Coating foreign matter defect on-line monitoring method based on polarized light imaging analysis

Abstract

The invention provides a coating foreign matter defect online monitoring method based on polarized light imaging analysis, and relates to the technical field of imaging analysis, and the method comprises the following steps: determining each monitoring point position of a coating production line, and obtaining a coating process parameter table; carrying out parameter verification on the dual-optical-path polarized imaging module and the linear array camera according to a polarized imaging standard, and carrying out equipment initial configuration; drawing up a coating state association map based on the coating process parameter table, performing association analysis on the coating state association map, foreign matter defects and coating layer defects, and setting a defect identification matrix; and generating a morphological filtering script by combining coating surface imaging data, performing feature extraction on polarization state imaging data acquired by the double-light-path polarization imaging module, and triggering foreign matter defect classification and positioning marking according to a feature matching result. According to the method and the device, the technical problem of poor online monitoring accuracy of the coating foreign matter defect in polarized light imaging analysis in the prior art can be solved, and the technical effect of improving the online monitoring accuracy of the coating foreign matter defect is achieved.

Description

Technical Field

[0001] This application relates to the field of imaging analysis technology, and in particular to an online monitoring method for coating foreign matter defects based on polarized light imaging analysis. Background Technology

[0002] With the widespread application of functional coating materials in high-precision manufacturing fields such as optoelectronic displays, lithium battery separators, and flexible base films, the requirements for surface defect detection in coating production lines are constantly increasing.

[0003] Currently, most existing coating surface inspection technologies are based on single-path visible light imaging or infrared optical detection modes, identifying abnormal areas by analyzing reflected light intensity, texture distribution, or local brightness changes. However, these methods are limited by the optical anisotropy and reflective properties of the coating material. When transparent particles, low-contrast impurities, or internal interlayer bubbles are present on the surface, the dynamic range of the image signal is insufficient, making it difficult to form stable defect features.

[0004] In summary, existing technologies suffer from technical problems such as difficulty in accurately identifying transparent foreign objects and low-contrast foreign objects due to the influence of the reflectivity, color uniformity, and optical interference of the coating material itself. Furthermore, foreign objects are easily confused with defects such as bubbles and wrinkles inside the coating layer, resulting in a high rate of misjudgment and a large rate of missed detection. This further affects the surface quality control accuracy and online defect response capability of the coating production line. Summary of the Invention

[0005] The purpose of this application is to provide an online monitoring method for coating foreign object defects based on polarized light imaging analysis, in order to solve the technical problems in the prior art, which are that visual monitoring is affected by the reflection of the coating material itself, color uniformity and optical interference, making it difficult to accurately identify transparent foreign objects and low-contrast foreign objects, and foreign objects are easily confused with defects such as bubbles and wrinkles inside the coating layer, resulting in a high misjudgment rate and a large missed detection rate, which further affects the surface quality control accuracy and online defect response capability of the coating production line.

[0006] In view of the above problems, this application provides an online monitoring method for coating foreign object defects based on polarized light imaging analysis, including: determining each monitoring point corresponding to the coating production line and assigning weights, and obtaining a coating process parameter table including coating speed, coating thickness, and drying temperature; verifying the parameters of the dual-path polarized light imaging module and the line scan camera using polarized light imaging standards, and performing initial equipment configuration based on the difference with the polarized light imaging standards; based on the coating process parameter table, formulating a coating state correlation map, performing correlation analysis with foreign object defects and coating layer defects, and setting a defect identification matrix; generating a morphological filtering script based on the defect identification matrix and combined with the coating surface imaging data, extracting features from the polarization state imaging data collected by the dual-path polarized light imaging module, and triggering foreign object defect classification and location marking based on the feature matching results.

[0007] Preferably, the online monitoring method for coating foreign object defects based on polarized light imaging analysis further includes: determining the foreign object morphological features, polarization features, and grayscale features according to the defect identification matrix; and generating a morphological filtering script that includes adaptive threshold segmentation, morphological opening and closing operations, and polarization feature enhancement based on the foreign object morphological features, polarization features, and grayscale features, combined with coating surface imaging data.

[0008] Preferably, the online monitoring method for coating foreign object defects based on polarized light imaging analysis further includes: automatically triggering dual-path polarization state switching and imaging parameter fine-tuning when low-contrast foreign objects are detected; simultaneously, integrating a digital twin simulation engine, combining coating surface imaging data to pre-simulate foreign object imaging scenarios under different coating conditions, and a morphological filtering script that meets the detection efficiency requirements of the coating production line.

[0009] Preferably, the online monitoring method for coating foreign object defects based on polarized imaging analysis further includes: setting up a defect feature federated learning framework driven by edge computing nodes, protecting the local imaging training data of the coating production line through differential privacy; and configuring a hardware-level polarization calibration environment based on the anti-ambient light interference capability of the defect feature federated learning framework.

[0010] Preferably, the online monitoring method for coating foreign matter defects based on polarized light imaging analysis further includes: real-time transmission protection of polarization state imaging data based on a dual-optical-path synchronously triggered imaging data transmission channel; the transmission rate of the imaging data transmission channel meets the detection efficiency requirements of the coating production line, and a data packet integrity verification mechanism based on hash check is used to perform feedback verification on each frame of polarization state imaging data.

[0011] Preferably, the online monitoring method for coating foreign object defects based on polarized light imaging analysis further includes: deploying a lightweight image compression module to generate lossless compressed data whose signal-to-noise ratio meets the defect detection standard; and based on the lossless compressed data, constructing a dual polarization state image fusion protocol to perform feature complementary fusion of s-polarization state imaging data and p-polarization state imaging data with time stamps at the same monitoring point.

[0012] Preferably, the online monitoring method for coating foreign object defects based on polarized light imaging analysis further includes: training a lightweight foreign object recognition model under a federated knowledge distillation architecture at edge computing nodes, and recognizing foreign object features across production lines through knowledge transfer; and adjusting dynamic filtering parameters based on coating material properties and coating process parameters to improve the accuracy of foreign object defect differentiation.

[0013] Preferably, the online monitoring method for coating foreign matter defects based on polarized light imaging analysis further includes: dynamically adjusting the imaging device parameters according to the coating defect risk level based on the digital twin simulation engine, wherein the update delay of the imaging device parameters conforms to the online monitoring delay threshold of the coating production line; and simultaneously integrating homomorphically encrypted defect detection logs to support defect statistical analysis in encrypted state.

[0014] Preferably, the online monitoring method for coating foreign matter defects based on polarized light imaging analysis further includes: using a generative adversarial network to classify and correct misjudgments in real time the imaging data of suspected defect areas on the coating surface; training the generator to simulate the imaging mode of a normal coating surface, the discriminator to distinguish foreign matter defects, bubble defects and wrinkle defects, and marking misjudged areas for secondary verification.

[0015] Preferably, the online monitoring method for coating foreign object defects based on polarized light imaging analysis further includes: simulating foreign object imaging modes under different coating materials by generating adversarial networks to obtain imaging parameter adjustment characteristics under different material scenarios; and updating the priority queue based on the imaging parameter adjustment characteristics under different material scenarios with the goal of minimizing the imaging parameter update delay, so that the update delay of the imaging device parameters always meets the online monitoring delay threshold of the coating production line.

[0016] The technical solution provided in this application has at least the following technical effects or advantages: by achieving the technical goal of high-precision real-time identification and classification of foreign object defects on the coating surface through the collaborative mechanism of dual-path polarized imaging module and edge computing, the technical effect of significantly improving the detection accuracy of transparent and low-contrast foreign objects, reducing the false judgment rate, and ensuring the continuous detection stability of the coating production line in complex coating environments is achieved.

[0017] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the online monitoring method for coating foreign matter defects based on polarized light imaging analysis proposed in this application.

[0020] Figure 2 This is a schematic diagram of the process for generating morphological filtering scripts in the online monitoring method for coating foreign matter defects based on polarized light imaging analysis in this application. Detailed Implementation

[0021] This application provides an online monitoring method for coating foreign object defects based on polarized light imaging analysis. It addresses the technical problems in existing technologies where visual monitoring is affected by the reflectivity, color uniformity, and optical interference of the coating material itself, making it difficult to accurately identify transparent and low-contrast foreign objects. Furthermore, foreign objects are easily confused with defects such as bubbles and wrinkles within the coating layer, leading to high false positive and false negative rates, which further affect the surface quality control accuracy and online defect response capability of the coating production line. The method achieves high-precision real-time identification and classification of foreign object defects on the coating surface through a dual-path polarized light imaging module and an edge computing collaborative mechanism. This significantly improves the detection accuracy of transparent and low-contrast foreign objects, reduces the false positive rate, and ensures the continuous detection stability of the coating production line in complex coating environments.

[0022] The technical solutions of this application will now be clearly and completely described 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. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0023] Please see Figure 1 and Figure 2 This application provides an online monitoring method for coating foreign matter defects based on polarized light imaging analysis, which specifically includes the following steps: S1: Determine the monitoring points corresponding to the coating production line and assign weights to them, and obtain a coating process parameter table including coating speed, coating thickness, and drying temperature.

[0024] Specifically, identifying and assigning weights to various monitoring points along the coating production line involves selecting key inspection locations, including the coating roller, drying section, and winding section. The coating roller controls the uniform distribution of coating on the substrate surface, the drying section evaporates the solvent in the coating using hot air or infrared heating, and the winding section winds up the coated and dried material. At different monitoring points, the likelihood and impact of defects vary, necessitating weight assignment. For example, the coating roller is more sensitive to foreign matter contamination and can be assigned a higher weight, while the winding section primarily affects surface smoothness and can be assigned a slightly higher weight. By setting these weights, the monitoring system can more accurately reflect the critical status of the entire production line during data fusion.

[0025] Obtaining a table of coating process parameters, including coating speed, coating thickness, and drying temperature, represents the collection of key process data from production equipment and sensor systems during the coating process. Coating speed refers to the linear speed of the substrate passing through the coating roller, directly affecting coating thickness and film quality. Coating thickness indicates the thickness of the layer formed by the coating on the substrate surface; excessive thickness may lead to uneven drying, while insufficient thickness affects hiding power and adhesion. Drying temperature refers to the temperature controlling coating curing and solvent evaporation in the drying section; excessively high temperatures can easily cause bubbles or cracks, while excessively low temperatures result in high coating moisture content and incomplete drying. The weighting of monitoring points needs to be adjusted based on changes in process parameters. For example, when the coating speed increases from 50 meters per minute to 100 meters per minute, the monitoring weight of the drying section should be increased accordingly to address the increased thermal stress and defect risk associated with higher speeds. Similarly, when the drying temperature increases by 10 degrees Celsius, it may be necessary to increase the monitoring accuracy of the winding section to prevent uneven surface tension or foreign matter adhesion due to overheating.

[0026] S2: Verify the parameters of the dual-path polarizing imaging module and the line scan camera using the polarizing imaging standard, and perform initial equipment configuration based on the difference between the polarizing imaging standard and the actual polarizing imaging.

[0027] Specifically, parameter verification of the dual-path polarizing imaging module and line scan camera using polarizing imaging standards indicates that before the coating defect detection system is put into use, the operating parameters of the imaging device need to be compared with the calibrated polarizing imaging standard. The polarizing imaging standard is a reference system based on an optical benchmark, including specific polarization angles, light intensity distribution, and grayscale response, used to determine whether the device can accurately distinguish changes in polarized light during imaging. The dual-path polarizing imaging module refers to acquiring images of different polarization states through two optical paths to enhance the contrast between coating surface texture and foreign objects. The line scan camera is a camera that images using a line scanning method, acquiring images of moving coating materials through continuous scanning, featuring high resolution and high-speed response. The purpose of parameter verification is to confirm that each component meets the standard requirements in terms of brightness, polarization response, and imaging stability, thereby ensuring the accuracy and repeatability of subsequent imaging results.

[0028] The polarization imaging standard provides a verification basis, and the initial equipment configuration is calibrated based on the verification results. That is, the initial equipment configuration is based on the discrepancy between the current equipment configuration and the polarization imaging standard. This means that after verification, parameters are adjusted according to the deviation between the current performance of the equipment and the standard to bring the imaging equipment to its optimal working state. The initial equipment configuration includes parameters such as polarization angle, exposure time, and resolution. The polarization angle determines the selectivity of the direction of light vibration; it can be adjusted between 0 and 90 degrees, with different angles highlighting different types of surface defects. Exposure time represents the cumulative acquisition time of light signals by the camera sensor; too short an exposure time will result in an overly dark image, while too long an exposure time will produce halos or ghosting. Resolution represents the spatial size of each pixel in the image, determining the sharpness of the image details.

[0029] S3: Based on the coating process parameter table, formulate a coating state correlation map, perform correlation analysis with foreign matter defects and coating layer defects, and set a defect identification matrix.

[0030] Specifically, based on the coating process parameter table, a coating state correlation map is proposed. This map represents a visual mapping relationship that reflects the dynamic state of the coating process, based on key process parameters such as coating speed, coating thickness, and drying temperature. The coating state correlation map is a multi-dimensional data model that combines coating process parameters with time, space, and material properties to create a behavioral profile of the coating process. For example, when the coating speed increases from 60 meters per minute to 90 meters per minute, the coating state correlation map will show the correlation trend between drying temperature and coating thickness, thus helping to identify potential risk points that may lead to uneven coating or foreign matter adhesion.

[0031] This study analyzes the correlation between coating state correlation maps and foreign matter defects and coating layer defects, statistically modeling the relationship between different types of defects and coating conditions. Foreign matter defects refer to abnormal reflective areas formed by foreign particles, dust, fibers, etc., mixed into the coating; coating layer defects include surface problems such as streaks, bubbles, and wrinkles caused by process instability or equipment deviation. The correlation analysis identifies high-risk operating ranges by calculating the correlation coefficient between changes in process parameters and the frequency of defect occurrence. For example, when the drying temperature exceeds 120 degrees Celsius, the probability of wrinkle defects may increase by 20%; while when the coating speed is below 50 meters per minute, the possibility of foreign matter adhesion may increase by 15%.

[0032] Setting up a defect identification matrix involves transforming the correlation analysis results into a digital identification model, representing the mapping relationship between different feature variables and defect categories in matrix form. Each row in the defect identification matrix represents a defect category, such as foreign objects, bubbles, or stripes, while each column corresponds to a feature parameter, such as polarization angle, grayscale value, or texture direction. The numerical values ​​in the defect identification matrix represent the correlation strength between the feature and the defect. For example, when the polarization angle is 45 degrees, the response value for a foreign object defect may reach 0.85, while the response value for a bubble defect is only 0.35, meaning that this angle is more suitable for identifying foreign objects than bubbles.

[0033] S4: Based on the defect identification matrix, a morphological filtering script is generated by combining the coating surface imaging data. The polarization state imaging data collected by the dual-path polarization imaging module is used to extract features, and the foreign object defect classification and location marking are triggered by the feature matching results.

[0034] Furthermore, this application also includes: determining foreign object morphological features, polarization features, and grayscale features based on the defect identification matrix; and generating a morphological filtering script that includes adaptive threshold segmentation, morphological opening and closing operations, and polarization feature enhancement based on the foreign object morphological features, polarization features, and grayscale features, combined with coating surface imaging data.

[0035] Furthermore, this application also includes: automatically triggering dual-path polarization state switching and imaging parameter fine-tuning when low-contrast foreign objects are detected; simultaneously, integrating a digital twin simulation engine to pre-simulate foreign object imaging scenarios under different coating conditions based on coating surface imaging data, and a morphological filtering script that meets the detection efficiency requirements of the coating production line.

[0036] Specifically, based on the defect identification matrix, the morphological characteristics, polarization characteristics, and grayscale characteristics of foreign objects are determined. This means extracting important visual and physical features from the defect identification matrix that can distinguish foreign objects from normal coatings. The morphological characteristics of foreign objects refer to their geometric shape, edge contours, and size characteristics in imaging; for example, foreign objects may be circular, elliptical, or irregular in shape. Polarization characteristics refer to the difference in reflection or transmission of light under polarized illumination, used to reveal differences in the microstructure of the material surface. For example, the reflection intensity of polarized light by metal particles in a coating is significantly higher than that of organic dust. Grayscale characteristics refer to the difference in brightness values ​​between the foreign object area and the background area. For example, the mean grayscale value of the foreign object area may be 20% higher than that of the surrounding area, and its variance is also larger, indicating non-uniform optical response.

[0037] Furthermore, based on the morphological features, polarization features, and grayscale features of foreign objects, combined with the coating surface imaging data, a morphological filtering script is generated, which includes adaptive threshold segmentation, morphological opening and closing operations, and polarization feature enhancement. This represents the design of a computational script for automatically recognizing and processing images. The coating surface imaging data is a high-resolution image sequence acquired by a polarized imaging system, reflecting the texture and defect distribution of the coating surface. Adaptive threshold segmentation is an algorithm that automatically adjusts the segmentation threshold based on the local brightness of the image, used to distinguish between defective and background areas. It can dynamically adjust the threshold according to changes in ambient light intensity; for example, when the overall brightness increases by 30%, the segmentation threshold will also increase accordingly. Morphological opening and closing operations are an image processing method based on structuring elements. Opening operations can remove isolated noise points, and closing operations can fill small holes to optimize the boundary coherence of foreign object regions. Polarization feature enhancement improves the saliency of foreign objects in the polarization direction by adjusting the polarization channel weights or calculating the polarization contrast, thereby enhancing the stability and sensitivity of detection.

[0038] In summary, identifying the key feature dimensions for foreign object identification guides the design and implementation of image processing algorithms. For example, as the complexity of foreign object morphology increases, the morphological filtering script automatically selects larger structuring elements to improve edge preservation; when the polarization feature difference increases, the polarization enhancement module increases the weight of the corresponding channel, making the foreign object region more prominent in the imaging result; if the grayscale feature contrast decreases, the sensitivity of the adaptive threshold is automatically adjusted to maintain detection accuracy.

[0039] Furthermore, when a low-contrast foreign object is detected, the system automatically triggers dual-path polarization state switching and imaging parameter fine-tuning. This means that when a small difference in brightness between the foreign object and the background is detected during imaging analysis, the polarization state switching of the dual-path imaging system is automatically initiated, and key imaging parameters are adjusted simultaneously. Low-contrast foreign objects refer to objects whose brightness or polarization reflection differs little from the background in the image, such as transparent resin particles or semi-transparent fibers, which are difficult to identify by conventional thresholding algorithms. Dual-path polarization state switching refers to the imaging system alternately acquiring images with different polarization directions between two optical paths, using the difference in polarization reflection to enhance the distinguishability of foreign objects and coatings. Imaging parameter fine-tuning includes dynamically adjusting exposure time, gain, polarization angle, etc., to improve the image signal-to-noise ratio. For example, when the exposure time increases from 5 milliseconds to 7 milliseconds, the image brightness increases by 15%, and the edge sharpness of the foreign object improves accordingly, thereby achieving a more stable detection effect.

[0040] Simultaneously, an integrated digital twin simulation engine is used to pre-simulate foreign object imaging scenarios under different coating conditions, combining coating surface imaging data. This means that while the system is detecting foreign objects, a virtual simulation model synchronized with the actual coating production line is constructed using digital twin technology. The digital twin simulation engine is a virtual-real fusion platform that uses real-world coating surface imaging data and process parameters to simulate the optical imaging changes under different conditions in real time on a computer. The coating surface imaging data includes polarization reflection intensity distribution, grayscale levels, and texture direction information, used to generate the optical response in the virtual scene. Pre-simulating foreign object imaging scenarios under different coating conditions means that the visibility and morphological characteristics of foreign objects can be predicted in the simulation environment when coating speed, drying temperature, or coating thickness changes. For example, when the coating speed increases from 60 meters per minute to 100 meters per minute, the imaging blur increases by 10%, and the simulation model can assess in advance that the polarization angle needs to be adjusted to 70 degrees to maintain detection clarity.

[0041] A morphological filtering script that meets the detection efficiency requirements of a coating production line means that the generated image processing algorithm must not only be accurate but also ensure real-time performance at actual production line speeds. Detection efficiency requirements are typically measured in the number of imaging frames that can be detected per unit time or in terms of processing latency. For example, it may require completing single-frame image analysis within 10 milliseconds to match a production line speed of 90 meters per minute.

[0042] Furthermore, feature extraction is performed on the polarization-state imaging data acquired by the dual-path polarization imaging module. This demonstrates that a polarization imaging module with two independent optical channels can simultaneously acquire image information of the same coated surface at different polarization angles. Polarization-state imaging data refers to information reflecting the directional distribution of light waves at different polarization angles, revealing microscopic reflection differences, scattering characteristics, and optical response differences of foreign objects on the coated surface. Image processing or deep learning algorithms are used to extract key features reflecting defect properties from the polarization-state imaging data, such as texture direction, polarization intensity distribution, and polarization angle change rate. For example, when the polarization angle is 45 degrees, the polarization degree of reflected light from foreign objects on the surface may be 15% higher than that of normal areas.

[0043] The foreign object defect classification and localization marking are triggered by feature matching results. This means matching the extracted features with a pre-established defect feature library, determining the presence of foreign object defects through similarity judgment, and marking the defect location in the image. Feature matching involves comparing the feature vector of the current image with the feature vector of a standard sample to determine if the similarity exceeds a set threshold. Foreign object defect classification determines the type of defect based on the matching results, such as dust, fibers, coating bubbles, or metal shavings. Different types of defects show significant differences in polarization features. Localization marking marks the detected defect area in the image in the form of coordinates or bounding boxes, providing a basis for subsequent repair, removal, or statistical analysis.

[0044] Furthermore, this application also includes: setting up a defect feature federated learning framework driven by edge computing nodes, protecting the local imaging training data of the coating production line through differential privacy; and configuring a hardware-level polarization calibration environment based on the anti-ambient light interference capability of the defect feature federated learning framework.

[0045] Specifically, the defect feature federated learning framework driven by edge computing nodes involves deploying intelligent nodes with computing and learning capabilities on the edge devices of the coating production line to jointly participate in model training. Edge computing nodes are computing units located near the data generation source, which can reduce data transmission latency and improve real-time processing capabilities. The defect feature federated learning framework is a distributed intelligent learning structure. Each edge node independently trains its model locally using its own data, and then uploads the model parameters, rather than the raw data, to the central server for aggregation, thereby achieving global model sharing and optimization. This enables knowledge sharing among multiple devices while avoiding the direct uploading of sensitive imaging data from the production site. Protecting the local imaging training data of the coating production line through differential privacy means introducing digital random perturbation technology during the federated learning process to inject noise into the uploaded model parameters, making it impossible for external parties to infer a specific sample, thus protecting data privacy.

[0046] Based on the robustness of the defect feature federated learning framework to ambient light interference, a hardware-level polarization calibration environment was configured to demonstrate that the framework can stably extract defect features even under complex environments such as varying illumination and reflection interference. This robustness against ambient light interference reflects the framework's robustness to illumination changes, relying on feature alignment and dynamic compensation mechanisms under different illumination conditions. The hardware-level polarization calibration environment is an experimental or production environment that uses physical means to control imaging optical conditions to ensure accurate acquisition of polarization information. It consists of a constant light source and an anti-reflective coating. The constant light source provides stable and controllable illumination conditions, ensuring consistent light intensity throughout the detection process; the anti-reflective coating reduces specular reflection of light on the equipment surface and surrounding materials, preventing polarization angle shifts.

[0047] Furthermore, this application also includes: a dual-optical-path synchronously triggered imaging data transmission channel for real-time transmission protection of polarization state imaging data; the transmission rate of the imaging data transmission channel meets the detection efficiency requirements of the coating production line, and a hash-based data packet integrity verification mechanism is used to perform feedback verification on each frame of polarization state imaging data.

[0048] Specifically, based on a dual-optical-path synchronously triggered imaging data transmission channel, real-time transmission protection of polarization-state imaging data is implemented. This means that after images are simultaneously acquired using a dual-optical-path polarization imaging device on the coating production line, polarization image data from both optical paths are simultaneously transmitted through a dedicated high-speed data channel. Dual-optical-path synchronous triggering refers to the two optical channels starting imaging simultaneously at the same time point, ensuring complete temporal correspondence of the acquired images and avoiding polarization information shifts caused by time differences. The imaging data transmission channel refers to the communication path for image data transmission between devices, consisting of industrial Ethernet, fiber optic cables, or high-speed buses. Real-time transmission protection means employing redundancy checks, encryption, and packet loss retransmission mechanisms during data transmission to ensure that polarization-state imaging data is not tampered with or lost during high-speed transmission.

[0049] The transmission rate of the imaging data transmission channel meets the detection efficiency requirements of the coating production line. A hash-based data packet integrity verification mechanism is used to perform feedback verification on each frame of polarization imaging data, indicating that the bandwidth and rate design of the imaging data channel can support real-time detection tasks during high-speed production line operation. Meeting the detection efficiency requirements means that the number of image frames that can be transmitted per second matches the operating speed of the production line. For example, when the production line is running at a speed of 50 meters per minute, the transmission channel needs to achieve a transmission capacity of 60 frames per second to avoid detection delays. The hash-based data packet integrity verification mechanism is a technique that detects whether errors or packet loss occur during transmission by calculating the hash value of the data packet content. If the hash value calculated by the receiving end does not match that of the sending end, data retransmission or an alarm is immediately triggered. Feedback verification refers to returning the verification result to the sending end after each frame of image transmission is completed, ensuring data accuracy in a closed-loop manner. For example, when 1% of the data packets fail verification, a retransmission mechanism is initiated within 100 milliseconds to ensure the consistency of the entire frame of data.

[0050] Furthermore, this application also includes: deploying a lightweight image compression module to generate lossless compressed data whose signal-to-noise ratio meets the defect detection standard; based on the lossless compressed data, constructing a dual polarization state image fusion protocol to perform feature complementary fusion of s-polarization state imaging data and p-polarization state imaging data with time stamps at the same monitoring point.

[0051] Specifically, deploying a lightweight image compression module generates lossless compressed data whose signal-to-noise ratio (SNR) meets defect detection standards. This signifies the introduction of a computationally low-load, real-time-efficient image compression algorithm into the imaging system's data processing flow, reducing data transmission and storage pressure. The lightweight image compression module is a data compression unit that achieves high compression ratios and low latency through algorithm optimization while maintaining image quality. It is deployed at the edge to quickly respond to the real-time needs of the production line. The signal-to-noise ratio (SNR) of imaging data refers to the ratio of useful information to noise in an image signal, directly affecting image clarity and detection accuracy. Lossless compressed data refers to the compression result where no image information is lost during compression and decompression; that is, the compressed data can be completely restored to its original state after decoding. Meeting defect detection standards means that the compressed image still retains the ability to identify minute defects. For example, when compressing a polarized image with a SNR of 45 dB, the compression ratio can reach 3:1, while still accurately identifying surface foreign objects with a diameter of 0.05 mm.

[0052] Based on lossless compressed data, a dual-polarization image fusion protocol is constructed. This protocol performs feature complementarity fusion on s-polarization and p-polarization imaging data with time stamps from the same monitoring point. This means that while maintaining data integrity, a specific fusion algorithm integrates optical information from different polarization directions to enhance the imaging quality of the coated surface. The dual-polarization image fusion protocol is a communication and computation rule for synchronously processing two polarization states of images. The s-polarization and p-polarization states represent the perpendicular and parallel components of the light polarization direction to the surface normal, respectively. These two polarization states respond differently to surface structure, scattering characteristics, and reflection intensity. Time stamping refers to attaching precise time information to each frame of imaging data, used for time synchronization of the two images during fusion to prevent mismatches caused by acquisition delays. Feature complementarity fusion extracts and fuses feature information such as brightness differences, polarization degree distribution, and texture direction from the two polarization images to form a composite image with higher information density. For example, when fusing images of the same monitoring point, s-light images can enhance surface microstructure, while p-light images can enhance reflective features. The detail resolution of the fused image is improved by about 30%, and the defect identification accuracy is improved by about 10%.

[0053] Furthermore, this application also includes: training a lightweight foreign object recognition model under a federated knowledge distillation architecture on edge computing nodes, and recognizing foreign object features across production lines through knowledge transfer; and adjusting dynamic filtering parameters based on coating material properties and coating process parameters to improve the accuracy of foreign object defect differentiation.

[0054] Specifically, training a lightweight foreign object recognition model under a federated knowledge distillation architecture on edge computing nodes means deploying an intelligent model combining federated learning and knowledge distillation mechanisms on computing devices near the coating production site to identify different types of foreign object defects. Edge computing nodes are terminal devices located near the data source with certain computing capabilities, used for data processing and model training without relying on a central server. The federated knowledge distillation architecture is a model training method that combines the advantages of federated learning and knowledge distillation. Federated learning protects production line privacy by sharing model parameters through multi-node collaborative training without sharing raw data; knowledge distillation reduces model size and improves inference efficiency by "distilling" the recognition capabilities of complex models to lightweight models. The lightweight foreign object recognition model refers to a deep learning model with optimized structure and reduced computational load, capable of quickly identifying surface defects in real-time monitoring scenarios. Identifying foreign object features across production lines through knowledge transfer means that the federated knowledge distillation architecture can transfer learning experience between different production lines to identify foreign object types with differentiated characteristics.

[0055] Based on the properties of the coating material and coating process parameters, dynamic filtering parameter adjustment is used to improve the accuracy of foreign object defect differentiation. This means that the filtering parameters of image processing are adaptively optimized according to the optical characteristics of different coating materials and production process conditions, thereby improving the accuracy of defect detection. Coating material properties include light reflectivity, refractive index, and surface roughness, which affect the polarization and scattering behavior of light. Coating process parameters refer to process variables such as coating speed, coating thickness, and drying temperature, which determine the optical conditions and surface texture appearance during imaging. Dynamic filtering parameter adjustment refers to continuously optimizing key parameters in the image filtering algorithm based on real-time acquired data during monitoring, such as adjusting the filter kernel size or enhancing specific frequency responses to adapt to the imaging characteristics under different working conditions. The purpose of improving the accuracy of foreign object defect differentiation is to more accurately distinguish transparent foreign objects, low-contrast foreign objects, and background texture, avoiding false detections or missed detections. For example, under high-reflectivity coatings, the filtering algorithm can increase the threshold by 20% to reduce background noise interference, while in low-brightness scenes, the threshold is dynamically lowered to enhance the visibility of foreign objects.

[0056] Furthermore, this application also includes: dynamically adjusting the imaging equipment parameters according to the coating defect risk level based on the digital twin simulation engine, wherein the update delay of the imaging equipment parameters conforms to the online monitoring delay threshold of the coating production line; and simultaneously integrating homomorphically encrypted defect detection logs to support defect statistical analysis in encrypted state.

[0057] Furthermore, this application also includes: simulating foreign object imaging modes under different coating materials by generating adversarial networks to obtain imaging parameter adjustment features under different material scenarios; and updating the priority queue based on the imaging parameter adjustment features under different material scenarios with the goal of minimizing the imaging parameter update delay, so that the update delay of the imaging device parameters always meets the online monitoring delay threshold of the coating production line.

[0058] Specifically, based on a digital twin simulation engine, the imaging equipment parameters are dynamically adjusted according to the coating defect risk level. This means using virtual simulation technology to digitally map the actual coating production line, and achieving predictive adjustments to equipment parameters by synchronizing the virtual model with the physical system in real time. The coating defect risk level refers to the result of a graded assessment of potential defect types such as foreign objects, bubbles, and wrinkles based on historical detection data and model prediction results, divided into three levels: low, medium, and high. The process of dynamically adjusting the imaging equipment parameters refers to automatically changing key parameters such as polarization angle, exposure time, and resolution under different risk levels to optimize the sensitivity and contrast of defect detection.

[0059] Furthermore, by using a generative adversarial network (GAN), the imaging patterns of foreign objects under different coating materials are simulated, and the imaging parameter adjustment features under different material scenarios are obtained. This means that a deep learning model composed of a generator and a discriminator is used to learn and reproduce the changes in the optical properties of different material surfaces during imaging. A GAN is a neural network structure composed of two opposing modules, where the generator is responsible for generating simulated imaging data, and the discriminator is responsible for distinguishing the differences between these data and real images. Through continuous game training, the generator can gradually approximate the imaging features of real coated surfaces. The foreign object imaging pattern refers to the visual manifestation of foreign objects under polarized light imaging conditions, including features such as brightness differences, texture morphology, and polarization angle response. These features change significantly with the different physical properties of the coating material, such as reflectivity, transparency, and refractive index. Different coating material scenarios include, for example, transparent polymer films, metal coatings, and paper-based coatings. Each material has different light scattering and absorption characteristics under polarized light. Imaging parameter adjustment features refer to the parameter patterns that need to be modified in different scenarios to obtain the best defect recognition effect, such as exposure time, polarization angle, and gain coefficient. For example, under a highly reflective metallic coating, a generative adversarial network may simulate the problem of excessive image brightness, thereby extracting parameter adjustment features to reduce the exposure time by 0.2 seconds.

[0060] Based on the imaging parameter adjustment characteristics under different material scenarios, and with the goal of minimizing the imaging parameter update delay, a priority queue is updated to ensure that the update delay of the imaging equipment parameters always conforms to the online monitoring delay threshold of the coating production line. This means that after identifying different material characteristics, the algorithm optimizes and determines the order of imaging equipment parameter updates, thereby ensuring that the imaging equipment can quickly respond to changes in different detection environments. Minimizing the imaging parameter update delay means that the parameter adjustment process should shorten the adjustment cycle as much as possible, making the time from detection to parameter effectiveness as short as possible. The priority queue is a scheduling mechanism that sorts tasks according to their importance, placing the parameter adjustment tasks that have the greatest impact on imaging quality at the front of the queue for priority execution. The online monitoring delay threshold refers to the maximum allowable response delay of the coating production line, for example, specified as within 0.1 seconds, to avoid monitoring lag affecting defect identification. For example, when a switch from a transparent film to a metal substrate material is detected, the priority queue is immediately updated, prioritizing the adjustment of the polarization angle and exposure time, reducing the delay from the original 0.12 seconds to 0.08 seconds.

[0061] Furthermore, integrating homomorphically encrypted defect detection logs to support defect statistical analysis in encrypted mode signifies the introduction of secure computing technology into the detection system, enabling direct computation in an encrypted state to protect the privacy of detection data. Defect detection logs are structured data recording the type, location, time, and risk level of defects found in each detection cycle, used for subsequent quality traceability and statistical analysis. Homomorphic encryption is a special encryption algorithm that allows statistical or computational operations to be performed without decrypting the data, thus avoiding the exposure of raw data during analysis. Defect statistical analysis in encrypted mode means that all calculations are completed in an encrypted environment; for example, the number of high-risk defects within a certain period can be directly counted without decrypting the log data. This ensures both the confidentiality of production line data and supports data aggregation and analysis across multiple factories.

[0062] Furthermore, this application also includes: using a generative adversarial network to classify and correct misjudgments in real time the imaging data of suspected defect areas on the coated surface; training the generator to simulate the imaging mode of a normal coated surface, the discriminator to distinguish foreign matter defects, bubble defects and wrinkle defects, and marking misjudged areas for secondary verification.

[0063] Specifically, for imaging data of suspected defect areas on the coated surface, a generative adversarial network (GAN) is used for real-time classification and misclassification correction. This involves analyzing images of potentially abnormal areas captured during the coating process. The coated surface refers to the imaging area of ​​the material surface after uniform coating treatment, while suspected defect areas are parts of the image that show abnormalities in brightness, texture, or reflectivity. Real-time classification means automatic judgment during imaging. Misclassification correction refers to the model re-correcting by comparing with generated reference samples when it makes a mistake. For example, when normal bubble reflection is misidentified as a defect, the GAN will correct it based on learned real patterns, thereby reducing false alarms.

[0064] The generator is trained to simulate the imaging pattern of a normal coated surface, while the discriminator distinguishes between foreign matter defects, bubble defects, and wrinkle defects. It also marks misclassified areas for secondary verification—this is the internal working mechanism of the Generative Adversarial Network (GAN). The generator's task is to learn the optical characteristics of a normal coated surface, such as brightness uniformity and polarization reflection intensity, to generate simulated samples that closely resemble real images. The discriminator is responsible for determining whether the input image is real or generated, and further subdividing it into different types of defects, such as foreign matter defects, bubble defects, and wrinkle defects. Marking misclassified areas for secondary verification involves identifying potentially misclassified regions in the detection results and re-verifying them using the model, thereby improving recognition accuracy. For example, if the generator detects an image highly similar to a normal surface but the discriminator still classifies it as a defect, the region is re-detected to avoid false positives.

[0065] In summary, the online monitoring method for coating foreign matter defects based on polarized light imaging analysis provided in this application has the following technical effects: by achieving the technical goal of high-precision real-time identification and classification of foreign matter defects on the coating surface through the collaborative mechanism of dual-path polarized light imaging module and edge computing, it achieves the technical effects of significantly improving the detection accuracy of transparent and low-contrast foreign matter, reducing the false judgment rate, and ensuring the continuous detection stability of the coating production line in complex coating environments.

[0066] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0067] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for online monitoring of coating foreign matter defects based on polarized light imaging analysis, characterized in that, The method includes: Determine the monitoring points corresponding to the coating production line and assign weights to them to obtain a coating process parameter table including coating speed, coating thickness, and drying temperature; The parameters of the dual-path polarizing imaging module and the line scan camera are verified using the polarizing imaging standard, and the initial equipment configuration is performed based on the difference between the polarizing imaging standard and the actual polarizing imaging standard. Based on the coating process parameter table, a coating state correlation map is proposed, and correlation analysis is performed with foreign matter defects and coating layer defects to set a defect identification matrix. Based on the defect identification matrix, a morphological filtering script is generated by combining the coating surface imaging data. Feature extraction is performed on the polarization state imaging data collected by the dual-path polarization imaging module, and the feature matching results trigger the classification and localization marking of foreign objects and defects.

2. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 1, characterized in that, Based on the defect identification matrix, a morphological filtering script is generated by combining the coating surface imaging data. The method includes: Based on the defect identification matrix, the morphological characteristics, polarization characteristics, and grayscale characteristics of the foreign object are determined; Based on the aforementioned foreign object morphological features, polarization features, and grayscale features, combined with the coating surface imaging data, a morphological filtering script is generated that includes adaptive threshold segmentation, morphological opening and closing operations, and polarization feature enhancement.

3. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 2, characterized in that, By combining imaging data of the coated surface, a morphological filtering script is generated that includes adaptive threshold segmentation, morphological opening and closing operations, and polarization feature enhancement. The method further includes: When a low-contrast foreign object is detected, the dual-path polarization state switching and imaging parameter fine-tuning are automatically triggered. Simultaneously, an integrated digital twin simulation engine is used to pre-simulate foreign object imaging scenarios under different coating conditions by combining coating surface imaging data, and a morphological filtering script is provided to meet the detection efficiency requirements of the coating production line.

4. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 1, characterized in that, An edge computing node-driven federated learning framework for defect features is set up, and local imaging training data of the coating production line is protected by differential privacy. Based on the immunity to ambient light interference of the aforementioned defect feature federated learning framework, a hardware-level polarization calibration environment is configured.

5. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 4, characterized in that, Real-time transmission protection of polarization state imaging data is achieved through a dual-optical-path synchronously triggered imaging data transmission channel. The transmission rate of the imaging data transmission channel meets the detection efficiency requirements of the coating production line, and a data packet integrity verification mechanism based on hash check is used to perform feedback verification on each frame of polarization state imaging data.

6. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 5, characterized in that, The method further includes: Deploy a lightweight image compression module to generate lossless compressed data whose signal-to-noise ratio meets the defect detection standards after compression; Based on the lossless compressed data, a dual polarization state image fusion protocol is constructed to perform feature complementary fusion of s-polarization state imaging data and p-polarization state imaging data with time stamps at the same monitoring point.

7. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 3, characterized in that, The method includes: A lightweight foreign object recognition model under a federated knowledge distillation architecture is trained on edge computing nodes, and foreign object features are identified across production lines through knowledge transfer. Based on the properties of the coating material and the coating process parameters, dynamic filtering parameters are adjusted to improve the accuracy of foreign object defect differentiation.

8. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 7, characterized in that, The method includes: Based on the digital twin simulation engine, the imaging equipment parameters are dynamically adjusted according to the coating defect risk level, and the update delay of the imaging equipment parameters meets the online monitoring delay threshold of the coating production line. Simultaneously, it integrates homomorphically encrypted defect detection logs to support defect statistical analysis in encrypted form.

9. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 8, characterized in that, For imaging data of suspected defect areas on the coated surface, a generative adversarial network is used for real-time classification and misjudgment correction. The training generator simulates the imaging mode of a normal coated surface, and the discriminator distinguishes foreign matter defects, bubble defects, and wrinkle defects, and marks misjudged areas for secondary verification.

10. The online monitoring method for coating foreign matter defects based on polarized light imaging analysis as described in claim 9, characterized in that, The update delay of the imaging device parameters conforms to the online monitoring delay threshold of the coating production line, and the method includes: By using generative adversarial networks, we simulate foreign object imaging modes under different coating materials and obtain imaging parameter adjustment features under different material scenarios. Based on the imaging parameter adjustment characteristics under different material scenarios, with the goal of minimizing the imaging parameter update delay, the priority queue is updated so that the update delay of the imaging device parameters always meets the online monitoring delay threshold of the coating production line.

Images