Flexible textile quality inspection system and device based on deep learning
By combining multimodal sensors and deep learning inference modules, along with human-machine collaborative verification and feedback correction from the learning module, the problem of balancing detection speed and accuracy in textile quality inspection has been solved. This enables efficient and accurate detection of flexible textiles, adapting to diverse defect types and complex texture backgrounds, and meeting the rapid detection needs of industrial sites.
Patent Information
- Application Number
- CN202511222541.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-29
- Publication Date
- 2025-12-05
AI Technical Summary
Existing deep learning-based textile quality inspection technologies struggle to balance detection speed and recognition accuracy, and are not well-suited to the diverse defect types and complex textures of flexible textiles, resulting in high rates of missed and false detections, making it difficult to meet the rapid inspection needs of industrial sites.
By employing multimodal sensors and a deep learning inference module, combined with the human-machine collaborative sampling inspection of the verification module and the feedback correction mechanism of the learning module, the system can accurately identify and classify multiple types and forms of textile defects. The detection module integrates an image collection unit, a sensing data unit, and a timing control unit to achieve synchronous acquisition and unified format conversion of multi-source data. The inference module performs feature extraction and result generation, the verification module performs manual sampling verification and feedback correction, and the learning module continuously optimizes the model.
It improves the detection accuracy and efficiency of textile quality inspection, reduces the rate of missed and false detections, meets the real-time detection needs of industrial production lines, achieves a high recognition rate for subtle and complex defects, and enhances the stability and adaptability of the system.
Smart Images

Figure CN121074577A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of automatic quality inspection of textiles, in particular to a flexible textile quality inspection system and device based on deep learning. BACKGROUND
[0002] Flexible textiles are important materials for clothing, home textiles and industrial fabrics. The quality of flexible textiles directly affects the use performance and market competitiveness of products. Traditional flexible textile quality inspection work relies on manual visual inspection, which has low efficiency, high fatigue misjudgment rate and strong subjectivity, and cannot meet the needs of modern textile industry for efficient and accurate quality control. With the development of computer vision and deep learning technology, automatic textile quality inspection systems based on image recognition have gradually emerged. Such systems collect fabric images and use deep neural networks to identify and classify fabric defects, greatly improving the automation and accuracy of detection. At the same time, the application of flexible detection devices enables the system to adapt to the diversified detection needs of different types of textiles.
[0003] Although existing deep learning-based textile quality inspection technology has made significant progress in defect recognition accuracy, it generally has the problem of balancing detection speed and recognition accuracy. On the one hand, in order to improve recognition accuracy, the model structure tends to be complex and the calculation amount is large, resulting in a long detection process and unsuitable for fast real-time detection on the production line. On the other hand, to meet the needs of fast detection, the system often uses lightweight models or reduces image resolution, resulting in a decrease in recognition accuracy and difficulty in accurately capturing subtle or complex defects. In addition, existing systems lack adaptability to the diverse defect types and complex texture backgrounds of flexible textiles, which can easily lead to missed detection and false detection, affecting the stability and reliability of quality control. SUMMARY
[0004] In order to solve the problem of balancing automatic detection speed and recognition accuracy, and to meet the needs of fast detection of flexible textiles in industrial fields, the present application provides a flexible textile quality inspection system and device based on deep learning.
[0005] On the one hand, the flexible textile quality inspection system based on deep learning provided by the present application adopts the following technical solution: A flexible textile quality inspection system based on deep learning, comprising: a detection module that collects data through a connected sensor and a scanning camera device, and uses the data for machine training and machine quality inspection; an inference module that communicates with the detection module and uses the data collected by the detection module to make inference and determine the defect type; a verification module that manually samples and reviews the inference results of the inference module; The learning module feeds back the manual sampling review result of the verification module to the reasoning module for optimized training and learning.
[0006] The system can efficiently and automatically collect multi-modal data on the surface and inside of the fabric by connecting various sensors and scanning camera devices through the detection module, intelligently analyze the data by combining the deep learning reasoning module, accurately identify and classify multiple types and forms of textile defects, greatly improve the detection accuracy, reduce manual misjudgment and missed detection, the verification module reviews the reasoning result by manual sampling, effectively monitors the system operation state and identification accuracy, the learning module uses the verification feedback to realize continuous iteration optimization and autonomous learning of the model, enhances the system's ability to identify new or complex defects, ensures long-term stable and efficient operation of the system, the system uses a hierarchical reasoning and feedback learning mechanism to effectively balance detection speed and identification accuracy, meets the real-time detection needs of industrial production lines, can realize high-speed mass detection while maintaining high identification rate for subtle and complex defects, and improves the overall efficiency of textile quality control.
[0007] Further, the detection module includes an image collection unit, a sensing data unit, a timing control unit and a data conversion unit, the image collection unit is installed on the textile equipment and collects target images, and all pixel data is exported after image pixel segmentation, the image collection unit includes multiple groups and collects image information in order according to the arrangement order, and superimposes multiple groups of image information to eliminate blurred image information, the image collection unit is built-in camera driver, image acquisition control software and automatic exposure and gain adjustment algorithm, the sensing data unit is installed and connected to the textile equipment, the sensing data unit is used to collect sensing data of the environment and the object, the sensing data includes thickness parameter, surface quality, deformation data, tensile strength and static constant, the sensing data unit is built-in sensor driver, sensing data filtering and preprocessing program and data synchronization timing control program, the timing control unit is communicatively connected to the image collection unit and the sensing data unit, the timing control unit is used to coordinate the collection time of image and sensing signal, the timing control unit is built-in timestamp recording system, device synchronization control protocol and trigger signal management program, the data conversion unit is communicatively connected to the image collection unit, the sensing data unit and the timing control unit, the data conversion unit converts multi-source raw data into a unified format, including data structuring, unit conversion and encoding conversion, the data conversion unit is built-in data analysis program, format conversion tool and standardization and normalization algorithm library.
[0008] The detection module integrates an image collection unit and a sensing data unit to realize multi-modal synchronous acquisition of the target area. The image collection unit uses multiple sets of arranged cameras, pixel segmentation, and blur elimination technology to obtain high-resolution and clear fabric images, effectively improving the detail and accuracy of defect detection. The sensing data unit synchronously collects multi-dimensional physical parameters such as thickness, surface quality, deformation, tensile strength, and static constant, providing rich feature information for subsequent deep learning models, realizing comprehensive recognition of complex defects. The image collection unit has built-in exposure and gain automatic adjustment algorithms that can automatically optimize imaging parameters according to different environmental light conditions, ensuring stable image quality in variable production environments, reducing false positives and false negatives caused by environmental changes. The sensing data unit has built-in data filtering and preprocessing programs that effectively suppress noise and abnormal data, improving the reliability of collected data. The timing control unit coordinates the acquisition timing of images and sensing data, records and synchronizes control protocols through timestamps, ensuring that all data is collected and recorded under a unified time reference, providing high temporal consistency for multi-source data fusion, multi-layer superposition, and subsequent deep learning processing, avoiding detection errors caused by asynchronous data. The data conversion unit structures, unit converts, and encodes the multi-source raw data, automatically completes format standardization and normalization processing, and the built-in special algorithm library can improve data compatibility and processing efficiency, providing high-quality, unified format data input for backend inference, analysis, and model training, reducing data preprocessing manpower and time cost.
[0009] Further, the inference module includes a feature extraction unit, a fuzzy processing unit, and a result generation unit. The feature extraction unit is used to extract key features from raw data. The feature extraction unit has built-in feature extraction algorithms, data dimensionality reduction tools, and numerical feature encoders. The fuzzy processing unit is sequentially connected to the feature extraction unit. The fuzzy processing unit is used to process uncertain or fuzzy information. The fuzzy processing unit has built-in fuzzy logic algorithm library, Bayesian inference framework, and data confidence calculation program. The result generation unit is sequentially connected to the feature extraction unit and the fuzzy processing unit, respectively. The result generation unit converts the results obtained by inference into system output, including classification labels, detection positions, and predicted values. The result generation unit has built-in result encoding and visualization software.
[0010] The feature extraction unit is built-in advanced feature extraction algorithm and data dimension reduction tool, which can automatically mine the core features related to textile defects from massive original multi-modal data, reduce redundant information, improve data processing efficiency, the numerical feature encoder normalizes the extracted features for subsequent model processing and cross-system data compatibility, the fuzzy processing unit uses fuzzy logic algorithm and Bayesian inference framework to scientifically process fuzzy or uncertain data in the detection process, calculate data confidence, reduce the risk of misjudgment caused by sensing error, environmental change, etc., and improve the credibility and stability of the detection results, the reasoning process is designed in the order of feature extraction, fuzzy processing and result generation, which effectively ensures the progressive data flow and orderly information transmission, avoids information loss and misuse, and improves the accuracy and reliability of the overall detection reasoning, the result generation unit not only converts the reasoning result into specific classification label, defect detection position and prediction value, but also is built-in result encoding and visualization software to support graphical display, which is convenient for operators to quickly understand the detection results and improve the response speed and decision-making efficiency of the production line.
[0011] Further, the verification module includes a random sampling unit, a task allocation unit, a manual verification unit, and a feedback correction unit. The random sampling unit is in communication with the reasoning module and selects sampling data from the reasoning results of the reasoning module. The random sampling unit uses a sampling algorithm for sampling and is connected to the reasoning module through a data retrieval interface. The task allocation unit is connected to and controls the output of the random sampling unit. The task allocation unit is used to allocate verification tasks to an automatic verification system or a human inspector. The task allocation unit is built-in a workflow management system and a task scheduling and allocation algorithm. The manual verification unit is connected to the task allocation unit and receives data sent by the task allocation unit. The manual verification unit is used to process data that cannot be automatically judged by the machine and to review and compare data automatically identified by the machine. The manual verification unit includes a verification work interface and a data labeling tool. The feedback correction unit is connected to the manual verification unit and the reasoning module and sends the review results of the manual verification unit to the reasoning module for verification and correction training. The feedback correction unit includes a model update module and a data backwriting and synchronization program.
[0012] The random sampling unit selects sampling data from the results of the reasoning module through a built-in sampling algorithm, ensures that the verification samples cover diversity and representativeness, efficiently calls reasoning data through a data retrieval interface, reduces data redundancy, improves verification efficiency, reduces the overall system operation load, and the task allocation unit is built-in with a workflow management system and a task scheduling algorithm, which can flexibly allocate verification tasks to the automatic verification system or manual inspectors, realize man-machine cooperation, optimize the verification process, ensure automatic processing of automatically determinable tasks, and preferentially hand over complex or abnormal tasks to manual verification, improve the overall verification accuracy and efficiency, the manual verification unit is equipped with a friendly verification work interface and an efficient data labeling tool, so that the manual inspector can conveniently review the machine reasoning results and handle complex samples that the automatic system cannot determine, effectively compensate for the shortcomings of the automatic system, and improve the overall recognition accuracy and reliability of the system, the feedback correction unit timely returns the review results of manual verification to the reasoning module, and through the model updating module and data backwriting program, the reasoning model is continuously verified and trained based on the true verification results, promotes the self-evolution and performance continuous improvement of the model, enhances the adaptability and robustness of the system, and through the efficient cooperation of each unit through communication connection and interface, ensures smooth transmission and accurate feedback of verification data, forms a closed-loop verification mechanism, improves the stability and reliability of the entire detection system, reduces the risk caused by data island and information lag, verifies and feeds back the reasoning results through the verification module, ensures the accuracy of the detection results, provides a solid data foundation for production quality control and decision-making, and helps the intelligent management and continuous improvement of the textile production process.
[0013] Further, the learning module includes a feedback extraction unit, a comparative analysis unit, a data updating unit, and a simulation verification unit. The feedback extraction unit is communicatively connected to the reasoning module and the verification module and extracts data fed back from the verification module to the reasoning module. The comparative analysis unit is connected to the feedback extraction unit and receives the feedback data extracted by the feedback extraction unit. The comparative analysis unit is configured to compare the reasoning result data and the verification data, and analyze the difference between the two sets of data through a performance evaluation tool and an error analysis program. The data updating unit is connected to the comparative analysis unit and the reasoning module. The data updating unit is configured to add new labeled data or improved features to the training database. The simulation verification unit is connected to the data updating unit. The simulation verification unit verifies the improvement effect of the updated model data through simulation running.
[0014] The learning module communicates with the reasoning module and the verification module through the feedback extraction unit, can efficiently extract verification feedback data, form a data closed loop, the system collects and utilizes the verification results in real time, ensures that the model training is always based on the latest and most real data, continuously corrects and optimizes the recognition ability of the reasoning module, the comparative analysis unit utilizes the performance evaluation tool and the error analysis program, systematically compares the reasoning results with the verification data, accurately locates the deficiencies of the model in the actual application, scientifically guides the optimization direction of the model through in-depth analysis of the data difference, effectively improves the accuracy and stability of the model, the data updating unit can add the new labeled data or optimized features obtained by manual verification to the training database in time, maintains the freshness and diversity of the model training data, continuously supplements the data and improves the features, so that the model can adapt to the changes of the business scene, improve the generalization ability, reduce the recognition errors caused by outdated data, the simulation verification unit verifies the performance improvement effect of the model after data updating through simulation operation, detects the performance of the model before formal deployment, reduces the risk caused by model updating, ensures that each optimization can bring actual performance improvement to the system, enhances the safety and reliability of the system, through the automatic feedback extraction, error analysis, data updating and simulation verification of the learning module, the system has self-learning and evolution ability, the model can be continuously self-adapted and optimized according to the actual business feedback, cope with complex and variable textile detection scenes, continuously improve the overall intelligent level, the learning module provides data support for quality control and production management, through the analysis of historical errors and performance improvement, assists the managers to optimize the process and adjust the strategy, promotes the fine and intelligent management of the textile production process.
[0015] In another aspect, the application provides a flexible textile quality inspection device based on deep learning, which adopts the following technical scheme: A flexible textile quality inspection device based on deep learning, comprising: A detection assembly installed on a textile equipment and connected to a control system, the detection assembly is used to collect raw data of fabric morphology and its environment state; A data collection assembly communicatively connected to the detection assembly, the data collection assembly is used for data storage, management, transmission and preliminary processing; A display interaction assembly communicatively connected to the data collection assembly and interacting with the data collection assembly to send data, the display interaction assembly is used for information display and man-machine interaction control.
[0016] The detection component is closely mounted on the textile equipment, and real-time original data of the fabric form and environmental state is collected to provide high-quality and rich input information for the deep learning model. The system can accurately identify fabric defects and abnormalities by using deep learning algorithms, significantly improving the accuracy and reliability of the inspection, reducing manual missed and false detections. The data collection component is responsible for data storage, management, transmission and preliminary processing tasks, effectively organizing and optimizing the original data flow to ensure the real-time and stability of the inspection process. The display and interaction component communicates with the data collection component to realize intuitive information display and effective human-computer interaction control.
[0017] Further, the detection component includes a camera, a laser scanning head, an electromagnetic scanning head, and a light transmission scanning head. The camera, laser scanning head, electromagnetic scanning head, and light transmission scanning head are respectively mounted on the textile equipment and aimed at the textile in transmission for shooting and scanning. The camera includes an industrial camera, a lens, an optical filter, and an image acquisition card. The laser scanning head includes a laser emitter, a photodetector, and a scanning controller. The electromagnetic scanning head includes an electromagnetic generator, signal amplification, a collection circuit, and an electromagnetic shielding shell. The light transmission scanning head includes a high-brightness light source and a photoelectric sensor.
[0018] The detection component integrates four different types of sensors, including industrial cameras, laser scanning, electromagnetic scanning, and light transmission scanning. It can obtain the form, structure, and internal state information of the textile from multiple dimensions and angles. The multi-sensor data is complementary, which significantly improves the comprehensiveness and accuracy of the detection and reduces the blind spots and misjudgments that may exist in a single sensor. The industrial camera is equipped with a high-performance lens, an optical filter, and an image acquisition card, which can obtain clear and high-resolution textile surface images, adapt to complex texture and color recognition requirements. The laser scanning head measures the small morphological changes on the fabric surface through laser emission and photodetection, enhancing the spatial resolution of defect recognition. The electromagnetic scanning head uses an electromagnetic generator, signal amplification, and a collection circuit, combined with an electromagnetic shielding shell, to effectively capture the material distribution and structural features inside the fabric, especially for detecting special materials such as metal fibers or conductive fibers, enhancing the ability to acquire deep information during detection. The light transmission scanning head uses a high-brightness light source and a photoelectric sensor to detect the light transmission and fiber density of the fabric, assisting in identifying the uniformity and defects of the fabric. Each sensor component is equipped with a dedicated controller and a signal conditioning module to effectively suppress the influence of environmental light, electromagnetic interference, and other factors, ensuring the stability and accuracy of data acquisition, adapting to complex production environments, and providing a solid foundation for subsequent deep learning model training and reasoning with diverse, rich, and high-quality original data. This improves the model's ability to recognize different defect types and complex fabric patterns, achieving accurate and intelligent quality detection.
[0019] Further, the plurality of cameras, laser scanning heads, electromagnetic scanning heads and light transmission scanning heads are arranged at intervals respectively, and the scanning time interval of adjacent devices is controlled according to the conveying speed of the textile.
[0020] The scanning time interval of adjacent sensors is dynamically controlled according to the actual conveying speed of the textile, so that each device can collect at the appropriate time point, ensuring the time sequence and synchronization of the data, facilitating the computer to automatically superimpose and coincide multiple recognition patterns, avoiding recognition distortion caused by fast conveying, and improving the effectiveness and accuracy of the detection data.
[0021] Further, the data collection assembly includes a memory, a communication device and a data processor, the memory is connected to the detection assembly and the interaction assembly through the communication device, the data processor is connected to the memory through communication and extracts, processes and rewrites the data stored in the memory, the memory includes temporary storage, fixed storage and power failure protector, the communication device includes a wired communication module and a wireless communication module, and the data processor includes an industrial computer, a graphics processing unit and an AI training module processor.
[0022] The memory is subdivided into temporary storage, fixed storage and power failure protector, meeting the storage needs of different types of data, the temporary storage supports cache and real-time data processing, the fixed storage ensures long-term data preservation, and the power failure protector prevents data loss caused by accidental power failure, improving system data security and stability, this layered storage structure optimizes data read-write efficiency, improves system response speed and reliability, the communication device integrates wired and wireless communication modules, supporting data transmission requirements in various network environments, the data processor integrates industrial computers, graphics processing units and AI training module processors, with rich computing resources and parallel processing capabilities, the industrial computer is responsible for overall control and task scheduling, the graphics processing unit accelerates image and laser scanning data processing, and the AI training module realizes machine learning model training and reasoning, improving the intelligent level of detection and analysis, the data processor can efficiently extract the required data from the memory, perform preprocessing, feature extraction, data fusion and AI reasoning, complete rapid feedback and decision support, meet the real-time requirements of online textile quality detection, and the processed data is rewritten into the memory to form a closed-loop management, facilitating subsequent tracing, analysis and optimization, by integrating AI training modules and various communication methods, the system can be seamlessly connected with the cloud, big data platforms and remote monitoring systems, realizing data sharing, remote diagnosis and intelligent optimization, and promoting the upgrading of textile production to intelligent manufacturing.
[0023] Further, the display interaction assembly comprises a display connected to the data collection assembly and displaying data processing results, and a control input connected to the data collection assembly and the display and performing human-computer interaction control operation, wherein the display comprises an interactive display and a monitoring display, and the control input comprises a touch screen, a keyboard and a mouse.
[0024] The interactive display can realize dynamic display of operation interface, parameter setting, alarm information and the like, so as to facilitate an operator to intuitively master system running state, the monitoring display is dedicated to real-time display of detection results, trend analysis and historical data, different information types are effectively distinguished, monitoring efficiency and data readability are improved, a plurality of control inputs such as the touch screen, the keyboard and the mouse meet different operation habits and environmental requirements, a user can flexibly select according to actual scenes, and operation comfort and efficiency are improved, the touch screen supports quick operation and graphical interface interaction, is suitable for quick response and on-site adjustment, and the keyboard and the mouse are suitable for fine parameter setting and complex data input, and operation precision and flexibility are improved.
[0025] In summary, the present application has the following beneficial technical effects: 1. The multi-modal sensing and deep learning inference module are used to realize efficient automatic identification of multiple types of defects on the surface and inside of the fabric, improve detection accuracy, and significantly reduce the false negative and false positive rates.
[0026] 2. The detection module and the data collection assembly cooperate to uniformly convert and fuse analyze image, physical parameter and environmental data, improve the comprehensiveness of defect identification and the robustness of the system.
[0027] 3. The human-computer cooperative sampling of the verification module is combined with the feedback correction and continuous self-learning mechanism of the learning module to realize dynamic optimization of the model and ensure continuous improvement of the system detection capability.
[0028] 4. The time sequence control unit cooperates with the efficient data conversion unit to realize synchronous acquisition and structured processing of multi-sensor data, guarantee real-time performance and data consistency, and meet the high timeliness requirement of online quality inspection.
[0029] 5. The detection assembly, the data collection assembly and the display interaction assembly are modularly designed, which facilitates equipment expansion, function upgrade and rapid integration of new algorithms or hardware in the later period, and adapts to different textile production scene requirements.
[0030] 6. The display interaction assembly supports interactive display and monitoring split screen, integrates a plurality of input modes such as touch, keyboard and mouse, and reduces operation threshold in combination with a graphical interface to improve use efficiency and user experience.
[0031] 7、Memory has power-off protection and multi-layer storage mechanism, and communication device supports wired and wireless dual channels, which guarantees the integrity and security of data transmission and storage, and reduces system risk.
[0032] 8、Verification module automatically allocates sampling inspection tasks, realizes efficient cooperation of automatic verification and manual verification, timely corrects algorithm error, and further guarantees the accuracy of detection results.
[0033] 9、Data collection and processing module supports historical data accumulation, trend analysis and result visualization, and provides data support for process improvement, quality traceability and production decision.
[0034] 10、Multi-sensing assembly can be combined as needed, scanning interval is automatically adjusted, and different textile types, widths and conveying speeds are adapted, so that errors can be eliminated through superposition, and the universality and application range of the system are improved. BRIEF DESCRIPTION OF DRAWINGS
[0035] Figure 1 is a connection diagram of the quality inspection system of the present application; Figure 2 is a connection diagram of the quality inspection device of the present application.
[0036] REFERENCE SIGNS: 1、Detection module, 11, image collection unit, 12, sensing data unit, 13, time sequence control unit, 14, data conversion unit, 2, reasoning module, 21, feature extraction unit, 22, fuzzy processing unit, 23, result generation unit, 3, verification module, 31, random sampling unit, 32, task allocation unit, 33, manual verification unit, 34, feedback correction unit, 4, learning module, 41, feedback extraction unit, 42, comparison and analysis unit, 43, data update unit, 44, simulation verification unit; 5、Detection assembly, 51, camera, 52, laser scanning head, 53, electromagnetic scanning head, 54, light transmission scanning head, 6, data collection assembly, 61, memory, 62, communication device, 63, data processor, 7, display and interaction assembly, 71, display, 72, control input device. DETAILED DESCRIPTION
[0037] The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings. Figures 1-2 The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings.
[0038] In the description of the present application, it should be understood that the terms "longitudinal", "transverse", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.
[0039] Textile defects refer to undesirable phenomena affecting the appearance, structure or performance of textiles due to raw materials, processes, equipment or human factors during textile production, processing and flow. Common textile defects can be divided into appearance defects, structure defects, size and shape defects, and performance defects. Appearance defects include color difference, stains, spots, fuzz, pilling, uneven gloss, structure defects include broken warp and weft, skip stitches, missed stitches, defects, uneven density, uneven evenness, size and shape defects include inaccurate width, deformation, twisting, edge curling, performance defects include insufficient strength and easy breakage, shrinkage, easy deformation, poor air permeability.
[0040] Color difference refers to the inconsistency in color depth of different parts of the same batch or the same piece of textile, which is often caused by uneven dyeing, raw material problems or improper process control. Stains and spots are visible stains caused by oil stains, chemicals, dust or impurities adhering to the surface of the textile. Fuzz and pilling are small spherical objects formed on the surface after fiber breakage or wear, which are commonly seen in synthetic fiber fabrics. Uneven gloss is the abnormal gloss of some areas of the fabric, which is usually caused by uneven fiber arrangement, weaving or finishing.
[0041] Broken warp and weft refers to the breakage of warp or weft yarn, resulting in gaps or sparse areas on the fabric. Skip stitches and missed stitches are holes or discontinuous structures formed in knitted fabrics due to needle skipping or missing. Defects (knots, neps, yarn defects) are uneven thicknesses caused by fiber aggregation or broken ends in yarn, affecting the uniformity of the fabric. Uneven density and uneven evenness are uneven distribution of fabric density or yarn thickness, resulting in abnormal hand feeling and appearance.
[0042] Inaccurate width refers to the discrepancy between the actual width of the fabric and the specified size, which is often caused by improper tension control or shrinkage. Deformation, twisting and edge curling are caused by uneven weaving or post-processing, resulting in bending, wavy or edge curling of the fabric.
[0043] Insufficient strength and easy breakage are caused by substandard fiber quality or yarn twist, resulting in decreased fabric strength. Shrinkage and easy deformation are caused by substandard fiber structure or post-finishing process, affecting the dimensional stability of the fabric after washing. Poor air permeability is caused by improper textile structure or raw material selection, affecting the functionality of the fabric.
[0044] Example 1: The embodiment of the application discloses a flexible textile quality inspection system based on deep learning, referring to Figure 1 , comprising: a detection module 1, which collects data by connecting sensors and scanning cameras, and uses the data for machine training and machine quality inspection; an inference module 2, which is communicatively connected to the detection module 1 and uses the data collected by the detection module 1 to make inference judgments to determine the defect type; a verification module 3, which manually samples and reviews the inference judgment results of the inference module 2; a learning module 4, which feeds back the manual sampling review results of the verification module 3 to the inference module 2 for optimization training and learning.
[0045] The detection module 1 integrates a high-resolution scanning camera and a specific type of sensor to take real-time photos and scan the textile on the production line, collecting multi-dimensional data including color, texture, structure, etc. The inference module 2 has a defect recognition model trained by deep learning, which can automatically recognize defects in the data collected by the detection module, including but not limited to color difference, holes, pilling, stains, unevenness, etc., and output the specific defect type and location. The verification module 3 allows quality inspectors to periodically sample and check the automatic recognition results of the inference module to form a "manual review report" for subsequent model optimization. The learning module 4 uses the manual review results output by the verification module as feedback to automatically correct and optimize the recognition model of the inference module, achieving continuous self-learning and self-optimization.
[0046] The detection module 1 is arranged at the key nodes of the textile production line to ensure that each batch / meter of fabric can be detected. High-resolution images, color information, reflectivity, and other physical parameters are collected, and the target defect types include color difference, stains and spots, broken warp and weft, pilling and missed stitches, fuzzing and unevenness.
[0047] The inference module 2 is trained based on the collected defect sample images using convolutional neural networks and target detection networks, which can distinguish different types of defects. The inference process includes: Input real-time images or data collected by the detection module; the model automatically analyzes image features and detects defects such as streaks, color blocks, and holes; and outputs information such as defect type, location, and area.
[0048] The verification module 3 extracts a certain proportion of samples from the automatic recognition results according to production batches or timing, and manually compares them with the model results. The review content includes whether the defect type is correct, and whether the defect location and range are accurate. Verification records are used to calculate the accuracy, missed detection rate, and false detection rate of automatic recognition.
[0049] The learning module 4 re-labels the recognition errors found in manual verification as difficult examples into the library, periodically trains the deep learning model incrementally, realizes continuous evolution of the model, and improves the defect recognition capability of different batches and different varieties of textile fabrics.
[0050] After the system is started, the camera and sensor state are automatically detected to ensure that the data acquisition equipment is normal. When the textile passes through the detection module 1, the image and sensing data are collected in real time and automatically uploaded to the inference module 2. The inference module 2 identifies the defects of the collected data, and the identification result is automatically recorded and prompted to the operator. The quality inspector samples and reviews part of the identification result, confirms the defect type and position, feeds back the manual verification result to the system, automatically iterates and optimizes the model, and the system can automatically generate a detection report including defect type, quantity, position distribution, automatic identification accuracy, etc. for quality traceability and process improvement.
[0051] Embodiment 2: Based on embodiment 1, the following is added: Referring to Figure 1 , the detection module 1 includes an image collection unit 11, a sensing data unit 12, a time sequence control unit 13, and a data conversion unit 14. The image collection unit 11 is installed on a textile equipment and collects target images, and performs pixel segmentation on the images and exports all pixel point data. The image collection unit 11 includes multiple groups and sequentially collects image information according to the arrangement order, and superimposes the multiple groups of image information to eliminate blurred image information. The image collection unit 11 is built-in with a camera driver program, an image acquisition control software, and an exposure and gain automatic adjustment algorithm. The sensing data unit 12 is installed and connected to the textile equipment. The sensing data unit 12 is used to collect sensing data of the environment and objects. The sensing data includes thickness parameters, surface quality, deformation data, tensile strength, and static constant. The sensing data unit 12 is built-in with a sensor driver, a sensing data filtering and preprocessing program, and a data synchronization time sequence control program. The time sequence control unit 13 is communicatively connected to the image collection unit 11 and the sensing data unit 12. The time sequence control unit 13 is used to coordinate the collection time of the image and the sensing signal. The time sequence control unit 13 is built-in with a time stamp recording system, an equipment synchronization control protocol, and a trigger signal management program. The data conversion unit 14 is communicatively connected to the image collection unit 11, the sensing data unit 12, and the time sequence control unit 13. The data conversion unit 14 converts the multi-source raw data into a unified format, including data structuring, unit conversion, and encoding conversion. The data conversion unit 14 is built-in with a data analysis program, a format conversion tool, and a standardization and normalization algorithm library.
[0052] Referring to Figure 1, the reasoning module 2 comprises a feature extraction unit 21, a fuzzy processing unit 22 and a result generation unit 23, the feature extraction unit 21 is used for extracting key features from original data, the feature extraction unit 21 is built-in feature extraction algorithm, data dimension reduction tool and numerical feature encoder, the fuzzy processing unit 22 is sequentially connected with the feature extraction unit 21, the fuzzy processing unit 22 is used for processing uncertain or fuzzy information, the fuzzy processing unit 22 is built-in fuzzy logic algorithm library, Bayesian inference framework and data confidence calculation program, the result generation unit 23 is sequentially connected with the feature extraction unit 21 and the fuzzy processing unit 22 respectively, the result generation unit 23 converts the result of reasoning into system output, including classification label, detection position and prediction value, the result generation unit 23 is built-in result encoding and visualization software.
[0053] With reference to Figure 1 , the verification module 3 comprises a random inspection unit 31, a task allocation unit 32, a manual verification unit 33 and a feedback correction unit 34, the random inspection unit 31 is communicatively connected with the reasoning module 2 and selects inspection data from the reasoning result of the reasoning module 2, the random inspection unit 31 uses sampling algorithm for inspection and is connected with the reasoning module 2 through a data retrieval interface, the task allocation unit 32 is connected with and controls the output of the random inspection unit 31, the task allocation unit 32 is used for allocating verification tasks to automatic verification system or manual inspectors, the task allocation unit 32 is built-in workflow management system and task scheduling and allocation algorithm, the manual verification unit 33 is connected with the task allocation unit 32 and receives data sent by the task allocation unit 32, the manual verification unit 33 is used for processing data that cannot be automatically judged by the machine and rechecking and comparing data automatically identified by the machine, the manual verification unit 33 comprises a verification work interface and a data labeling tool, the feedback correction unit 34 is connected with the manual verification unit 33 and the reasoning module 2 and sends the rechecking result of the manual verification unit 33 to the reasoning module 2 for verification and correction training, the feedback correction unit 34 comprises a model updating module and a data backwriting and synchronization program.
[0054] With reference to Figure 1The learning module 4 includes a feedback extraction unit 41, a comparative analysis unit 42, a data updating unit 43, and a simulation verification unit 44. The feedback extraction unit 41 is communicatively connected to the reasoning module 2 and the verification module 3 and extracts data fed back by the verification module 3 to the reasoning module 2. The comparative analysis unit 42 is connected to the feedback extraction unit 41 and receives the feedback data extracted by the feedback extraction unit 41. The comparative analysis unit 42 is used to compare the reasoning result data with the verification data, and analyze the difference between the two sets of data through a performance evaluation tool and an error analysis program. The data updating unit 43 is connected to the comparative analysis unit 42 and the reasoning module 2, respectively. The data updating unit 43 is used to add new labeled data or improved features to the training database. The simulation verification unit 44 is connected to the data updating unit 43. The simulation verification unit 44 verifies the improvement effect of the model data updating through simulation running.
[0055] The image collection unit 11 adopts multiple groups of linear / surface array industrial cameras combined with high-brightness LED array illumination, automatically adjusts exposure and gain, and collects high-resolution images of the fabric in real time through a camera driver program and image acquisition control software. When detecting defects such as color difference, stains, broken warp and weft, multiple cameras are used to collect images from different angles and under different lighting conditions, and the images are pixel-level segmented to extract all pixel point data. Multiple images are superimposed to remove motion blur and improve defect detection accuracy.
[0056] The sensing data unit 12 integrates multiple types of sensors such as laser thickness gauges, surface roughness sensors, tension sensors, and electrostatic sensors, and has built-in sensing data filtering and preprocessing software to ensure data accuracy and consistency. It is used to detect structural or performance defects such as thickness abnormalities, surface deformation, and tensile strength abnormalities.
[0057] The timing control unit 13 uses high-precision clock chips and synchronous trigger circuits, and has built-in timestamp management and synchronization protocols to ensure that image acquisition and physical parameter acquisition times are consistent, enabling multi-source data collaboration.
[0058] The data conversion unit 14 uses data parsing libraries and format conversion tools to convert multi-source data such as images, thickness, and tensile strength into a unified structured format, and performs standardization and normalization processing to facilitate subsequent model analysis.
[0059] The feature extraction unit 21 uses pre-trained convolutional neural networks (such as ResNet, EfficientNet, etc.) for image feature extraction, combined with dimensionality reduction algorithms such as principal component analysis (PCA) to extract key texture, color, and structure features.
[0060] The fuzzy processing unit 22 integrates fuzzy logic reasoning (Fuzzy Logic), Bayesian inference (Bayesian Inference), etc., to improve the recognition ability of the fuzzy or uncertain area, and to deal with color difference, slight stains, edge skipping, and other defects that are difficult to define manually.
[0061] The result generation unit 23 uses a defect classifier to output results including defect categories (such as color difference, stains, broken warp and weft, etc.), defect positions (coordinates / length), confidence scores, and real-time display through a visual interface.
[0062] The random sampling unit 31 calls a random sampling algorithm, such as stratified sampling, to select a certain proportion of defect samples from the inference module results to ensure that the sampling is representative.
[0063] The task allocation unit 32 automatically allocates sampling tasks to designated quality inspectors or automatic verification systems based on a workflow engine, supporting task priority and workload balancing.
[0064] The manual verification unit 33 allows quality inspectors to view system detection results on a special verification interface and manually confirm or correct difficult defects through labeling tools.
[0065] The feedback correction unit 34 automatically compares the manual verification results with the model results, marks the misjudged samples, and automatically returns them to the model training library to drive model optimization.
[0066] The feedback extraction unit 41 automatically collects verification module feedback data, including misidentified samples and correct samples.
[0067] The comparative analysis unit 42 uses error analysis algorithms to quantitatively evaluate the differences between inference results and manual verification data, generating false positive and false negative statistics.
[0068] The data update unit 43 automatically adds new labeled data and feature vectors to the training data set, supporting incremental learning of the model.
[0069] The simulation verification unit 44 performs simulation detection on the updated model to evaluate its recognition ability for new defect samples, ensuring that the model is effectively improved before being put into production.
[0070] For example, the color difference defect detection step: The image collection unit 11 collects high-resolution images at a 10cm interval above the fabric surface and automatically adjusts the exposure, and the sensing data unit 12 synchronously records the light intensity, thereby excluding the influence of environmental light changes. The feature extraction unit 21 of the inference module 2 extracts the color feature channel of the image, quantizes the color difference, and the fuzzy processing unit 22 performs fuzzy inference on the color difference critical region (such as the gradual change zone) to determine whether it constitutes a defect. The result generation unit 23 outputs the position and area of the color difference region. The random sampling unit 31 of the verification module 3 extracts part of the color difference alarm samples, and the task allocation unit 32 allocates them to the artificial verification unit. The artificial verification unit 33 performs artificial observation and color card comparison on the color difference detected by the system to confirm whether it is a defect. The feedback correction unit 34 returns the difference between the artificial judgment and the model result to the inference module. The feedback extraction unit 41 of the learning module 4 collects the color difference difficult cases determined by artificial judgment, the comparison and analysis unit 42 analyzes the consistency of the model and artificial judgment, the data updating unit 43 integrates the newly labeled data into the training data, and the simulation verification unit 44 evaluates the accuracy of the new model in color difference detection and confirms whether it is improved and put into use.
[0071] For example, the broken warp and weft defect detection step: The image collection unit 11 of the detection module 1 collects fabric structure images at multiple angles, the sensing data unit 12 records tension changes and surface deformation parameters, the feature extraction unit 21 of the inference module 2 identifies warp and weft arrangement and gap changes, the fuzzy processing unit 22 performs confidence calculation on suspected broken warp and weft regions, the result generation unit 23 outputs the defect category (broken warp / broken weft), position and length, the random sampling unit 31 of the verification module 3 checks part of the broken warp and weft alarm points, the task is allocated to the artificial verification unit 33, the artificial verification unit 33 confirms the actual broken yarn condition through a magnifying glass and manual stretching comparison, the feedback correction unit 34 feeds back the results to the inference module, optimizes feature detection, and the learning module 4 records the differences between the model and artificial detection through feedback extraction and comparison and analysis units. The data updating unit 43 supplements new broken warp and weft samples, and the simulation verification unit 44 checks the notification accuracy of the new model.
[0072] Embodiment 3: The embodiment of the application discloses a flexible textile quality inspection device based on deep learning, referring to Figure 2 , comprising: A detection component 5 is installed on a textile equipment and connected to a control system, and the detection component 5 is used to collect raw data of fabric morphology and its environmental state; A data collection component 6 is communicatively connected to the detection component 5, and the data collection component 6 is used for data storage, management, transmission and preliminary processing; A display interaction component 7 is communicatively connected to the data collection component 6 and interacts with the data collection component 6 to send data, and the display interaction component 7 is used for information display and man-machine interaction control.
[0073] The detection assembly 5 is installed at a key position of the textile equipment, such as at the outlet of the loom or beside the fabric conveying channel, to ensure continuous and stable acquisition of the shape image of the fabric and the environmental state data. The detection assembly 5 is connected with the control system through a wired (such as industrial Ethernet) or wireless communication interface to realize information synchronization and command reception.
[0074] The data collection assembly 6 is configured with a large-capacity storage unit (SSD or industrial-grade storage card) and a built-in high-performance computing platform (such as an embedded GPU or FPGA), which communicates with the detection assembly 5 through a high-speed bus interface and supports multiple industrial protocols (such as Modbus, OPC UA) to ensure system compatibility.
[0075] The display and interaction assembly 7 is equipped with an industrial-grade touch screen or display device, supports high-resolution image display and multi-point touch operation, and is connected with the data collection assembly 6.
[0076] The detection assembly 5 continuously acquires the surface image of the fabric and environmental sensing data and sends the data to the data collection assembly 6. The data collection assembly 6 performs preliminary preprocessing and abnormality screening, stores and transmits the data. The data collection assembly 6 or the background server calls a deep learning-based detection model to automatically identify the fabric defect type (such as color difference, broken warp and weft, stains, etc.) and its position. The processing result is displayed in real time through the display and interaction assembly 7, and the operator can view the defect image and obtain detailed information.
[0077] Embodiment 4: On the basis of embodiment 3, the following is added: Referring to Figure 2 , the detection assembly 5 includes a camera 51, a laser scanning head 52, an electromagnetic scanning head 53, and a light-transmitting scanning head 54. The camera 51, the laser scanning head 52, the electromagnetic scanning head 53, and the light-transmitting scanning head 54 are respectively installed on the textile equipment and face the textile fabric in conveying for shooting and scanning. The camera 51 includes an industrial camera, a lens, an optical filter, and an image acquisition card. The laser scanning head 52 includes a laser emitter, a photodetector, and a scanning controller. The electromagnetic scanning head 53 includes an electromagnetic generator, signal amplification, an acquisition circuit, and an electromagnetic shielding shell. The light-transmitting scanning head 54 includes a high-brightness light source and a photoelectric sensor.
[0078] Referring to Figure 2 , a plurality of camera 51, laser scanning head 52, electromagnetic scanning head 53 and light-transmitting scanning head 54 are respectively arranged at intervals, and the scanning time interval of adjacent devices is controlled according to the conveying speed of the textile.
[0079] Referring to Figure 2The data collection assembly 6 comprises a memory 61, a communicator 62 and a data processor 63. The memory 61 is connected to the detection assembly 5 and the display interaction assembly 7 through the communicator 62. The data processor 63 is communicatively connected to the memory 61 and extracts, processes and rewrites the data stored in the memory 61. The memory 61 comprises temporary storage, fixed storage and power protection. The communicator 62 comprises a wired communication module and a wireless communication module. The data processor 63 comprises an industrial computer, a graphics processing and an AI training module processor.
[0080] With reference to Figure 2 The display interaction assembly 7 comprises a display 71 and a control input 72. The display 71 is connected to the data collection assembly 6 and displays the data processing result. The control input 72 is connected to the data collection assembly 6 and the display 71 and performs human-computer interaction control operation. The display 71 comprises an interactive display and a monitoring display. The control input 72 comprises a touch screen, a keyboard and a mouse.
[0081] Each scanning head in the detection assembly 5 starts scanning in order according to the fabric conveying speed. The camera 51 collects the fabric surface image, the laser scanning head 52 obtains the three-dimensional profile, the electromagnetic scanning head 53 detects the internal structure, and the light transmission scanning head 54 measures the fabric density and yarn integrity. The data collected by each device is transmitted in real time to the data processor 63 through the communicator 62. The data processor 63 synchronously fuses the multi-source data, performs image preprocessing, signal filtering and feature extraction, automatically identifies defects based on a deep learning model, classifies defect types such as color difference, warp breakage, stain, unevenness, etc., and locates the defect position. The processed result is written into the memory 61 and uploaded to the display interaction assembly 7 at the same time. The display 71 displays the detection result in real time. The defect area is marked with high light. When serious or critical defects are detected, the system automatically triggers an alarm to prompt the operator to handle it in time.
[0082] The operator adjusts the detection parameters through the control input 72 (touch screen, keyboard, mouse), including the collection frequency and scanning time interval, sensitivity threshold, view switching, and viewing multi-angle collection images and sensor data. The operator can also start the defect verification module to manually confirm or correct the automatic detection result.
[0083] The above content is only an example and description of the structure of the present application. Those skilled in the art can make various modifications or supplements or use similar ways to replace the described specific embodiments without deviating from the scope defined by the structure of the present application, which shall fall within the protection scope of the present application.
Claims
1. A deep learning based flexible textile quality inspection system, characterized in that, Include: Detection module (1) through the connection sensor and scanning camera data acquisition, and the data for machine training and machine quality inspection; Inference module (2), communication connection detection module (1) and use detection module (1) to collect data for reasoning judgment, determine the defect type; Verification module (3), through the artificial sampling review of inference module (2) reasoning judgment result; Learning module (4), the feedback of artificial sampling review result of verification module (3) to inference module (2) optimization training learning.
2. The deep learning-based flexible textile quality inspection system according to claim 1, characterized in that: The detection module (1) includes image collection unit (11), sensing data unit (12), time sequence control unit (13) and data conversion unit (14), the image collection unit (11) is installed on the textile equipment and collects target image, and the image is pixel segmented and all pixel point data is exported, the image collection unit (11) includes multiple groups and sequentially collects image information according to arrangement order, and the multiple groups of image information are superimposed to eliminate fuzzy image information, the image collection unit (11) is built-in camera driver, image acquisition control software and exposure and gain automatic adjustment algorithm, the sensing data unit (12) is installed and connected with textile equipment, the sensing data unit (12) is used for collecting environmental and object sensing data, the sensing data includes thickness parameter, surface quality, deformation data, tensile strength and static constant, the sensing data unit (12) is built-in sensor driver, sensing data filtering and preprocessing program and data synchronization time sequence control program, the time sequence control unit (13) is connected with image collection unit (11) and sensing data unit (12), the time sequence control unit (13) is used for coordinating the acquisition time of image and sensing signal, the time sequence control unit (13) is built-in timestamp recording system, equipment synchronization control protocol and trigger signal management program, the data conversion unit (14) is connected with image collection unit (11), sensing data unit (12) and time sequence control unit (13), the data conversion unit (14) converts multiple source original data into unified format, including data structuring, unit conversion and encoding conversion, the data conversion unit (14) is built-in data analysis program, format conversion tool and standardization and normalization algorithm library.
3. The deep learning-based flexible textile quality inspection system according to claim 1, characterized in that: The reasoning module (2) comprises a feature extraction unit (21), a fuzzy processing unit (22) and a result generation unit (23), the feature extraction unit (21) is used for extracting key features from original data, the feature extraction unit (21) is built-in with a feature extraction algorithm, a data dimension reduction tool and a numerical feature encoder, the fuzzy processing unit (22) is sequentially connected with the feature extraction unit (21), the fuzzy processing unit (22) is used for processing uncertain or fuzzy information, the fuzzy processing unit (22) is built-in with a fuzzy logic algorithm library, a Bayesian inference framework and a data confidence calculation program, the result generation unit (23) is sequentially connected with the feature extraction unit (21) and the fuzzy processing unit (22) respectively, the result generation unit (23) converts the result derived by reasoning into system output, including classification labels, detection positions and prediction values, the result generation unit (23) is built-in with result encoding and visualization software.
4. The deep learning-based flexible textile quality inspection system according to claim 1, characterized in that: The verification module (3) comprises a random sampling unit (31), a task allocation unit (32), a manual verification unit (33) and a feedback correction unit (34), the random sampling unit (31) is communicatively connected with the reasoning module (2) and selects sampling data from the reasoning result of the reasoning module (2), the random sampling unit (31) uses a sampling algorithm for sampling and is connected with the reasoning module (2) through a data retrieval interface, the task allocation unit (32) is connected with and controls the output of the random sampling unit (31), the task allocation unit (32) is used for allocating verification tasks to an automatic verification system or a manual inspector, the task allocation unit (32) is built-in with a workflow management system and a task scheduling and allocation algorithm, the manual verification unit (33) is connected with the task allocation unit (32) and receives data sent by the task allocation unit (32), the manual verification unit (33) is used for processing data that cannot be automatically judged by a machine and rechecking and comparing data automatically reasoned and identified by the machine, the manual verification unit (33) comprises a verification work interface and a data labeling tool, the feedback correction unit (34) is connected with the manual verification unit (33) and the reasoning module (2) and sends the rechecking result of the manual verification unit (33) to the reasoning module (2) for verification and correction training, the feedback correction unit (34) comprises a model updating module and a data backwriting and synchronization program.
5. The deep learning-based flexible textile quality inspection system according to claim 1, characterized in that: The learning module (4) comprises a feedback extraction unit (41), a comparative analysis unit (42), a data updating unit (43) and a simulation verification unit (44), the feedback extraction unit (41) is in communication connection with the reasoning module (2) and the verification module (3) and extracts the data fed back by the verification module (3) to the reasoning module (2), the comparative analysis unit (42) is connected with the feedback extraction unit (41) and receives the feedback data extracted by the feedback extraction unit (41), the comparative analysis unit (42) is used for comparing the reasoning result data with the verification data, and the difference between the two groups of data is analyzed through a performance evaluation tool and an error analysis program, the data updating unit (43) is connected with the comparative analysis unit (42) and the reasoning module (2) respectively, the data updating unit (43) is used for adding new labeled data or improved features to the training database, and the simulation verification unit (44) is connected with the data updating unit (43), and the simulation verification unit (44) verifies the improvement effect of the model data updating by simulation running.
6. A deep learning-based flexible textile quality inspection device, characterized in that, It comprises: A detection assembly (5) installed on a textile equipment and connected with a control system, the detection assembly (5) is used for collecting original data of fabric morphology and its environment state; A data collection assembly (6) in communication connection with the detection assembly (5), the data collection assembly (6) is used for data storage, management, transmission and preliminary processing; A display interaction assembly (7) in communication connection with the data collection assembly (6) and interactive data transmission with the data collection assembly (6), the display interaction assembly (7) is used for information display and man-machine interaction control.
7. The deep learning-based flexible textile quality inspection device according to claim 6, characterized in that: The detection assembly (5) comprises a camera (51), a laser scanning head (52), an electromagnetic scanning head (53) and a light transmission scanning head (54), the camera (51), the laser scanning head (52), the electromagnetic scanning head (53) and the light transmission scanning head (54) are respectively installed on the textile equipment and shoot and scan the textile fabric in transmission, the camera (51) comprises an industrial camera, a lens, an optical filter and an image acquisition card, the laser scanning head (52) comprises a laser emitter, a photoelectric detector and a scanning controller, the electromagnetic scanning head (53) comprises an electromagnetic generator, signal amplification, an acquisition circuit and an electromagnetic shielding shell, and the light transmission scanning head (54) comprises a high-brightness light source and a photoelectric sensor. 8.The deep learning based flexible textile quality inspection device according to claim 7, characterized in that: A plurality of the camera (51), the laser scanning head (52), the electromagnetic scanning head (53) and the light transmission scanning head (54) are respectively arranged at intervals, and the scanning time interval of adjacent equipment is controlled according to the conveying speed of the textile product. 9.The deep learning based flexible textile quality inspection device of claim 6, wherein: The data collection component (6) includes a memory (61), a communicator (62) and a data processor (63), the memory (61) is connected with the detection component (5) and the display interaction component (7) through the communicator (62), the data processor (63) is communicatively connected with the memory (61) and extracts, processes and rewrites the data stored in the memory (61), the memory (61) includes temporary storage, fixed storage and power protector, the communicator (62) includes a wired communication module and a wireless communication module, and the data processor (63) includes an industrial computer, a graphics processing and an AI training module processor.
10. The deep learning-based flexible textile quality inspection device according to claim 6, characterized in that: The display interaction component (7) includes a display (71) and a control input device (72), the display (71) is connected with the data collection component (6) and displays the data processing result, the control input device (72) is connected with the data collection component (6) and the display (71) and performs human-computer interaction control operation, the display (71) includes an interactive display and a monitoring display, and the control input device (72) includes a touch screen, a keyboard and a mouse.
Citation Information
Cited By
Textile fabric image segmentation and feature extraction processing method and system
CN121639707A
A processing method and system for textile fabric image segmentation and feature extraction
CN121639707B
