System and method for detecting surface defects in moulded foam cups of lingerie
Patent Information
- Application Number
- HK32026123799
- Authority / Receiving Office
- HK · HK
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-05-21
- Publication Date
- 2026-08-28
Smart Images

Figure 00000015_0000 
Figure 00000015_0001 
Figure 00000015_0002
Abstract
Description
Instruction Manual 1 A System and Method for Detecting Surface Defects in Cotton Cups of Women's Underwear Technical Field This Hong Kong short-term invention relates to the field of textile industry inspection technology, and particularly to a system and method for detecting surface defects in cotton cups of women's underwear. Background Technology The cotton cups of women's underwear (bra) are a key component, with complex manufacturing processes and zero tolerance for surface defects (such as micro-stains, residues, etc.). Currently, the industry generally uses manual visual inspection for quality control. Some use traditional unidirectional visual recognition (relying on RNG color for judgment). However, manual inspection has many drawbacks: not only is the detection efficiency low, but the results are also easily affected by personnel fatigue and differences in subjective judgment standards, resulting in unstable detection quality and poor consistency. Once a customer spot check finds defects, the entire batch of products faces the risk of being returned for re-inspection or scrapped, bringing huge cost pressure and time loss to manufacturers, making it difficult to meet the high-efficiency and demanding requirements of modern industrial production. While traditional visual recognition technology performs well in the detection of high-contrast and similar-sized targets, the detection effect is not ideal for objects such as underwear with low contrast and large size differences, and missed detections often occur. Women's bras (busts) often feature smooth cotton cups, low contrast in terms of defects (such as stains), significant size variations, and height differences due to their three-dimensional structure. These height differences can cause defects to be hidden in curved surfaces or folds. To address these issues, this short-term research proposes a surface defect detection system and method for women's bra cotton cups. The system involves multi-angle sampling and photography of the cotton cups, followed by image recognition and detection based on deep learning algorithms. The system is standardized, provides stable detection quality, and effectively improves detection efficiency. The specific technical solution is as follows: A surface defect detection system for women's underwear cotton cups, comprising a start switch, a frame, a dark box mounted on the frame, a six-axis robot, a mechanical transmission module, an image acquisition module, a control module, and a recognition module. The six-axis robot is housed inside the dark box. The mechanical transmission module includes a conveyor belt mounted on the frame, a drive motor, a speed controller, an underwear cotton cup fixture mounted on the conveyor belt, and multiple photoelectric sensors mounted on the conveyor belt. The photoelectric sensors are used to detect the position of the cotton cup. The image acquisition module includes an industrial camera, a lens, a light source, and a spotlight cover. The image acquisition module is fixed to the end of the six-axis robot via a fixed bracket. The six-axis robot is positioned at the end of the conveyor belt. The movement of the six-axis robot drives the image acquisition module to perform multi-angle shooting. The light source provides a stable, uniform, and high-contrast lighting environment for image acquisition. The control module includes a main control unit, a motion control subunit, a vision acquisition subunit, a light source drive subunit, an I / O and sensor subunit, and a human-machine interface. The motion control subunit includes a motion control card and encoder input, used for conveyor belt control, workstation positioning, and camera control.The vision acquisition subunit includes a camera trigger circuit, a synchronization interface, and an industrial camera interface for camera trigger / exposure synchronization control. The light source driving subunit drives the light source. The I / O and sensor subunit includes limit switches, photoelectric sensors, and an emergency stop switch for cotton cup positioning detection and safe switching. The human-machine interface (HMI) includes a UI display interface located on the outside of the darkroom and a host monitoring software interface for parameter setting, status display, and manual intervention. The main control unit enables real-time control, task scheduling, and safety and error handling between the subunits. The real-time control and task scheduling refer to controlling the drive motor through a motion control card, causing the conveyor belt to transport the cotton cup to be inspected to the dark box at a set speed. An encoder converts the linear displacement of the cotton cup fixture into a digital electrical signal, which is fed back to the main control unit. When the cotton cup blocks the photoelectric sensor at the end of the conveyor belt, a trigger signal is generated, indicating that the cotton cup has reached the shooting area. The main control unit receives the trigger signal and sends it to the motion control subunit, vision acquisition subunit, and light source drive subunit, controlling the six-axis robot to move and triggering the camera to start image acquisition. Strict synchronization between camera triggering, exposure control, and six-axis robot movement is ensured. The light source is controlled to set the light intensity / zone according to the recognition module requirements, acquiring multiple images of the same cotton cup at different light intensities and incident angles. After the images are taken, the inspected cotton cup is transported out of the dark box, and the images are sent to the main control unit. The main control unit sends the images, station ID, shooting parameters, timestamp, and pre-configured cotton cup model along with the corresponding detection rules to the recognition module, and then receives the recognition data. (HK 30136970 A Instruction Manual 3) The module receives detection results, cutting diagrams, timestamps, and operator information. Based on the detection results, it issues rejection or sorting instructions and records the results. It handles abnormal situations and logs them, while displaying the detection results on the human-machine interface. The abnormal situations include re-shooting, alarms, pauses, and emergency workpiece stops. The detection results include whether there are defects, defect type, defect coordinates, and confidence level. The safety and error handling refers to the emergency stop response, grating trigger stop, abnormal retry, and manual confirmation process for conveyor belts or underwear cotton cup fixtures. The recognition module receives raw images, workstation IDs, shooting parameters, timestamps, cotton cup models, and corresponding detection rules from the control module in real time. It preprocesses the images, extracts features from the preprocessed images based on a combination of traditional image processing analysis and / or deep learning features, and then uses a multi-branch detector for defect detection. It integrates the results of each detector, confidence level, and business rules into a final judgment, and generates and saves the detection results, cutting diagrams, timestamps, and operator information, which are then sent to the control module. The multi-branch detector includes a surface defect detector, a size and shape detector, and a color / texture anomaly detector. The size and shape detector is based on contour fitting, template matching, or key point detection techniques.To achieve precise measurement of geometric parameters and compliance assessment, the color / texture anomaly detector automatically evaluates the color consistency and texture uniformity of the product surface using color difference thresholds, texture statistics, or self-supervised anomaly detection methods. Furthermore, the light source can be an LED dome light source, and the shooting parameters include exposure and trigger time; the detection rules include setting the region of interest (ROI). Furthermore, the recognition module also includes offline model training, defect sample management, and a manual annotation feedback mechanism. This mechanism involves annotating suspected defects or low-confidence items and displaying them on the human-machine interface for manual review; the review results are then used for model retraining. Furthermore, the detection results and images from the recognition module can be associated with and stored using cotton cup batch numbers and timestamps, and reports can be exported for quality traceability. Furthermore, the cotton cup model and corresponding detection rules can be changed in the control module. Furthermore, image preprocessing includes denoising, gamma correction, perspective correction, ROI cropping, background segmentation, and brightness equalization. The surface defects refer to black spots, indentations, yellow spots, and / or foreign objects. The surface defect detector (HK 30136970 A specification 4) can be a model based on semantic segmentation or object detection. The business rules include an acceptable defect area threshold and defect location priority. Furthermore, the main control unit can be an embedded industrial controller or an industrial PC. This Hong Kong short-term study also proposes a detection method based on the above-mentioned detection system, which includes the following steps: S1 Material loading: Receive the upper-level task or manual start command, and place the cotton cup to be detected on the underwear cotton cup fixture; S2 Start the switch, and the photoelectric sensor determines that the cotton cup is at the starting point; S3 The transmission module transports the cotton cup to be detected. When the cotton cup blocks the photoelectric sensor at the end of the conveyor belt, a trigger signal is generated, indicating that the cotton cup has reached the shooting area; S4 Based on the trigger signal, the six-axis robot, camera, and light source are started. The light source is set with light intensity / zone according to the recognition requirements. The six-axis robot moves to drive the camera to start acquiring images, ensuring strict synchronization between camera triggering, exposure control, and six-axis robot movement. Multiple images are acquired for the same cotton cup using different light sources and different incident angles. After image acquisition is completed, the conveyor belt transports the cotton cup back to the starting point; S5 The acquired images are preprocessed, and features are extracted from the preprocessed images based on a combination of traditional image processing analysis and / or deep learning features. A multi-branch detector is used for defect detection. The results of each detector, confidence level, and business rules are integrated into a final judgment, generating and saving the detection results, cropped images, timestamps, and operator information. S6 issues rejection or sorting instructions based on the detection results, records the results, and displays the results on the human-machine interface. The output is OK indicating no defects, and NG indicating defects. Furthermore, the detection method also includes handling abnormal situations and logging.The abnormal situations include re-shooting, alarms, pauses, and workpiece emergency stops. This short-term Hong Kong-proposed women's underwear cotton cup surface defect detection system and method uses a conveyor belt to automatically transport the cotton cup to the imaging area. A six-axis robot drives a camera to acquire images from multiple angles. The recognition module uses deep learning algorithms to extract features and identify defects from the images, generating detection results and action suggestions. Through multi-angle imaging, the system can capture product appearance information 360 degrees without blind spots, overcoming the blind spot problem of traditional single-view inspection and significantly improving the defect detection rate (especially defects on the sides and in recesses). Combined with deep learning algorithms, the system can learn deep features of defects from massive amounts of data, not only stably identifying known defect types but also possessing strong generalization ability to recognize unknown, minute, and low-contrast complex defects. It transforms subjective and easily fatigued human eye judgment into objective and consistent algorithmic standards, eliminating the pain points of inconsistent manual quality inspection standards and efficiency fluctuations, ensuring extremely high product quality consistency and traceability. It significantly reduces the quality risks caused by missed or false detections. Brief Description of the Drawings To more clearly illustrate the technical solutions in 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. The drawings described below are only some embodiments of the present invention, wherein: Figure 1A-B is a perspective view of the surface defect detection system for cotton cups of women's underwear according to the present application; Figure 2A-B is a schematic diagram of the mechanical transmission module; Figure 3A-D is a schematic diagram of the image acquisition module; Figure 4 is a schematic diagram of an embodiment of the control module; Figure 5 is a flowchart of the defect detection method. Detailed Description of the Embodiments The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The structures, proportions, and sizes shown in the accompanying drawings are solely for illustrative purposes to aid those skilled in the art and to facilitate understanding. They are not intended to limit the scope of implementation in Hong Kong in the short term and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, provided they do not affect the effectiveness and purpose achievable in Hong Kong in the short term, should still fall within the scope of the technical content disclosed in Hong Kong in the short term. Furthermore, the terms "above," "below," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of implementation in Hong Kong in the short term. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of implementation in Hong Kong in the short term. As shown in Figure 1...A surface defect detection system for women's underwear cotton cups includes a start switch 1, a frame 12, a dark box 6 mounted on the frame 12, a six-axis robot 11, a mechanical transmission module 4, an image acquisition module, a control module, and a recognition module. The six-axis robot 11 is housed inside the dark box 6. As shown in Figure 1, the dark box 6 and the frame 12 are an integral unit. In one embodiment, a relatively enclosed enclosure can be used, along with a dark acrylic panel, to create a completely enclosed environment, achieving the effect of a dark box. As shown in Figure 2, the mechanical transmission module includes a conveyor belt 43 mounted on the frame 12, a drive motor 41, a speed controller, an underwear cotton cup fixture 2 mounted on the conveyor belt 43, and multiple photoelectric sensors 42 and 421 mounted on the conveyor belt 43. The photoelectric sensors 42 and 421 are used to detect the position of the cotton cup 3. In one embodiment, when the cotton cup 3 blocks the photoelectric sensor 421, a trigger signal is immediately generated to notify the camera to take a picture, thereby ensuring the synchronization and accuracy of image acquisition. As shown in Figure 3, the image acquisition module includes a camera module 9, a ring light source 7, and a spotlight cover 8. The camera module 9 includes an industrial camera and a lens. The image acquisition module is fixed to the end of a six-axis robot 11 via a fixed bracket 10, and the six-axis robot 11 is positioned at the end of the conveyor belt 43. When the cotton cup 3 moves with the conveyor belt 43 to the designated shooting area, the six-axis robot 11 moves to carry the image acquisition module to perform multi-angle sampling (because of the three-dimensional structure of the cotton cup, the sampling will be performed in segments from multiple angles, and the data will be summarized and judged). The industrial camera and lens are responsible for capturing high-definition images of the surface of the cotton cup 3 and converting physical optical signals into digital electrical signals. The light source 7 provides a stable, uniform, and high-contrast lighting environment for image acquisition. Through a specific optical angle design, it is designed to highlight small stains, bumps, or foreign objects while suppressing interference from ambient light. The control module is primarily responsible for the hardware driver, workstation scheduling, motion control, data acquisition triggering, lighting / camera configuration, communication with the host software, and security protection of the entire machine. The main functions of the control module are as follows: Workstation scheduling: Implementing automatic process control for material loading → positioning → photography / inspection → qualified sorting; Timing synchronization: Strict timing control of camera triggering and light source exposure to ensure stable image quality; Parameter distribution / acquisition: Receiving the detection configuration (ROI, exposure, trigger time) from the recognition module and transmitting back the acquisition status and original image ID / timestamp; Safety and error handling: Emergency stop response, raster trigger stop, abnormal retry, and manual confirmation process; Remote maintenance: Log recording, OTA parameter uploading, and remote diagnostic interface; In one embodiment,The control module mainly includes a main control unit, a motion control subunit, a vision acquisition subunit, a light source driving subunit, an I / O and sensor subunit, and a human-machine interface (HMI). The motion control subunit includes a motion control card and encoder input for conveyor belt control, workstation positioning, and camera control. The encoder converts the linear displacement of the underwear fixture into digital electrical signals, which are fed back to the main control unit, enabling the main control unit to control the speed, position, and motion (including direction control) of the mechanical conveyor module. The vision acquisition subunit includes a camera trigger circuit, a synchronization interface, and an industrial camera interface for camera trigger / exposure synchronization control. The light source driving subunit drives the light source. In one embodiment, the light source driving subunit drives an LED dome light source. The I / O and sensor subunit includes limit switches, photoelectric sensors, and an emergency stop switch for cotton cup positioning detection and safety switching. The HMI includes a UI display interface 5 located on the outside of the darkroom and a host monitoring software interface for parameter setting, status display, and manual intervention. The main control unit implements real-time control, task scheduling, and safety and error handling between the subunits. In one embodiment, the main control unit can be an embedded industrial controller or an industrial PC. The real-time control and task scheduling refer to controlling the drive motor via a motion control card, causing the conveyor belt to transport the cotton cup to be inspected to the dark box at a set speed. When the cotton cup blocks the photoelectric sensor at the end of the conveyor belt, a trigger signal is generated, indicating that the cotton cup has reached the shooting area. The main control unit receives the trigger signal and sends it to the motion control subunit, vision acquisition subunit, and light source drive subunit, controlling the six-axis robot to move and triggering the camera to start acquiring images. Strict synchronization is ensured between camera triggering, exposure control, and the six-axis robot's movement. The light source is controlled to set the light intensity / zone according to the recognition module's requirements, acquiring multiple images of the same cotton cup at different light intensities and incident angles. After the images are taken, the inspected cotton cup is transported out of the dark box, and the images are sent to the main control unit. The main control unit sends the images, workstation ID, shooting parameters, timestamp, and pre-configured cotton cup model along with the corresponding detection rules to the recognition module, and then receives the detection results from the recognition module. (HK 30136970 A Specification 8) The system collects the results, cutting diagram, timestamp, and operator information. Based on the inspection results, it issues rejection or sorting instructions and records the results. It handles abnormal situations and logs them, while displaying the inspection results on the human-machine interface. Abnormal situations include re-shooting, alarms, pauses, and emergency stops of the workpiece. The inspection results include whether there are defects, defect type, defect coordinates, and confidence level. The safety and error handling refers to the emergency stop response, grating-triggered stop, abnormal retry, and manual confirmation process for conveyor belts or underwear cotton cup fixtures. In one embodiment,Examples of interface and data formats are as follows: Control → Recognition: Send image, workstation ID, shooting parameters, timestamp (JSON / Protobuf / binary stream); Recognition → Control: Return detection results (OK / NG / defect type / coordinates / confidence), suggested actions, and retake request. Figure 4 shows an exemplary control module. In Figure 4, an industrial computer (host computer, i.e., main control unit) is used instead of a PLC to control and process data for the entire module, serving as the centralized control core. The industrial computer (host computer) uses built-in IO and bus interfaces to establish direct signal connections with the six-axis robot, the transmission module, the camera trigger of the image acquisition module, the controller of the light source module, and the data output of the defect recognition module. Status feedback signals from each module, sensor trigger signals, and motion control commands from the six-axis robot are all aggregated to the industrial computer (host computer), where its internal real-time operating system performs unified signal acquisition, logical operations, and command issuance. By converging all signals to the industrial computer (host computer), centralized timing control and data interaction for all system actions are achieved. This setup eliminates the reliance on a standalone PLC, simplifies the control hierarchy, reduces system hardware costs, and facilitates direct modification of control logic, recording of all process data, and information integration with the upper-level production management system via host computer software. The defect identification module is primarily used for preprocessing, defect detection, size / appearance determination, result output, and learning optimization of the acquired raw images of the bra cups. It supports the identification of various bra cup styles and defect types. The main software components of the identification module are as follows: - Interface layer: Receives raw images and metadata from the control module and returns detection results; manages configuration files and versions. - Image preprocessing module: Denoising, gamma correction, perspective correction, ROI cropping, background segmentation, and brightness equalization. - Feature extraction module: Based on a combination of traditional image processing contour analysis and / or deep learning features (CNN features, semantic segmentation). - Defect Detection Module: Multi-branch detectors, including: - Surface defect (black spots, indentations, yellow spots, foreign objects) detectors (semantic segmentation or object detection, such as U-Net / Mask R-CNN or lightweight models); - Size and shape determiners (based on contour fitting, template matching, or keypoint detection); - Color / texture anomaly detection (color difference threshold, texture statistics, or self-supervised anomaly detection). - Decision Fusion and Rule Engine: Fuses the results of each detector, confidence scores, and business rules (e.g., acceptable defect area threshold, defect location priority) into a final decision.It generates action suggestions. - Learning and Update Module (optional): Offline model training, defect sample management, manual annotation feedback mechanism (continuous improvement). - Log and Traceability Module: Saves detection results, cropped images, coordinates, confidence levels, timestamps, and operator information, supporting traceability and statistical reports. - Visualization and Interaction Layer: Defect visualization, manual review interface, model performance monitoring. In one embodiment, the recognition module receives the original image, workstation ID, shooting parameters, timestamp, cotton cup model, and corresponding detection rules from the control module in real time. It preprocesses the image, extracts features from the preprocessed image based on a combination of traditional image processing analysis and / or deep learning features, and then uses a multi-branch detector for defect detection. It integrates the results of each detector, confidence level, and business rules into a final judgment, and generates and saves the detection results, cropped images, timestamps, and operator information, sending them to the control module. HK 30136970 A Specification 10 The multi-branch detector includes a surface defect detector, a size and shape detector, and a color / texture anomaly detector. The size and shape detector, based on contour fitting, template matching, or key point detection technology, achieves accurate measurement of geometric parameters and compliance judgment. The color / texture anomaly detector automatically evaluates the color consistency and texture uniformity of the product surface through color difference thresholds, texture statistics, or self-supervised anomaly detection methods. In one embodiment, the shooting parameters include exposure and trigger time, and the detection rules include setting the region of interest (ROI). In one embodiment, the detection results and images of the recognition module can be associated with and stored with the cotton cup batch number and timestamp, and reports can be exported for quality traceability. In one embodiment, the cotton cup model and corresponding detection rules can be changed in the control module. As shown in Figure 5, this Hong Kong short-term study also proposes a detection method based on the above-mentioned detection system, which includes the following steps: S1 Material placement: Receive the upper-level task or manual start command and place the cotton cup to be detected on the underwear cotton cup fixture; S2 Start the switch, and the photoelectric sensor determines that the cotton cup is at the starting point; S3 The transmission module transports the cotton cup to be detected. When the cotton cup blocks the photoelectric sensor at the end of the conveyor belt, a trigger signal is generated, indicating that the cotton cup has reached the shooting area; S4 Based on the trigger signal, the six-axis robot, camera, and light source are started. The light source is set with light intensity / zone according to the recognition requirements. The six-axis robot moves to drive the camera to start acquiring images, ensuring strict synchronization between camera triggering, exposure control, and six-axis robot movement. Multiple images are acquired for the same cotton cup using different light sources and different incident angles. After image acquisition is completed, the conveyor belt transports the cotton cup back to the starting point; S5 Preprocess the acquired images. Based on a combination of traditional image processing analysis and / or deep learning features, feature extraction is performed on the preprocessed images. A multi-branch detector is used for defect detection. The results of each detector, confidence level, and business rules are integrated into the final judgment.The system generates and saves inspection results, cutting diagrams, timestamps, and operator information; S6 issues rejection or sorting instructions based on the inspection results, records the results, and displays the results on the human-machine interface. HK 30136970 A Specification 11: An OK output indicates no defects, and an NG output indicates defects. In one embodiment, the inspection method further includes handling abnormal situations and recording logs, including re-taking, alarms, pauses, and workpiece emergency stops. This Hong Kong short-term system integrates automated conveying, multi-angle imaging, and deep learning recognition to construct a high-precision, high-efficiency, and highly adaptive intelligent inspection system for the surface defects of women's underwear cotton cups. It fundamentally changes the traditional quality inspection mode that relies on human eyes or fixed rule algorithms, liberating workers from repetitive and tiring visual inspection, and realizing the standardization and unmanned operation of the quality inspection process. Multi-angle photography overcomes the blind spots of single-view inspection, ensuring full coverage of complex curved surfaces and internal structures; deep learning algorithms endow the system with powerful feature learning and defect discrimination capabilities, not only accurately identifying known defects but also possessing a certain generalization potential for new defect types. The resulting technological benefits are significant: In terms of accuracy, the system's false negative and false alarm rates are far lower than manual inspection, effectively improving product quality consistency. In terms of efficiency, it operates 24 / 7 with millisecond-level detection speeds, significantly increasing production line cycle time and capacity. In terms of economics, it reduces labor and management costs in the long term, while maintaining brand reputation by minimizing defective products. In terms of traceability, all test results are digitally stored, providing comprehensive data support for production process optimization and quality analysis, driving continuous improvement and intelligent upgrading of the manufacturing process. The above description is merely a preferred embodiment of this invention and is not intended to limit this invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention. HK 30136970 A Claim 1 1. A surface defect detection system for women's underwear cotton cups, comprising a start switch, a frame, a dark box mounted on the frame, a six-axis robot, a mechanical transmission module, an image acquisition module, a control module, and a recognition module, wherein the six-axis robot is housed within the dark box; characterized in that: the mechanical transmission module comprises a conveyor belt mounted on the frame, a drive motor, a speed controller, an underwear cotton cup fixture mounted on the conveyor belt, and multiple photoelectric sensors mounted on the conveyor belt, the photoelectric sensors being used to detect the position of the cotton cup; the image acquisition module comprises an industrial camera, a lens, a light source, and a spotlight cover, the image acquisition module being fixed to the end of the six-axis robot via a fixed bracket, the six-axis robot being positioned at the end of the conveyor belt, and the movement of the six-axis robot driving the image acquisition module to perform multi-angle imaging.The light source provides a stable, uniform, and high-contrast lighting environment for image acquisition. The control module includes a main control unit, a motion control subunit, a vision acquisition subunit, a light source driver subunit, an I / O and sensor subunit, and a human-machine interface (HMI). The motion control subunit includes a motion control card and encoder input for conveyor belt control, workstation positioning, and camera control. The vision acquisition subunit includes a camera trigger circuit, a synchronization interface, and an industrial camera interface for camera trigger / exposure synchronization control. The light source driver subunit drives the light source. The I / O and sensor subunit includes limit switches, photoelectric sensors, and emergency stop switches for cotton cup positioning detection and safe switching. The HMI includes a UI display interface located outside the darkroom and a host monitoring software interface for parameter setting, status display, and manual intervention. The main control unit enables real-time control, task scheduling, and safety and error handling between the subunits. The real-time control and task scheduling refer to controlling the drive motor through a motion control card, causing the conveyor belt to transport the cotton cup to be detected into the dark box at a set speed. An encoder converts the linear displacement of the cotton cup fixture into a digital electrical signal, which is then fed back to the main control unit. When the cotton cup blocks the photoelectric sensor at the end of the conveyor belt, a trigger signal is generated, indicating that the cotton cup has reached the shooting area. The main control unit receives the trigger signal. (HK 30136970 A Claim 2) The signal is sent to the motion control subunit, vision acquisition subunit, and light source drive subunit to control the movement of the six-axis robot and trigger the camera to start acquiring images. Strict synchronization between camera triggering, exposure control, and the six-axis robot's movement is ensured. The light source is controlled to set the light intensity / zone according to the recognition module's requirements, acquiring multiple images of the same cotton cup at different light intensities and incident angles. After the images are taken, the cotton cup is transported out of the dark box, and the images are sent to the main control unit. The main control unit sends the images, workstation ID, shooting parameters, timestamp, and pre-configured cotton cup model along with the corresponding detection rules to the recognition module. It then receives the recognition module's detection results, cutting diagram, timestamp, and operator information. Based on the detection results, it issues rejection or sorting instructions and records the results. Abnormal situations are handled and logged, and the detection results are displayed on the human-machine interface. Abnormal situations include re-shooting, alarm, pause, and workpiece emergency stop. The detection results include whether there is a defect, defect type, defect coordinates, and confidence level. The aforementioned safety and error handling refers to the emergency stop response, grating-triggered stop, abnormal retry, and manual confirmation process for the conveyor belt or underwear cotton cup fixture; the recognition module receives raw images, workstation IDs, shooting parameters, timestamps, cotton cup models, and corresponding detection rules from the control module in real time, preprocesses the images, extracts features from the preprocessed images based on a combination of traditional image processing analysis and / or deep learning features, and then uses a multi-branch detector for defect detection.The system integrates the results and confidence levels of each detector with business rules to form a final judgment, and generates and saves the detection results, cropping images, timestamps, and operator information, which are then sent to the control module. The multi-branch detectors include surface defect detectors, size and shape detectors, and color / texture anomaly detectors. The size and shape detectors, based on contour fitting, template matching, or key point detection techniques, achieve accurate measurement of geometric parameters and compliance judgment. The color / texture anomaly detectors automatically evaluate the color consistency and texture uniformity of the product surface through color difference thresholds, texture statistics, or self-supervised anomaly detection methods. 2. The detection system according to claim 1 is further characterized in that the light source can be an LED dome light source, the shooting parameters include exposure and trigger time, and the detection rules include setting a region of interest (ROI). 3. The detection system according to claim 1 is further characterized in that the identification module also includes offline model training, defect sample management, and a manual annotation feedback mechanism. The manual annotation feedback mechanism refers to annotating suspected defects or low-confidence items in claim 3 of HK 30136970 A and displaying them on the human-machine interface for manual review. The review results are used for model retraining. 4. The detection system according to claim 1, characterized in that the detection results and images of the identification module can be associated with and stored with the cotton cup batch number and timestamp, and reports can be exported for quality traceability. 5. The detection system according to claim 1, characterized in that the cotton cup model and corresponding detection rules can be changed in the control module. 6. The detection system according to claim 1, characterized in that image preprocessing includes denoising, gamma correction, perspective correction, ROI cropping, background segmentation, and brightness equalization; the surface defects refer to black spots, indentations, yellow spots, and / or foreign objects; the surface defect detector can be a model based on semantic segmentation or target detection; and the business rules include an acceptable defect area threshold and defect location priority. 7. The detection system according to claim 1, characterized in that the main control unit can be an embedded industrial controller or an industrial PC. 8. A detection method based on the detection system according to any one of claims 1-7, comprising the following steps: S1 Unloading: receiving a host task or manual start command, placing the cotton cup to be detected on the underwear cotton cup fixture; S2 Activating the switch, the photoelectric sensor determines that the cotton cup is at the starting point; S3 The transmission module conveys the cotton cup to be detected, and when the cotton cup blocks the photoelectric sensor at the end of the conveyor belt, a trigger signal is generated, indicating that the cotton cup has reached the shooting area; S4 Based on the trigger signal, the six-axis robot, camera, and light source are activated. The light source is set with light intensity / zone according to the recognition requirements. The six-axis robot moves to drive the camera to start acquiring images, ensuring strict synchronization between camera triggering, exposure control, and the movement of the six-axis robot.Multiple images are acquired from the same cotton cup using different light sources and incident angles. After image acquisition, the conveyor belt transports the cotton cup to the starting point. S5: The acquired images are preprocessed. Feature extraction is performed on the preprocessed images based on a combination of traditional image processing analysis and / or deep learning features. A multi-branch detector is used for defect detection. The results of each detector, confidence level, and business rules are integrated into a final judgment. Detection results, cut images, timestamps, and operator information are generated and saved. S6: Based on the detection results, rejection or sorting instructions are issued, results are recorded, and results are displayed on the human-machine interface. An output of OK indicates no defects, and an output of NG indicates defects. 9. The detection method according to claim 1, further characterized in that the detection method also includes handling abnormal situations and recording logs, wherein the abnormal situations include re-taking, alarm, pause, and workpiece emergency stop. HK 30136970 A Instruction Manual Appendix 1; HK 30136970 A Instruction Manual Appendix 2; Appendix 3; Appendix 4; HK 30136970 A Instruction Manual Appendix 5; HK 30136970 A,
