Real-time detection system and compensation method for defects of RFID tag production line based on machine vision

Abstract

The application discloses a kind of based on machine vision's RFID label production line defect real-time detection system and compensation method, belong to intelligent manufacturing and quality control technical field.System includes image acquisition module, defect detection module, compensation execution module and feedback control module, based on defect type and position, trigger hierarchical compensation strategy, including real-time adjustment printing parameter, laser precision repair defect area or mechanical arm automatic rejection waste, and through industrial internet of things platform synchronous update process parameter library.Innovative point is: ① propose multimodal defect data fusion method, improve the detection robustness under complex conditions;② design lightweight double-branch detection network, consider detection speed and accuracy;③ develop compensation instruction cooperative control algorithm, realize defect repair and production line beat seamless convergence.This scheme can reduce defect miss rate to 0.5% or less, compensation response time is shortened to 200ms, significantly improve production line good product rate and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and quality control technology, and in particular to a real-time defect detection system and compensation method for RFID tag production lines based on machine vision. Background Technology

[0002] As a key component of the Internet of Things (IoT), the production of Radio Frequency Identification (RFID) tags involves multiple precision processes, including printing, chip mounting, and antenna etching. Current mainstream inspection methods rely on manual visual inspection or traditional machine vision technology, which suffers from low efficiency, high false negative rates (approximately 3%-5%), and delayed defect compensation. Especially in high-speed production lines (≥200 tags / minute), existing technologies struggle to capture micron-level defects (such as antenna breakage or chip misalignment) in real time, and lack dynamic compensation mechanisms after defect detection, resulting in persistently high scrap rates (approximately 4%-8%), severely restricting production yield and cost control.

[0003] Referring to the Chinese patent "A Method and System for Defect Detection of RFID Tags," a defect detection scheme based on multispectral imaging and machine learning algorithms is proposed, which improves the ability to identify surface and internal defects through the fusion imaging of visible and infrared light. However, this scheme has significant limitations: 1) The detection model is not optimized for high-speed production lines, and the processing time for a single tag is ≥500ms, which is difficult to meet real-time requirements; 2) No defect compensation mechanism is integrated, and the detection results are only used for offline sorting, which cannot dynamically adjust production parameters or repair defects, resulting in defect compensation lagging behind the production line cycle time; 3) The multispectral data fusion method is singular, and the false detection rate is relatively high (approximately 2.5%) under complex lighting or material differences.

[0004] Based on existing technologies, current RFID tag production line defect management faces the following core problems: 1) Disconnect between detection and compensation: Traditional systems only achieve defect identification, failing to form a closed loop of "perception-decision-execution," leading to defect accumulation due to compensation lag; 2) Poor dynamic adaptability: Fixed threshold or rule-driven compensation strategies cannot adapt to multiple types of defects (such as blurred printing and broken antennas requiring differentiated processing); 3) Low production line collaboration efficiency: The lack of timing synchronization algorithms between the detection module and the execution mechanism (such as laser repair and robotic arms) easily causes production line rhythm disorder. This invention aims to overcome the above bottlenecks and construct an integrated solution of real-time detection and intelligent compensation. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a machine vision-based RFID tag production line defect real-time detection system and compensation method that solves the problems of disconnect between detection and compensation, poor dynamic adaptability, and low production line collaboration efficiency.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a machine vision-based real-time defect detection system for RFID tags on a production line, comprising:

[0009] Image acquisition module: The multispectral imaging unit acquires surface and internal structure images of RFID tags transmitted on the production line. The imaging spectrum covers visible light (400-700nm) and near-infrared (800-1200nm), and the spatial resolution is not less than 20μm / pixel.

[0010] Defect detection module: It is equipped with a lightweight dual-branch deep learning network. The first branch uses an improved YOLOv5 model for defect localization, and the second branch uses ResNet-18 for defect classification, outputting defect type and coordinate information.

[0011] Compensation execution module: Triggers a graded compensation strategy based on the defect type, including a dynamic adjustment unit for printing parameters, a laser repair unit, and a robotic arm rejection unit;

[0012] Feedback control module: Feeds back defect data and compensation results to the production line PLC through the industrial IoT platform, and updates the process parameter library in real time.

[0013] As a preferred embodiment of the machine vision-based real-time defect detection system for RFID tags on a production line according to the present invention, the imaging parameters of the multispectral imaging unit satisfy the following:

[0014] The surface imaging bands are 450nm, 630nm, and 850nm, which are used to detect defects in the printed layer.

[0015] The internal imaging band is 1050nm, with a penetration depth ≥200μm, used to detect antenna breakage and chip misalignment; the imaging frame rate is synchronized with the production line speed, satisfying the following relationship:

[0016]

[0017] Where F is the imaging frame rate (fps), v is the production line transmission speed (mm / s), d is the minimum defect size of the tag (mm), and η = 1.2 to 1.5 is the redundancy coefficient.

[0018] As a preferred embodiment of the machine vision-based real-time defect detection system for RFID tags on a production line according to the present invention, the loss function of the lightweight dual-branch deep learning network is defined as:

[0019] L total =αL loc +βL cls +γL reg

[0020] Among them, L loc For localization loss (using CIoU loss), L cls For classification loss (focus loss), L reg The weights are regularization terms for model complexity, and their weights satisfy α:β:γ = 3:2:0.5.

[0021] As a preferred embodiment of the machine vision-based RFID tag production line defect real-time detection system described in this invention, the triggering condition for the hierarchical compensation strategy is:

[0022] Level 1 compensation: When the defect area accounts for S defect / S tag When the percentage is ≤5%, the laser repair unit is activated, and the repair path planning meets the following requirements:

[0023]

[0024] Secondary compensation: when 5% defect / S tag When the content is ≤15%, adjust the printing parameters (temperature T±3℃, pressure P±5%).

[0025] Level 3 Compensation: S defect / S tag When the defect rate is >15%, the robotic arm removal unit removes the defect label within t≤200ms;

[0026] Among them, S defect S represents the defect area. tag This represents the label area.

[0027] As a preferred embodiment of the machine vision-based real-time defect detection system for RFID tags on a production line according to the present invention, the network structure optimization of the improved YOLOv5 model includes:

[0028] Replacing the C3 module in the backbone network with the Ghost module reduces the number of parameters by 40%.

[0029] With the addition of coordinate attention (CA), the feature map weights are calculated as follows:

[0030]

[0031] Among them, F c δ(·) represents the channel characteristics and is the Sigmoid function.

[0032] In a preferred embodiment of the machine vision-based real-time defect detection system for RFID tag production lines described in this invention, the power adjustment formula of the laser repair unit is as follows: ​

[0033]

[0034] Among them, A defect Let A0 be the defect area, which is 0.1 mm. 2 k = 12 W / mm 2 b = 5W.

[0035] As a preferred embodiment of the machine vision-based RFID tag production line defect real-time detection system described in this invention, the feedback control module dynamically optimizes production line parameters using a PID algorithm.

[0036]

[0037] Where e(t) is the defect rate deviation, and the proportionality coefficient K p =0.8, integral coefficient K i =0.2, differential coefficient K d =0.1.

[0038] As a preferred embodiment of the machine vision-based RFID tag production line defect compensation method described in this invention, it includes the following steps:

[0039] Step S1: The multispectral imaging unit simultaneously acquires images of the label's surface and interior;

[0040] Step S2: The dual-branch deep learning network outputs the defect type and location. If the confidence level conf≥0.9, it is determined to be a defect.

[0041] Step S3: Select the compensation method according to the defect level, and re-inspect through an industrial camera after execution. If the defect still exists, trigger secondary compensation or rejection.

[0042] As a preferred embodiment of the machine vision-based RFID tag production line defect compensation method described in this invention, wherein: the confidence calculation in step S2 employs an improved Softmax function:

[0043]

[0044] Where ρ = 0.5 is a temperature coefficient used to enhance the classification confidence and discrimination, z i This represents the original output value of the i-th category (i.e., the unnormalized logits), with the smoothness of the confidence distribution adjusted by the temperature coefficient ρ.

[0045] As a preferred embodiment of the machine vision-based RFID tag production line defect compensation method described in this invention, wherein: the response time t of the compensation command in step S3 is... response satisfy:

[0046]

[0047] Among them, L tag v represents the tag length. line For production line speed, t process ≤50ms is the data processing delay.

[0048] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the machine vision-based RFID tag production line defect compensation method described in the first aspect of the present invention.

[0049] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the machine vision-based RFID tag production line defect compensation method described in the first aspect of the present invention.

[0050] The beneficial effects of this invention are:

[0051] 1. High-precision real-time detection and low false negative rate

[0052] By fusing visible and near-infrared bands (400–1200 nm) with multispectral imaging technology and combining it with a lightweight dual-branch deep learning network (improved YOLOv5+ResNet-18), the system can simultaneously detect surface printing defects and internal structural defects (such as antenna breakage and chip misalignment). The defect identification accuracy is over 99.5%, the false negative rate is reduced to below 0.5%, and the single-tag detection time is ≤50ms, meeting the real-time requirements of high-speed production lines (≥300 tags / minute).

[0053] 2. Dynamic closed-loop compensation and efficient repair

[0054] A graded compensation strategy (parameter adjustment, laser repair, and automatic rejection) is proposed. By combining PID feedback control and laser power adjustment algorithm, the compensation response time is shortened to within 200ms, the defect repair success rate is ≥95%, and the scrap rate is reduced from 4% to 8% of the traditional method to less than 1%.

[0055] 3. Intelligent production line collaboration and process optimization

[0056] The feedback control module, based on an industrial IoT platform, synchronizes defect data with production parameters (temperature, pressure, speed) in real time, and dynamically adjusts the PID control (K... p =0.8,K i =0.2,K d=0.1) It enables production line self-optimization, reduces the frequency of manual intervention by more than 80%, and supports continuous iterative updates of the process parameter library to adapt to the flexible production needs of multiple types of RFID tags (high frequency, ultra-high frequency).

[0057] 4. Cost savings and scalability

[0058] By adopting a lightweight network (replacing the C3 module with the Ghost module, reducing the number of parameters by 40%) and a multimodal data fusion method, hardware resource consumption is reduced by 35%, while being compatible with the upgrade and transformation of existing production line equipment. The deployment cost is only 60% to 70% of that of traditional solutions, making it suitable for large-scale mass production scenarios.

[0059] In summary, this invention significantly improves the yield, efficiency, and economy of RFID tag production through closed-loop management of the entire process of detection, compensation, and feedback, providing a highly reliable quality control solution for intelligent manufacturing. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram of a machine vision-based real-time defect detection system for RFID tag production lines in Example 1.

[0062] Figure 2 This is a flowchart of a machine vision-based RFID tag production line defect compensation method in Example 2. Detailed Implementation

[0063] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0064] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0065] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0066] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a machine vision-based real-time defect detection system for RFID tag production lines, comprising:

[0067] Image acquisition module: The multispectral imaging unit 101 is used to acquire surface and internal structure images of RFID tags transmitted on the production line. The imaging spectrum range covers visible light (400-700nm) and near-infrared (800-1200nm), and the spatial resolution is not less than 20μm / pixel.

[0068] Defect detection module: Equipped with a lightweight dual-branch deep learning network 201. The first branch uses an improved YOLOv5 model for defect localization, and the second branch uses ResNet-18 for defect classification, outputting defect type and coordinate information.

[0069] Compensation execution module: Triggers a graded compensation strategy based on the defect type, including a printing parameter dynamic adjustment unit 301, a laser repair unit 302, and a robotic arm rejection unit 303;

[0070] Feedback control module: Feeds back defect data and compensation results to the production line PLC through the industrial IoT platform 401, and updates the process parameter library in real time.

[0071] The imaging parameters of the multispectral imaging unit 101 satisfy:

[0072] The surface imaging bands are 450nm, 630nm, and 850nm, which are used to detect defects in the printed layer.

[0073] The internal imaging band is 1050nm, with a penetration depth ≥200μm, used to detect antenna breakage and chip misalignment; the imaging frame rate is synchronized with the production line speed, satisfying the following relationship:

[0074]

[0075] Where F is the imaging frame rate (fps), v is the production line transmission speed (mm / s), d is the minimum defect size of the tag (mm), and η = 1.2 to 1.5 is the redundancy coefficient.

[0076] The loss function of the lightweight two-branch deep learning network 201 is defined as follows:

[0077] L total =αL loc +βL cls +γL reg

[0078] Among them, L loc For localization loss (using CIoU loss), Lcls For classification loss (focus loss), L reg The weights are regularization terms for model complexity, and their weights satisfy α:β:γ = 3:2:0.5.

[0079] The triggering conditions for the tiered compensation strategy are:

[0080] Level 1 compensation: When the defect area accounts for S defect / S tag When the percentage is ≤5%, the laser repair unit 302 is activated, and the repair path planning satisfies the following:

[0081]

[0082] Secondary compensation: when 5% defect / S tag When the content is ≤15%, adjust the printing parameters (temperature T±3℃, pressure P±5%).

[0083] Level 3 Compensation: S defect / S tag When the defect rate is >15%, the robotic arm removal unit 303 removes the defect label within t≤200ms;

[0084] Among them, S defect S represents the defect area. tag This represents the label area.

[0085] The improved YOLOv5 model's network structure optimizations include:

[0086] Replacing the C3 module in the backbone network with the Ghost module reduces the number of parameters by 40%.

[0087] With the addition of coordinate attention (CA), the feature map weights are calculated as follows:

[0088]

[0089] Among them, F c δ(·) represents the channel characteristics and is the Sigmoid function.

[0090] The power adjustment formula for the laser repair unit 302 is as follows:

[0091]

[0092] Among them, A defect Let A0 be the defect area, which is 0.1 mm. 2 k = 12 W / mm 2 b = 5W.

[0093] Feedback control module 401 dynamically optimizes production line parameters using a PID algorithm: ​

[0094]

[0095] Where e(t) is the defect rate deviation, and the proportionality coefficient K p =0.8, integral coefficient K i =0.2, differential coefficient K d =0.1.

[0096] Example 2, refer to Figure 2 This is the second embodiment of the present invention, which provides a machine vision-based method for defect compensation in RFID tag production lines, including the following steps:

[0097] Step S1: The multispectral imaging unit 101 simultaneously acquires images of the label's surface and interior;

[0098] 1. Hardware Configuration and Light Source Design

[0099] The multispectral imaging unit consists of two optical modules:

[0100] Surface imaging module: It adopts a high-resolution visible light camera (resolution ≥20μm / pixel) and is equipped with a three-band LED array light source of 450nm (blue light), 630nm (red light) and 850nm (near infrared) to cover the needs of printing layer color and surface morphology detection.

[0101] Internal imaging module: It adopts a near-infrared camera (wavelength 1050nm, penetration depth ≥200μm) and a coaxial light source, and uses an optical filter to shield stray light. It is dedicated to capturing the antenna copper foil layer and the internal structure of the chip package.

[0102] Synchronous triggering mechanism: The transmission position of the production line is monitored in real time by photoelectric sensors (or encoders). When the RFID tag enters the imaging area, the two modules are triggered to expose synchronously (time deviation ≤ 1ms) to ensure that the surface and internal images are aligned in time and space.

[0103] 2. Dynamic adaptation of imaging parameters

[0104] Frame rate matching with production line speed: The imaging frame rate is dynamically adjusted based on the production line transmission speed v (unit: mm / s) and the minimum defect size d of the label (unit: mm).

[0105]

[0106] For example, when v = 300 mm / s and d = 0.2 mm, F = 1800 fps is calculated, ensuring the overlap rate between adjacent frames.

[0107] ≥20%, to avoid missed detection.

[0108] Adaptive exposure control: Based on surface reflectivity and material light transmission characteristics, the light source intensity and camera exposure time (range: 10μs~10ms) are adjusted in real time through PID algorithm to ensure that the imaging signal-to-noise ratio (SNR) of different material labels (such as PET, paper) is ≥40dB.

[0109] 3. Image preprocessing and calibration

[0110] Spatial alignment: A checkerboard calibration plate is used to geometrically calibrate the field of view of the surface and internal cameras, and pixel-level alignment (error ≤ 2μm) is achieved through an affine transformation matrix.

[0111] Multispectral fusion: Weighted fusion of visible light three-band images (weighting coefficient: w) 450 =0.4, w 630 =0.3, w 850 =0.3), enhancing the contrast of the printed pattern; the internal near-infrared image is enhanced with an unsharp mask to improve antenna edge details.

[0112] Data encapsulation: The synchronously acquired multispectral images (surface RGB + internal NIR) are bound with timestamps and production line speed information and transmitted to the defect detection module 201 via gigabit Ethernet, with a single frame transmission delay of ≤5ms.

[0113] 4. Anti-interference and stability assurance

[0114] Enclosed imaging cavity: Light-shielding materials are used to isolate ambient light, and the interior is filled with nitrogen to prevent mirror condensation.

[0115] Vibration compensation: In high-speed production line scenarios, mechanical vibration is monitored in real time by inertial sensors, and the position of the optical platform is adjusted by servo motors (compensation accuracy ±5μm) to eliminate motion blur.

[0116] Self-cleaning system: Integrates air curtain device and periodic ultraviolet lamp irradiation to prevent dust from adhering to the lens and ensure long-term imaging stability.

[0117] It should be noted that, through the above implementation method, step S1 can simultaneously acquire surface and internal images with a spatial resolution of ≤0.1mm on a high-speed production line of 300 tags / minute, providing high-precision, low-noise input data for subsequent defect detection and compensation, while ensuring strict synchronization with the production line cycle time to avoid missed detections or false triggers due to imaging delays.

[0118] Step S2: The dual-branch deep learning network 201 outputs the defect type and location. If the confidence level conf≥0.9, it is determined to be a defect.

[0119] In step S2, the confidence score is calculated using a modified Softmax function:

[0120]

[0121] Where ρ = 0.5 is a temperature coefficient used to enhance the classification confidence and discrimination, z i This represents the original output value of the i-th category (i.e., the unnormalized logits), with the smoothness of the confidence distribution adjusted by the temperature coefficient ρ.

[0122] 1. Dual-branch network architecture design

[0123] Input data:

[0124] The multispectral image (RGB three channels) of the tag surface and the internal near-infrared image (NIR single channel) acquired synchronously in step S1 are received and stitched together into a four-channel input (size: 640×640×4).

[0125] Branching structure:

[0126] Locating branches (improved YOLOv5):

[0127] Backbone Network: The C3 module of YOLOv5 is replaced with the Ghost module (reducing the number of parameters by 40%), and a coordinate attention (CA) mechanism is embedded to enhance sensitivity to minor defects. Output Layer: Outputs the bounding box coordinates (x, y, w, h) and location confidence (conf). loc .

[0128] Classification branch (lightweight ResNet-18):

[0129] Feature extraction: The candidate region (ROI) output by the localization branch is received, adjusted to a size of 224×224 through adaptive pooling, and then input into ResNet-18 for feature extraction.

[0130] Classification header: Outputs the probability p of the defect category i (Including 6 categories such as blurry printing, broken antenna, and misaligned chip) and classification confidence level conf cls .

[0131] 2. Confidence fusion and defect determination

[0132] Joint confidence calculation:

[0133] Based on the combined localization and classification results, calculate the final confidence score: conf = conf loc ×conf cls

[0134] in:

[0135] conf locThe probability of target existence output by the localization branch (after Sigmoid activation);

[0136] conf cls Calculated from the classification branch using the improved Softmax function:

[0137]

[0138] In the formula z i ρ represents the logits value of the i-th class. ρ reduces the smoothness of the classification probability distribution and enhances the discriminative power of high-confidence classes.

[0139] Decision logic:

[0140] If conf ≥ 0.9, it is determined to be a valid defect, and the type and coordinates are output.

[0141] If 0.7 ≤ conf < 0.9, trigger the verification mechanism (such as secondary detection of local images);

[0142] If conf < 0.7, it is considered a normal tag.

[0143] 3. Network Training and Optimization

[0144] Loss function:

[0145] Total loss L total Includes positioning loss L loc Classification loss L cls and regularization term L reg :

[0146] L total =3L loc +2L cls +0.5L reg

[0147] L loc CIoU loss is used to measure the overlap between the predicted bounding box and the ground truth bounding box, as well as the distance between their center points.

[0148] L cls Focal loss is used to alleviate the class imbalance problem;

[0149] L reg L2 regularization constraint model complexity.

[0150] Training data:

[0151] The dataset contains 100,000 labeled RFID tag images (surface and interior aligned), covering 6 types of defects and normal samples;

[0152] Data augmentation: random multispectral channel perturbation, motion blur simulation, and Gaussian noise injection.

[0153] Training strategy:

[0154] Pre-training: The localization branch was pre-trained on the COCO dataset, and the classification branch was pre-trained on ImageNet;

[0155] Joint fine-tuning: Freeze the main branch of the localization branch, train only the CA module and the classification branch, and set the learning rate...

[0156] 1×10 -4 ;

[0157] End-to-end optimization: Unfreeze the entire network, use the AdamW optimizer, and set a learning rate of 5×10⁻⁶. -5 .

[0158] 4. Real-time guarantee

[0159] Hardware acceleration:

[0160] The dual-branch network was quantized (FP16 precision) using the TensorRT engine and deployed on NVIDIA Jetson AGXXavier, with a single-frame inference time of ≤15ms.

[0161] Dynamic resource allocation:

[0162] Adjust the ROI quantity based on the production line speed (v):

[0163]

[0164] Where t cls The time taken for classifying a single ROI is approximately 1.2ms, ensuring that the total processing time is ≤50ms.

[0165] It should be noted that this invention has the advantages of high-precision judgment, low false detection rate and real-time response. The confidence threshold of 0.9 is verified by ROC curve, balancing precision (Precision≥98%) and recall (Recall≥96%). Under complex lighting and material variation scenarios, the false detection rate is ≤0.3%. The end-to-end latency from image input to defect judgment is ≤30ms, supporting the high-speed production line requirements of 300 tags / minute.

[0166] Step S3: Select the compensation method according to the defect level, and re-inspect through an industrial camera after execution. If the defect still exists, trigger secondary compensation or rejection.

[0167] The response time t of the compensation command in step S3 response satisfy:

[0168]

[0169] Among them, L tagv represents the tag length. line For production line speed, t process ≤50ms is the data processing delay.

[0170] 1. Defect level classification and compensation strategy matching

[0171] Grading standards:

[0172] Based on the defect type and area ratio (S) output in step S2 defect / S tag Based on location sensitivity (e.g., higher weighting for defects in chip regions), defect levels are dynamically assigned.

[0173] Level 1 Defect (Minor):

[0174] Condition: S defect / S tag ≤5% and located in non-critical areas (such as antenna edges).

[0175] Compensation method: Laser repair unit 302 performs local repair.

[0176] Level 2 Defect (Moderate):

[0177] Condition: 5% defect / S tag ≤15% or located in critical areas (such as chip mounting areas).

[0178] Compensation method: Dynamically adjust printing parameters (temperature T±3℃, pressure P±5%).

[0179] Level 3 Defect (Severe):

[0180] Condition:S defect / S tag >15% or functional failure (such as antenna breakage).

[0181] Compensation method: The robotic arm rejection unit 303 removes the defect label and marks it as scrap.

[0182] Dynamic priority scheduling:

[0183] If multiple defects exist for the same label, compensation will be performed in the order of "Level 3 > Level 2 > Level 1" to ensure that critical defects are handled first.

[0184] 2. Compensation Execution and Parameter Adaptation

[0185] Laser repair unit 302:

[0186] Path planning: Generate the optimal laser path based on the defect shape to minimize repair time.

[0187] ​

[0188] Power adjustment: The laser power is dynamically adjusted according to the defect area Adefect.

[0189]

[0190] Ensure the repair depth matches the material properties (such as PET substrate or copper foil) to avoid overheating or insufficient repair.

[0191] Parameter adjustment unit 301:

[0192] Printing parameters are adjusted in real time using a PID algorithm.

[0193]

[0194] Where e(t) is the deviation between the current defect rate and the target value, and the parameter adjustment period is ≤100ms.

[0195] Robotic arm rejection unit 303:

[0196] Using a Delta robotic arm, the response time is ≤200ms, and the motion trajectory is calculated based on the production line speed v.

[0197]

[0198] Where t delay To compensate for the delay, d tag To ensure accurate capture, the label spacing is adjusted.

[0199] 3. Re-inspection process and secondary compensation logic

[0200] Re-inspection trigger conditions:

[0201] After completing the first or second level compensation, the industrial camera (of the same model as the main imaging unit) is triggered to perform a second imaging of the repair / adjustment area.

[0202] The re-inspection image is transmitted to the dual-branch network (201), and only the original defect area is locally detected, which takes ≤10ms.

[0203] Re-inspection judgment rules:

[0204] Pass: If the confidence level of the retest is conf recheck If the value is less than 0.7, the compensation is considered successful, and the label proceeds to the next process.

[0205] Failure: If conf recheck ≥0.7, triggered based on residual defect level:

[0206] Primary residual defects: Initiate secondary laser repair, increasing power by 10%–20% (e.g., P). laser ×1.15).

[0207] Secondary residual defects: Adjust printing parameters again (such as temperature T±5℃, pressure P±8%) and extend the PID integral time to 200ms.

[0208] Level 3 residual defects or failed secondary repairs: forcibly removed and recorded in the process parameter library for subsequent production optimization.

[0209] Attempt limit:

[0210] A single tag is allowed a maximum of two compensation attempts. If the target is still not met, the tag will be forcibly removed and an alarm will be triggered.

[0211] 4. Timing synchronization and production line collaboration

[0212] Beat matching:

[0213] The total time for compensation and re-inspection must meet the production line transfer constraints:

[0214]

[0215] Among them, t comp To compensate for time, t recheck For the re-inspection time, L tag v represents the tag length. line For production line speed, t process ≤50ms is the data processing delay.

[0216] Exception handling:

[0217] If the compensation timeout or robotic arm malfunctions, the emergency stop protocol will be activated immediately, the MES system will be notified simultaneously, and the system will switch to the redundant execution unit.

[0218] It should be noted that this invention has the advantages of high repair success rate, low false rejection rate, production line compatibility, and process self-optimization. The first repair success rate of first-level defects is ≥95%, and it reaches 99% after the second repair. Through the re-inspection mechanism, the false rejection rate is ≤0.2%. It supports production line speed of ≤400 labels / minute, and the compensation and re-inspection process is seamlessly embedded in the production cycle. Accumulated defect data is fed back to the parameter library to realize continuous iterative improvement of printing and mounting processes.

[0219] This embodiment also provides a computer device applicable to a machine vision-based RFID tag production line defect compensation method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the machine vision-based RFID tag production line defect compensation method proposed in the above embodiment.

[0220] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0221] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements a machine vision-based RFID tag production line defect compensation method as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0222] This invention integrates visible and near-infrared bands (400–1200 nm) using multispectral imaging technology, combined with a lightweight dual-branch deep learning network (improved YOLOv5+ResNet-18), to achieve simultaneous detection of surface printing defects and internal structural defects (antenna breakage, chip misalignment, etc.). The defect identification accuracy reaches over 99.5%, the false negative rate is reduced to below 0.5%, and the single-tag detection time is ≤50ms, meeting the real-time requirements of high-speed production lines (≥300 tags / minute). A graded compensation strategy (parameter adjustment, laser repair, automatic rejection) is proposed, combined with PID feedback control and a laser power adjustment algorithm, shortening the compensation response time to within 200ms, achieving a defect repair success rate ≥95%, and reducing the scrap rate from 4%–8% of traditional methods to below 1%. Based on an industrial IoT platform, a feedback control module synchronizes defect data with production parameters (temperature, pressure, speed) in real time, and dynamically adjusts the PID control (K... p =0.8,K i =0.2,K d =0.1) Achieves production line self-optimization, reducing manual intervention frequency by over 80%, and supports continuous iterative updates of the process parameter library, adapting to the flexible production needs of various types of RFID tags (high frequency, ultra-high frequency). Employing a lightweight network (replacing the C3 module with a Ghost module, reducing parameter count by 40%) and a multimodal data fusion method, hardware resource consumption is reduced by 35%, while remaining compatible with upgrades to existing production line equipment. Deployment costs are only 60%–70% of traditional solutions, making it suitable for large-scale mass production scenarios. This invention significantly improves the yield, efficiency, and economy of RFID tag production through closed-loop management of the entire process of detection-compensation-feedback, providing a highly reliable quality control solution for intelligent manufacturing.

[0223] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A machine vision-based real-time defect detection system for RFID tags on a production line, characterized in that, include: Image acquisition module: The multispectral imaging unit (101) is used to acquire surface and internal structure images of RFID tags transmitted on the production line. The imaging spectrum range covers visible light of 400-700nm and near infrared of 800-1200nm, and the spatial resolution is not less than 20μm / pixel. Defect detection module: Equipped with a lightweight dual-branch deep learning network (201), the first branch uses the YOLOv5 model for defect localization, and the second branch uses ResNet-18 for defect classification, outputting defect type and coordinate information; the network structure optimization of the YOLOv5 model includes: Replacing the C3 module in the backbone network with the Ghost module reduces the number of parameters by 40%. With the addition of coordinate attention mechanism, the feature map weights are calculated as follows: in, As a channel feature, For the Sigmoid function; Compensation execution module: Triggers a graded compensation strategy based on the defect type, including a printing parameter dynamic adjustment unit (301), a laser repair unit (302), and a robotic arm rejection unit (303). The triggering condition for the tiered compensation strategy is: Level 1 compensation: When the defect area accounts for... When the laser repair unit (302) is activated, the repair path planning satisfies: Secondary compensation: When Adjust printing parameters as needed; Level 3 compensation: At that time, the robotic arm removal unit (303) removes the defect label within t≤200ms; in, The defect area is... For the label area; Feedback control module: Feeds back defect data and compensation results to the production line PLC through the industrial IoT platform (401) and updates the process parameter library in real time.

2. The real-time defect detection system for RFID tags on a production line based on machine vision as described in claim 1, characterized in that, The imaging parameters of the multispectral imaging unit (101) satisfy: The surface imaging bands are 450nm, 630nm, and 850nm, which are used to detect defects in the printed layer. The internal imaging band is 1050nm, and the penetration depth is ≥200μm, which is used to detect antenna breakage and chip misalignment. The imaging frame rate is synchronized with the production line speed, satisfying the following relationship: Where F is the imaging frame rate, v is the production line transmission speed, d is the minimum defect size of the tag, and η = 1.2~1.5 is the redundancy coefficient.

3. The real-time defect detection system for RFID tags on a production line based on machine vision as described in claim 1, characterized in that, The loss function of the lightweight dual-branch deep learning network (201) is defined as: in, To pinpoint the loss, For classifying losses, The model complexity regularization term has weight coefficients that satisfy α:β:γ=3:2:0.

5.

4. The real-time defect detection system for RFID tags on a production line based on machine vision as described in claim 1, characterized in that, The power adjustment formula of the laser repair unit (302) is as follows: in, The defect area is... =0.1mm 2 k=12W / mm 2 b=5W.

5. The real-time defect detection system for RFID tags on a production line based on machine vision as described in claim 1, characterized in that, The feedback control module (401) dynamically optimizes production line parameters using a PID algorithm: in, For defect rate deviation, the proportionality coefficient =0.8, integral coefficient =0.2, differential coefficient =0.

1.

6. A machine vision-based method for defect compensation in an RFID tag production line, which is implemented based on a machine vision-based real-time defect detection system for an RFID tag production line as described in any one of claims 1 to 5, characterized in that, Includes the following steps: Step S1: The multispectral imaging unit (101) simultaneously acquires images of the label surface and interior; Step S2: The dual-branch deep learning network (201) outputs the defect type and location, and the confidence level is... If the value is ≥0.9, it is considered a defect; Step S3: Select the compensation method according to the defect level, and re-inspect through an industrial camera after execution. If the defect still exists, trigger secondary compensation or rejection.

7. The method for defect compensation in an RFID tag production line based on machine vision as described in claim 6, characterized in that, The confidence level calculation in step S2 uses the Softmax function: in, =0.5 is a temperature coefficient used to enhance classification confidence and discrimination. Represents the i-th category The original output value, through the temperature coefficient Adjust the smoothness of the confidence distribution.

8. The method for defect compensation in an RFID tag production line based on machine vision as described in claim 6, characterized in that, The response time of the compensation command in step S3 satisfy: in, For the label length, For production line speed, ≤50ms is the data processing delay.

Images