An automatic clothing pattern detection method and system based on visual detection

Abstract

This invention discloses an automatic detection method and system for clothing patterns based on visual inspection, belonging to the field of visual inspection technology for clothing patterns. The invention includes acquiring the continuous detection range of the visual inspection system over the clothing pattern, matching the capture frequency to the clothing pattern transmission rate, synchronously marking the collected pattern data according to the timeline, and generating a raw dataset with spatiotemporal coordinates. Based on a decision-linked processing module, the invention ensures the reliability of the analysis and evaluation signals through a verification mechanism. When no errors are found, the detection equipment parameters are automatically adjusted; when discrepancies exist, abnormal steps are inferred or interference sources are identified, forming a closed loop of acquisition-analysis-evaluation-control. This not only unifies the detection standards but also enhances the ability to cope with interference factors, ultimately ensuring the stability of clothing pattern quality, reducing production costs, and improving product market competitiveness.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology for clothing patterns, specifically to an automatic detection method and system for clothing patterns based on visual inspection. Background Technology

[0002] As a core design element of clothing, clothing patterns encompass various forms such as printing, embroidery, and sewing textures. Their visual effects directly affect the market competitiveness of products. Traditional manual inspection methods are inefficient and highly subjective, making it difficult to meet the needs of large-scale production. With the development of computer vision technology, automatic inspection methods based on deep learning have gradually become the mainstream.

[0003] The application of visual inspection technology in automatic detection of clothing patterns extracts clothing areas through semantic segmentation models, converts clothing images in the wearing state into tiled display images, eliminates human posture and background interference, and provides standardized data for subsequent analysis.

[0004] Traditional manual inspection relies on visual judgment, which is highly subjective. Moreover, the inspection results are mostly stored in paper records or simple electronic files, lacking spatiotemporal coordinate marking and standardized classification of pattern data. This makes data traceability difficult, and when quality disputes arise, it is impossible to accurately trace the time, location, and related factors of the defects. Limited by labor costs and fatigue, it is difficult to adapt to the high-speed inspection needs of large-scale production lines. At the same time, the risk assessment of pattern defects relies heavily on experience-based judgment, lacking early screening of low-risk suspected defects and trend analysis of high-risk defects. Often, the defects are only dealt with passively after they have expanded, leading to increased rework costs.

[0005] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for automatic detection of clothing patterns based on visual inspection, in order to solve the problems mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an automatic detection method for clothing patterns based on visual detection, comprising the following steps;

[0008] S1. Obtain the continuous detection range of the visual detection pair passing through the clothing pattern, match the capture frequency to the clothing pattern transmission rate, synchronously mark the collected pattern data according to the timeline, and generate the original dataset with spatiotemporal coordinates.

[0009] S2. After performing grayscale processing and noise reduction on the original dataset, construct an active cache file library. Classify the processed original data according to the pre-stored quality rules: data of the type that meets the standard is stored in the qualified data sub-library, and new pattern data that can be referenced in the historical database is added. Data of the type that is suspected of being defective is stored in the data sub-library to be inspected.

[0010] S3. Retrieve the low-risk and high-risk suspected defect sets from the data sub-database to be inspected, and combine them with the historical database and the current cache file library for joint analysis to generate quality analysis signals and high-risk assessment signals.

[0011] S4. After obtaining the quality analysis signal and the high-risk assessment signal, combine the qualified data sub-library and the data sub-library to be inspected for joint verification. If there is no error, adjust the visual inspection equipment to correct the pattern defect. If there is a difference, reverse reasoning is used to mark the abnormal step or analyze the interference source, and generate control and warning instructions.

[0012] Furthermore, the process of obtaining the original dataset in step S1 is as follows:

[0013] The system continuously and dynamically detects and collects patterns of clothing passing on the conveyor belt within a continuous detection range. The acquisition frequency is matched with the clothing transmission speed to avoid data omission. A unique timestamp is added to each frame of pattern data collected according to the timeline. The timestamp information includes the acquisition time, device number, and clothing transmission position coordinates, generating a raw dataset with spatiotemporal coordinates.

[0014] Furthermore, the processing procedure for the original dataset in step S2 is as follows:

[0015] The original dataset is obtained and grayscale processing is performed. Color pattern data in the original dataset is retrieved and converted into grayscale images to simplify the data dimensions. Gaussian filtering algorithm is used to denoise the grayscale images to eliminate noise points caused by ambient light interference, equipment jitter, etc. The processed image data is built into an active cache file library. The files are hashed and marked to ensure data uniqueness and integrity.

[0016] The system retrieves pre-stored pattern defect judgment criteria from the existing cache file library and divides the data into qualified data sub-libraries and data to be inspected sub-libraries: the qualified data sub-library stores pattern data that is free of defects or meets the standards, and is used to supplement newly stored pattern data that can be referenced in the historical database, serving as a benchmark for subsequent analysis; the data to be inspected sub-library stores pattern data that is suspected of having defects, and is divided into low-risk suspected defect sets and high-risk suspected defect sets according to the defect risk level.

[0017] Furthermore, the analysis process for the low-risk suspected defect set in step S3 is as follows:

[0018] The system retrieves the low-risk suspected defect set from the data sub-database to be inspected, and compares it with the pre-stored standard pattern templates and newly stored pattern data from the historical database. The difference between the standard pattern templates and the newly stored pattern data is used as the error tolerance coefficient. When the comparison between the low-risk suspected defect set and the standard pattern template is close and the difference is less than the error tolerance coefficient, a callback classification signal is generated. When the comparison between the low-risk suspected defect set and the standard pattern template is close and the difference is greater than or equal to the error tolerance coefficient, a quality analysis signal is generated.

[0019] Furthermore, the analysis and processing procedure for the high-risk suspected defect set in step S3 is as follows:

[0020] The high-risk suspected defect set is obtained from the sub-database to be inspected. At the same time, the historical records of the same type of clothing patterns in the historical database are extracted. The high-risk difference between the high-risk suspected defect set and the corresponding standard pattern sample in the historical records is calculated and marked as the historical difference. The collection period of the high-risk suspected defect set is divided into N time nodes according to the timeline. Each time node corresponds to a sampling point of the original dataset. The original parameters, historical records parameters and historical differences of the high-risk suspected defect set at that time node are recorded to construct the node dataset.

[0021] Furthermore, a node dataset is constructed with time nodes as the X-axis and clothing pattern difference reference values ​​pre-stored in the historical database as the Y-axis to construct a Cartesian coordinate system;

[0022] Plot points in the coordinate system and connect them to draw three sets of floating curves: the first set of floating curves represents the high-risk floating curves of the high-risk suspected defect set at each time node; the second set of floating curves represents the historical floating curves of the historical record parameters; and the third set represents the historical floating curves of the historical-current difference. Calculate the deviation of each set of curves from the Y-axis reference value. If the high-risk floating curve deviates from the reference value by more than 30% in a single instance, or if the historical-current floating curve shows an upward trend for three consecutive time nodes and the cumulative deviation exceeds 50%, then the high-risk suspected defect set is determined to have significant defect risk, and a high-risk assessment signal is generated. If the deviations of both the high-risk floating curve and the historical-current floating curve do not reach the difference reference value of 30%, then the high-risk suspected defect set is determined to have no significant risk, and a no-significant-risk assessment signal is generated.

[0023] Furthermore, the processing procedure for the acquired quality analysis signal and high-risk assessment signal in step S4 is as follows:

[0024] The system acquires quality analysis signals and high-risk assessment signals, and generates defect sets for both sets of signals. It then retrieves standard pattern templates of similar garment patterns from the historical database as verification benchmarks. Temporary sample comparison parameters are constructed from the low-risk and high-risk suspected defect sets in the quality analysis and high-risk assessment signals. These temporary parameters are compared with the standard pattern templates. If the parameter matching degree between the temporary sample comparison parameters and the standard pattern templates is ≥95% and there are no logical conflicts, the verification is deemed correct, and control instructions are generated to adjust the operating parameters of the visual inspection equipment to correct potential pattern defects in subsequent inspections. If the parameter matching degree between the temporary sample comparison parameters and the standard pattern templates is <95% or there are logical conflicts, the verification anomaly handling mechanism is activated.

[0025] Furthermore, step S4 performs reverse reasoning on the processing procedure of step S3, as follows:

[0026] The key steps in processing the original dataset are traced, including the timestamp matching of the original data collection, the threshold setting for grayscale processing, and the calculation logic for historical differences. If reverse reasoning reveals that the parameter setting of a certain step deviates from the pre-stored threshold, the step is marked as an abnormal step, and an abnormal warning instruction containing the step number, abnormal parameters, and correction suggestions is generated. If no abnormal step is found, possible interference sources are analyzed, including ambient light intensity fluctuations, equipment mechanical vibration, and clothing wrinkles. By comparing the environmental parameters in the collection log with the equipment operation data, the type of interference source is determined and an interference warning instruction is generated. The abnormal warning instruction and the interference warning instruction are synchronized to the terminal display.

[0027] An automatic garment pattern detection system based on vision detection includes the following steps:

[0028] The visual data acquisition module communicates with the visual inspection equipment and is used to continuously acquire raw datasets of clothing patterns and mark them according to a timeline.

[0029] The shallow data analysis module communicates with the visual data acquisition module and the pattern cache core platform to process the raw dataset and generate grade analysis signals.

[0030] The deep risk assessment module communicates with the pattern cache core platform to jointly analyze data and generate high-risk assessment signals.

[0031] The decision-making linkage processing module is connected to the shallow data analysis module, the deep risk assessment module, and the pattern cache core platform, respectively. It is used to process the grade analysis signal and the high-risk assessment signal and generate control commands, abnormal warning commands, and interference warning commands.

[0032] The pattern caching core platform communicates with each of the above modules to store data, thresholds, and models, providing data support and triggering operation with signals and instructions.

[0033] The beneficial effects of this invention are:

[0034] 1. This invention continuously acquires clothing patterns with spatiotemporal coordinates through a visual data acquisition module, and combines timestamps, device numbers, and location coordinate markers to ensure that each frame of data can be accurately traced. At the same time, the pattern caching core platform classifies, stores, and hashes the data, enabling the correlation and comparison of historical data and current data. This solves the problems of data chaos and difficulty in traceability in traditional testing, provides a standardized data foundation for quality analysis, and improves the accuracy and traceability of test data.

[0035] 2. This invention works in collaboration between a shallow data analysis module and a deep risk assessment module. It optimizes data quality through grayscale processing and noise reduction, performs hierarchical analysis on suspected defects, and constructs a floating curve model based on historical data to predict the trend of high-risk defects. Compared with manual inspection, this invention significantly improves efficiency, can identify potential defects in advance, reduces rework rate, and improves inspection efficiency and risk prediction capabilities. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0038] Figure 2 This is a flowchart of the system of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1: Please refer to Figure 1 - Figure 2As shown, this embodiment is a method and system for automatic detection of clothing patterns based on visual inspection. It includes a pattern caching core platform, which is connected to each module for communication. It is used to store historical databases and pre-stored thresholds. The pre-stored thresholds include pattern difference reference values, verification matching degree thresholds, equipment operating parameter thresholds, etc. It supports manual updates and provides data support. Each module realizes data interaction through industrial Ethernet or IoT, with a response latency of ≤100ms. The specific steps are as follows:

[0041] S1. The visual data acquisition module communicates with the visual inspection equipment to continuously acquire raw datasets of clothing patterns and mark them according to a timeline. It obtains the continuous detection range of the visual inspection pair passing through the clothing pattern to match the capture frequency with the clothing pattern transmission rate. It synchronously marks the acquired pattern data according to the timeline, generates raw datasets with spatiotemporal coordinates, and stores them in the pattern cache core platform. The raw dataset acquisition and processing process in step S1 is as follows:

[0042] This system continuously and dynamically acquires data on clothing patterns passing along a conveyor belt within a continuous detection range. It utilizes an industrial camera and image sensor; when using an industrial camera, the vision data acquisition module employs a 20-megapixel camera with LED lighting. It continuously acquires 30 frames per second of clothing patterns at a transmission speed of 0.1-1 m / s. Each frame is stamped with a timestamp containing "YYYY-MM-DD-HH-MM-SS-Device ID-Conveyor Belt Position," for example, "2024-07-18-10-30-22-CAM01-POS05." The acquisition frequency is matched to the clothing transmission speed to avoid data omissions. A unique timestamp is added to each frame of acquired pattern data according to the timeline, containing the acquisition time, device number, and clothing transmission position coordinates, generating a raw dataset with spatiotemporal coordinates.

[0043] The labeled raw dataset is transmitted in real time to the historical database of the pattern caching core platform, and archived according to clothing type and production batch, retaining at least 6 months of records. At the same time, a collection log is generated to record data integrity, such as whether there are missing or blurred frames, providing traceable basic data support for subsequent data retrieval and analysis.

[0044] S2, the shallow data analysis module communicates with the visual data acquisition module and the pattern caching core platform to process the raw dataset and generate quality analysis signals. After grayscale processing and noise reduction, the raw dataset is used to construct an active cache file library. This library is used for temporary storage and is automatically archived to the historical database after 24 hours. The processed raw data is classified according to pre-stored quality rules: data that conforms to the classification criteria are stored in the qualified data sub-library, and newly stored pattern data for reference in the historical database is added; data suspected of defects and that do not conform to the classification criteria are stored in the pending inspection data sub-library. The processing procedure for the raw dataset in step S2 is as follows:

[0045] The original dataset is first processed to convert RGB images into 8-bit grayscale images. Color pattern data in the original dataset is retrieved and converted into grayscale images to simplify the data dimensions and reduce computational complexity.

[0046] The Gaussian filtering algorithm is used to denoise grayscale images. The filter kernel size can be set to 3×3 to eliminate noise points caused by ambient light interference, equipment jitter, etc. The processed image data is used to build an active cache file library. The files are MD5 hashed and marked to ensure data uniqueness and integrity. Files with the same first 8 bits of hash value are grouped into the same qualified data sub-library or the data sub-library to be inspected.

[0047] Retrieve the pre-stored pattern defect judgment criteria from the existing cache file library to divide the qualified data sub-library and the data sub-library to be inspected:

[0048] The qualified data sub-database stores pattern data that has been manually verified or screened by comparison with historical conforming templates. The pattern data that is defect-free or conforms to the standards is stored in the qualified data sub-database. It can be used to supplement the newly stored pattern data that can be referenced in the historical database when archiving, and serve as a benchmark for subsequent analysis.

[0049] The data to be inspected sub-database stores pattern data that are suspected of having defects, and is divided into a low-risk suspected defect set and a high-risk suspected defect set according to the defect risk level. Specifically, the defect risk level can be defined as follows: data with a size deviation of <1mm is judged as slightly suspected defect data, and data with a size deviation ≥1mm or a color deviation ≥10% is high-risk suspected defect data.

[0050] S3, the deep risk assessment module communicates with the pattern cache core platform to jointly analyze data and generate high-risk assessment signals. It retrieves low-risk and high-risk suspected defect sets from the data sub-library to be inspected, and combines them with the historical database and the current cache file library for joint analysis to generate quality analysis signals and high-risk assessment signals. The analysis process of the low-risk suspected defect set in step S3 is as follows:

[0051] The low-risk suspected defect set is retrieved from the data sub-database to be inspected. The standard pattern templates and newly stored pattern data in the historical database are retrieved and compared with the low-risk suspected defect set. The difference between the standard pattern templates and the newly stored pattern data is used as the error tolerance coefficient.

[0052] When the low-risk suspected defect set is close to the standard pattern template and the difference value is less than the pattern difference tolerance coefficient, a callback classification signal is generated. When the pattern cache core platform receives the callback classification signal, it can mark the low-risk suspected defect set that generated the callback classification signal in the data sub-database to be inspected with color, generate a callback archive log, and at the same time re-edit the name of the low-risk suspected defect set, add the archive time and comparison results and merge it into the historical database as the new pattern data of the low-risk suspected defect set of the same type of clothing pattern.

[0053] When the low-risk suspected defect set is close to the standard pattern template and the difference value is greater than or equal to the pattern difference tolerance coefficient, a defect analysis signal is generated. The defect analysis signal contains the defect location and type.

[0054] The analysis and processing procedure for the high-risk suspected defect set in step S3 is as follows:

[0055] Obtain the high-risk suspected defect set from the sub-database to be inspected, and at the same time extract 100 sets of historical data parameters of the same style of clothing in the past 3 months from the historical database. The historical data parameters include standard pattern templates and defect repair cases of this type of pattern in different batches and different inspection scenarios.

[0056] Calculate the high-risk difference between the high-risk suspected defect set and the corresponding standard pattern template in the historical record, and label it as the historical difference. The high-risk difference includes pattern size deviation, color saturation deviation, pattern position offset, etc.

[0057] The high-risk suspected defect set is divided into N time nodes according to the timeline. N is a natural number greater than zero and can take the value N≥5. Each time node corresponds to a sampling point of the original dataset. The original parameters, historical parameters and historical difference of the high-risk suspected defect set at that time node are recorded to construct the node dataset and provide a data foundation for subsequent curve plotting.

[0058] Using the time node as the X-axis and retrieving the clothing pattern difference reference value pre-stored in the historical database as the Y-axis, a rectangular coordinate system is constructed. The clothing pattern difference reference value is pre-stored in the pattern cache core platform and is set according to the pattern type. For example, the difference reference value for printed patterns is 0.5mm size deviation and 5% color deviation.

[0059] Plot the points in the coordinate system and connect them to draw three sets of floating curves:

[0060] The first set of floating curves represents the high-risk floating curves of the high-risk suspected defect set at each time point.

[0061] The second set of floating curves are the historical floating curves of historical record parameters;

[0062] The third group consists of the historical fluctuation curves of the historical-current difference;

[0063] Calculate the degree of deviation of each group of curves from the Y-axis reference value. If a high-risk floating curve deviates from the reference value by more than 30% in a single instance, or if the historical floating curve shows an upward trend for three consecutive time nodes and the cumulative deviation exceeds 50%, then the group of high-risk suspected defects is determined to have significant defect risk. A high-risk assessment signal is generated, which includes the defect level and risk location. The defect level can include minor, moderate, and severe. It is trained based on data from the historical database and combined with the deviation percentage to determine the range of the level. It is not limited to this and is updated and confirmed based on each garment pattern and data from the historical database.

[0064] For example, if the historical average size of a set of clothing patterns is 30cm, and the size of a high-risk suspected defect set at a certain time node is 29.5cm, then the historical-to-current difference is 0.5cm. Dividing the data into 5 time nodes with a 10-second collection period, after plotting the curve, it was found that the historical-to-current difference curve increased from 0.3cm to 0.8cm for 3 consecutive nodes, with a cumulative deviation of more than 50%, generating a high-risk assessment signal. The high-risk assessment signal contains information such as medium defect risk and location on the right edge of the pattern. After receiving the high-risk assessment signal, the pattern caching core platform immediately locks and marks the relevant data of the high-risk assessment signal with color, generates a processing log, converts it into text information about the high-risk assessment signal, and sends it to the smart device of the supervisor for display.

[0065] If the deviations of the high-risk floating curve and the historical floating curve do not reach the difference reference value of 30%, then the high-risk suspected defect set is judged to have no significant risk, a no-significant-risk assessment signal is generated, and it is synchronously stored in the pattern cache core platform.

[0066] Example 2

[0067] S4, the decision-making linkage processing module, communicates with the shallow data analysis module, the deep risk assessment module, and the pattern cache core platform, respectively. It processes the quality analysis signal and the high-risk assessment signal, generating control commands, anomaly warning commands, and interference warning commands. After acquiring the quality analysis signal and the high-risk assessment signal, it verifies them jointly with the qualified data sub-library and the data sub-library to be inspected. If there are no errors, the visual inspection equipment is adjusted to correct the pattern defects. If discrepancies exist, reverse reasoning is used to mark abnormal steps or analyze interference sources, generating control and warning commands. The processing procedure for the acquired quality analysis signal and high-risk assessment signal in step S4 is as follows:

[0068] Obtain quality analysis signals and high-risk assessment signals, and generate defect sets for the two sets of signals. Retrieve standard pattern templates of similar clothing patterns from the historical database as verification benchmarks.

[0069] Temporary sample comparison parameters are constructed from the low-risk and high-risk suspected defect sets in the quality analysis and high-risk assessment signals. These temporary sample comparison parameters are then compared with the standard pattern template.

[0070] If the temporary sample comparison parameters match the standard pattern template parameters by ≥95% and there is no logical conflict, the verification is deemed correct, and a control command is generated. No logical conflict can be interpreted as the defect location marked by the quality analysis signal in the defect set of the same type being consistent with the high-risk assessment signal. When the pattern cache core platform receives the control command, it adjusts the operating parameters of the visual inspection equipment, such as adjusting the lens focal length, supplementary light intensity, and transmission speed, to correct potential pattern defects in subsequent inspections. It should be noted that the quality analysis signal in the temporary sample comparison parameters shows that the allowable deviation of the right edge of the pattern is ≤0.6cm. Verification reveals that the 0.8cm deviation of the high-risk assessment signal is logically correct, thus generating a control command. When the pattern cache core platform receives the control command, it adjusts the supplementary light intensity from 500 lux to 600 lux, fine-tunes the lens focal length by 0.5mm, and keeps the transmission speed constant at 1m / s. If the deviation is found to decrease to 0.4cm in subsequent inspections, the adjustment is deemed effective.

[0071] If the temporary sample parameters match the standard pattern template parameters by less than 95% or there is a logical conflict, the verification anomaly handling mechanism will be activated.

[0072] S4 performs reverse reasoning on the processing procedure of step S3 and initiates the verification exception handling mechanism. The specific process is as follows:

[0073] The key steps in processing the original dataset are traced, including the timestamp matching of the original data collection, the threshold setting for grayscale processing, and the calculation logic for the difference between past and present values.

[0074] If reverse reasoning finds that the parameter settings of a certain step deviate from the pre-stored threshold, such as abnormal filtering coefficients in the noise reduction process, then the step is marked as an abnormal step, and an abnormal warning instruction containing the step number, abnormal parameters, and correction suggestions is generated.

[0075] If no abnormal steps are found, possible sources of interference are analyzed. These sources include ambient light intensity fluctuations, equipment vibration, and clothing wrinkles. Ambient light intensity fluctuations are defined as light intensity changes exceeding 200 lux during the data collection period. Equipment vibration is defined as conveyor belt vibration frequency > 5 Hz. Clothing wrinkles are defined as wrinkles covering more than 10% of the total pattern area. By comparing environmental parameters in the data collection log with equipment operating data, the type of interference source is determined and an interference warning instruction is generated. For example, the interference warning instruction may suggest adjusting the position of the supplementary lighting equipment to stabilize the lighting. The abnormal warning instruction and the interference warning instruction are synchronized to the pattern cache core platform. After parsing the abnormal warning instruction and the interference warning instruction, the pattern cache core platform generates text information and pushes it along with relevant comparison data to the smart device terminal of the supervisor for display.

[0076] Combining Embodiment 1 and Embodiment 2, the decision-linked processing module ensures the reliability of the analyzed and evaluated signals through a verification mechanism. When there are no errors, the parameters of the detection equipment are automatically adjusted. When there are differences, abnormal steps are inferred in reverse or interference sources are identified, forming a closed loop of acquisition-analysis-evaluation-control. This not only unifies the detection standards but also enhances the ability to cope with interference factors, ultimately ensuring the stability of the garment pattern quality, reducing production costs, and improving the product's market competitiveness.

[0077] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the described embodiments or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, they should all fall within the protection scope of the present invention.

[0078] In the description of this specification, the references to terms such as "an embodiment," "example," and "specific example" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Related accessories include commonly used mechanical connection components in this field such as couplings, lead screws, gears, and gaskets, but are not limited to these. Specific replacements and adaptations are made according to actual use.

[0079] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for automatic detection of a garment pattern based on visual inspection, characterized in that, Includes the following steps; S1. Obtain the continuous detection range of the visual detection pair passing through the clothing pattern, match the capture frequency to the clothing pattern transmission rate, synchronously mark the collected pattern data according to the timeline, and generate the original dataset with spatiotemporal coordinates. S2. After grayscale processing and noise reduction of the original dataset, construct an active cache file library. Classify the processed original data according to the pre-stored quality rules: data that meets the standards is stored in the qualified data sub-library, and new pattern data that can be referenced in the historical database is added. Data that is suspected of being defective is stored in the data to be inspected sub-library. The data to be inspected sub-library stores pattern data that is suspected of having defects, and is split into low-risk suspected defect set and high-risk suspected defect set according to the defect risk level. S3. Retrieve the low-risk and high-risk suspected defect sets from the data sub-database to be inspected, and combine them with the historical database and the current cache file library for joint analysis to generate quality analysis signals and high-risk assessment signals. The analysis process for the low-risk suspected defect set is as follows: The difference between the standard pattern template and the newly stored pattern data is calibrated as the pattern difference tolerance coefficient. When the low-risk suspected defect set is close to the standard pattern template and the difference is greater than or equal to the pattern difference tolerance coefficient, a quality analysis signal is generated. The analysis process for the high-risk suspected defect set is as follows: The high-risk suspected defect set is obtained from the sub-database to be inspected. At the same time, the historical records parameters of the same type of clothing patterns in the historical database are extracted. The high-risk difference between the high-risk suspected defect set and the corresponding standard pattern sample in the historical records is calculated and marked as the historical difference. The collection period of the high-risk suspected defect set is divided into N time nodes according to the timeline. Each time node corresponds to a sampling point of the original dataset. The original parameters, historical records parameters and historical differences of the high-risk suspected defect set at that time node are recorded to construct the node dataset. A node dataset is constructed with time nodes as the X-axis and clothing pattern difference reference values ​​pre-stored in the historical database as the Y-axis to construct a Cartesian coordinate system. Points are plotted and connected to draw three sets of floating curves: the first set of floating curves represents the high-risk suspected defect set at each time node; the second set represents the historical floating curves of historical parameters; and the third set represents the historical floating curves of historical-current differences. The deviation of each curve from the Y-axis reference value is calculated. If a high-risk floating curve deviates from the reference value by more than 30% in a single instance, or if the historical floating curve shows an upward trend for three consecutive time nodes and the cumulative deviation exceeds 50%, then the high-risk suspected defect set is determined to have a significant defect risk, and a high-risk assessment signal is generated. S4. After obtaining the quality analysis signal and the high-risk assessment signal, combine the qualified data sub-library and the data sub-library to be inspected for joint verification. If there is no error, adjust the visual inspection equipment to correct the pattern defect. If there is a difference, reverse reasoning is used to mark the abnormal step or analyze the interference source, and generate control and warning instructions.

2. The automatic detection method of garment pattern based on visual detection according to claim 1, characterized in that, The process of obtaining and processing the original dataset in step S1 is as follows: The system continuously and dynamically detects and collects patterns of clothing passing on the conveyor belt within a continuous detection range. The acquisition frequency is matched with the clothing transmission speed to avoid data omission. A unique timestamp is added to each frame of pattern data collected according to the timeline. The timestamp information includes the acquisition time, device number, and clothing transmission position coordinates, generating a raw dataset with spatiotemporal coordinates.

3. The automatic detection method of garment pattern based on visual detection according to claim 1, characterized in that, The processing procedure for the original dataset in step S2 is as follows: The original dataset is obtained and processed in grayscale. Color pattern data in the original dataset is retrieved and converted into grayscale images to simplify the data dimensions. Gaussian filtering algorithm is used to denoise the grayscale images to eliminate noise points caused by ambient light interference and device jitter. The processed image data is used to build an active cache file library. The files are hashed and marked. The system retrieves pre-stored pattern defect judgment criteria from the existing cache file library and divides the data into qualified data sub-libraries and data to be inspected sub-libraries: the qualified data sub-library stores pattern data that is defect-free or conforms to the standards, and is used to supplement newly stored pattern data that can be referenced in the historical database, serving as a benchmark for subsequent analysis.

4. The method for automatic detection of clothing patterns based on visual inspection according to claim 3, characterized in that, The analysis process for the low-risk suspected defect set in step S3 is as follows: The low-risk suspected defect set is retrieved from the data sub-database to be inspected. The pre-stored standard pattern templates and newly stored pattern data in the historical database are retrieved and compared with the low-risk suspected defect set. When the low-risk suspected defect set and the standard pattern template are close in comparison and the difference value is less than the error tolerance coefficient, a callback classification signal is generated.

5. The automatic detection method of garment pattern based on visual detection according to claim 4, characterized in that, The S3 step further includes: if the deviation of the high-risk floating curve and the historical floating curve does not reach the difference reference value of 30%, then it is determined that the high-risk suspected defect set has no significant risk and a no-significant-risk assessment signal is generated.

6. The automatic detection method of garment pattern based on visual detection according to claim 1, characterized in that, The processing procedure for the acquired grade analysis signal and high-risk assessment signal in step S4 is as follows: Obtain quality analysis signals and high-risk assessment signals, and generate defect sets for the two sets of signals. Retrieve standard pattern templates of similar clothing patterns from the historical database as verification benchmarks. Temporary sample comparison parameters are constructed from the low-risk suspected defect set and high-risk suspected defect set in the quality analysis signal and high-risk assessment signal. The temporary sample comparison parameters are compared with the standard pattern template. If the parameter matching degree between the temporary sample comparison parameters and the standard pattern template is ≥95% and there is no logical conflict, the verification is deemed correct, and control instructions are generated to adjust the operating parameters of the visual inspection equipment to correct any pattern defects that may exist in subsequent inspections. If the temporary sample parameters match the standard pattern template parameters by less than 95% or there is a logical conflict, the verification anomaly handling mechanism will be activated.

7. The method for automatic detection of clothing patterns based on visual inspection according to claim 1, characterized in that, S4 performs reverse reasoning on the processing procedure of step S3, and the specific process is as follows: The key steps in the processing of the original dataset are traced, including the timestamp matching of the original data collection, the threshold setting for grayscale processing, and the calculation logic of the historical difference. If reverse reasoning reveals that the parameter setting of a certain step deviates from the pre-stored threshold, the step is marked as an abnormal step, and an abnormal warning instruction containing the step number, abnormal parameters, and correction suggestions is generated. If no abnormal step is found, possible interference sources are analyzed. By comparing the environmental parameters in the collection log with the device operation data, the type of interference source is determined and an interference warning instruction is generated. The abnormal warning instruction and the interference warning instruction are synchronized to the terminal display.

8. A visual detection-based automatic garment pattern detection system, used in the visual detection-based automatic garment pattern detection method according to any one of claims 1-7, characterized in that, Includes the following steps: The visual data acquisition module communicates with the visual inspection equipment and is used to continuously acquire raw datasets of clothing patterns and mark them according to a timeline. The shallow data analysis module communicates with the visual data acquisition module and the pattern cache core platform to process the raw dataset and generate grade analysis signals. The deep risk assessment module communicates with the pattern cache core platform to jointly analyze data and generate high-risk assessment signals. The decision-making linkage processing module is connected to the shallow data analysis module, the deep risk assessment module, and the pattern cache core platform, respectively. It is used to process the grade analysis signal and the high-risk assessment signal and generate control commands, abnormal warning commands, and interference warning commands. The pattern caching core platform communicates with each of the above modules to store data, thresholds, and models, providing data support and triggering operation with signals and instructions.

Images