter. The technical problem that defect detection is disconnected with process parameters in the prior art, resulting in unstable bag manufacturing quality, is solved, and the technical effect of optimizing bag manufacturing process parameters based on defect detection results and improving bag manufacturing quality is achieved.

Description

Technical Field

[0001] This invention relates to the field of image detection technology, and specifically to a method and system for detecting double-outlet square zipper bags based on image detection. Background Technology

[0002] In the production of flexible packaging bags, double-sided square zipper bags are widely used in the food and daily chemical industries due to their large capacity and convenient sealing. These bags are typically made simultaneously from two symmetrical zippered bags through heat sealing and cutting processes, requiring high precision and stability from the bag-making equipment. However, in actual production, factors such as material tension fluctuations, heat sealing deviations, and mechanical errors often lead to quality defects in the double bags, affecting product consistency and yield. Existing technologies mainly rely on single-channel imaging or manual sampling to inspect the quality of finished bags, which suffers from low detection accuracy, response delays, and inability to effectively feed back to the process control system, making it difficult to achieve closed-loop control of defects and automatic optimization of the production process. Summary of the Invention

[0003] This application provides a method and system for detecting double-outlet square zipper bags based on image detection, which solves the technical problem in the prior art where defect detection and process parameters are disconnected, leading to unstable bag quality.

[0004] The first aspect of this application provides a bag-making detection method for double-outlet square zipper bags based on image detection, the method comprising:

[0005] Simultaneous image acquisition is performed on the surfaces of the two bags of a double-sided square zippered bag to generate a dual-channel image dataset. Symmetry analysis is performed on the dual-channel image dataset to extract structural feature vectors of the two bags. Defect detection is then performed based on these structural feature vectors to generate a defect label set. Bag-making process parameters are dynamically matched with the defect label set to generate sorting instructions. These sorting instructions are then executed to detect bag-making defects and determine defect distribution data. Based on the defect distribution data, the bag-making process parameters are correlated and optimized to obtain optimized bag-making process parameters.

[0006] A second aspect of this application provides an image detection-based bag making and inspection system for double-outlet square zipper bags, the system comprising:

[0007] Image acquisition module: Simultaneously acquires images of the surfaces of the two bags of the double-outlet square zipper bag, generating a dual-channel image dataset; Defect detection module: Performs symmetry analysis based on the dual-channel image dataset, extracts structural feature vectors of the two bags, performs defect detection based on the structural feature vectors, and generates a defect label set; Sorting instruction generation module: Introduces bag-making process parameters and dynamically matches them with the defect label set to generate sorting instructions, executes the sorting instructions to perform bag-making defect detection, and determines defect distribution data; Parameter update module: Based on the defect distribution data, performs correlation optimization on the bag-making process parameters to obtain optimized bag-making process parameters.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] First, simultaneous image acquisition is performed on the surfaces of both bags of the double-outlet square zippered bag to generate a dual-channel image dataset. Next, symmetry analysis is performed based on the dual-channel image dataset to extract structural feature vectors of the two bags. Defect detection is then performed based on these feature vectors to generate a defect label set. Then, bag-making process parameters are dynamically matched with the defect label set to generate sorting instructions. These instructions are then executed to detect bag-making defects and determine the defect distribution data. Finally, the bag-making process parameters are correlated and optimized based on the defect distribution data to obtain optimized bag-making process parameters. This solves the technical problem in existing technologies where defect detection and process parameters are disconnected, leading to unstable bag-making quality. It achieves the technical effect of optimizing bag-making process parameters based on defect detection results, thereby improving bag-making quality. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic flowchart of the image detection-based bag making and detection method for double-outlet square zipper bags provided in this application embodiment;

[0012] Figure 2 This is a schematic diagram of the structure of the image detection-based double-outlet square zipper bag making and inspection system provided in an embodiment of this application.

[0013] Figure labeling: Image acquisition module 11, defect detection module 12, sorting instruction generation module 13, parameter update module 14. Detailed Implementation

[0014] This application provides a bag-making inspection method and system for double-outlet square zipper bags based on image detection, which solves the technical problem in the prior art where defect detection and process parameters are disconnected, leading to unstable bag quality.

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0016] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0017] Example 1, as Figure 1 As shown, this application provides a bag-making detection method for double-outlet square zipper bags based on image detection, wherein the method includes:

[0018] Simultaneous image acquisition was performed on the surfaces of both zippered pockets of a double-outlet square bag to generate a dual-channel image dataset.

[0019] In this embodiment of the application, an industrial camera deployed on the production line is used to synchronously acquire images of the two surfaces of the double-sided square zipper bag, generating a dual-channel image dataset that includes images of both sides of the bag.

[0020] Furthermore, the method involves simultaneously acquiring images of the surfaces of both zippered pockets of a double-sided square bag to generate a dual-channel image dataset, including:

[0021] A first industrial camera and a second industrial camera are symmetrically deployed on the dual parallel conveyor belts of a bag-making production line. The optical axes of the first and second industrial cameras are perpendicular to the corresponding bag surfaces. The speed signals of the dual parallel conveyor belts in the bag-making production line are acquired in real time. When the speed signals reach a preset speed threshold, synchronous trigger pulse data is generated. Based on the synchronous trigger pulse data, the first industrial camera is controlled to expose and acquire images of the first bag to obtain the original surface image of the first bag. Based on the synchronous trigger pulse data, the second industrial camera is controlled to expose and acquire images of the second bag to obtain the original surface image of the second bag. The original surface images of the first bag and the original surface images of the second bag are aligned with pixel-level timestamps to generate the dual-channel image dataset.

[0022] On the dual parallel conveyor belts of the bag making production line, two high-resolution industrial cameras, namely the first industrial camera and the second industrial camera, are symmetrically deployed. The installation positions of these two cameras ensure that their optical axes are perpendicular to the outer surfaces of the corresponding first and second bags, thereby avoiding image distortion caused by shooting angle deviation.

[0023] The system acquires the speed signals of the dual parallel conveyor belts in real time via speed sensors and compares them with a preset speed threshold. Once the conveyor speed stabilizes and reaches the set threshold, the system immediately generates a synchronous trigger pulse signal. Based on this signal, the system precisely controls the first industrial camera to expose and capture an image of the first bag (the left bag of the double-outlet square zipper bag). Simultaneously, the second industrial camera, under the same trigger signal, exposes and captures an image of the second bag (the right bag of the double-outlet square zipper bag). The acquired original surface images of the first and second bags are then aligned using pixel-level timestamps to ensure that both images are acquired within the same frame. The aligned images are then integrated to generate a dual-channel image dataset, with each channel corresponding to an image of one bag.

[0024] Symmetry analysis is performed on the dual-channel image dataset to extract the structural feature vectors of the double-bag body. Defect detection is then performed based on the structural feature vectors to generate a defect label set.

[0025] Furthermore, based on the dual-channel image dataset, symmetry analysis is performed to extract the structural feature vectors of the double-bag body. The method includes:

[0026] Spatiotemporal registration extraction is performed on the dual-channel image dataset to generate image registration results. Mapping analysis is then performed on the original surface images of the first and second bags according to the image registration results to construct pixel mapping relationships. Edge enhancement processing is then performed on the dual-channel image dataset based on the pixel mapping relationships to extract key point sets of bag geometric contours. Dynamic symmetry fitting is then performed based on the key point sets of bag geometric contours to determine a reference symmetry axis. The structure of the dual-bag region is calculated along the reference symmetry axis to obtain multi-scale structural similarity. Structural feature calculation is then performed based on the multi-scale structural similarity to obtain the structural feature vectors of the dual-bags.

[0027] Specifically, spatiotemporal registration is performed on the original surface images of the first and second bags. This registration includes synchronous correction of shooting timestamps and image perspective errors. Feature point matching-based image registration algorithms, such as SIFT or ORB, are used to obtain the spatial registration relationship between the two images, generating image registration results. Based on the image registration results, the first bag image is mapped to the coordinate space of the second bag image, constructing a pixel-level mapping relationship to describe the spatial correspondence between the two bags. Based on the pixel mapping relationship, edge enhancement processing is performed on the dual-channel image dataset. Edge detection algorithms such as Sobel or Canny are used to extract image edges, and gradient direction information and morphological filtering operations are combined to further enhance the boundary contours, thereby extracting a clear set of geometric contour key points for each bag. Using the set of bag geometric contour key points, a reference symmetry axis with better overall structural symmetry for the two bags is calculated using a minimum mean square error fitting algorithm or a symmetry fitting method. This symmetry axis serves as a symmetry reference line for the dual-bag structure analysis. The left and right bag regions are divided and matched along the reference axis of symmetry. A multi-scale sliding window and region feature extraction strategy is adopted to extract the structural features of the symmetrical regions at multiple scales. The structural similarity index between regions is calculated, including contour shape matching degree, edge strength consistency and texture mean difference, etc., to obtain the multi-scale structural similarity of the two bags.

[0028] Based on multi-scale structural similarity, the system extracts structural similarity scores for symmetrical regions of the two bags within different scale windows along the baseline axis of symmetry. These scores can include multiple dimensions such as contour matching rate, edge gradient difference, texture uniformity, and symmetry intensity. For each symmetrical window, the similarity between the left and right bags in terms of pixel structure, intensity distribution, and texture direction is calculated to form a local similarity matrix. Next, a feature fusion strategy is used to statistically summarize the local similarity matrices, extracting multiple structural parameters including average structural similarity, minimum similarity, maximum similarity, symmetry deviation, and edge discontinuity rate. Outliers are filtered for confidence levels to ensure feature dimension stability. Based on this, a multi-dimensional structural feature space is constructed by combining global structural indicators such as the distribution of geometric key points of the bags, contour balance of the regions on both sides of the symmetry axis, consistency of zipper line direction, and difference in bottom seal length. The system uses principal component analysis (PCA) for feature dimensionality reduction to compress redundant information and finally vectorizes the structural feature space to output the structural feature vector of the double-bag body. This structural feature vector can accurately characterize the symmetry integrity, geometric matching degree and key structural differences of the double-bag body.

[0029] Furthermore, the method for calculating the double-bag region structure along the aforementioned reference axis of symmetry includes:

[0030] Based on the reference symmetry axis and the key point set of the bag's geometric contour, a correlation analysis is performed to construct a spatial topology map of the two bags. The zipper is then transformed and detected based on this map to determine its linear features. Using these linear features as a baseline, the map is expanded to both sides. A pixel width parameter is set based on the expansion result, and the zipper detection area structure is determined based on this parameter. Bag contour parameters are extracted according to the key point set of the bag's geometric contour, and identification is performed along these parameters to the bottom edge of the bag to determine the edge sealing area structure. Finally, the zipper detection area structure and the edge sealing area structure are added to the dual-bag area structure.

[0031] Based on a defined reference axis of symmetry, the key point set of the bag's geometric contour is mapped onto a symmetric coordinate system. Point-to-point relationships are then used to construct the corresponding geometric element sets for the left and right bags. Based on the topological connections between these key points, a spatial topological graph reflecting the dual-bag structure is generated. Nodes represent geometric feature points, and edges represent structural connection paths. This graph clearly describes the relative positions and topological dependencies of key areas such as zippers and edge sealing.

[0032] Based on the spatial topology diagram of the dual-bag body, linear fitting and transformation detection are performed on the zipper section in the middle region of the dual-bag body. Linear feature extraction algorithms such as Hough transform are used to extract the straight-line features of the continuous zipper path. The optimal fitting result is selected based on path continuity and gray-level gradient stability to determine the final zipper linear features. These zipper linear features serve as a detection reference baseline, expanding pixel-level regions to both sides of the axis of symmetry at a set step size to generate the detection area structure around the zipper. The expansion width is determined by empirical parameters of the zipper width and image resolution. Simultaneously, based on key points of the bag's geometric contour, bag contour parameters are extracted according to the contour direction. Combined with the bottom boundary features of the bag, this is further extended downwards to locate the sealing area. By detecting the continuity and gray-level changes of the bottom edge segments of the bag, the actual boundary of the sealing area is identified, constructing the sealing area structure. Finally, the system integrates the zipper detection area structure and the sealing area structure at the structural level and adds them to the dual-bag body area structure as the base area set for subsequent defect detection and structural analysis.

[0033] Furthermore, the method for detecting defects and generating a defect label set based on the structural feature vector includes:

[0034] Based on the structural feature vector and the structure of the zipper detection area, offset analysis is performed to obtain the centerline offset. When the centerline offset exceeds a first misalignment threshold, zipper horizontal misalignment defect data is determined. Based on the structural feature vector and the structure of the zipper detection area, tooth pitch distribution is calculated to obtain adjacent tooth pitch distribution data. When the adjacent tooth pitch distribution data exceeds a second misalignment threshold, zipper tooth pitch misalignment defect data is determined. The zipper horizontal misalignment defect data and the zipper tooth pitch misalignment defect data are associated and integrated to generate a zipper misalignment defect label, and the zipper misalignment defect label is added to the defect label set.

[0035] Based on the structural feature vectors and the structure of the zipper detection area obtained above, the system performs fitting analysis on the centerline of the double-bag body. By comparing the offset between the theoretical axis of symmetry and the actual zipper centerline, the centerline offset is calculated. When this centerline offset exceeds a preset first misalignment threshold, it indicates that the zipper has a significant horizontal offset relative to the ideal symmetrical position, and the system records the zipper horizontal misalignment defect data accordingly.

[0036] Based on the structural features of the zipper detection area, the tooth pitch distribution is calculated. Edge detection and grayscale gradient analysis algorithms are used to accurately identify the center point of the zipper teeth. Then, the distance between the centers of adjacent teeth is traversed and statistically analyzed to obtain complete adjacent tooth pitch distribution data. When an anomaly is found in the adjacent tooth pitch distribution data where the tooth pitch exceeds the second misalignment threshold, the system determines it as zipper tooth pitch misalignment and extracts the corresponding tooth pitch misalignment defect data.

[0037] The system integrates and correlates zipper horizontal misalignment defect data with zipper tooth pitch misalignment defect data. Through spatial location mapping and event time identification, it forms a zipper misalignment defect label with comprehensive descriptive capabilities. This label uniformly records information such as misalignment type, location area, and offset value, and identifies it as a zipper misalignment defect label. Finally, the zipper misalignment defect label is added to the defect label set.

[0038] Furthermore, the method for detecting defects and generating a defect label set based on the structural feature vector includes:

[0039] Based on the structural feature vector and the edge sealing area structure, pixel grayscale scanning is performed along the sealing line direction to generate a sealing line strength distribution curve. Fluctuation analysis is performed across the sealing line strength distribution interval to extract curve trough location information, which includes trough depth data. When the trough depth data is lower than a preset sealing line strength threshold, the curve trough location information is marked as a virtual sealing point. Morphological closing operations are performed on the virtual sealing points to obtain virtual sealing line parameters. Based on the virtual sealing line parameters, continuous calculations are performed to obtain continuous virtual sealing area parameters. A safe virtual sealing area threshold is set according to the structural feature vector of the double-bag body. When the continuous virtual sealing area parameter exceeds the safe virtual sealing area threshold, a sealing defect label is generated and added to the defect label set.

[0040] Based on the extracted structural feature vectors and the structure of the sealing area, a pixel-level grayscale scanning operation is performed along the sealing line direction of the bag to obtain the grayscale intensity distribution data at the sealing edge of the bag, and a sealing line intensity distribution curve is constructed. The sealing line intensity distribution curve reflects the spatial change of the heat-pressing or adhesive strength of the sealing area during the sealing process.

[0041] The system traverses the distribution range of sealing strength, analyzes the local fluctuations of the curve, and focuses on extracting the location information of curve troughs with significant sinking. Each trough is recorded with its relative position and corresponding minimum grayscale intensity as trough depth data. If the trough depth data is lower than a preset sealing strength threshold, the system automatically identifies this location as a potentially incompletely bonded sealing anomaly and marks it as a false seal point. For multiple false seal points, the system further uses morphological closing operations to aggregate their spatial distribution, generating false seal line parameters reflecting the connection relationships between multiple false seal points. Based on the false seal line parameters, the system performs continuous region identification and area calculation on the two-dimensional image of the bag, obtaining the area parameters of the continuous false seal area. The system sets a safety threshold for the false seal area based on the bag size and sealing specification information contained in the structural feature vector. When the area parameter of the detected continuous false seal area exceeds this threshold, the bag is considered to have a serious false seal problem and cannot meet the sealing quality requirements. Based on this, the system generates a sealing defect label, which records the sealing location, area range, and strength deviation information, and adds the sealing defect label to the defect label set.

[0042] By introducing bag-making process parameters and combining them with the defect label set for dynamic matching, sorting instructions are generated, and the sorting instructions are executed to detect bag-making defects and determine defect distribution data.

[0043] Furthermore, the method involves dynamically matching bag-making process parameters with the defect label set to generate sorting instructions, including:

[0044] The bag-making process parameters of the bag-making machine are collected in real time. Based on these parameters, alignment analysis is performed according to the process time sequence to construct a bag-making process parameter vector. A process time axis is constructed, and the defect tag set is mapped to the process time axis to obtain a defect event sequence. The defect event sequence is dynamically matched with the bag-making process parameter vector to obtain a defect matching degree. Sorting analysis is performed according to the defect matching degree. When the defect matching degree in the sorting analysis result is greater than the defect similarity threshold, an immediate sorting instruction is generated. When the defect matching degree in the sorting analysis result is less than the defect similarity threshold, but the number of defect matches is greater than a preset defect matching limit, a batch sorting instruction is generated.

[0045] The system collects bag-making process parameters in real time during the bag-making process, including but not limited to heat-sealing temperature, pressing pressure, forming speed, zipper tension, sealing time, die-cutting frequency, and material batch information. The system normalizes these parameters based on the collected timestamps and performs alignment analysis according to the bag-making sequence to construct a bag-making process parameter vector. This vector reflects the process status changes of each bag during the forming process. Simultaneously, based on the defect tag set generated during image recognition, the system constructs a time-series-based defect event sequence, mapping the bag number corresponding to each defect tag to the process parameter collection timeline to form a unified process timeline representation. Subsequently, the system dynamically matches and compares the defect event sequence with the corresponding process parameter vector, calculating the matching degree between each defect event and its corresponding parameter, and generating a defect matching degree for sorting analysis.

[0046] During the sorting analysis phase, the system makes judgments based on the defect matching degree and the preset defect similarity threshold: when the defect matching degree of a bag is higher than the defect similarity threshold, it is considered that the defect of the current bag has a process causal relationship or severity, and the system immediately generates a corresponding immediate sorting instruction, instructing the execution mechanism to perform a rejection operation on the bag; if the current defect matching degree is lower than the similarity threshold, but the cumulative number of matching times exceeds the preset defect matching limit within several consecutive cycles, it is determined that the defect has a batch fluctuation trend, and the system generates a batch sorting instruction to centrally sort or warn the bags in the current and adjacent ranges.

[0047] Furthermore, mapping the defect tag set to the process timeline to obtain a defect event sequence, and dynamically matching the defect event sequence with the bag-making process parameter vector to obtain the defect matching degree, the method includes:

[0048] The zipper misalignment defect label is mapped to the process time axis to obtain a zipper misalignment defect event sequence, which has spatiotemporal coordinates of zipper misalignment. The sealing defect label is also mapped to the process time axis to obtain a sealing defect event sequence, which has spatiotemporal coordinates of sealing defects. The bag-making process parameter vector and the zipper misalignment defect event sequence are dynamically normalized according to the zipper misalignment spatiotemporal coordinates to obtain a zipper misalignment matching degree. The bag-making process parameter vector and the sealing defect event sequence are also dynamically normalized according to the sealing defect spatiotemporal coordinates to obtain a sealing defect matching degree. The zipper misalignment matching degree and the sealing defect matching degree are then overlaid and fused to obtain the defect matching degree.

[0049] First, the zipper misalignment defect labels are mapped to the process timeline of the bag-making process according to the timestamp information, forming a zipper misalignment defect event sequence. Each event in this sequence includes spatiotemporal coordinate data such as the time of occurrence of the zipper misalignment, the bag number, and the zipper offset direction and amount. Similarly, the sealing defect labels are mapped to the same process timeline, forming a sealing defect event sequence. Each defect event also includes typical characteristics of sealing defects such as the location of the sealing trough and the area of ​​the false seal, as well as their spatiotemporal coordinates. Subsequently, the system uses the spatiotemporal coordinates of the defect events as anchor points to perform dynamic event normalization on the bag-making process parameter vector at the corresponding time. Specifically: Using time points in the zipper misalignment defect event sequence as a reference, relevant process data such as zipper tension parameters, conveyor speed, and heat sealing temperature of the bag-making equipment within a certain range before and after that time point are extracted. This parameter sequence is dynamically normalized and statistically analyzed using a sliding time window method to calculate the zipper misalignment matching degree. Similarly, using the spatiotemporal coordinates of the sealing defect event sequence as a benchmark, key process parameters such as sealing temperature, pressing pressure, and sealing duration are normalized, and the sealing defect matching degree is calculated. Finally, the system superimposes and merges the zipper misalignment matching degree and the sealing defect matching degree according to a set weight or temporal distribution pattern to calculate a unified defect matching degree. The defect matching degree is used to measure the correlation strength between the current process state and the occurrence of the defect.

[0050] Furthermore, the method for executing the sorting instruction to detect bag-making defects and determine defect distribution data includes:

[0051] When the sorting instruction is the immediate sorting instruction, the double-parallel conveyor belts in the bag-making production line track the double-outlet square bag zipper in real time, locating the real-time position of the double-outlet square bag zipper. Conveyor belt speed data is combined with the real-time position of the double-outlet square bag zipper for mechanical sorting response, capturing the sorting response result image. Based on the sorting response result image, bag-making defects are detected and recorded to determine the defect density parameter. A three-dimensional space is defined, and the defect density parameter is mapped to the three-dimensional space to construct a defect type distribution histogram. The defect type distribution histogram is added to the defect distribution data.

[0052] When the sorting instruction is an immediate sorting instruction, the system activates the real-time tracking module to continuously track the position of the double-outlet square zipper bags conveyed by the dual parallel conveyor belts in the bag-making production line, and accurately locates the real-time spatial coordinates of the current bag based on visual recognition algorithms or coded tags. Subsequently, the system introduces the speed data of the conveyor belts and matches it with the real-time position data of the bags, controlling mechanical execution units (such as pneumatic levers, robotic arms, etc.) to complete high-response sorting operations. At the same time, it triggers the image acquisition module to capture the corresponding sorting response result image, which is used to record the bag status and defect manifestations during the sorting process. Based on the sorting response result image, the system further calls the defect recognition algorithm for verification and identification, extracts the defect type and location of the current sample bag, counts the number and types of defects per unit time or per batch, and then calculates the defect density parameter. On this basis, the system defines a three-dimensional spatial coordinate system, where the horizontal axis represents the defect type category, the vertical axis represents the defect location (such as the zipper area, sealing area, etc.), and the height axis represents the density value of the corresponding defect; the defect density parameter is mapped to this three-dimensional space to generate a defect type distribution histogram, which is used to visualize the distribution characteristics of various defects on the bag. Finally, the defect type distribution histogram results are stored and added to the defect distribution data.

[0053] Based on the defect distribution data, the bag-making process parameters are correlated and optimized to obtain optimized bag-making process parameters.

[0054] First, the defect distribution data collected during the bag-making process is mapped to bag-making process parameters. Based on the defect type, occurrence time, and location indicated by the defect labels, these parameters are mapped onto the bag-making process timeline to establish a correspondence between defect occurrence and process flow, thus constructing a defect-process correlation sample set. Then, for different defect types (such as zipper misalignment and poor sealing), relevant key bag-making process parameters are extracted, such as pressing temperature, feeding speed, tension control value, and zipper positioning time. Based on this defect-process sample set, multivariate regression modeling, Bayesian causal inference, or neural network methods are used to construct a defect-process causal mapping model to evaluate the influence of various process parameters on defect type and frequency. Next, using a defect-process causal mapping model, the current process parameter settings are compared with historical defect distributions to identify abnormal parameter combinations that cause high defect density, and an optimization function with defect minimization as its objective is constructed. Based on this optimization function, combined with the adjustment range and constraints of the bag-making equipment parameters, gradient descent, genetic algorithms, or Bayesian optimization strategies are used to search for a set of process parameters with the lowest probability of defect occurrence, under the premise of ensuring process safety, as the process optimization parameters corresponding to the current bag-making task. Finally, the bag-making process optimization parameters are pushed to the bag-making control system to realize dynamic adjustment of core links such as pressing, feeding, tension control, and zipper synchronization, thereby improving bag consistency and finished product quality stability.

[0055] In summary, the embodiments of this application have at least the following technical effects:

[0056] First, simultaneous image acquisition is performed on the surfaces of both bags of the double-outlet square zippered bag to generate a dual-channel image dataset. Next, symmetry analysis is performed based on the dual-channel image dataset to extract structural feature vectors of the two bags. Defect detection is then performed based on these feature vectors to generate a defect label set. Then, bag-making process parameters are dynamically matched with the defect label set to generate sorting instructions. These instructions are then executed to detect bag-making defects and determine the defect distribution data. Finally, the bag-making process parameters are correlated and optimized based on the defect distribution data to obtain optimized bag-making process parameters. This solves the technical problem in existing technologies where defect detection and process parameters are disconnected, leading to unstable bag-making quality. It achieves the technical effect of optimizing bag-making process parameters based on defect detection results, thereby improving bag-making quality.

[0057] Example 2 is based on the same inventive concept as the image detection-based bag making detection method for double-outlet square zipper bags in the previous examples, such as... Figure 2 As shown, this application provides an image detection-based bag making and inspection system for double-sided square zipper bags, wherein the system includes:

[0058] Image acquisition module 11: Simultaneously acquires images of the surfaces of the two bags of the double-outlet square zipper bag, generating a dual-channel image dataset; Defect detection module 12: Performs symmetry analysis based on the dual-channel image dataset, extracts structural feature vectors of the two bags, performs defect detection based on the structural feature vectors, and generates a defect label set; Sorting instruction generation module 13: Introduces bag-making process parameters and dynamically matches them with the defect label set to generate sorting instructions, executes the sorting instructions to perform bag-making defect detection, and determines defect distribution data; Parameter update module 14: Based on the defect distribution data, performs correlation optimization on the bag-making process parameters to obtain optimized bag-making process parameters.

[0059] Furthermore, the image acquisition module 11 is used to perform the following methods:

[0060] A first industrial camera and a second industrial camera are symmetrically deployed on the dual parallel conveyor belts of a bag-making production line. The optical axes of the first and second industrial cameras are perpendicular to the corresponding bag surfaces. The speed signals of the dual parallel conveyor belts in the bag-making production line are acquired in real time. When the speed signals reach a preset speed threshold, synchronous trigger pulse data is generated. Based on the synchronous trigger pulse data, the first industrial camera is controlled to expose and acquire images of the first bag to obtain the original surface image of the first bag. Based on the synchronous trigger pulse data, the second industrial camera is controlled to expose and acquire images of the second bag to obtain the original surface image of the second bag. The original surface images of the first bag and the original surface images of the second bag are aligned with pixel-level timestamps to generate the dual-channel image dataset.

[0061] Furthermore, the defect detection module 12 is used to perform the following method:

[0062] Spatiotemporal registration extraction is performed on the dual-channel image dataset to generate image registration results. Mapping analysis is then performed on the original surface images of the first and second bags according to the image registration results to construct pixel mapping relationships. Edge enhancement processing is then performed on the dual-channel image dataset based on the pixel mapping relationships to extract key point sets of bag geometric contours. Dynamic symmetry fitting is then performed based on the key point sets of bag geometric contours to determine a reference symmetry axis. The structure of the dual-bag region is calculated along the reference symmetry axis to obtain multi-scale structural similarity. Structural feature calculation is then performed based on the multi-scale structural similarity to obtain the structural feature vectors of the dual-bags.

[0063] Furthermore, the defect detection module 12 is used to perform the following method:

[0064] Based on the reference symmetry axis and the key point set of the bag's geometric contour, a correlation analysis is performed to construct a spatial topology map of the two bags. The zipper is then transformed and detected based on this map to determine its linear features. Using these linear features as a baseline, the map is expanded to both sides. A pixel width parameter is set based on the expansion result, and the zipper detection area structure is determined based on this parameter. Bag contour parameters are extracted according to the key point set of the bag's geometric contour, and identification is performed along these parameters to the bottom edge of the bag to determine the edge sealing area structure. Finally, the zipper detection area structure and the edge sealing area structure are added to the dual-bag area structure.

[0065] Furthermore, the defect detection module 12 is used to perform the following method:

[0066] Based on the structural feature vector and the structure of the zipper detection area, offset analysis is performed to obtain the centerline offset. When the centerline offset exceeds a first misalignment threshold, zipper horizontal misalignment defect data is determined. Based on the structural feature vector and the structure of the zipper detection area, tooth pitch distribution is calculated to obtain adjacent tooth pitch distribution data. When the adjacent tooth pitch distribution data exceeds a second misalignment threshold, zipper tooth pitch misalignment defect data is determined. The zipper horizontal misalignment defect data and the zipper tooth pitch misalignment defect data are associated and integrated to generate a zipper misalignment defect label, and the zipper misalignment defect label is added to the defect label set.

[0067] Furthermore, the defect detection module 12 is used to perform the following method:

[0068] Based on the structural feature vector and the edge sealing area structure, pixel grayscale scanning is performed along the sealing line direction to generate a sealing line strength distribution curve. Fluctuation analysis is performed across the sealing line strength distribution interval to extract curve trough location information, which includes trough depth data. When the trough depth data is lower than a preset sealing line strength threshold, the curve trough location information is marked as a virtual sealing point. Morphological closing operations are performed on the virtual sealing points to obtain virtual sealing line parameters. Based on the virtual sealing line parameters, continuous calculations are performed to obtain continuous virtual sealing area parameters. A safe virtual sealing area threshold is set according to the structural feature vector of the double-bag body. When the continuous virtual sealing area parameter exceeds the safe virtual sealing area threshold, a sealing defect label is generated and added to the defect label set.

[0069] Furthermore, the sorting instruction generation module 13 is used to execute the following method:

[0070] The bag-making process parameters of the bag-making machine are collected in real time. Based on these parameters, alignment analysis is performed according to the process time sequence to construct a bag-making process parameter vector. A process time axis is constructed, and the defect tag set is mapped to the process time axis to obtain a defect event sequence. The defect event sequence is dynamically matched with the bag-making process parameter vector to obtain a defect matching degree. Sorting analysis is performed according to the defect matching degree. When the defect matching degree in the sorting analysis result is greater than the defect similarity threshold, an immediate sorting instruction is generated. When the defect matching degree in the sorting analysis result is less than the defect similarity threshold, but the number of defect matches is greater than a preset defect matching limit, a batch sorting instruction is generated.

[0071] Furthermore, the sorting instruction generation module 13 is used to execute the following method:

[0072] The zipper misalignment defect label is mapped to the process time axis to obtain a zipper misalignment defect event sequence, which has spatiotemporal coordinates of zipper misalignment. The sealing defect label is also mapped to the process time axis to obtain a sealing defect event sequence, which has spatiotemporal coordinates of sealing defects. The bag-making process parameter vector and the zipper misalignment defect event sequence are dynamically normalized according to the zipper misalignment spatiotemporal coordinates to obtain a zipper misalignment matching degree. The bag-making process parameter vector and the sealing defect event sequence are also dynamically normalized according to the sealing defect spatiotemporal coordinates to obtain a sealing defect matching degree. The zipper misalignment matching degree and the sealing defect matching degree are then overlaid and fused to obtain the defect matching degree.

[0073] Furthermore, the sorting instruction generation module 13 is used to execute the following method:

[0074] When the sorting instruction is the immediate sorting instruction, the double-parallel conveyor belts in the bag-making production line track the double-outlet square bag zipper in real time, locating the real-time position of the double-outlet square bag zipper. Conveyor belt speed data is combined with the real-time position of the double-outlet square bag zipper for mechanical sorting response, capturing the sorting response result image. Based on the sorting response result image, bag-making defects are detected and recorded to determine the defect density parameter. A three-dimensional space is defined, and the defect density parameter is mapped to the three-dimensional space to construct a defect type distribution histogram. The defect type distribution histogram is added to the defect distribution data.

[0075] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0076] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0077] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A bag-making detection method for double-outlet square zipper bags based on image detection, characterized in that, The method includes: Simultaneous image acquisition was performed on the surfaces of both zippered pockets of a double-outlet square bag to generate a dual-channel image dataset; Symmetry analysis is performed on the dual-channel image dataset to extract the structural feature vectors of the double-bag body. Defect detection is then performed based on the structural feature vectors to generate a defect label set. By introducing bag-making process parameters and combining them with the defect label set for dynamic matching, sorting instructions are generated, and the sorting instructions are executed to detect bag-making defects and determine defect distribution data. Based on the defect distribution data, the bag-making process parameters are correlated and optimized to obtain optimized bag-making process parameters; The method involves simultaneously acquiring images of the surfaces of both zippered pockets of a double-sided square bag to generate a dual-channel image dataset. A first industrial camera and a second industrial camera are symmetrically deployed on the double parallel conveyor belts of the bag making production line, with the optical axes of the first industrial camera and the second industrial camera perpendicular to the corresponding bag surface. The speed signal of the dual parallel conveyor belts in the bag making production line is acquired in real time, and synchronous trigger pulse data is generated when the speed signal reaches a preset speed threshold. Based on the synchronous trigger pulse data, the first industrial camera is controlled to expose and acquire the first bag body to obtain the original surface image of the first bag body. Based on the synchronous trigger pulse data, the second industrial camera is controlled to expose and acquire the second bag body to obtain the original surface image of the second bag body; The original surface images of the first bag and the original surface images of the second bag are aligned with pixel-level timestamps to generate the dual-channel image dataset. Based on the aforementioned dual-channel image dataset, symmetry analysis is performed to extract the structural feature vectors of the double-bag structure. The method includes: Spatiotemporal registration extraction is performed based on the dual-channel image dataset to generate image registration results; Based on the image registration results, a mapping analysis is performed on the original surface image of the first bag and the original surface image of the second bag to construct a pixel mapping relationship; Based on the pixel mapping relationship, edge enhancement processing is performed on the dual-channel image dataset to extract the key point set of the bag's geometric contour. Dynamic symmetry fitting is performed based on the key point set of the bag's geometric contour to determine the reference symmetry axis; The structure of the double-bag region is calculated along the reference axis of symmetry to obtain multi-scale structural similarity. Structural features are calculated based on the multi-scale structural similarity to obtain the structural feature vector of the double-bag body.

2. The image detection-based bag making detection method for double-outlet square zipper bags as described in claim 1, characterized in that, The method for calculating the structure of the double-bag region along the reference axis of symmetry includes: Based on the reference symmetry axis and the key point set of the bag's geometric contour, a correlation analysis is performed to construct a spatial topology diagram of the two bags. Based on the spatial topology diagram of the dual-pocket body, the zipper is transformed and its linear characteristics are determined. Using the linear feature of the zipper as a baseline, the expansion is carried out to both sides. The pixel width parameter is set according to the expansion result, and the zipper detection area structure is determined based on the pixel width parameter. Extract bag contour parameters according to the key point set of the bag's geometric contour, and extend the bag contour parameters to the bottom edge of the bag for identification to determine the structure of the edge sealing area. The zipper detection area structure and the edge sealing area structure are added to the double-bag area structure.

3. The image detection-based bag making detection method for double-outlet square zipper bags as described in claim 2, characterized in that, Defect detection is performed based on the structural feature vectors to generate a defect label set, the method including: Based on the structural feature vector and the structure of the zipper detection area, offset analysis is performed to obtain the centerline offset. When the centerline offset exceeds the first misalignment threshold, zipper horizontal misalignment defect data is determined. Based on the structural feature vector and the structure of the zipper detection area, the tooth pitch distribution is calculated to obtain adjacent tooth pitch distribution data. When the adjacent tooth pitch distribution data exceeds the second misalignment threshold, zipper tooth pitch misalignment defect data is determined; The zipper horizontal misalignment defect data and the zipper tooth pitch misalignment defect data are associated and integrated to generate a zipper misalignment defect label, and the zipper misalignment defect label is added to the defect label set.

4. The image detection-based bag making detection method for double-outlet square zipper bags as described in claim 3, characterized in that, Defect detection is performed based on the structural feature vectors to generate a defect label set, the method including: Based on the structural feature vector and the edge sealing area structure, pixel grayscale scanning is performed according to the sealing line direction to generate the sealing line intensity distribution curve. The fluctuation change analysis is performed by traversing the distribution range of the sealing line strength, and the location information of the curve trough is extracted, which includes the trough depth data. When the valley depth data is lower than the preset sealing strength threshold, the valley location information of the curve is marked as a virtual sealing point; Morphological closing operations are performed on the virtual sealing points to obtain virtual sealing line parameters. Based on the virtual sealing line parameters, continuous calculations are performed to obtain the area parameters of the continuous virtual sealing region. Based on the structural feature vector of the double bag body, a critical value for the safe false sealing area is set. When the area parameter of the continuous false sealing area exceeds the critical value for the safe false sealing area, a sealing defect label is generated and added to the defect label set.

5. The image detection-based bag making detection method for double-outlet square zipper bags as described in claim 4, characterized in that, The method involves dynamically matching bag-making process parameters with the defect label set to generate sorting instructions, including: The bag-making process parameters of the bag-making machine are collected in real time, and the bag-making process parameters are aligned and analyzed according to the process time sequence to construct a bag-making process parameter vector. Construct a process timeline, map the defect label set to the process timeline to obtain a defect event sequence, and dynamically match the defect event sequence with the bag making process parameter vector to obtain the defect matching degree; Sorting analysis is performed based on the defect matching degree. When the defect matching degree in the sorting analysis result is greater than the defect similarity threshold, an immediate sorting instruction is generated. When the defect matching degree in the sorting analysis result is less than the defect similarity threshold, but the number of defect matching is greater than the preset defect matching limit, a batch sorting instruction is generated.

6. The image detection-based bag making detection method for double-outlet square zipper bags as described in claim 5, characterized in that, Mapping the defect tag set to the process time axis to obtain a defect event sequence, and dynamically matching the defect event sequence with the bag-making process parameter vector to obtain the defect matching degree, the method includes: The zipper misalignment defect label is mapped to the process time axis to obtain a zipper misalignment defect event sequence, which has spatiotemporal coordinates of zipper misalignment. The sealing defect label is mapped to the process time axis to obtain a sealing defect event sequence, which has sealing defect spatiotemporal coordinates. According to the spatiotemporal coordinates of zipper misalignment, the bag-making process parameter vector and the zipper misalignment defect event sequence are dynamically event-normalized to obtain the zipper misalignment matching degree. According to the spatiotemporal coordinates of the sealing defect, the bag-making process parameter vector and the sealing defect event sequence are dynamically normalized to obtain the sealing defect matching degree. The defect matching degree is obtained by overlaying and fusing the data of the zipper misalignment matching degree and the sealing defect matching degree.

7. The image detection-based bag making detection method for double-outlet square zipper bags as described in claim 5, characterized in that, The method for executing the sorting instruction to detect bag-making defects and determine defect distribution data includes: When the sorting instruction is the immediate sorting instruction, the double parallel conveyor belts in the bag making production line track the double-outlet square bag zipper bag in real time and locate the real-time position of the double-outlet square bag zipper bag. The mechanical sorting response is performed by combining the conveyor belt speed data with the real-time position of the double-outlet square bag zipper bag, and the sorting response result image is captured. Based on the sorting response result image, bag-making defects are detected and recorded to determine the defect density parameter; Define a three-dimensional space, map the defect density parameter to the three-dimensional space, and construct a defect type distribution histogram; Add the defect type distribution histogram to the defect distribution data.

8. A bag-making inspection system for double-outlet square zipper bags based on image detection, characterized in that, The system is used to implement the image detection-based bag making and inspection method for double-sided square zipper bags according to any one of claims 1-7, the system comprising: Image acquisition module: Simultaneously acquires images of the surfaces of the two zippered pockets of the double-outlet square bag, generating a dual-channel image dataset; Defect detection module: Based on the dual-channel image dataset, symmetry analysis is performed to extract the structural feature vector of the double-bag body, and defect detection is performed based on the structural feature vector to generate a defect label set; Sorting instruction generation module: Introduces bag making process parameters and combines them with the defect label set for dynamic matching to generate sorting instructions, executes the sorting instructions to detect bag making defects, and determines defect distribution data; Parameter update module: Based on the defect distribution data, the bag making process parameters are correlated and optimized to obtain optimized bag making process parameters.

Images