Softwood wafer surface quality detection system

Abstract

The invention discloses a softwood wafer surface quality detection system, belongs to the technical field of automatic sorting, and solves the problems that the existing softwood wafer detection technology depends on low-efficiency and subjective manual visual inspection or is limited by traditional machine vision and is poor in robustness, weak in adaptability and the like. The system comprises a vibration feeding device used for arranging cork wood round pieces in order and outputting the cork wood round pieces in a single-piece mode; the rotary positioning device is used for receiving the cork wood round pieces and rotating the cork wood round pieces to all stations; the double-sided image acquisition device is used for synchronously acquiring top surface and bottom surface images of the cork wafer; the defect detecting and grading device is used for processing the image data, identifying surface defects and outputting grade results; the laser marking device is used for marking a corresponding mark on the surface of the cork wafer; the pneumatic sorting device is used for directionally blowing the cork wood discs to the corresponding channels; and the PLC control device is used for coordinating the action time sequence of each device. The method is suitable for scenes such as shuttlecock head production.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of softwood product quality detection and automatic sorting, and particularly relates to a softwood disc surface quality detection technology. BACKGROUND

[0002] The performance of a badminton head directly affects the hitting feel and flight stability. Natural cork skin is a key raw material for manufacturing high-end badminton heads and has characteristics such as good elasticity, sealing, heat insulation, sound insulation, electrical insulation, and friction resistance. However, during the growth and processing of cork, pores and fractures are easily formed at different positions, resulting in substandard weight of the same specification head and affecting the subsequent hair planting process, which seriously affects the user experience. Currently, the industry mainly relies on manual visual method for grade sorting of cork discs. This method is low in efficiency, high in missed detection rate, strong in subjectivity, and easily affected by the experience and fatigue of the operator, making it difficult to ensure the consistency of the sorting quality. Therefore, developing an intelligent detection device based on machine vision to realize automatic detection and grade classification of cork disc defects is of great significance to improve the production quality of badminton heads.

[0003] In recent years, surface detection technology based on machine vision has gradually become a research hotspot. For example, Chinese patent application CN221351180U proposes a wood chip surface defect detection device, and Chinese patent application CN117214191A proposes a plate surface defect double-sided detection device and detection method, both of which propose to capture images by a conveyor belt in cooperation with a CCD camera and realize defect recognition by combining image processing algorithms, but there are the following problems: Detection blind area: when a traditional clamp fixes a workpiece, the blocked area cannot be completely photographed, resulting in missed detection; Pollution interference: an open detection environment easily causes dust or oil stains to adhere to the surface of the wood, affecting the image quality; Insufficient adaptability: existing systems are mostly designed for regular plates and are difficult to adapt to the irregular curved surface characteristics of cork discs.

[0004] In view of the above defects, some research attempts to introduce deep learning technology. For example, a cork defect detection model based on YOLOv5 improves the classification accuracy, but the model has large number of parameters and high computational complexity, making it difficult to realize real-time detection on embedded devices. In addition, existing technologies lack specificity in recognizing cork-specific defects (such as local warping caused by bark layering), and do not solve key problems such as pollution interference and multi-dimensional defect coordination during the detection process.

[0005] Therefore, it is urgent to develop a high-efficiency, non-contact surface quality detection device that adapts to the characteristics of cork discs, realizes accurate positioning, classification, and real-time sorting of defects through structural innovation and algorithm optimization, and provides reliable quality assurance for high-end products such as badminton heads. Summary of the Invention

[0006] This invention provides a cork disc surface quality inspection system, which aims to solve the problems of existing cork disc inspection technologies that either rely on inefficient and subjective manual visual inspection, or are limited by the poor robustness and weak adaptability of traditional machine vision, while existing intelligent solutions suffer from insufficient real-time performance, complex deployment, or lack of complete industrial-grade integration solutions.

[0007] The cork disc surface quality inspection system proposed in this invention includes: A vibrating feeder is used to arrange cork discs in an orderly manner and output them one by one; A rotary positioning device is used to receive cork discs and rotate them at a preset angle to the inspection station, marking station, and sorting station. A double-sided image acquisition device is used to simultaneously acquire images of the top and bottom surfaces of a cork disc to obtain image data; A defect detection and grading device is used to process image data, identify surface defects, and output grading results. Laser marking device, used to mark corresponding marks on the surface of cork discs according to grade results; A pneumatic sorting device is used to directionally blow cork discs into the corresponding channels according to the grade results; A PLC control unit is used to coordinate the timing of actions of various devices.

[0008] Furthermore, a preferred embodiment is provided: the vibrating feeding device includes a vibrating plate and a controller, the vibrating plate is equipped with a spiral track, and the controller is used to control the vibration amplitude of the vibrating plate.

[0009] Furthermore, a preferred solution is provided: the vibratory feeding device also includes a photoelectric detection unit for monitoring the feeding status of the cork discs and is linked with the PLC control module to provide real-time feedback on the conveying status.

[0010] Furthermore, a preferred embodiment is provided: the rotary positioning device includes a mechanical turntable, a feeding unit, and a photoelectric switch; the mechanical turntable is configured as a dial structure with a locking position, used to drive the cork disc to make a circular motion; the feeding unit is used to control the cork disc to enter the locking position at a uniform speed; the photoelectric switch is used to detect the cork disc in place and trigger the turntable to rotate.

[0011] Furthermore, a preferred embodiment is provided: the feeding unit includes: a push rod for pushing cork discs; a pressing cylinder for driving the push rod to perform a telescopic action via air pressure, controlling the cork discs to enter the locking position; a solenoid valve for receiving pulse signals from the PLC control device and switching the air circuit on and off; an air circuit adjustment unit for controlling the push rod telescopic speed and thrust; and a position sensor for monitoring the push rod stroke in real time and feeding back to the PLC control device.

[0012] Furthermore, a preferred embodiment is provided: the dual-sided image acquisition device includes an industrial camera and an illumination unit symmetrically arranged vertically, wherein the industrial camera is a CCD camera and the illumination unit is a ring-shaped uniform light source.

[0013] Furthermore, a preferred embodiment is provided: the dual-sided image acquisition device further includes a locking mechanism for fixing and locking the industrial camera.

[0014] Furthermore, a preferred embodiment is provided: the defect detection and grading device includes a deep learning module, which uses the YOLOv5s algorithm for defect identification and the CCGS grading strategy for grading the cork discs.

[0015] Furthermore, a preferred solution is provided: the defect identification using the YOLOv5s algorithm includes the following steps: Mosaic data augmentation is employed to improve the robustness of small target detection through four-image stitching and random scaling, and the initial box size is optimized by combining adaptive anchor box calculation. The backbone network introduces the C3 module and the SPPF module. The C3 module is based on the CSPNet design with a dual-branch residual structure to reduce computational redundancy. The SPPF module improves the traditional spatial pyramid pooling into a serial 5×5 max pooling layer. The feature fusion layer integrates a bidirectional feature pyramid and a path aggregation network, enhancing multi-target localization capabilities through cross-scale bidirectional feature interaction. The detection head outputs three sets of feature maps: 20×20, 40×40, and 80×80. Each grid has three preset anchor boxes and predicts target box parameters. The number of positive samples is expanded through cross-grid and cross-branch matching strategies. The localization loss uses CIoU loss to comprehensively evaluate the similarity of overlap rate, center distance and aspect ratio. The classification and confidence loss are calculated based on binary cross-entropy. In the post-processing stage, redundant detection boxes are removed by non-maximum suppression.

[0016] Furthermore, a preferred embodiment is provided: the pneumatic sorting device includes: An air pump is used to provide an air source; The gas distribution unit includes a pressure regulating valve and a flow meter, which are used to control the gas volume and are connected to a solenoid valve array through the main gas pipe; The solenoid valve array contains a group of independent solenoid valves, each corresponding to a different grade of cork disc, and each group of solenoid valves is connected to the jet nozzle via an air tube. The jet nozzle, located above the sorting station, is used to focus high-pressure airflow onto the edge of the cork disc, blowing the cork disc into the corresponding channel for sorting.

[0017] Compared with the prior art, the advantages of the present invention are: The existing technology has the following problems: Bottlenecks of manual inspection: The industry generally relies on manual visual inspection, which has drawbacks such as low efficiency, strong subjectivity of inspection results, and susceptibility to operator experience and fatigue. This leads to inconsistent product grading standards and difficulty in ensuring quality consistency.

[0018] Limitations of traditional machine vision methods: Although traditional feature extraction-based detection methods have achieved a certain degree of automation, they usually require complex manual feature design and have poor adaptability (robustness) to changes in lighting and the diversity of defect morphology, resulting in limited accuracy.

[0019] The shortcomings of existing intelligent detection solutions: Although deep learning methods have been applied to this field in recent years, most of them are two-stage detection models (such as Faster R-CNN), which have room for optimization in terms of real-time performance and ease of hardware deployment; or the solutions only stay at the algorithm model level, lacking a proven overall solution that fully integrates efficient detection, automatic labeling and automatic sorting functions and can be directly used in industrial production.

[0020] To address the aforementioned problems, the innovation of this application lies in proposing a quality inspection system for cork discs that incorporates deep learning technology. Unlike traditional research, this invention not only optimizes the algorithm model but also designs and develops a complete intelligent inspection system to address key technical issues in cork disc quality inspection. This system uses machine vision technology to automatically detect defects in cork discs, classifies them according to the CCGS strategy, and marks them with a laser, thereby achieving efficient and accurate sorting of cork discs. The system proposed in this invention not only improves inspection accuracy but also significantly increases production efficiency, providing a reliable technical guarantee for the efficient production of badminton shuttlecock heads.

[0021] Specifically, this invention constructs an intelligent detection and sorting system for cork discs that integrates hardware system integration and deep learning algorithms: at the hardware level, it innovatively designs a high-precision rotary positioning mechanism and a dual industrial CCD camera visual inspection module. The laser marking system and pneumatic sorting device, through a compact disc-shaped mechanical structure and PLC control system, achieve automatic feeding, double-sided image acquisition, defect detection, grade marking, and rapid sorting of cork discs. On the software side, a defect detection algorithm is built based on a lightweight YOLOv5s deep learning model, combined with the unique Cork Chip Grading Strategy (CCGS). Using a defect size exceeding 3 mm as the unacceptable standard, cork discs are classified into three grades: A, B, and C. This overcomes the bottlenecks of traditional manual inspection, which suffers from high subjectivity, low efficiency, and insufficient accuracy of traditional feature extraction algorithms. Compared to existing two-stage deep learning detection methods, the single-stage YOLOv5s model significantly improves real-time performance while maintaining an average detection accuracy of 98%. Furthermore, the system's robustness is enhanced through multi-light environment data training. Laser marking technology achieves high-contrast marking. The entire solution combines the advantages of high detection efficiency, accurate classification, and production automation, providing an intelligent solution for the quality control of cork products.

[0022] This invention is applicable to defect detection, grading, and automatic sorting of cork discs in scenarios such as badminton shuttlecock head production. It integrates high-precision mechanical positioning, industrial camera vision acquisition, and... The laser marking and pneumatic sorting hardware module, combined with the YOLOv5s deep learning detection model and customized grading strategy, enables automated detection, quality grade classification, and efficient sorting and marking of surface defects in cork discs. It is suitable for industrial intelligent detection and automation equipment technology scenarios. Attached Figure Description

[0023] Figure 1 This is a structural diagram of the cork disc surface quality detection system according to a specific embodiment of the present invention; Figure 2 These are different angle views of the cork disc surface quality detection system according to a specific embodiment of the present invention; Figure 3 This is a schematic diagram of the rotary positioning system structure according to a specific embodiment of the present invention; Figure 4 This is a 3D diagram of the rotary positioning device and the jet nozzle according to a specific embodiment of the present invention; Figure 5 This is a rendering of the laser marking device according to a specific embodiment of the present invention; Figure 6 This is a schematic diagram of the cork disk quality inspection model architecture according to a specific embodiment of the present invention; Figure 7 This is a flowchart of the cork disc quality classification procedure according to a specific embodiment of the present invention; Figure 8 This is a diagram showing the training results as described in a specific embodiment of the present invention; Figure 9 This is a PR curve diagram as described in a specific embodiment of the present invention; Figure 10 This is a schematic diagram illustrating the three main defect types described in a specific embodiment of the present invention; Figure 11 This is a schematic diagram of the detection results for the three main defect types described in a specific embodiment of the present invention; Figure 12 This is a schematic diagram illustrating the impact of the external environment on the detection of the machine vision system according to a specific embodiment of the present invention; Figure 13 This is a schematic diagram of the three-stage identification and detection results of a Class A product according to a specific embodiment of the present invention; Figure 14 This is a schematic diagram of the three-marking results of a Grade A product according to a specific embodiment of the present invention; Figure 15 This is a schematic diagram of the three-stage identification and detection results of a Class B product according to a specific embodiment of the present invention; Figure 16 This is a schematic diagram of the three-marking results of the Grade B product according to a specific embodiment of the present invention; Figure 17 This is a schematic diagram of the three-stage identification and detection results of a Class C product according to a specific embodiment of the present invention; Figure 18 This is a schematic diagram of the three-marking results of a Class C product according to a specific embodiment of the present invention. Detailed Implementation

[0024] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0025] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0027] Implementation Method 1: A cork disc surface quality inspection system, the system comprising: A vibrating feeding device is used to arrange cork discs in an orderly manner and output them one by one. Specifically, the vibrating feeding device includes a vibrating plate and a controller. The vibrating plate is equipped with a spiral track, and the controller is used to control the vibration amplitude of the vibrating plate. The controller can precisely adjust the vibration amplitude of the vibrating plate, thereby achieving flexible control of the sample conveying speed and ensuring a stable and efficient feeding process. The vibrating feeding device also includes a photoelectric detection unit for monitoring the feeding status of the cork discs and is linked with a PLC control module to provide real-time feedback on the conveying status. The vibrating feeding device also includes a height-adjustable bracket for supporting the vibrating plate and the controller, and ensuring that the outlet of the vibrating plate and the mechanical turntable are on the same plane.

[0028] A rotary positioning device is used to receive cork discs and rotate them to the inspection station, marking station, and sorting station at a preset angle. Specifically, the rotary positioning device includes a mechanical turntable, a feeding unit, and a photoelectric switch. The mechanical turntable is configured as a dial structure with locking positions to drive the cork discs in a circular motion. The feeding unit controls the cork discs to enter the locking positions at a uniform speed. The photoelectric switch detects the cork discs in place and triggers the turntable to rotate. This design fully meets the functional requirements of the system while greatly saving space resources and effectively improving the system's integration and practicality. The feeding unit includes: a push rod for pushing the cork discs; a pressing cylinder for driving the push rod to perform extension and retraction actions via air pressure, controlling the cork discs to enter the locking positions; a solenoid valve for receiving pulse signals from the PLC control device and switching the air circuit on and off; an air circuit adjustment unit for controlling the extension and retraction speed and thrust of the push rod; and a position sensor for real-time monitoring of the push rod stroke and feeding back to the PLC control device. To ensure that the sample can pass evenly through the feed port and smoothly enter the corresponding station of the chuck, this embodiment is designed as follows: Figure 4 The cylinder switch structure is shown. By precisely controlling the opening and closing time of the air valve, the sample enters the chuck at a uniform speed, effectively avoiding jamming. In the classification stage, this study selected a pneumatic system, which has significant advantages such as low cost, rapid response, and convenient operation, and can efficiently assist in the grading of cork discs.

[0029] A double-sided image acquisition device is used to simultaneously acquire images of the top and bottom surfaces of a cork disc to obtain image data. Specifically, the double-sided image acquisition device includes an industrial camera and an illumination unit symmetrically arranged vertically. The industrial camera is a CCD camera, and the illumination unit is a ring-shaped uniform light source. The double-sided image acquisition device also includes a locking mechanism for fixing and locking the industrial camera.

[0030] A defect detection and grading device is used to process image data, identify surface defects, and output grading results. The device includes a deep learning module that uses the YOLOv5s algorithm for defect identification. YOLOv5 (You Only Look Once version 5) is a high-efficiency single-stage object detection model that achieves a significant balance between real-time performance and detection accuracy through a modular architecture design. It dynamically adjusts its size using depth coefficients (controlling the number of network layers) and width coefficients (adjusting the number of feature map channels). The YOLOv5s version has only 7.2M parameters, and its lightweight structure facilitates deployment on hardware devices while meeting the rapid response requirements of real-time detection scenarios. The core architecture of the YOLOv5s model is as follows: Figure 6 As shown. The input layer employs Mosaic data augmentation, enhancing robustness for small target detection through four-image stitching and random scaling, and optimizing the initial bounding box size using adaptive anchor box calculation. The backbone network introduces the C3 module and the SPPF module. The former is based on a CSPNet-designed dual-branch residual structure to reduce computational redundancy, while the latter improves the traditional Spatial Pyramid Pooling (SPP) into a serial 5×5 max-pooling layer, further improving running speed while preserving multi-scale receptive fields. The feature fusion layer (Neck) fuses the Bidirectional Feature Pyramid (BiFPN) and the Path Aggregation Network (PANet), enhancing multi-target localization capabilities through cross-scale bidirectional feature interaction. The detection head outputs three sets of feature maps: 20×20, 40×40, and 80×80. Each grid pre-sets three anchor boxes and predicts target box parameters (coordinate offset constrained by Sigmoid, width and height scaling factors limited to 0~4). Cross-grid and cross-branch matching strategies are used to expand the number of positive samples to optimize training stability. Regarding the loss function, the localization loss uses CIoU loss to comprehensively evaluate the overlap rate, center distance, and aspect ratio similarity, while the classification and confidence loss are calculated based on binary cross-entropy (BCE loss). In the post-processing stage, redundant detection boxes are removed through non-maximum suppression (NMS).

[0031] Furthermore, the deep learning module employs a CCGS grading strategy to classify the cork discs. Specifically, when a defect with a diameter greater than 3mm is detected in the image, it is considered a defect that is unacceptable. The absence of unacceptable defects is considered "good," and the side with the unacceptable defect is considered "bad." Based on these criteria, the quality grades of the cork discs are divided into: Grade A: "Good" on both sides; Grade B: "Bad" on both sides; Grade C: "Good" on one side and "Bad" on the other.

[0032] A laser marking device is used to mark corresponding marks on the surface of cork discs according to grade results. Specifically, the laser marking device includes a laser marking machine and a marking machine stand, with the laser marking machine placed on the marking machine stand. Based on the need for efficient ablation to remove material and generate clearly visible marks on the sample surface, this embodiment selects a carbon dioxide laser marking machine. For cork disc samples with a diameter of Φ27mm, the marking machine was used to simultaneously mark three samples, with a total time of 0.75s. Calculations show that the average marking time for a single sample is 0.25s, and the marking effect is as follows. Figure 5 As shown. From Figure 5 It can be seen that the marks produced by this laser marking machine are clear and highly recognizable. Furthermore, based on the marking characteristics, the samples were classified into three grades: A, B, and C.

[0033] A pneumatic sorting device is used to directionally blow cork discs to corresponding channels based on their grade. Specifically, the pneumatic sorting device includes: an air pump for providing air supply; in this embodiment, a 24L oil-free silent air pump is used; an air distribution unit, including a pressure regulating valve and a flow meter, for controlling the air volume and connected to a solenoid valve array via a main air pipe; a solenoid valve array containing several independent solenoid valves, each corresponding to a different grade of cork disc, with each group of solenoid valves connected to an air nozzle via an air pipe; and an air nozzle located above the sorting station for focusing high-pressure airflow onto the edge of the cork disc, blowing it to the corresponding channel for sorting. This device uses an air pump to precisely control the air volume; the air pump is installed on the outside of the cabinet, facilitating adjustment and data reading by the operator. The solenoid valves of the cylinder operate in a pulse manner when receiving extension and retraction commands, offering significant advantages such as fast response speed and long service life.

[0034] The collection device includes several channels located around the periphery of the mechanical turntable. The center of the inlet of each channel is on the same plane as the plane of the mechanical turntable. The number of channels is determined according to the grade of the cork discs, and each channel corresponds to a collection slot. In this embodiment, the cork discs are divided into three grades: A, B, and C, thus corresponding to three channels.

[0035] The PLC control unit is used to coordinate the timing of the actions of various devices. The PLC control unit is housed in a PLC control box, which serves as an operator console to house the various device components.

[0036] The PLC control device includes: a core controller, which adopts a Siemens S7-200smart PLC, equipped with a digital input module (16×24VDC input), a digital output module (16×24VDC output), an analog input module (8 channels), and an analog output module (4 channels); and a communication interface, which integrates an RS485 / RS232 serial communication interface and 8 RJ45 Ethernet interfaces for real-time interaction with the host computer software system to exchange detection data and instructions.

[0037] Specifically, the signal chain of the PLC control device is as follows: enter: Receive the position signal from the photoelectric switch to trigger the industrial camera to acquire images; Receive the grading results (A / B / C character commands) from the grading subsystem and trigger laser marking and pneumatic sorting; Output: Send pulse commands to the mechanical turntable motor to control the rotation angle and speed; Send a grade marking command to the laser marking machine to drive the laser marking machine to mark; Send a sorting channel selection signal to the pneumatic sorting solenoid valve array to control the directional injection of high-pressure airflow.

[0038] The PLC control device also monitors the air circuit pressure in real time through the analog input module, triggering the safety relief valve when the pressure is over-pressurized; it also receives fault feedback signals from the vibratory feeder controller and automatically suspends the system operation.

[0039] Implementation Method Two: This embodiment is a further illustrative example of a cork disc surface quality inspection system described in Embodiment 1.

[0040] The cork disc surface quality inspection system described in Embodiment 1 is used to inspect the cork discs, and the steps are as follows: Step 1: The cork disc samples are precisely sorted by a vibratory feeder and then conveyed to the feeding port; Step 2: By controlling the solenoid valve switch, the feed port can be in a conducting or closed state; Step 3: Using a mechanical turntable, the sample is transferred to the corresponding testing station. Two industrial cameras, one above and one below, are used to photograph the front surface of the sample, and the resulting images are transmitted to the computer. Step 4: The computer uses deep learning algorithms to calculate the number and size of defects, thereby determining the grade of the cork disc.

[0041] Step 5: Send the determined level result to the PLC; Step Six: The PLC then sends relevant instructions to the laser and air valve to perform marking and air blowing (material selection) operations, thereby completing the entire sorting process of the cork discs; Step 7: After the test, shut down the computer, laser, automatic feeding system and rotary positioning platform in sequence, and properly store the cork disc specimens in the designated storage container.

[0042] This implementation method involves testing cork discs used in the manufacture of high-end badminton shuttlecock heads, the quality of which directly affects the performance and user experience of the final product. The sample has a diameter of 27 mm and a thickness of 6-7 mm. The system aims to identify and evaluate natural defects such as porosity and cracks on both sides of the sample, and classify the sample into three quality levels (A, B, and C) based on a pre-defined CCGS (Cork Chip Grading Strategy). This strategy uses whether the defect diameter is greater than 3 mm as the core criterion.

[0043] The core workflow of the automated system constructed in this embodiment is based on, as follows: Figure 7 The program logic shown is expanded, and the specific steps are explained below: Step 1: Automatic feeding and sorting The cork discs to be tested are placed in batches into a vibratory feeder. By activating the vibratory feeder controller and adjusting its vibration frequency and amplitude, the samples are arranged in an orderly manner and conveyed stably along a spiral track, and then enter the subsequent conveying station through the discharge port. This step provides the system with a continuous and uniform material flow.

[0044] Step Two: Precise Positioning and Delivery Cork discs are conveyed along a track and positioned on a dial of a high-precision rotary positioning system. To avoid jamming, a cylinder at the feeding port is precisely controlled by a PLC program to ensure that only one sample enters the station at a time. Once a photoelectric switch detects that the sample is in place, it triggers a high-precision rotary motor to drive the mechanical turntable to rotate by a preset angle, sequentially conveying the sample to the vision inspection, laser marking, and pneumatic sorting areas.

[0045] Step 3: Double-sided image acquisition As the sample moves into the visual inspection area, the industrial CCD cameras positioned on its upper and lower sides are simultaneously triggered to perform image acquisition. The upper camera captures the top surface of the sample, while the lower camera captures its bottom surface through the center hole of the dial, thus achieving complete capture of information from both sides of the sample. The cameras employ a global exposure mode, supplemented by a uniform illumination system, to ensure high image clarity and signal-to-noise ratio.

[0046] Step 4: Defect Identification and Grading After receiving the image data, the PC-based software system calls a detection model based on the deep learning YOLOv5s algorithm for analysis. This model is used to automatically identify defect features on the surface of the cork disc. Subsequently, the system executes the CCGS grading strategy: based on the criterion that defects with a diameter greater than 3mm are unacceptable, the system counts the number of unacceptable defects on each surface and completes the grading assessment.

[0047] Grade A: No defects on either side.

[0048] Grade B: Both sides have defects.

[0049] Grade C: One side has a defective product.

[0050] Step 5: Command Transmission The computer completes the rating assessment within 0.1 seconds and sends the assessment result (A, B, or C) to the PLC controller via the communication interface.

[0051] Step Six: Laser Marking and Pneumatic Sorting After receiving the level command, the PLC controller coordinates the actions of subsequent execution units: (1) Laser marking: When the sample is transferred to the laser marking area, the PLC controls 30W according to the grade instruction. Laser marking machine marks the corresponding grade mark on the sample surface (effect as shown). Figure 5 The marking process for a single sample takes approximately 0.25 seconds.

[0052] (2) Pneumatic sorting: The marked samples continue to be conveyed to the sorting area. The PLC controls the pneumatic valves and nozzles of the corresponding channels to open instantaneously according to their grade, generating high-pressure airflow to accurately blow the samples into the corresponding A, B, and C grade channels and drop them into the corresponding collection tanks.

[0053] Step 7: Cyclic Operation After completing the entire process for a single sample, the system automatically enters the next cycle, forming a continuous and efficient automated operation. Once all samples have been processed, the sorted finished products can be retrieved from each collection tank.

[0054] Results and Analysis To verify the performance of the device and method described in this invention, their key technical indicators were quantitatively tested and analyzed.

[0055] Core Algorithm Performance Verification In this embodiment, the mAP@0.5 metric is used as the main evaluation standard to evaluate the overall performance of the defect detection model. This metric is defined as the mean average precision when the IoU between the predicted bounding box and the ground truth bounding box is ≥ 0.5, which can intuitively reflect the detection accuracy of the model under a set threshold.

[0056] During the model training process of the deep learning module in the defect detection and grading device, the convergence speed and training quality are evaluated by monitoring the changing trend of the loss function. Figure 8 The graph shows the model's performance on the training and validation sets after 300 training epochs, with the training set results in the upper half and the validation set results in the lower half. The loss function includes: (1) Box_loss: used to measure the positional deviation between the predicted box and the ground truth box. The smaller the value, the higher the localization accuracy. (2) Obj_loss: used to measure the confidence prediction error of the model in whether the target exists in the candidate region. Its continuous decrease indicates that the model’s ability to judge whether the target appears or not is gradually improving. (3) Cls_loss: used to measure the multi-class classification error, but since the dataset involved in this embodiment is only labeled with a single class "Defect", Cls_loss is always zero and no classification learning is required.

[0057] In the model performance evaluation phase, in addition to mAP@0.5, mAP@0.5:0.95 was calculated as a supplementary metric. This metric represents the average mAP value across IoU thresholds from 0.5 to 0.95 (step size 0.05) to reflect the model's detection stability under more stringent conditions. Validation set results show that the model achieves approximately 98% average accuracy under IoU ≥ 0.5 conditions, while maintaining a high mAP value even at higher IoU thresholds, fully demonstrating its excellent detection performance.

[0058] In addition, precision-recall (PR) curves were plotted. Figure 9 The curves illustrate the trade-off between precision and recall at different confidence thresholds. As can be seen from the curves, the model achieves both high recall of potential defects and high precision in the detection of surface defects on cork discs, meeting the dual requirements of accuracy and reliability in practical applications.

[0059] Defects in cork discs mainly include common defects, machine-damaged defects, and rot defects. See [link to relevant documentation]. Figure 10 Based on application requirements, common defects are classified into different levels, while machine-damaging and corrosive defects are judged as "unusable".

[0060] Figure 11The detection results for three defect types are shown. The detection process is as follows: First, a neural network marks suspected defective areas with bounding boxes. Then, the defect size is calculated, and the number of defects larger than Φ3 mm is counted. If the number exceeds a preset threshold, the wood chip is deemed unqualified. Specific details are as follows: (1) Common defects: All defects were correctly marked, but due to the presence of defects larger than Φ3 mm, they were ultimately judged as “Bad”; (2) Machine destructive defects: Their color characteristics are similar to those of ordinary defects, which can easily lead to misjudgment; In this case, although a "Bad" judgment was obtained, the credibility was insufficient because it was not based on stable feature extraction. (3) Rotten defects: The color contrast is obvious, but due to the dense and complex distribution, some are missed, affecting the accuracy of the judgment.

[0061] In summary, although different defect types vary in color and shape, the embodiments of the present invention can still achieve effective identification and classification of various defects.

[0062] like Figure 12 As shown, the same batch of cork discs was inspected under low, medium, and high brightness environments. The results show that as the ambient brightness increases, the background brightness strengthens, and the contrast between defects and the background improves, which is beneficial for defect identification and localization. However, small, light-colored defects are easily "masked" by the background under high brightness, leading to missed detections or underestimation of their size. To enhance the system's robustness to changes in illumination, this implementation method collected samples of the same wood discs under various brightness conditions during the model training phase to enrich the dataset. Experiments verified that dataset diversification significantly reduced network overfitting and improved the model's generalization ability. Therefore, although changes in illumination have a slight impact on the detection rate of some small, light-colored defects, the overall detection results remain stable and do not produce substantial deviations.

[0063] To verify the stability of this implementation method, three repeated inspections and marking tests were conducted on the same sample. The test subjects were one sample each of Grade A, Grade B, and Grade C, which were manually classified. The identification results are shown below. Figure 13 , Figure 15 and Figure 17 The labeling results are shown below. Figure 14 , Figure 16 and Figure 18 .

[0064] Comparative analysis shows that the identification results and marking positions of the same grade of samples were highly consistent in three repeated tests, proving that the system has excellent stability and reliability and meets the requirements of industrial applications.

[0065] The system described in this embodiment enables efficient and intelligent identification and automatic sorting of cork discs.

[0066] The system proposed in this embodiment constructs a typical defect detection model for cork disc surfaces based on the YOLOv5s convolutional neural network, and establishes a cork disc grading standard (CCGS) accordingly. Both are integrated into the cork disc surface quality inspection system. Experimental results show that the system achieves an mAP of 98% under the condition of IoU ≥ 0.5, accurately identifying and labeling defect features of varying sizes and shapes on the cork disc surface, significantly improving detection accuracy and robustness.

[0067] To meet the requirements of classifying cork discs into three quality grades (A, B, and C), this system integrates electrical and pneumatic control. Through the coordinated control of rotary positioning, conveying, sorting cylinders, and air nozzles by a PLC, it achieves automatic classification of cork discs of different grades. It has the advantages of compact structure, high efficiency, high degree of automation, and simple on-site maintenance.

[0068] This system uses 30W Laser marking machines grade the surfaces of sorted cork discs. The marking process features rapid material removal, high marking speed (approximately 0.25 s / disc), high contrast, fast communication response, and customizable identifiers. This design meets the high-throughput requirements of industrial production lines while ensuring marking quality and durability.

[0069] In summary, this invention combines deep learning detection algorithms, disc-type mechanical conveying, and PLC pneumatic sorting. By employing laser marking and other technologies, a complete solution has been developed for cork discs, from defect detection to automated sorting and surface marking. This system can complete single-disc grading and simultaneously perform sorting and marking within 0.1 seconds, significantly improving production efficiency while ensuring classification accuracy and marking standardization, demonstrating broad prospects for industrial applications.

Claims

1. A cork disc surface quality inspection system, characterized in that, The system includes: A vibrating feeder is used to arrange cork discs in an orderly manner and output them one by one; A rotary positioning device is used to receive cork discs and rotate them at a preset angle to the inspection station, marking station, and sorting station. A double-sided image acquisition device is used to simultaneously acquire images of the top and bottom surfaces of a cork disc to obtain image data; A defect detection and grading device is used to process image data, identify surface defects, and output grading results. Laser marking device, used to mark corresponding marks on the surface of cork discs according to grade results; A pneumatic sorting device is used to directionally blow cork discs into the corresponding channels according to the grade results; A PLC control unit is used to coordinate the timing of actions of various devices.

2. The cork disc surface quality inspection system according to claim 1, characterized in that, The vibratory feeding device includes a vibratory feeder and a controller. The vibratory feeder is equipped with a spiral track, and the controller is used to control the vibration amplitude of the vibratory feeder.

3. The cork disc surface quality inspection system according to claim 1, characterized in that, The vibratory feeding device also includes a photoelectric detection unit for monitoring the feeding status of cork discs and is linked with the PLC control module to provide real-time feedback on the conveying status.

4. The cork disc surface quality inspection system according to claim 1, characterized in that, The rotary positioning device includes a mechanical turntable, a feeding unit, and a photoelectric switch. The mechanical turntable is configured as a dial structure with a locking position, used to drive the cork disc to make a circular motion. The feeding unit is used to control the cork disc to enter the locking position at a uniform speed. The photoelectric switch is used to detect the cork disc in place and trigger the turntable to rotate.

5. The cork disc surface quality inspection system according to claim 4, characterized in that, The feeding unit includes: a push rod for pushing cork discs; a pressing cylinder for driving the push rod to extend and retract via air pressure, controlling the cork discs to enter the locking position; a solenoid valve for receiving pulse signals from the PLC control device and switching the air circuit on and off; an air circuit adjustment unit for controlling the extension and retraction speed and thrust of the push rod; and a position sensor for monitoring the push rod stroke in real time and feeding back to the PLC control device.

6. The cork disc surface quality inspection system according to claim 1, characterized in that, The dual-sided image acquisition device includes an industrial camera and an illumination unit arranged symmetrically at the top and bottom. The industrial camera is a CCD camera, and the illumination unit is a ring-shaped uniform light source.

7. The cork disc surface quality inspection system according to claim 6, characterized in that, The dual-sided image acquisition device also includes a locking mechanism for fixing and locking the industrial camera.

8. The cork disc surface quality inspection system according to claim 1, characterized in that, The defect detection and grading device includes a deep learning module, which uses the YOLOv5s algorithm for defect identification and the CCGS grading strategy for grading cork discs.

9. The cork disc surface quality inspection system according to claim 8, characterized in that, The defect identification using the YOLOv5s algorithm includes the following steps: Mosaic data augmentation is employed to improve the robustness of small target detection through four-image stitching and random scaling, and the initial box size is optimized by combining adaptive anchor box calculation. The backbone network introduces the C3 module and the SPPF module. The C3 module is based on the CSPNet design with a dual-branch residual structure to reduce computational redundancy. The SPPF module improves the traditional spatial pyramid pooling into a serial 5×5 max pooling layer. The feature fusion layer integrates a bidirectional feature pyramid and a path aggregation network, enhancing multi-target localization capabilities through cross-scale bidirectional feature interaction. The detection head outputs three sets of feature maps: 20×20, 40×40, and 80×80. Each grid has three preset anchor boxes and predicts target box parameters. The number of positive samples is expanded through cross-grid and cross-branch matching strategies. The localization loss uses CIoU loss to comprehensively evaluate the similarity of overlap rate, center distance and aspect ratio. The classification and confidence loss are calculated based on binary cross-entropy. In the post-processing stage, redundant detection boxes are removed by non-maximum suppression.

10. The cork disc surface quality inspection system according to claim 1, characterized in that, The pneumatic sorting device includes: An air pump is used to provide an air source; The gas distribution unit includes a pressure regulating valve and a flow meter, which are used to control the gas volume and are connected to a solenoid valve array through the main gas pipe; The solenoid valve array contains a group of independent solenoid valves, each corresponding to a different grade of cork disc, and each group of solenoid valves is connected to the jet nozzle via an air tube. The jet nozzle, located above the sorting station, is used to focus high-pressure airflow onto the edge of the cork disc, blowing the cork disc into the corresponding channel for sorting.

Images