Zipper visual quality inspection method and system for directional defect detection

Through the zipper visual quality inspection method and system for directional defect detection, multi-angle image acquisition and scanning probes, combined with historical quality inspection examples and computer-aided evaluation, the existing zipper quality inspection problems are solved, and the existing zipper quality inspection problems are achieved. Intelligent quality inspection and efficient inspection.

CN118671001BActive Publication Date: 2024-12-06PING HU SHI WU XING GONG MAO YOU XIAN GONG SI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410910728.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-12-06
Estimated Expiration
2044-07-09

AI Technical Summary

Technical Problem

The existing zipper quality inspection methods rely on manual labor, have low efficiency and low accuracy, and lack of intelligent quality inspection, resulting in high false inspection and missed inspection rates.

Method used

Using zipper visual quality inspection methods and systems for directional defect detection, computer-aided performance evaluation is carried out by obtaining basic zipper information, multi-angle image acquisition, combining historical quality inspection examples and scanning probes, computer-aided performance evaluation is carried out, visual quality inspection models are trained, and integrated into the production line for Synchronous quality inspection.

Benefits of technology

It improves the quality inspection efficiency and defect detection accuracy, realizes intelligent quality inspection, and reduces false inspection and missed inspection rates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118671001B_ABST
    Figure CN118671001B_ABST
Patent Text Reader

Abstract

The present invention discloses a zipper visual quality inspection method and system for directional defect detection, which relates to the technical field of visual quality inspection. The method comprises: obtaining basic information of the zipper, obtaining multi-angle image data through a zipper visual acquisition device, analyzing and locking the first directional defect position and the corresponding first abnormal failure type and second abnormal failure type through the multi-angle image data with defect identification; obtaining defect impact indicators and failure probability based on the scanning probe on the zipper visual acquisition device; conducting directional defect reinforcement training on the zipper visual quality inspection model, and integrating it into the zipper visual acquisition device to perform defect quality inspection on the zipper production line. The present invention solves the technical problems of low zipper quality inspection efficiency, lack of intelligent quality inspection, and low defect detection accuracy in the prior art, and achieves the technical effect of improving quality inspection efficiency and defect detection accuracy and realizing intelligent quality inspection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of visual quality inspection, and in particular to a zipper visual quality inspection method and system for directional defect detection. Background Art

[0002] In the zipper production process, the traditional defect detection method mainly relies on manual inspection, using the naked eye and measuring tools to check the defects of the zipper. However, this method has many problems: on the one hand, manual inspection is inefficient and requires a lot of manpower and time costs. Faced with large-scale zipper production, manual inspection is difficult to meet the production speed requirements. On the other hand, the inspection results are not accurate enough and are easily affected by the subjective factors of the inspectors, resulting in a high rate of false detection and missed detection.

[0003] The existing technology has technical problems such as low zipper quality inspection efficiency, lack of intelligent quality inspection, and low defect detection accuracy. Summary of the invention

[0004] The present application provides a zipper visual quality inspection method and system for directional defect detection, which is used to solve the technical problems in the prior art of low zipper quality inspection efficiency, lack of intelligent quality inspection, and low defect detection accuracy.

[0005] In view of the above problems, the present application provides a zipper visual quality inspection method and system for directional defect detection.

[0006] In a first aspect of the present application, a zipper visual quality inspection method for directional defect detection is provided, the method comprising:

[0007] The basic information of the zipper is obtained, and the components corresponding to the basic information of the zipper include chain teeth, sliders, latches, and pull tabs; multi-angle acquisition is performed through the zipper visual acquisition equipment to obtain a multi-angle image group, directional defect classification and defect identification are added; based on the historical quality inspection instance corresponding to the slider, in the zipper visual quality inspection model, the first directional defect position under the locked pull mode and the corresponding first abnormal failure type, the second directional defect position under the unlocked pull mode and the corresponding second abnormal failure type are analyzed through the multi-angle image group with defect identification; based on the first directional defect position under the locked pull mode and the corresponding first abnormal failure type The zipper visual acquisition device is connected to add a scanning probe; based on the scanning probe on the zipper visual acquisition device and combined with the basic information of the zipper, a computer-aided performance evaluation is performed to obtain defect impact indicators and failure probabilities; based on the defect impact indicators and failure probabilities, the smoothness of the slider is evaluated according to the high fixed tip defect set and low fixed tip defect set corresponding to the slider, the zipper visual quality inspection model is trained for directional defects, and is integrated into the zipper visual acquisition device to perform synchronous quality inspection for directional defects on the zipper production line.

[0008] A second aspect of the present application provides a zipper visual quality inspection system for directional defect detection, the system comprising:

[0009] A zipper basic information acquisition module, the zipper basic information acquisition module is used to acquire the zipper basic information, and the components corresponding to the zipper basic information include sprockets, sliders, latches, and pull tabs; a multi-angle image group acquisition module, the multi-angle image group acquisition module is used to perform multi-angle acquisition through a zipper visual acquisition device, acquire a multi-angle image group, directional defect classification and add defect identification; a zipper visual quality inspection model analysis module, the zipper visual quality inspection model analysis module is based on the historical quality inspection instance corresponding to the slider. In the zipper visual quality inspection model, the zipper visual quality inspection model analysis module analyzes the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode through a multi-angle image group with defect identification; a scanning probe adding module, the scanning probe adding module is used to classify the zipper visual quality inspection model based on the historical quality inspection instance corresponding to the slider. The first directional defect position and the corresponding first abnormal failure type in the unlocked pulling mode, the second directional defect position and the corresponding second abnormal failure type in the unlocked pulling mode, connect the zipper vision acquisition device, and add a scanning probe; a defect influence index and failure probability acquisition module, the defect influence index and failure probability acquisition module is based on the scanning probe on the zipper vision acquisition device, combined with the zipper basic information, to perform computer-aided performance evaluation, and obtain defect influence indicators and failure probabilities; a directional defect synchronous quality inspection module, the directional defect synchronous quality inspection module is based on the defect influence indicators and failure probabilities, according to the high fixed tip defect set and low fixed tip defect set corresponding to the pull head to evaluate the smoothness of the pull head, perform directional defect reinforcement training on the zipper vision quality inspection model, and integrate it into the zipper vision acquisition device to perform directional defect synchronous quality inspection on the zipper production line.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] Obtain basic information about the zipper; perform multi-angle acquisition through the zipper visual acquisition device, obtain multi-angle image groups, directional defect classification and add defect identification; based on the historical quality inspection examples corresponding to the slider, analyze the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode in the zipper visual quality inspection model; based on the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode, connect the zipper visual acquisition device and add a scanning probe; perform computer-aided performance evaluation based on the scanning probe on the zipper visual acquisition device to obtain defect impact indicators and failure probability; conduct directional defect reinforcement training on the zipper visual quality inspection model, and integrate it into the zipper visual acquisition device to perform synchronous quality inspection of directional defects on the zipper production line. The technical effect of improving quality inspection efficiency and defect detection accuracy and realizing intelligent quality inspection is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic diagram of a flow chart of a zipper visual quality inspection method for directional defect detection provided in an embodiment of the present application;

[0014] Figure 2 A schematic diagram of the structure of a zipper visual quality inspection system for directional defect detection provided in an embodiment of the present application.

[0015] Explanation of the reference numerals: zipper basic information acquisition module 10, multi-angle image group acquisition module 20, zipper visual quality inspection model analysis module 30, scanning probe adding module 40, defect impact index and failure probability acquisition module 50, directional defect synchronous quality inspection module 60. DETAILED DESCRIPTION

[0016] The present application provides a zipper visual quality inspection method and system for directional defect detection, which is used to solve the technical problems in the prior art of low zipper quality inspection efficiency, lack of intelligent quality inspection, and low defect detection accuracy.

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0018] Embodiment 1

[0019] like Figure 1 As shown, the present application provides a zipper visual quality inspection method for directional defect detection, the method comprising:

[0020] Step S100: acquiring basic information of a zipper, wherein each component corresponding to the basic information of the zipper includes a sprocket, a slider, a latch, and a pull tab.

[0021] Specifically, by establishing a connection with a relevant database, storage device or other data source and starting the data acquisition operation, the basic information of the zipper obtained covers several key components of the zipper. For the chain teeth, the information obtained includes its tooth shape (such as round, square, etc.), tooth pitch, tooth height and thickness, material characteristics (such as metal, plastic, etc.) and surface treatment (such as electroplating, spray painting, etc.). Regarding the slider, the basic information collected includes the shape of the slider (such as ordinary type, automatic lock type, etc.), size (length, width, height), material (zinc alloy, aluminum alloy, etc.), internal structural characteristics (such as the smoothness of the slide rail, the elasticity of the spring, etc.) and appearance design details (such as lettering, pattern, etc.), among which the slider includes a high fixed pin, a low fixed pin and a groove. The high fixed pin and the low fixed pin refer to the parts in the slider used to fix the zipper tape to keep it stable during the pulling process. The difference in their height will affect the performance and stability of the slider. The groove is the space or channel inside the slider used to accommodate and guide the zipper teeth. Whether the shape and size of the groove are appropriate will directly affect the smoothness and bite effect of the zipper. For the pin, the information obtained includes the length, diameter, material, insertion method, and matching accuracy with the chain teeth. The basic information of the pull tab includes the shape (such as round, square, special shape), size, material (same or different from the slider), connection method (such as welding, riveting, etc.) and decorative features (such as the presence or absence of inlays, patterns, etc.) of the pull tab. By fully obtaining the detailed basic information of these components, accurate standards and reference bases are provided for the subsequent quality inspection process, which helps to more accurately detect and identify possible defects in the zipper.

[0022] Step S200: Use the zipper vision acquisition device to perform multi-angle acquisition, obtain a multi-angle image group, classify the directional defects and add defect identification.

[0023] Specifically, specialized zipper visual acquisition equipment is used to perform image acquisition tasks. These visual acquisition devices are placed in multiple different positions and angles to ensure that all parts of the zipper can be fully captured. The equipment includes high-definition cameras, industrial cameras, etc., and is equipped with appropriate lighting equipment to provide clear and uniform light conditions. During the acquisition process, the image of the zipper is acquired from different perspectives to form a multi-angle image group, which covers the front, side, top, bottom, etc., to present the details of the zipper as completely as possible. After the acquisition is completed, the acquired images are analyzed. Image processing algorithms and machine learning models are used to classify the existing directional defects, which can be skew, missing, or damaged chain teeth, stuck and deformed pull heads, loose and bent pins, broken and worn pull tabs, etc. Once the type of defect is determined, a specific mark is added to each defect, which can be a digital code, a specific color mark, or a text description, so that the defect can be quickly and accurately identified and located in subsequent processing and analysis. Through this step, the foundation is laid for further in-depth analysis of the zipper's defects and evaluation of its quality. By collecting images from multiple angles, it is possible to more comprehensively cover all parts of the zipper, reduce blind spots in detection, increase the probability of defect discovery, and improve detection accuracy.

[0024] Step S300: Based on the historical quality inspection instances corresponding to the slider, in the zipper visual quality inspection model, the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode are analyzed through a multi-angle image group with defect identification.

[0025] Specifically, make full use of the data accumulated from the historical quality inspection examples of the slider. The multi-angle image group with defect identification is input into the specially constructed zipper visual quality inspection model. Inside the model, the locked pulling mode is first analyzed. Through in-depth analysis of the image group, the specific location of the first directional defect that occurs in this mode is accurately found. At the same time, based on historical experience and the judgment ability of the model, the corresponding first abnormal failure type is determined, which includes the slider stuck and unable to be pulled, the slider and the chain teeth are not properly engaged, resulting in delinking, etc. Then, the unlocking pulling mode is processed in the same way. The second directional defect position in this mode is identified, and the corresponding second abnormal failure type is clarified, such as the slider is not unlocked smoothly, the slider does not slide smoothly in the unlocked state, etc. Accurately finding the defect position under different pulling modes provides precise targets for subsequent repairs and improvements, which helps to improve the design and production process of the slider in a targeted manner, thereby improving the quality of zipper products as a whole.

[0026] Step S400: Based on the first directional defect position and the corresponding first abnormal failure type in the locked pulling mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pulling mode, the zipper vision acquisition device is connected and a scanning probe is added.

[0027] Specifically, after clarifying the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode, the scanning probe is connected to the zipper visual acquisition device. After the connection is completed, the zipper is further inspected and analyzed using the scanning probe. The scanning probe can directly contact or approach the specific position of the zipper, and detect the conditions of these positions more accurately according to the previously determined first directional defect position and second directional defect position. Through the use of the scanning probe, more detailed and accurate information can be obtained, such as the depth, shape, size and other characteristics of the defect, as well as the material conditions around the defect, which helps to understand the nature of the defect more comprehensively and provide a more powerful basis for subsequent processing and repair. In addition, the scanning probe can also feedback the detection data in real time, cooperate with the zipper visual acquisition device and related analysis system, further verify and calibrate the accuracy of the defect position and type obtained by image analysis before, and improve the precision and accuracy of the detection. In actual operation, it is necessary to pay attention to the reasonable setting and adjustment of the position, angle, contact force and other parameters of the scanning probe according to the specific specifications, styles and detection requirements of the zipper to ensure that directional defects in various situations can be effectively detected. At the same time, it is also necessary to ensure that the connection between the scanning probe and the zipper vision acquisition device is stable and the data transmission is accurate so that accurate analysis and processing can be carried out later.

[0028] Step S500: Based on the scanning probe on the zipper vision acquisition device, combined with the basic information of the zipper, a computer-aided performance evaluation is performed to obtain defect impact indicators and failure probability.

[0029] Specifically, the scanning probe on the zipper visual acquisition device will conduct a comprehensive and detailed inspection of the zipper. It will obtain a variety of data from various parts of the zipper, such as the shape and size of the chain teeth, the specific situation of surface defects (such as the location, size, depth, etc. of the defects), the smoothness of the slider, the connection status of the latch and the pull tab, etc. Then, these detailed data are combined with the basic information of the zipper obtained previously, which includes the design specifications of the zipper, the characteristics of the materials used, the standard dimensions and allowable tolerances of each component, etc. Then, the computer-aided performance evaluation system begins to analyze, which can be done in the following ways: data comparison and analysis, comparing the actual data obtained by the scanning probe with the standard data in the basic information of the zipper. For example, compare the detected chain tooth size with the standard chain tooth size to determine whether there is a dimensional deviation; mechanical analysis, based on the principles of material mechanics, considering the characteristics of the zipper material and the nature and location of the defects, analyze the impact of the defects on the overall strength and durability of the zipper. For example, if a defect on a sprocket causes a reduction in its local strength, the system will evaluate the impact of this reduction on the load-bearing capacity of the entire zipper; statistical analysis uses statistical methods to analyze a large amount of historical inspection data of similar zippers to determine the degree of performance degradation and failure probability range that the currently detected defect usually causes under similar circumstances. Through such a specific analysis process, quantitative defect impact indicators and failure probabilities can be obtained, thereby providing a basis for zipper quality assessment and decision-making, such as determining whether the zipper is qualified, whether it needs to be repaired or replaced, etc. It can accurately quantify the degree of impact of defects on zipper performance and the possibility of failure, thereby improving the overall quality of zipper products.

[0030] Step S600: Based on the defect impact index and failure probability, the smoothness of the slider is evaluated according to the high fixed tip defect set and the low fixed tip defect set corresponding to the slider, the zipper visual quality inspection model is trained for directional defects, and is integrated into the zipper visual acquisition equipment to perform synchronous quality inspection for directional defects on the zipper production line.

[0031] Specifically, based on the defect impact indicators and failure probabilities obtained previously, these data can reflect the degree of impact of the defects in the zipper on its performance and the probability of possible failure. Next, based on the high fixed tip defect set and low fixed tip defect set corresponding to the slider, these two sets contain various defect types and characteristics that may appear in high fixed tips and low fixed tips. Then, we begin to evaluate the smoothness of the slider, and use some methods or equipment to detect the smoothness of the slider when it moves on the zipper. For example, observe the resistance during the movement of the slider, whether there is any jamming or unsmoothness, etc. After obtaining the evaluation results of the slider smoothness, use this information to conduct targeted defect reinforcement training on the zipper visual quality inspection model. It is necessary to strengthen the directional defect training of the zipper visual quality inspection model, clarify the defect type, and determine the specific defect type and characteristics that need to be detected in detail based on the high fixed tip defect set and the low fixed tip defect set; collect defect sample data, collect a large number of slider image samples containing various directional defects, and these samples should cover defects of different degrees and forms as much as possible; data annotation, accurately annotate the collected sample images, and mark the specific defect type and location in each image; select appropriate models and algorithms, such as convolutional neural networks. Adjust the training parameters and optimize the training parameters of the model, such as learning rate, number of iterations, etc., to improve training efficiency and model performance. Perform model training and train the model using labeled defect sample data. During the training process, the model learns the characteristics and patterns of defects. Introduce enhancement technology, such as using data enhancement methods such as random rotation, scaling, flipping, etc., to increase the diversity of samples to prevent model overfitting and improve its generalization ability for different situations; Real-time evaluation and adjustment, during the training process, evaluate the performance of the model on the validation set in real time, such as accuracy, recall rate and other indicators, analyze the problems of the model according to the evaluation results, adjust the model structure, parameters or add more samples for retraining; Combine defect impact indicators and failure probability, take into account the previously obtained defect impact indicators and failure probability, and conduct targeted training for different types of defects according to their importance, so that the model is more sensitive and accurate to important defects. Integration and optimization, integrate the trained model into the zipper visual acquisition equipment, and test and optimize it in the actual production environment. According to the actual detection results, further adjust the model parameters to improve its accuracy and stability in practical applications; Continuous improvement, as new defect types appear or production processes change, continue to collect new data and update the model to ensure that the model can always effectively detect directional defects. After training, the optimized zipper visual quality inspection model is integrated into the zipper visual acquisition equipment. When the zipper production line is actually running, the visual acquisition equipment integrated with the quality inspection model will perform synchronous quality inspection of directional defects on the zippers on the production line in real time.When the zipper passes through the equipment, the equipment will immediately use the model to detect whether the slider has related problems in the high fixed pin defect set or the low fixed pin defect set, and give the corresponding quality inspection results. Through this directional defect synchronous quality inspection, specific defects of zippers can be discovered and handled in a timely manner, improving product quality and production efficiency.

[0032] In a possible implementation, step S200 further includes:

[0033] Step S210: Detecting ambient light through a light sensor to obtain current lighting conditions.

[0034] Step S220: Based on the current lighting conditions, a dark channel priori algorithm is used to output a transmittance map to initialize the configuration of the zipper vision acquisition device.

[0035] Step S230: After completing the initialization configuration, perform image enhancement processing.

[0036] Specifically, first, start the light sensor and the sensor starts working. The light sensor can sense multiple parameters such as light intensity, color spectrum, and incident angle of light in the surrounding environment. During operation, the sensor continuously samples and measures ambient light. It converts the received light signal into an electrical signal, and processes and analyzes these electrical signals through internal circuits and algorithms. After processing, the light sensor obtains a series of data about the current ambient light, such as the value of light intensity (which can be expressed in units such as lux), the color temperature value of light (in Kelvin), and the direction angle of light. The data obtained from these measurements and analyses are combined to form a detailed description and definition of the current lighting conditions. Acquiring and storing these data provides a key ambient light information basis for subsequent steps.

[0037] First, get the input foggy image. For each pixel x of the image, perform the following operations: calculate the minimum value of each pixel in the local window centered on x in the three RGB channels to obtain a grayscale image. The value of each pixel in this grayscale image is the minimum value of the RGB channel in the local window of the corresponding original image. Then, perform minimum value filtering on this grayscale image, that is, use a rectangular convolution kernel of a specific size (such as 2*r+1, where r is the radius) to slide on the grayscale image, and take the minimum value in the area covered by the convolution kernel as the output value of the position each time it slides. After processing through the above steps, the dark channel image is obtained. Next, estimate the atmospheric light A. In the dark channel image, select the top 0.1% of the pixels with the largest brightness value. These pixels usually correspond to the most opaque areas of fog. In the original foggy image, find the value of the point with the highest brightness at these pixel positions and use it as the estimated value of atmospheric light A. With the estimated value of atmospheric light A, the transmittance map can be calculated. Assuming that the transmittance t(x) of each window is a constant t'(x), the calculation formula for the estimated value of the transmittance t'(x) can be derived based on the atmospheric scattering model and dark channel prior theory. However, since there are some particles even on sunny days in reality, in order to retain a certain degree of fog to ensure the authenticity of the visual effect, a factor w between 0 and 1 (generally 0.95) is introduced to correct the estimated transmittance. In addition, it is necessary to set a lower limit t0 (generally 0.1) for the transmittance. When the calculated t value is less than t0, t=t0 is taken. After the above processing, the final result is the corrected transmittance map. This transmittance map describes the attenuation of light when passing through the atmosphere. Its value is between 0 and 1, 0 means completely opaque, and 1 means completely transparent. After obtaining the transmittance map, it can be used to initialize the configuration of the zipper visual acquisition device, including automatic exposure and white balance adjustment: automatic exposure adjustment, according to the transmittance map, adjust the exposure time and gain of the camera so that the brightness of the image is moderate, not too bright or too dark. White balance adjustment, based on the transmittance map and the known atmospheric light intensity A, sets the white balance to ensure the color of the image is true and accurately restores the color of the object. Through these adjustments, the collected zipper images can have good quality under different lighting conditions, providing more accurate data for subsequent image processing and analysis.

[0038] After completing the initial configuration of the zipper visual acquisition device, the image enhancement processing is immediately carried out. First, the contrast of the image is adjusted. By analyzing the grayscale histogram of the image, the darker and brighter areas in the image are identified. Then, appropriate algorithms, such as linear or nonlinear stretching, are used to increase the difference between the dark and bright areas in the image, making the details in the image more clear and discernible, highlighting the texture and features of the zipper. Next, the image sharpening operation is implemented. Edge enhancement algorithms, such as the Sobel operator and the Laplacian operator, are used to enhance the clarity and sharpness of the zipper edge in the image, so as to more accurately detect and analyze the shape, size and defects of the zipper. Subsequently, image denoising is performed, using methods such as median filtering and Gaussian filtering to remove the noise introduced in the image due to the acquisition equipment or environmental factors, while retaining the important details and edge information of the zipper as much as possible to improve the quality and readability of the image. Finally, color balance and saturation adjustment are performed to optimize the color distribution of the image according to actual needs and lighting conditions, making the color of the image more realistic, natural, and uniform, further improving the visual effect of the image and providing strong support for subsequent precise analysis and detection of zippers.

[0039] In a possible implementation, step S400 further includes:

[0040] Step S410: Based on the multi-angle image group, a disparity map is calculated by a feature matching algorithm, and converted into multiple groups of dense point clouds.

[0041] Step S420: aligning point cloud data at multiple angles through the multiple groups of dense point clouds to build a zipper three-dimensional model.

[0042] Step S430: configuring the scanning probe based on the zipper three-dimensional model in combination with the embedding depth and the impact range.

[0043] Specifically, first, we have a set of zipper images taken from different angles, and the feature matching algorithm starts working. It searches for feature points in these images that have unique properties, are easy to identify, and are relatively stable in different images. For each feature point, the algorithm looks for similar corresponding points in other images. By comparing the positions of these corresponding points in different images, the position difference between them, that is, the disparity, can be calculated. When the calculation of the disparity of all feature points is completed, the disparity map is obtained. The value of each pixel in the disparity map represents the disparity size corresponding to the pixel position. Next, using known camera parameters, such as the focal length and pixel size of the camera, and the disparity information just calculated, each pixel in the disparity map can be converted into a point coordinate in three-dimensional space through geometric relationships and mathematical calculations. By performing such a conversion operation on the entire disparity map, multiple sets of dense point clouds can be obtained. These point clouds contain the position information of the zipper in three-dimensional space, thus providing basic data for subsequent operations such as building three-dimensional models.

[0044] Since these point clouds are acquired from different perspectives and positions, their coordinate systems and positions are different. In order to align these point cloud data, common features or iconic points, lines, surfaces and other elements will be found in different point cloud groups. Then, through the iterative closest point (ICP) algorithm, the rotation and translation transformations between the point clouds are gradually calculated and adjusted so that these point clouds can overlap and match as much as possible. During the alignment process, the accuracy of the alignment will be continuously evaluated and optimized, and multiple iterative calculations may be required to achieve a good alignment effect. When the point cloud data from multiple angles are successfully aligned, the three-dimensional model of the zipper can be built. This process is similar to a jigsaw puzzle, which integrates the aligned point cloud data together to form a continuous and complete three-dimensional structure. Some surface reconstruction algorithms may be used to construct the surface shape of the zipper based on the distribution and density of the point cloud. At the same time, the model will be smoothed, noise and outliers will be removed to improve the quality and accuracy of the model.

[0045] Based on the built 3D model of the zipper, this model contains detailed information such as the precise shape, size and structure of the zipper. Consider the embedding depth factor, which refers to the depth to which the zipper component (such as the zipper teeth) is embedded in other components (such as the fabric tape). By analyzing the 3D model, the embedding depth data of different parts is determined. At the same time, the impact range must be clarified. The impact range refers to the area that the scanning probe can cover when performing detection or operation, which is limited by factors such as the physical size of the probe, the scanning method and the detection accuracy requirements. The scanning probe is configured based on comprehensive consideration of the embedding depth and the impact range. The specific configuration includes but is not limited to the following aspects: Determine the starting position and scanning path of the scanning probe to ensure that the area to be detected can be fully covered, and at the same time, fully consider the parts with a large embedding depth to avoid missing key information. Adjust the scanning speed and step size of the scanning probe to improve the detection efficiency while ensuring the detection accuracy. For areas with a deeper embedding depth or a more complex structure, it may be necessary to appropriately reduce the scanning speed and step size. Set the detection parameters of the scanning probe, such as detection intensity and frequency, to adapt to different detection requirements and zipper material characteristics. Through the above detailed configuration, the scanning probe can efficiently and accurately detect, measure or perform other related operations on the zipper, thereby obtaining valuable data and information, providing strong support for the quality assessment and design optimization of the zipper.

[0046] In a possible implementation, step S600 further includes:

[0047] Step S610: continuously collecting data on the target model zippers of the zipper production line to obtain model-limited defect sample images.

[0048] Step S620: Based on the model-limited defect sample image, taking the zipper visual quality inspection model as a starting point, using incremental learning technology corresponding to a specific defect mode, and optimizing model parameters.

[0049] Specifically, on the zipper production line, special data acquisition devices are set up, including but not limited to high-definition industrial cameras, image sensors, etc. These devices will continuously monitor and shoot the target model zippers being produced. During the shooting process, key parts of the zipper will be focused on, such as zipper teeth, sliders, and fabric tapes. The collected raw image data will be transmitted to the data processing system in real time. The data processing system will perform preliminary screening and preprocessing on these raw images, and identify zipper images that may have defects through image recognition technology and preset screening rules. For images that are initially determined to have possible defects, more accurate image processing algorithms and manual review methods are further used to finally determine them as sample images with model-limited defects. At the same time, in order to ensure the diversity and representativeness of the data, the collection work will be carried out in different production time periods, different production batches, and different production environment conditions to cover various possible situations. The collected model-limited defect sample images will be classified and labeled, for example, according to the type of defect (such as uneven tooth spacing, damaged zipper sliders, torn fabric tapes, etc.), the severity of the defect (mild, moderate, severe), etc., so as to be used for subsequent model training and optimization.

[0050] Get the previously collected model-limited defect sample images. These images have been carefully screened, classified and annotated, and clearly show various specific defect patterns. Taking the established and trained zipper visual quality inspection model as the initial starting point, this model may already have the ability to detect some common defects, but for newly emerging or specific model defect patterns, further optimization and improvement are needed. For these specific defect patterns, use incremental learning technology. The incremental learning process is as follows: Input new model-limited defect sample images into the model, and the model will analyze and learn these new data. During the learning process, the model will adjust the internal parameters according to the characteristics and rules of the specific defect pattern in the new data. For example, for the defect pattern related to uneven tooth spacing, the model will adjust the weight and sensitivity of the tooth spacing feature. By continuously inputting new data into the model and adjusting the parameters according to the feedback results, the model's detection ability for specific defect patterns is gradually optimized. During the parameter optimization process, some evaluation indicators are used to measure the performance improvement of the model, such as accuracy, recall rate, F1 value, etc. According to the evaluation results, if the model performance does not meet expectations, we will further adjust parameters such as learning rate and number of iterations, or try different optimization algorithms, until the model can accurately and efficiently detect specific defect patterns and maintain good performance and generalization ability overall. Finally, we get an optimized zipper visual quality inspection model that can better meet the quality inspection needs of the target model zippers.

[0051] In a possible implementation, step S610 further includes:

[0052] Step S611: Based on the target model zipper, an attention mechanism is introduced to identify frequent items.

[0053] Step S612: Based on the online data stream processing architecture of the zipper production line, locate the potential defective area in the multi-angle image group by comparing the frequent items.

[0054] Step S613: for the potential defect area in the multi-angle image group, use spectral imaging technology to identify hidden defect features, and associate and bind them with the frequent items.

[0055] Specifically, for the target model of zipper, relevant data and information are obtained, including but not limited to its design specifications, production process parameters, historical quality inspection data, and images and other sensor data collected from the production line. The introduction of the attention mechanism means that when processing this data, the model does not pay equal attention to all information. The attention mechanism will automatically allocate more computing resources and weights to those features and parts that are considered more important, more discriminative, or more relevant to quality defects according to certain rules. For example, for the image data of the zipper, the attention mechanism will pay more attention to features such as the edge clarity of the zipper teeth, the uniformity of the spacing between teeth, or the shape integrity of the slider. While introducing the attention mechanism, the data is analyzed to identify frequent items. Frequent items are feature combinations, parameter value ranges, image patterns, or other measurable elements that frequently appear in the production process of this model of zipper, including the frequency of occurrence of zipper teeth of a certain size, the frequency of use of a certain material of cloth tape under certain production conditions, and the number of times a slider of a certain color appears in a batch. By identifying these frequent items, we can gain a preliminary understanding of the common patterns and characteristic distribution of this model of zipper during the production process, providing an important reference for subsequent quality inspection and defect analysis.

[0056] The online data stream processing architecture of the zipper production line is a system that can receive, process and analyze a large number of continuous data streams from the production line in real time. This architecture has efficient data transmission, storage and computing capabilities. In this architecture, relevant information about the identified frequent items is obtained, including the specific characteristics, value ranges, and occurrence patterns of these frequent items. Then, this information is used as a reference standard to conduct a detailed analysis and comparison of the multi-angle image group. During the analysis process, the differences between the features and frequent items in the image are checked pixel by pixel and region by region. Those areas that are significantly different from the features or patterns of the frequent items are marked as potential defective areas. These areas show abnormal shapes, colors, textures, sizes and other characteristics. For example, if the spacing of the zipper teeth in the frequent items is usually within a certain range, and the spacing of the teeth in a certain area in the image at a certain angle is found to be significantly beyond this range, then this area will be located as a potential defective area. By comparing the frequent items in this way, areas with possible quality problems can be quickly and accurately screened out in the multi-angle image group, preparing for subsequent more in-depth defect detection and analysis.

[0057] Once the potential defect areas are identified in the multi-angle image group, spectral imaging technology is used to further analyze these areas. Spectral imaging technology can obtain the unique features of an object under different wavelengths of light. For the potential defect areas of the zipper, this technology can capture some subtle information that is difficult to detect with conventional imaging methods. For example, it can detect hidden defect features such as changes in the chemical composition of the material, subtle stress distribution, and subtle surface roughness differences. After identifying these hidden defect features, they are associated and bound with the frequent items determined in the previous step. This means establishing a relationship between these hidden defect features and common, observable frequent items. For example, if a certain type of zipper tooth shape deviation often appears in the frequent items, the hidden defect features discovered by spectral imaging technology (such as abnormal stress distribution inside the material) have a certain causal relationship with this shape deviation. Through this association and binding, the root causes and patterns of defects can be understood more deeply, providing valuable information and basis for subsequent quality improvement measures, production process adjustments, and defect prediction, and improving the accuracy and reliability of defect detection.

[0058] In a possible implementation, step S613 further includes:

[0059] Step S6131: using spectral imaging technology, in the process of multi-angle image acquisition, collect various multi-angle image groups at different wavelengths.

[0060] Step S6132: Perform spectral analysis on each multi-angle image group at different wavelengths to separate surface feature information and internal feature information.

[0061] Step S6133: Based on the surface feature information and the internal feature information, associate with the potential defect area to locate the surface defect associated partition and the internal defect associated partition.

[0062] Step S6134: wherein the surface defect associated partition matches a first spectral feature, and the first spectral feature includes an abnormal change in reflectivity, absorptivity or emissivity.

[0063] Step S6135: The internal defect associated partition is matched with a second spectral feature, wherein the second spectral feature includes a decrease in transmittance and an abnormal change in absorption peak.

[0064] Specifically, when collecting images of zippers, a spectral imaging device is enabled. This device can emit and receive light of different wavelengths. Image acquisition devices are set at multiple angles to ensure that all sides and details of the zipper can be fully captured. When collecting images, the device will switch to different wavelengths in sequence within the preset wavelength range for irradiation and shooting. For example, first use light with a wavelength of λ1 for irradiation to obtain a set of images from various angles; then switch to light with a wavelength of λ2, and obtain a set of multi-angle images again, and repeat this process until multiple preset different wavelengths are covered. In this way, various multi-angle image groups at different wavelengths can be collected.

[0065] After acquiring each multi-angle image group at different wavelengths, spectral analysis begins. Pixels in each image group are analyzed point by point. When light of different wavelengths interacts with an object, its penetration ability and reflection and absorption characteristics will be different. For the separation of surface feature information, it is mainly based on the reflection and scattering of shorter wavelength light. For the separation of internal feature information, the focus is on the penetration and absorption of longer wavelength light. Longer wavelength light has stronger penetration ability and can reflect the internal structure and component distribution of an object. By analyzing the propagation and attenuation of light inside the object at these wavelengths, internal feature information such as the uniformity of the internal material, whether there are cavities, foreign matter, etc. can be obtained. The separated surface feature information and internal feature information are sorted and stored separately to provide targeted data support for subsequent defect location and analysis.

[0066] After obtaining the surface feature information and internal feature information, this information is compared and linked with the potential defect areas that have been identified before. Through detailed comparative analysis, the correlation between the surface feature information and the internal feature information and the potential defect areas is found. For example, if in a potential defect area, the surface texture, color or roughness and other features are found to be significantly different from those in the normal area, then this area can be associated with the surface feature information and determined as a surface defect associated partition. Similarly, if the analysis finds that the structure, material distribution and other features inside the potential defect area are abnormal, it will be associated with the internal feature information and identified as an internal defect associated partition.

[0067] For the part determined as the surface defect associated partition, it matches the specific first spectral feature. These first spectral features mainly include abnormal changes in reflectivity, absorptivity or emissivity. For example, if the reflectivity of a surface area is significantly higher or lower than the reflectivity range of a normal surface, or if the absorptivity or emissivity fluctuates abnormally, then it can be judged that there is a surface defect in the area.

[0068] For the internal defect correlation partition, it is matched with the second spectral feature. The second spectral feature mainly covers the reduction of transmittance and abnormal changes in the absorption peak. For example, if the transmittance drops significantly when the light passes through a certain internal area, or the absorption peak position and intensity of the area are significantly different from the absorption peak of the normal internal structure, then it can be inferred that there is an internal defect in this area. Through this series of correlation and matching operations, the surface and internal defect areas of the zipper can be accurately located and identified, and the defects can be accurately judged and analyzed based on the corresponding spectral characteristics.

[0069] In a possible implementation, the zipper visual quality inspection method for directional defect detection further includes:

[0070] Based on the zipper production line, the environmental impact from the production start point to the production end point is evaluated to identify environmental pressure points. Based on the environmental pressure points, a two-way balance optimization is performed in combination with production efficiency and environmental impact, and an energy optimization scheduling strategy is formulated using mixed integer linear programming in combination with the zipper production plan. Based on the energy optimization scheduling strategy, efficiency optimization is performed while ensuring the quantity corresponding to the zipper production plan.

[0071] Specifically, first, for the zipper production line, a comprehensive assessment of the environmental impact of the entire process from the production start point (raw material acquisition, processing preparation, etc.) to the production end point (product packaging, transportation, etc.) is conducted. This includes but is not limited to the assessment of energy consumption, waste generation, greenhouse gas emissions, water resource use, etc. Through detailed analysis and data collection, the links that cause greater pressure on the environment in the entire production chain, namely the environmental pressure points, are identified.

[0072] After clarifying the environmental pressure points on the zipper production line, we started to carry out two-way balance optimization. This requires taking into account both the improvement of production efficiency and the reduction of environmental impact. On the one hand, we deeply analyze the factors that restrict efficiency in the production process, such as equipment operation efficiency, the smoothness of coordination between processes, and the proficiency of workers, and explore feasible ways to improve efficiency and room for improvement. On the other hand, for environmental pressure points, such as the operation of high-energy-consuming equipment and the process of generating a large amount of waste, we explore ways to reduce the adverse impact on the environment through technological innovation, process improvement, or the use of more environmentally friendly materials. At the same time, combined with the production plan of zippers, we clearly grasp the specific indicators such as the number, specifications, and delivery time of zippers to be produced. At this time, the multi-objective optimization algorithm is used to formulate an energy optimization scheduling strategy. The multi-objective optimization algorithm can handle multiple conflicting goals at the same time and find the optimal balance solution between them. First, the improvement of production efficiency and the reduction of environmental impact are set as the two main goals. For production efficiency, it can be measured by indicators such as output per unit time and equipment utilization rate; for environmental impact, it can be quantified by indicators such as energy consumption, waste emissions, and greenhouse gas emissions. Then, the production plan and various constraints (such as energy supply upper limit, equipment capacity limit, production time requirements, etc.) are converted into constraints in the algorithm. The multi-objective optimization algorithm searches the feasible solution space while satisfying these constraints to find a set of non-inferior solutions (Pareto optimal solutions) that represent different degrees of balance between production efficiency and environmental impact. For example, some solutions may focus on improving production efficiency while accepting relatively high environmental impacts, while other solutions may focus more on reducing environmental impacts but with slightly reduced production efficiency. Decision makers can choose the most appropriate solution from this set of non-inferior solutions as the final energy optimization scheduling strategy based on actual needs and preferences.

[0073] After establishing the energy optimization scheduling strategy, we will further carry out efficiency optimization work based on this strategy, and carefully analyze each link and operation in the production process, including the supply and storage of raw materials, the operation and maintenance of production equipment, the work arrangement and cooperation of workers, etc. While meeting the quantity requirements of the zipper production plan, we will maximize the efficiency of the entire production process and reduce production costs.

[0074] Embodiment 2

[0075] Based on the same inventive concept as the zipper visual quality inspection method for directional defect detection in the aforementioned embodiment, Figure 2 As shown, the present application provides a zipper visual quality inspection system for directional defect detection, and the system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0076] The zipper basic information acquisition module 10 is used to acquire the zipper basic information. The components corresponding to the zipper basic information include sprockets, sliders, latches, and pull tabs.

[0077] The multi-angle image group acquisition module 20 is used to perform multi-angle acquisition through a zipper visual acquisition device, acquire a multi-angle image group, classify directional defects and add defect identification.

[0078] The zipper visual quality inspection model analysis module 30 is based on the historical quality inspection instances corresponding to the slider. In the zipper visual quality inspection model, the module 30 analyzes the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode through a multi-angle image group with defect identification.

[0079] The scanning probe adding module 40 is used to connect the zipper vision acquisition device and add a scanning probe based on the first directional defect position and the corresponding first abnormal failure type in the locked pulling mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pulling mode.

[0080] The defect impact index and failure probability acquisition module 50 performs computer-aided performance evaluation based on the scanning probe on the zipper visual acquisition device and the zipper basic information to obtain the defect impact index and failure probability.

[0081] The directional defect synchronous quality inspection module 60 evaluates the smoothness of the slider based on the defect influencing index and failure probability according to the high fixed tip defect set and low fixed tip defect set corresponding to the slider, performs directional defect reinforcement training on the zipper visual quality inspection model, and integrates it into the zipper visual acquisition equipment to perform directional defect synchronous quality inspection on the zipper production line.

[0082] Furthermore, the multi-angle image group acquisition module 20 further includes:

[0083] The lighting condition acquisition unit is used to detect ambient light through a light sensor to acquire current lighting conditions.

[0084] An initialization configuration unit, wherein the initialization configuration unit outputs a transmittance map using a dark channel priori algorithm based on the current lighting conditions to initialize the configuration of the zipper vision acquisition device.

[0085] An image enhancement processing unit, wherein the image enhancement processing unit is used to perform image enhancement processing after completing the initialization configuration.

[0086] Furthermore, the scanning probe adding module 40 also includes:

[0087] A plurality of dense point cloud conversion units are used to calculate a disparity map based on the multi-angle image group by a feature matching algorithm and convert the disparity map into a plurality of dense point clouds.

[0088] A zipper three-dimensional model building unit is used to align point cloud data at multiple angles through the multiple groups of dense point clouds to build a zipper three-dimensional model.

[0089] A scanning probe configuration unit is provided, wherein the scanning probe configuration unit configures the scanning probe based on the zipper three-dimensional model in combination with the embedding depth and the influence range.

[0090] Furthermore, the directional defect synchronous quality inspection module 60 further includes:

[0091] A model-limited defect sample image acquisition unit is used to continuously collect data on the target model zipper of the zipper production line and acquire model-limited defect sample images.

[0092] A model parameter optimization unit, wherein the model parameter optimization unit is based on the model-limited defect sample image, takes the zipper visual quality inspection model as a starting point, uses incremental learning technology corresponding to a specific defect mode, and optimizes the model parameters.

[0093] Furthermore, the model-limited defect sample image acquisition unit further includes:

[0094] A frequent item recognition unit is provided, wherein the frequent item recognition unit introduces an attention mechanism and recognizes frequent items based on the target model zipper.

[0095] A potential defect area positioning unit is configured to locate the potential defect area in the multi-angle image group by comparing the frequent items based on the online data stream processing architecture of the zipper production line.

[0096] A frequent item associating and binding unit is used to identify hidden defect features using spectral imaging technology for potential defect areas in the multi-angle image group, and associate and bind them with the frequent items.

[0097] Furthermore, the frequent item association binding unit further includes:

[0098] The multi-angle image group collecting unit uses spectral imaging technology to collect various multi-angle image groups under different wavelengths during the multi-angle image acquisition process.

[0099] The spectral analysis unit is used to perform spectral analysis on each multi-angle image group at different wavelengths to separate surface feature information and internal feature information.

[0100] A potential defect area association unit is used to associate with the potential defect area based on the surface feature information and the internal feature information, and locate the surface defect association partition and the internal defect association partition.

[0101] The first spectral feature matching unit is used for the surface defect associated partition and matches with the first spectral feature, wherein the first spectral feature includes an abnormal change in reflectivity, absorptivity or emissivity.

[0102] The second spectral feature matching unit is used for the internal defect associated partition and matches with the second spectral feature, and the second spectral feature includes a decrease in transmittance and an abnormal change in absorption peak.

[0103] Furthermore, the frequent item association binding unit further includes:

[0104] Based on the zipper production line, the environmental impact from the start to the end of production is evaluated and the environmental pressure points are identified.

[0105] Based on the environmental pressure points, a two-way balance optimization is performed in combination with production efficiency and environmental impact, and combined with the zipper production plan, an energy optimization scheduling strategy is formulated using mixed integer linear programming.

[0106] Based on the energy optimization scheduling strategy, efficiency optimization is performed while ensuring the quantity corresponding to the zipper production plan.

[0107] Furthermore, the zipper visual quality inspection system for directional defect detection also includes:

[0108] An environmental pressure point identification unit is used to evaluate the environmental impact from the production start point to the production end point based on the zipper production line and identify the environmental pressure points.

[0109] A two-way balance optimization unit is provided, wherein the two-way balance optimization unit performs two-way balance optimization based on the environmental pressure points in combination with production efficiency and environmental impact, and combines the zipper production plan to formulate an energy optimization scheduling strategy using mixed integer linear programming.

[0110] An efficiency optimization unit, wherein the efficiency optimization unit performs efficiency optimization based on the energy optimization scheduling strategy while ensuring the quantity corresponding to the zipper production plan.

[0111] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0112] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0113] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A zipper visual quality inspection method for directional defect detection, characterized in that: The method comprises: Obtaining basic information of a zipper, wherein each component corresponding to the basic information of the zipper includes a sprocket, a slider, a latch, and a pull tab; Through the zipper vision acquisition device, multi-angle acquisition is performed to obtain multi-angle image groups, directional defect classification and defect identification are added; Based on the historical quality inspection examples corresponding to the slider, in the zipper visual quality inspection model, the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode are analyzed through a multi-angle image group with defect identification; Based on the first directional defect position and the corresponding first abnormal failure type in the locked pulling mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pulling mode, the zipper visual acquisition device is connected and a scanning probe is added; Based on the scanning probe on the zipper visual acquisition device, combined with the basic information of the zipper, a computer-aided performance evaluation is performed to obtain defect impact indicators and failure probability; Based on the defect impact index and failure probability, the smoothness of the slider is evaluated according to the high fixed tip defect set and the low fixed tip defect set corresponding to the slider, and the zipper visual quality inspection model is trained for directional defects, and integrated into the zipper visual acquisition equipment to perform synchronous quality inspection for directional defects on the zipper production line; Wherein, the method further comprises: Based on the multi-angle image group, a disparity map is calculated by a feature matching algorithm, and converted into multiple groups of dense point clouds; By using the multiple groups of dense point clouds, point cloud data at multiple angles are aligned to build a zipper three-dimensional model; Based on the zipper three-dimensional model, combined with the embedding depth and the impact range, the scanning probe is configured; Among them, the method is integrated into the zipper visual acquisition device to perform synchronous quality inspection of directional defects on the zipper production line, and includes: Continuously collecting data on the target model zippers of the zipper production line to obtain model-limited defect sample images; Based on the model-limited defect sample images, taking the zipper visual quality inspection model as a starting point, using incremental learning technology corresponding to specific defect patterns, and optimizing model parameters; Wherein, data collection is continuously performed on the target model zippers of the zipper production line, and the method further comprises: Based on the target model zipper, an attention mechanism is introduced and frequent items are identified; Based on the online data stream processing architecture of the zipper production line, locating potential defect areas in the multi-angle image group by comparing the frequent items; For the potential defect area in the multi-angle image group, spectral imaging technology is used to identify hidden defect features and associate and bind them with the frequent items.

2. The zipper visual quality inspection method for directional defect detection according to claim 1, characterized in that: By using a zipper vision acquisition device, multi-angle acquisition is performed to obtain a multi-angle image group, and the method further includes: Detect ambient light through light sensors to obtain current lighting conditions; Based on the current lighting conditions, a dark channel priori algorithm is used to output a transmittance map to initialize the configuration of the zipper vision acquisition device; After completing the initialization configuration, image enhancement processing is performed.

3. The zipper visual quality inspection method for directional defect detection according to claim 1, characterized in that: Using spectral imaging technology to identify hidden defect characteristics, the method also includes: Using spectral imaging technology, during the multi-angle image acquisition process, each multi-angle image group at different wavelengths is collected; Perform spectral analysis on each multi-angle image group at different wavelengths to separate surface feature information and internal feature information; Based on the surface feature information and the internal feature information, the potential defect area is associated with the potential defect area to locate the surface defect associated partition and the internal defect associated partition; Wherein, the surface defect associated partition matches a first spectral feature, and the first spectral feature includes an abnormal change in reflectivity, absorptivity or emissivity; The internal defect associated partition matches the second spectral feature, which includes a decrease in transmittance and an abnormal change in absorption peak.

4. The zipper visual quality inspection method for directional defect detection according to claim 1, characterized in that: The method further comprises: Based on the zipper production line, evaluate the environmental impact from the production start to the production end, and identify environmental pressure points; Based on the environmental pressure points, a two-way balance optimization is performed in combination with production efficiency and environmental impact, and combined with the zipper production plan, an energy optimization scheduling strategy is formulated using mixed integer linear programming; Based on the energy optimization scheduling strategy, efficiency optimization is performed while ensuring the quantity corresponding to the zipper production plan.

5. Zipper visual quality inspection system for directional defect detection, characterized in that: For executing the method according to any one of claims 1 to 4, the system comprises: A zipper basic information acquisition module, wherein the zipper basic information acquisition module is used to acquire the zipper basic information, wherein each component corresponding to the zipper basic information includes a sprocket, a slider, a latch, and a pull tab; A multi-angle image group acquisition module, which is used to perform multi-angle acquisition through a zipper visual acquisition device, acquire a multi-angle image group, classify directional defects and add defect identification; A zipper visual quality inspection model analysis module, which is based on historical quality inspection instances corresponding to the slider. In the zipper visual quality inspection model, the module analyzes the first directional defect position and the corresponding first abnormal failure type in the locked pull mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pull mode through a multi-angle image group with defect identification; A scanning probe adding module, the scanning probe adding module is used to connect the zipper visual acquisition device and add a scanning probe based on the first directional defect position and the corresponding first abnormal failure type in the locked pulling mode, and the second directional defect position and the corresponding second abnormal failure type in the unlocked pulling mode; A defect impact index and failure probability acquisition module, which performs computer-aided performance evaluation based on a scanning probe on the zipper visual acquisition device and the zipper basic information to obtain defect impact index and failure probability; A directional defect synchronous quality inspection module, which evaluates the smoothness of the slider based on the defect impact index and failure probability according to the high fixed tip defect set and low fixed tip defect set corresponding to the slider, performs directional defect reinforcement training on the zipper visual quality inspection model, and is integrated into the zipper visual acquisition equipment to perform directional defect synchronous quality inspection on the zipper production line.