High-speed lamination production line product quality detection system based on visual identification

Abstract

The invention provides a high-speed lamination production line product quality detection system based on visual identification, and belongs to the technical field of visual detection. Comprising a multispectral image acquisition module (visible light, near infrared and ultraviolet band synchronous imaging), a motion compensation module (encoder synchronous correction displacement deviation), a preprocessing module (denoising, enhancement and ROI coarse positioning), a multi-scale feature fusion module (geometric, texture and spectral feature fusion) and a multi-task deep learning model. And a production line is linked to realize hierarchical control (sudden stop, early warning and parameter adjustment). The system eliminates image dislocation and blurring under high-speed motion through a multispectral information complementation and motion compensation technology; the recognition capability of tiny and complex defects is improved by utilizing multi-scale feature fusion; detection and decision-making integration is realized through a multi-task model, a high-precision and low-delay quality detection effect in a high-speed scene is finally achieved, and the quality control level of a lamination production line is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of visual detection, in particular to a high-speed laminated film production line product quality detection system based on visual recognition. BACKGROUND

[0002] In the high-speed laminated film production line, product quality detection is a key link to ensure the yield. The traditional detection method mainly relies on manual inspection or single modal (such as visible light) automatic visual detection, but faces the following problems:

[0003] 1. Poor image quality caused by high-speed motion: The speed of the laminated film production line is usually dozens of meters per minute, and the image collected by the traditional camera is easy to lose details due to motion blur and trailing, making it difficult to identify small defects (such as 0.1mm level creases).

[0004] 2. Complex and diverse defect types: Laminated film product defects include surface (bubbles, impurities, scratches) and internal (delamination, uneven thickness) defects, and single-spectrum (only visible light) detection cannot penetrate the surface layer and has insufficient internal defect recognition capability.

[0005] 3. Real-time and accuracy contradiction: High-speed production lines require millisecond-level response from detection systems, while traditional algorithms (such as manual feature extraction) have low classification accuracy for complex defects (such as composite wrinkles + impurities), which are prone to missed detection or misjudgment.

[0006] 4. Detection and production are disconnected: Traditional detection only outputs defect results and cannot link to production line to adjust process parameters (such as temperature, coating speed), resulting in repeated defects and high quality control costs.

[0007] Therefore, a high-speed laminated film production line product quality detection system based on visual recognition is proposed. SUMMARY

[0008] The present application provides a high-speed laminated film production line product quality detection system based on visual recognition to solve the problems raised in the background art.

[0009] The specific technical solutions are as follows:

[0010] A high-speed laminated film production line product quality detection system based on visual recognition, comprising:

[0011] An image acquisition module for synchronously acquiring multispectral images of laminated film products during continuous operation of the high-speed laminated film production line;

[0012] A motion compensation module for compensating the collected images in space-time alignment according to the real-time speed of the production line and the image acquisition timestamp;

[0013] A preprocessing module is configured to denoise, enhance and coarsely locate a defect region of the compensated image;

[0014] A feature extraction module is configured to extract geometric features, texture features and spectral features of the defect based on a multi-scale feature fusion algorithm;

[0015] A defect classification module is configured to input the extracted features into a pre-trained multi-task deep learning model and output a defect type and a severity level;

[0016] A control execution module is configured to trigger a production line shutdown, marking or parameter adjustment instruction according to the defect classification result;

[0017] The image acquisition module adopts a combination of a linear array camera array and a area exposure technology, the line frequency of the linear array camera and the production line speed are synchronized through an encoder to ensure that there is no splicing misplacement between image rows; the multi-spectral image includes image data of visible light, near-infrared and ultraviolet wave bands.

[0018] The above-mentioned high-speed film-coated production line product quality detection system based on visual recognition, wherein the image acquisition module comprises:

[0019] At least two groups of linear array industrial cameras are arranged at longitudinal and transverse detection stations of the film-coated production line, the longitudinal camera is used to detect quality defects in the width direction of the film, and the transverse camera is used to detect continuous defects in the length direction of the film;

[0020] A stroboscopic light source group is synchronously triggered with the linear array camera, the light source wavelength covers the visible light 400-700 nm wave band, the near-infrared light 700-1000 nm wave band and the ultraviolet light 300-400 nm wave band, and the flash duration is ≤1 μs to ensure that the image is clear without trailing in high-speed motion;

[0021] An encoder synchronization device is installed on a transmission shaft of the production line, outputs pulse signals to the linear array camera and the stroboscopic light source, realizes frame synchronization of image acquisition and production line motion, and the synchronization error is ≤0.1 ms.

[0022] The above-mentioned high-speed film-coated production line product quality detection system based on visual recognition, wherein the motion compensation module comprises:

[0023] A speed acquisition unit acquires real-time linear speed υ(t) and acceleration a(t) of the production line through an encoder;

[0024] A timestamp recording unit records a timestamp t for each frame of image i ;

[0025] A displacement compensation calculation unit calculates a displacement offset of the image in the production line direction by using the following equation:

[0026]

[0027] an image registration unit, based on the calculated Δx, corrects the image by translation through bilinear interpolation to eliminate image misregistration caused by speed fluctuation;

[0028] where Δt is the image acquisition time interval; λ is the acceleration attenuation coefficient; sgn is the sign function; υ(t i ) is the real-time linear velocity at time t i ; a(t i ) is the real-time acceleration at time t i .

[0029] The high-speed film production line product quality detection system based on visual recognition, wherein the preprocessing module comprises:

[0030] a motion blur correction submodule, which uses Wiener filtering combined with a blind deconvolution algorithm to restore image blur caused by high-speed motion of the production line, and blur kernel parameters are adaptively estimated through image gradient variance;

[0031] a gray scale equalization submodule, which enhances the contrast between the defect area and the background based on the CLAHE (Contrast Limited Adaptive Histogram Equalization) algorithm;

[0032] a ROI coarse positioning submodule, which quickly extracts the area where defects may exist by adaptive threshold segmentation combined with morphological operation, thereby reducing the calculation range for subsequent feature extraction.

[0033] The high-speed film production line product quality detection system based on visual recognition, wherein the feature extraction module uses a multi-scale feature fusion network, comprising:

[0034] a shallow feature extraction branch, which uses a 3x3 convolution kernel to extract low-level geometric features such as edges and corner points of the image;

[0035] a deep semantic feature extraction branch, which uses a ResNet-50 backbone network to extract high-level semantic features such as texture and wrinkles;

[0036] a spectral feature fusion branch, which performs principal component analysis (PCA) dimensionality reduction on the multispectral image and extracts principal components of each waveband as spectral features;

[0037] a feature fusion unit, which splices feature maps of the three branches in the spatial dimension through a skip connection, and then reduces the dimension through a 1x1 convolution to obtain fused multi-dimensional defect features.

[0038] The high-speed film production line product quality detection system based on visual recognition, wherein the pre-trained multi-task deep learning model of the defect classification module comprises:

[0039] a shared feature extraction layer, inputting multi-dimensional defect features;

[0040] a classification head, including a defect type classification branch (using a Softmax activation function, outputting common defect categories such as bubbles, wrinkles, uneven thickness, impurities, etc.) and a severity level regression branch (using a Sigmoid activation function, outputting a severity score between 0 and 1);

[0041] The loss function is a weighted sum of the classification cross-entropy loss and the mean square error loss, and the weight is dynamically adjusted according to the importance of the defect type.

[0042] The above-mentioned visual recognition-based high-speed lamination production line product quality detection system, wherein the control execution module comprises:

[0043] A first alarm unit, when a fatal defect (such as a large-area wrinkle or a penetrating bubble) with a severity level ≥ 0.8 is detected, an emergency stop signal is immediately triggered to the production line PLC, and a marking device is controlled to spray a code at the defect position.

[0044] A second early warning unit, when a secondary defect (such as local uneven thickness or a small impurity) with a severity level of 0.3-0.8 is detected, the defect position is recorded and displayed on the human-machine interface, prompting the operator to check the upstream process parameters.

[0045] A parameter adjustment unit, which communicates with the production line control system through an industrial internet of things (IIoT) interface, automatically adjusts the lamination temperature, coating speed or pressing pressure according to historical defect data statistics.

[0046] The above-mentioned visual recognition-based high-speed lamination production line product quality detection system, wherein it further comprises a self-calibration module for periodically calibrating the precision of the detection system, specifically comprising:

[0047] A standard sample library storing standard lamination samples of known defect types and parameters;

[0048] An automatic feeding and unloading unit for conveying the standard sample to the detection station;

[0049] A calibration evaluation unit for calculating the defect recognition accuracy, missed detection rate and positioning error by comparing the system detection results with the labeled data of the standard sample, and triggering system parameter correction if the error exceeds the threshold.

[0050] The above-mentioned visual recognition-based high-speed lamination production line product quality detection system, wherein the multi-spectral image acquisition method comprises:

[0051] The time-sharing exposure strategy adopts short exposure 5 mu s to capture surface details in the visible light band, adopts medium exposure 20 mu s to penetrate the surface layer of the film to detect the internal structure in the near-infrared band, and adopts long exposure 50 mu s to excite fluorescent defects in the ultraviolet band.

[0052] The wave band registration unit aligns the multispectral images at the pixel level through a feature point matching algorithm (such as SIFT), and eliminates the spatial deviation caused by lens distortion or light difference in different wave bands.

[0053] The above-mentioned high-speed film production line product quality detection system based on visual recognition further comprises a data storage and traceability module, which is used for:

[0054] Real-time storage of detection results (defect type, position, timestamp, severity level) and corresponding production line process parameters (temperature, speed, coating amount) into a local database;

[0055] Upload data to the cloud platform through the OPCUA protocol, support defect distribution visualization analysis based on time axis or product batch, and process parameter and defect type association rule mining.

[0056] The above-mentioned high-speed film production line product quality detection system based on visual recognition, wherein the training method of the multi-task deep learning model comprises:

[0057] Data enhancement, rotation (±15°), scaling (0.8-1.2 times), and adding Gaussian noise (σ=0.01) operations are performed on the film defect samples to expand the training set scale;

[0058] Transfer learning, using the pre-trained ResNet-50 on the ImageNet dataset as the backbone network, freezing the first 3 layers of convolutional layers, and fine-tuning the last 5 layers and the classification head;

[0059] Small sample learning, for rare defect types (such as composite wrinkles + impurities), using meta-learning (MAML) algorithm to optimize the initial parameters of the model, only 10-20 samples are needed to complete fine-tuning.

[0060] The present application has the following beneficial effects:

[0061] 1. Full scene adaptability: multispectral image acquisition covers visible light (surface details), near-infrared (internal structure), and ultraviolet (fluorescent defects) bands, combined with a motion compensation module to eliminate the influence of speed fluctuations, which can simultaneously identify surface micro-defects (such as impurities, scratches) and internal hidden defects (such as delamination, bubbles), covering all types of film product defects.

[0062] 2. High-speed and high-robustness detection: The line array camera array and encoder synchronization technology ensure no error between image lines, the motion compensation module dynamically corrects displacement deviation, and the motion blur correction and gray scale enhancement preprocessing are combined, so that clear and aligned images can be output even under high-speed start-stop or variable-speed working conditions of the production line, and the detection stability is ensured.

[0063] 3. Intelligent decision-making and closed-loop control: The multi-task deep learning model synchronously outputs the defect type (such as bubble, wrinkle) and the severity level (such as slight, fatal), and combines the grading response mechanism (emergency stop, warning, parameter adjustment) to realize the closed loop from "detection" to "intervention", avoid the generation of batch unqualified products, and reduce the defect rate from the source.

[0064] 4. Long-term reliability and process optimization: The self-calibration module verifies the system accuracy through standard samples regularly, avoids performance degradation caused by equipment aging or environmental changes, and the data storage and traceability module associates defect data and process parameters, supports defect distribution analysis and process optimization, and promotes the quality control from "post-detection" to "pre-prevention". BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 The composition schematic diagram of the high-speed film coating production line product quality detection system based on visual recognition provided by the embodiments of the present application is shown.

[0066] Figure 2 The relationship diagram of the production line speed and synchronization error is shown.

[0067] Figure 3 The influence diagram of the motion compensation on the image clarity is shown.

[0068] Figure 4 The improvement diagram of the preprocessing module on the image contrast is shown. DETAILED DESCRIPTION

[0069] The technical solutions of the present application will be further described below by combining the drawings and through specific embodiments.

[0070] Among them, the drawings are only used for illustrative explanation, and the representation is only a schematic diagram, not a physical diagram, and cannot be understood as a limitation of the present patent; in order to better illustrate the embodiments of the present application, some components of the drawings will be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some known structures and their descriptions in the drawings can be omitted.

[0071] The same or similar reference numerals in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that, if the terms "upper", "lower", "left", "right", "inner", "outer" and the like indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the drawings are only used for exemplary illustration, and cannot be understood as a limitation on the present patent, for those of ordinary skill in the art, the specific meanings of the above terms can be understood according to the specific circumstances.

[0072] In the description of the present application, unless otherwise explicitly specified and limited, if the terms "connection" and the like indicating the connection relationship between components appear, the term should be broadly understood, for example, it can be fixedly connected, or it can be detachably connected, or it can be integrated; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication inside two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0073] Embodiment

[0074] Reference Figures 1-4 , wherein, Figure 1 The composition module of the high-speed film production line product quality detection system based on visual recognition is shown, Figure 2 The synchronization accuracy characteristic of the line array camera horizontal frequency and the production line speed is shown, the assumed data shows that the synchronization error is controlled within 0.1 ms at the maximum design speed of 300 m / min, and the error increases linearly with the speed, this characteristic ensures that there is no misalignment between images during high-speed production, and provides a spatial alignment basis for subsequent processing; Figure 3 The compensation effect of the motion compensation module in the speed range of 50-300 m / min is shown, the image definition sharply decreases with the increase of speed (0.8-0.1) without compensation, and remains above 0.85 after compensation, verifying the adaptability of the dynamic correction algorithm to high-speed start-stop and variable-speed working conditions, and ensuring that the defect features can be identified; Figure 4 The contrast change curve before and after preprocessing is shown, after Wiener filtering + blind deconvolution restoration and gray scale equalization, the image contrast is improved from the original 0.3-0.8 to 0.7-0.95, which effectively enhances the difference between the defect area and the background, and provides high-quality input for the feature extraction module; the high-speed film production line product quality detection system based on visual recognition provided in the embodiment comprises an image acquisition module, a motion compensation module, a motion compensation module, a preprocessing module, a feature extraction module, a defect classification module and a control execution module.

[0075] The image acquisition module is used for synchronously acquiring multispectral images of the laminated film product during continuous operation of the high-speed laminated film production line.

[0076] The motion compensation module is used for compensating the acquired images in space-time alignment according to the real-time speed of the production line and the image acquisition timestamp.

[0077] The preprocessing module is used for denoising, enhancing and coarsely positioning the defect area of the compensated images.

[0078] The feature extraction module is used for extracting geometric features, texture features and spectral features of the defects based on a multiscale feature fusion algorithm.

[0079] The defect classification module is used for inputting the extracted features into a pre-trained multi-task deep learning model to output the defect type and severity level.

[0080] The control execution module is used for triggering a production line shutdown, marking or parameter adjustment instruction according to the defect classification result.

[0081] The image acquisition module adopts a combination of a linear array camera array and area exposure technology, and the line frequency of the linear array camera and the production line speed are synchronized through an encoder to ensure that there is no splicing misplacement between image rows. The multispectral images include image data in the visible light, near-infrared and ultraviolet wave bands.

[0082] The high-speed laminated film production line product quality detection system based on visual recognition solves the detection problems caused by motion blur, complex defect types (surface / interior defects coexist) and high real-time requirements in the high-speed laminated film production line through multi-module collaborative design (image acquisition-motion compensation-feature processing-intelligent classification-linkage control), realizes full-process and high-robustness detection of the quality of the laminated film product, ensures that the detection result is synchronized with the production line operation state, and provides a reliable basis for subsequent quality control.

[0083] The image acquisition module acquires original images (such as camera and sensor data) and transmits them to the preprocessing module for basic processing such as denoising, grayscale conversion and size normalization.

[0084] The preprocessed images are input into the motion compensation module to correct image displacement caused by equipment vibration, transmission delay, etc. (such as electronic image stabilization and optical flow compensation).

[0085] The compensated stable images enter the feature extraction module to extract key features (such as edges, textures, geometric shapes and color features) to provide a basis for defect recognition.

[0086] Feature data input defect classification module (such as rule-based, machine learning or deep learning model), to determine whether there is a defect and the type of defect (such as cracks, stains, size deviation, etc.);

[0087] The classification result is output to the control execution module to drive the equipment to perform corresponding actions (such as sorting qualified / unqualified products, adjusting production parameters, alarming, etc.).

[0088] Specifically, in the present embodiment, the image acquisition module comprises: at least two groups of linear array industrial cameras, a stroboscopic light source group and an encoder synchronization device, wherein:

[0089] The at least two groups of linear array industrial cameras are arranged at longitudinal and transverse detection stations of the film production line respectively, the longitudinal camera is used for detecting quality defects in the width direction of the film, and the transverse camera is used for detecting continuous defects in the length direction of the film;

[0090] The stroboscopic light source group is synchronously triggered with the linear array camera, the wavelength of the light source covers the visible light 400-700nm band, the near-infrared light 700-1000nm band and the ultraviolet light 300-400nm band, and the flash duration is ≤1μs, so as to ensure that the image is clear without trailing in high-speed motion;

[0091] The encoder synchronization device is installed on the transmission shaft of the production line, outputs pulse signals to the linear array camera and the stroboscopic light source, realizes frame synchronization of image acquisition and production line motion, and the synchronization error is ≤0.1ms.

[0092] Through the above technical scheme, the linear array camera array (longitudinal + transverse arrangement) is adopted to cover the detection requirements in the width and length directions of the film, and the multi-spectrum stroboscopic light source (visible light / near-infrared / ultraviolet) is synchronously triggered, so as to solve the problem of insufficient recognition ability of the traditional single-camera single-band detection to the surface details (such as small impurities) and internal structure (such as delamination, bubbles) of the film; the encoder synchronization device ensures that the image acquisition and the production line motion are strictly synchronized, avoids image misplacement or trailing in high-speed motion, and provides a clear original data basis for subsequent image processing.

[0093] Specifically, in the present embodiment, the motion compensation module comprises: a speed acquisition unit, a timestamp recording unit and an image registration unit, wherein:

[0094] The speed acquisition unit acquires the real-time linear speed v(t) and acceleration a(t) of the production line through the encoder;

[0095] The timestamp recording unit records the acquisition timestamp t i for each frame of image;

[0096] The displacement compensation calculation unit calculates the displacement offset of the image in the direction of the production line by using the following equation:

[0097]

[0098] an image registration unit, which corrects the image translation based on Δx by bilinear interpolation;

[0099] wherein:

[0100] Δt is the image acquisition time interval;

[0101] λ is the acceleration attenuation coefficient (calibration value, typical range 0.1-0.5);

[0102] sgn is the sign function;

[0103] is the acceleration weight attenuation term.

[0104] Parameter definition

[0105]

[0106] Example

[0107] Scenario: the production line at t i momentary parameters:

[0108] υ(t i ) = 10 m / s;

[0109] a(t i ) = 2 m / s 2 (acceleration state);

[0110] Δt = 0.5 ms = 0.0005 s;

[0111] λ = 0.3 (calibration value).

[0112] Calculation process:

[0113]

[0114] Execution: the image is translated 0.005000137 m in the opposite direction of the motion.

[0115] Technical effects

[0116] (1) solving the displacement error caused by sudden change of speed

[0117] Limitations of traditional solutions: only linear compensation formula Δx = υ·Δt is used, which cannot adapt to the nonlinear displacement deviation caused by acceleration change.

[0118] This equation compensates for nonlinear displacement by introducing the acceleration term ​Dynamic adjustment of acceleration weight; displacement compensation accuracy is significantly improved, and image misalignment is significantly reduced in the scene of rapid acceleration / deceleration of the production line.

[0119] (2) Inhibit high-frequency vibration interference

[0120] Limitations of traditional solutions: completely ignore the influence of acceleration fluctuation, and the image misalignment is serious under vibration working condition.

[0121] The equation: through the exponential decay term Dynamic inhibition of high-frequency vibration (the greater |a| is, the smaller the weight is), avoid over-compensation caused by mechanical jitter; the image misalignment rate under high-frequency vibration working condition is significantly reduced, which can ensure the image stability during high-speed operation.

[0122] (3) Improve the adaptability of the scene of emergency stop / emergency start

[0123] Limitations of traditional solutions: uniform acceleration model When the error reaches the set value when |a|>5m / s 2 When the error reaches the set value when |a|>5m / s

[0124] (4) The sign function sgn(a) of the equation distinguishes the acceleration / deceleration direction and ensures the correct displacement compensation direction; the decay term Inhibit acceleration overshoot and avoid over-compensation when emergency stop.

[0125] By using the above technical solutions, the production line speed is collected in real time, and the image displacement offset is dynamically calculated and corrected by combining the image timestamp, which effectively eliminates the image spatial misalignment problem caused by production line speed fluctuation (such as start / stop, speed change), ensures that the images collected at different times are strictly aligned in the spatial dimension, and provides an accurate image basis for subsequent feature extraction and defect positioning.

[0126] Specifically, in the embodiment, the preprocessing module includes a motion blur correction submodule, a gray equalization submodule, and an ROI coarse positioning submodule, wherein:

[0127] The motion blur correction submodule uses Wiener filtering combined with blind deconvolution algorithm to restore the image blur caused by high-speed motion of the production line, and the blur kernel parameters are adaptively estimated by image gradient variance;

[0128] The gray equalization submodule enhances the contrast of the defect area and the background based on the CLAHE (Contrast Limited Adaptive Histogram Equalization) algorithm;

[0129] The ROI coarse positioning submodule quickly extracts the area where defects may exist by adaptive threshold segmentation combined with morphological operation, and reduces the calculation range of subsequent feature extraction.

[0130] The above technical scheme is adopted to solve the problems of image blurring (such as motion blurring) caused by high-speed motion and low contrast between defects and background, and to significantly improve the image quality, reduce the computational complexity of subsequent feature extraction, and avoid the interference of irrelevant areas on the detection result by motion blurring correction (restoring clear details), gray scale equalization (enhancing the difference between defects and background), and ROI coarse positioning (reducing the subsequent processing range).

[0131] Specifically, in the embodiment, the feature extraction module adopts a multi-scale feature fusion network, including a shallow feature extraction branch, a deep semantic feature extraction branch, a spectral feature fusion branch, and a feature fusion unit, wherein:

[0132] The shallow feature extraction branch uses a 3x3 convolution kernel to extract the edge and corner low-level geometric features of the image;

[0133] The deep semantic feature extraction branch uses the backbone network of ResNet-50 to extract texture and wrinkle high-level semantic features;

[0134] The spectral feature fusion branch performs principal component analysis (PCA) dimension reduction on the multispectral image, and extracts the principal components of each waveband as spectral features;

[0135] The feature fusion unit splices the feature maps of the three branches in the spatial dimension through a skip connection, and then reduces the dimension through a 1x1 convolution to obtain the fused multi-dimensional defect features.

[0136] Through the above technical scheme, the multi-scale feature fusion network (shallow geometric features + deep semantic features + spectral features) is adopted, which not only retains the edge, corner and other detail information of the defects (suitable for micro defect recognition), but also extracts texture, wrinkle and other high-level semantic features (suitable for complex defect classification), and combines multispectral features (supplementing internal defect information that cannot be captured by visible light), thereby comprehensively improving the feature expression ability of different types and sizes of defects and enhancing the adaptability of the detection system.

[0137] Specifically, in the embodiment, the pre-trained multi-task deep learning model of the defect classification module includes a shared feature extraction layer and a classification head, wherein:

[0138] The shared feature extraction layer input is a multi-dimensional defect feature;

[0139] The classification head includes a defect type classification branch (using a Softmax activation function, outputting common defect categories such as bubbles, wrinkles, uneven thickness, impurities, etc.) and a severity level regression branch (using a Sigmoid activation function, outputting a severity score between 0 and 1);

[0140] The loss function is a weighted sum of the classification cross-entropy loss and the mean square error loss, and the weight is dynamically adjusted according to the importance of the defect type.

[0141] By using the technical scheme, the defect type and the severity level are synchronously output by the multi-task deep learning model (classification + regression), so that the redundancy of training multiple models by the traditional single-task model is avoided; the dynamic loss function adjusts the weight according to the importance of the defect, so that the learning of the model on the key defect (such as a fatal defect) is more sufficient, and the classification accuracy on a complex defect scene is improved, thereby providing a reliable decision basis for subsequent control execution.

[0142] Specifically, in the embodiment, the control execution module comprises a first alarm unit, a second early warning unit and a parameter adjustment unit, wherein:

[0143] When a fatal defect (such as a large-area wrinkle or a penetrating bubble) with a severity level greater than or equal to 0.8 is detected, the first alarm unit immediately triggers an emergency stop signal to the production line PLC and controls the marking device to spray a code at the defect position.

[0144] When a secondary defect (such as local uneven thickness or a small impurity) with a severity level of 0.3-0.8 is detected, the second early warning unit records the defect position and displays it on the human-machine interface, prompting the operator to check the upstream process parameters.

[0145] The parameter adjustment unit communicates with the production line control system through an industrial Internet of Things (IIoT) interface and automatically adjusts the film coating temperature, coating speed or pressing pressure according to the historical defect data statistics.

[0146] Through the above technical scheme, the hierarchical response mechanism (first emergency stop + second early warning + parameter adjustment) based on the defect severity level is realized, so that the intelligent control of the production line is realized: fatal defects are quickly stopped and marked to avoid the production of a large number of unqualified products; secondary defects are prompted to check the upstream process to reduce invalid downtime; the production process is optimized by automatic parameter adjustment to reduce the defect rate from the source, thereby improving the timeliness and effectiveness of production control.

[0147] Specifically, in the embodiment, a self-calibration module is further included for periodically calibrating the precision of the detection system, specifically comprising: a standard sample library, an automatic feeding and unloading unit and a calibration evaluation unit, wherein:

[0148] The standard sample library stores standard film samples with known defect types and parameters;

[0149] The automatic feeding and unloading unit transports the standard sample to the detection station;

[0150] The calibration evaluation unit calculates the defect recognition accuracy, the missed detection rate and the positioning error by comparing the system detection result with the labeled data of the standard sample, and triggers the system parameter correction if the error exceeds the threshold.

[0151] Closed-loop control of the self-calibration module

[0152] Calibration trigger: The self-calibration module can actively trigger (or receive external instructions) to calibrate the image acquisition module (such as camera calibration and light source calibration), the motion compensation module (reference position calibration) or the feature extraction module (parameter optimization) when the system is started or the detection accuracy is abnormal.

[0153] Calibration feedback: The calibrated parameters are fed back to the image acquisition module (to adjust the acquisition parameters), the preprocessing module (to update the filtering rules) or the feature extraction module (to optimize the feature operator), forming a closed-loop optimization.

[0154] By using the above technical solutions, the system detection accuracy (accuracy, missed detection rate and positioning error) is calibrated regularly by the standard sample library and the automatic feeding and unloading unit, avoiding the detection performance decline caused by equipment aging, environmental changes and other factors; the calibration evaluation unit automatically corrects the system parameters, ensuring the long-term stable operation of the detection system, reducing the manual maintenance cost and ensuring the reliability of the detection result.

[0155] Specifically, in the embodiment, the multispectral image acquisition method includes a time-sharing exposure strategy and a waveband registration unit, wherein:

[0156] The time-sharing exposure strategy uses short exposure 5μs for visible light waveband to capture surface details, uses medium exposure 20μs for near-infrared waveband to penetrate the surface layer of the film to detect the internal structure, and uses long exposure 50μs for ultraviolet waveband to excite fluorescent defects;

[0157] The waveband registration unit aligns the multispectral images at the pixel level through a feature point matching algorithm (such as SIFT), eliminating the spatial offset caused by lens distortion or light difference in different wavebands.

[0158] By using the above technical solutions, the time-sharing exposure strategy (short exposure of visible light to capture surface details, medium exposure of near-infrared to penetrate the surface layer, and long exposure of ultraviolet to excite fluorescent defects) combined with waveband registration solves the problem that single waveband images are not sensitive to internal defects (such as bubbles and delamination) of the film, and eliminates the spatial offset caused by lens distortion or light difference in different wavebands, significantly improving the information integrity and consistency of the multispectral images, and providing more comprehensive defect information for subsequent feature extraction.

[0159] Specifically, in the embodiment, a data storage and traceability module is further included, which is configured to:

[0160] Real-time storage of detection results (defect type, location, timestamp, severity level) and corresponding production line process parameters (temperature, speed, coating amount) to local database;

[0161] Raw images, preprocessing results, feature data, classification labels, control instructions, and other full-process data are synchronized to the data storage traceability module for quality traceability, fault analysis, or algorithm optimization.

[0162] Data is uploaded to the cloud platform through the OPCUA protocol, supporting defect distribution visualization analysis based on the timeline or product batch, and process parameter and defect type association rule mining.

[0163] By using the above technical solutions, the detection results and process parameters are stored locally, supporting defect distribution visualization analysis based on the timeline or batch, facilitating quick problem identification. Cloud data upload supports process parameter and defect type association rule mining, providing data support for production process optimization (such as adjusting temperature and coating speed), promoting the upgrade from "post-detection" to "pre-prevention" quality control mode.

[0164] Specifically, in this embodiment, the training method of the multi-task deep learning model includes:

[0165] Data augmentation, rotating (±15°), scaling (0.8-1.2 times), and adding Gaussian noise (σ=0.01) operations are performed on the film defect samples to expand the training set size;

[0166] Transfer learning, using a pre-trained ResNet-50 on the ImageNet dataset as the backbone network, freezing the first 3 convolutional layers, and fine-tuning the last 5 layers and the classification head;

[0167] Small sample learning, for rare defect types (such as complex wrinkles + impurities), using meta-learning (MAML) algorithm to optimize the initial parameters of the model, only 10-20 samples are needed to complete fine-tuning.

[0168] By using the above technical solutions, data augmentation (rotation, scaling, and noise addition) expands the training set size, avoiding model overfitting in small sample scenarios; transfer learning (fine-tuning based on ImageNet pre-trained model) utilizes general visual features to accelerate model convergence, reducing model training cost; small sample learning (meta-learning MAML) requires only a small number of samples for rare defects to complete fine-tuning, improving the generalization ability for complex defect scenarios and ensuring the system's ability to quickly adapt to new defect types.

[0169] In summary, the high-speed film production line product quality detection system based on visual recognition provided in this embodiment has the following advantages:

[0170] Through the whole-link design of "multi-spectral acquisition-motion compensation-multi-scale feature fusion-multi-task intelligent decision-making-linkage control", the core problems of high-speed film detection are systematically solved, and the following technical effects are formed:

[0171] 1. Full-scene adaptability: Multi-spectral image acquisition covers visible light (surface details), near-infrared (internal structure), and ultraviolet (fluorescent defects) bands, combined with a motion compensation module to eliminate the effects of speed fluctuations, which can simultaneously identify surface micro-defects (such as impurities and scratches) and internal hidden defects (such as delamination and bubbles), covering all types of defects in film products.

[0172] 2. High-speed and high-robustness detection: Linear camera array and encoder synchronization technology ensures no misalignment between image rows, and the motion compensation module dynamically corrects displacement deviation, combined with motion blur correction and grayscale enhancement preprocessing, even under high-speed start-stop or variable-speed working conditions, it can still output clear and aligned images, ensuring detection stability.

[0173] 3. Intelligent decision-making and closed-loop control: Multi-task deep learning model simultaneously outputs defect types (such as bubbles and wrinkles) and severity levels (such as slight and fatal), combined with a grading response mechanism (emergency stop, warning, and parameter adjustment), achieving a closed loop from "detection" to "intervention" to avoid the production of batch unqualified products and reduce defect rates from the source.

[0174] 4. Long-term reliability and process optimization: The self-calibration module verifies system accuracy through standard samples regularly, avoiding performance degradation caused by equipment aging or environmental changes; the data storage and traceability module associates defect data with process parameters, supporting defect distribution analysis and process optimization, and upgrading quality control from "post-detection" to "prevention".

[0175] Working principle

[0176] The system is based on the intelligent detection logic of "perception-processing-decision-execution", and the specific workflow is as follows:

[0177] 1. Multi-spectral image perception:

[0178] After the production line starts, the encoder collects the transmission shaft speed (linear speed v(t)) in real time, synchronously triggers the longitudinal (width direction) and transverse (length direction) linear camera and multi-spectral stroboscopic light source (visible light / near-infrared / ultraviolet time-sharing exposure). The original images collected by the camera are transmitted to the preprocessing module through a gigabit network.

[0179] 2. Motion compensation and image preprocessing:

[0180] Motion compensation module calculates the displacement offset Δx of the image in the production line direction according to v(t) and image timestamp t_i, and corrects the image misalignment through bilinear interpolation; the preprocessing module sequentially performs motion blur restoration (Wiener filter), gray balance enhancement (CLAHE), and ROI coarse positioning (adaptive threshold segmentation), and outputs a clear defect candidate region image.

[0181] 3. Multi-dimensional feature extraction:

[0182] The feature extraction module performs multi-scale feature fusion on the preprocessed image: the shallow branch (3x3 convolution) extracts geometric features such as edges and corner points; the deep branch (ResNet-50 backbone) extracts semantic features such as texture and wrinkles; the spectral branch (PCA dimensionality reduction) extracts multi-spectral principal component features. The three-branch features are concatenated through a skip connection and then reduced in dimension to form a multi-dimensional defect feature vector.

[0183] 4. Intelligent defect classification and decision-making:

[0184] The multi-task deep learning model (shared feature extraction layer + classification / regression head) inputs multi-dimensional features and outputs defect types (Softmax classification) and severity levels (Sigmoid regression). The loss function dynamically adjusts the weights of the classification and regression tasks to ensure the learning priority of critical defects (such as fatal defects).

[0185] 5. Production line linkage control:

[0186] The control execution module triggers a hierarchical response according to the defect severity level: fatal defects (severity level ≥ 0.8) immediately trigger an emergency stop and mark with a code; secondary defects (0.3 ≤ level < 0.8) record the position and prompt to investigate the upstream process; the system automatically adjusts parameters such as film temperature and coating speed through the Industrial Internet of Things (IIoT) interface to reduce defects from the source.

[0187] 6. System maintenance and process optimization:

[0188] The self-calibration module periodically calls the standard sample library, transports the sample to the detection station through the automatic feeding and unloading unit, compares the detection results with the standard annotation data by the evaluation unit, and corrects the system parameters; the data storage module stores the detection results and process parameters in real time, and the cloud platform supports defect distribution visualization and process parameter-defect correlation analysis to guide process optimization.

[0189] Method of use

[0190] The use of the system should follow the following steps to ensure efficient and stable operation:

[0191] 1. System deployment and initialization:

[0192] Hardware installation: Fix the line array camera at the key stations of the production line in the longitudinal direction (width direction) and the transverse direction (length direction), arrange the multi-spectral stroboscopic light source (covering visible light / near-infrared / ultraviolet wave band), and install the encoder on the transmission shaft.

[0193] Software configuration: Set the production line parameters (such as standard width, thickness, target coating amount) through the human-machine interface (HMI), define the defect types (such as bubbles, wrinkles, etc.) and the severity threshold (such as grade ≥ 0.8 as fatal defect).

[0194] 2. Production line synchronous calibration:

[0195] Start the production line at low speed, calibrate the camera trigger time and the production line speed through the encoder synchronization module, ensure that there is no misalignment between image rows (synchronization error ≤ 0.1 ms), and verify the system detection accuracy (accuracy, missed detection rate) using standard samples (containing known defects), and adjust the model parameters until the standard is met.

[0196] 3. Normal production detection:

[0197] After the production line enters the high-speed running mode, the system automatically triggers the camera and light source to synchronously collect images; after the images are motion compensated, preprocessed, and feature extracted, they are input into the multi-task model for analysis, and the defect type and severity level are output; the control execution module triggers the emergency stop, marking or parameter adjustment (such as increasing the temperature to improve the uniformity of the coating) according to the level.

[0198] 4. Daily maintenance and data management:

[0199] Periodically (such as every week), use the standard sample library for system self-calibration, evaluate the detection accuracy (accuracy, positioning error), and if the error exceeds the threshold, automatically correct the model parameters.

[0200] Real-time view of the defect statistical report of the data storage module (such as the distribution of defect types in each period), analyze the correlation between process parameters (such as temperature, speed) and defects through the cloud platform, and optimize the production process (such as adjusting the coating speed to reduce the thickness unevenness defect).

[0201] Summary: The present application realizes the intelligentization, closed loop and high efficiency of the product quality detection of the high-speed curtain coating production line, solves the problems of poor image quality, incomplete defect recognition, and disconnection between detection and production in the traditional detection method in the high-speed scene, and provides a full-process solution for the quality control of curtain coating production.

[0202] The above is only the preferred embodiment of the present application, and does not limit the implementation and protection scope of the present application. For those skilled in the art, it should be realized that any equivalent replacement and obvious change made by applying the contents of the present application should be included in the protection scope of the present application.

Claims

1. A product quality detection system for a high-speed film-coating production line based on visual recognition, characterized in that, The application relates to a high-speed film production line quality defect detection system, which comprises the following: an image acquisition module for synchronously acquiring multispectral images of film products during continuous operation of a high-speed film production line; a motion compensation module for performing space-time alignment compensation on the acquired images according to a real-time speed of the production line and an image acquisition timestamp; a preprocessing module for denoising, enhancing and coarsely positioning a defect area of the compensated images; a feature extraction module for extracting geometric features, texture features and spectral features of the defect based on a multiscale feature fusion algorithm; a defect classification module for inputting the extracted features into a pre-trained multi-task deep learning model to output a defect type and a severity level; a control execution module for triggering a production line shutdown, marking or parameter adjustment instruction according to the defect classification result; wherein the image acquisition module adopts a combination of a linear array camera array and a area exposure technology, the line frequency of the linear array camera is synchronized with the production line speed through an encoder to ensure that there is no splicing dislocation between image rows; and the multispectral images comprise image data in the visible light, near-infrared and ultraviolet wave bands.

2. The product quality detection system for high-speed curtain coating production line based on visual recognition according to claim 1, characterized in that, The image acquisition module comprises: at least two groups of linear array industrial cameras arranged at longitudinal and transverse detection stations of the film production line, wherein the longitudinal cameras are used for detecting quality defects in the width direction of the film, and the transverse cameras are used for detecting continuous defects in the length direction of the film; a strobe light source group which is synchronously triggered with the linear array cameras, wherein the light source wavelength covers the visible light 400-700 nm wave band, the near-infrared light 700-1000 nm wave band and the ultraviolet light 300-400 nm wave band, and the flash duration is less than or equal to 1 mu s to ensure that the image is clear without trailing. An encoder synchronization device is installed on a transmission shaft of the production line, and outputs pulse signals to the linear array cameras and the strobe light source to realize frame synchronization of image acquisition and production line movement. 3.The vision recognition based product quality detection system for high-speed curtain coating production line according to claim 1, characterized in that, The motion compensation module comprises: a speed acquisition unit for acquiring the real-time line speed v(t) and acceleration a(t) of the production line through an encoder; a time stamp recording unit records a capture time stamp t for each frame of image i ; a displacement compensation calculation unit for calculating the displacement offset of the image in the production line direction by using the following equation: an image registration unit for correcting the image by bilinear interpolation based on the calculated Delta x to eliminate the image dislocation caused by speed fluctuation; where Δt is the image acquisition time interval; λ is the acceleration attenuation coefficient; sgn is the sign function; υ(t i ) is the real-time linear velocity at time t i ; a(t i ) is the real-time acceleration at time t i .

4. The product quality detection system for high-speed curtain coating production line based on visual recognition according to claim 1, characterized in that, The preprocessing module comprises: a motion blur correction submodule for restoring the image blur caused by high-speed movement of the production line by using a Wiener filter combined with a blind deconvolution algorithm, and adaptively estimating the blur kernel parameters through the image gradient variance; a gray scale equalization submodule for enhancing the contrast of the defect area and the background based on the CLAHE algorithm; an ROI coarse positioning submodule for quickly extracting the area where the defect may exist by using adaptive threshold segmentation combined with morphological operation, and reducing the calculation range of subsequent feature extraction.

5. The visual recognition-based product quality detection system for high-speed curtain coating production line according to claim 1, characterized in that, The feature extraction module adopts a multiscale feature fusion network, which comprises: a shallow feature extraction branch for extracting low-level geometric features of edges and corners of the image by using a 3*3 convolution kernel; a deep semantic feature extraction branch for extracting high-level semantic features of texture and wrinkles by using a ResNet-50 backbone network; a spectral feature fusion branch for performing principal component analysis dimension reduction on the multispectral images, and extracting principal components of each wave band as spectral features. The feature fusion unit splices the feature maps of the three branches in the spatial dimension through a skip connection, and then reduces the dimension through a 1x1 convolution to obtain multi-dimensional defect features after fusion. 6.The vision recognition based product quality detection system for high-speed curtain coating production line according to claim 1, wherein, The pre-training multi-task deep learning model of the defect classification module includes: A shared feature extraction layer with an input of multi-dimensional defect features; A classification head including a defect type classification branch and a severity level regression branch; The loss function is a weighted sum of classification cross-entropy loss and mean square error loss, and the weight is dynamically adjusted according to the importance of the defect type. 7.The vision recognition based product quality detection system for high-speed curtain coating production line according to claim 1, wherein, The control execution module includes: A first alarm unit that triggers an emergency stop signal to the production line PLC when a fatal defect with a severity level ≥ 0.8 is detected, and controls the marking device to spray a code at the defect position; A second early warning unit that records the defect position and displays it on the human-machine interface when a secondary defect with a severity level of 0.3-0.8 is detected, prompting the operator to check the upstream process parameters; A parameter adjustment unit that communicates with the production line control system through an industrial Internet of Things interface and automatically adjusts the film temperature, coating speed, or pressing pressure based on historical defect data statistics. 8.The vision recognition based product quality detection system for high-speed curtain coating production line according to claim 1, wherein, It also includes a self-calibration module for periodic precision calibration of the detection system, which specifically includes: A standard sample library that stores standard film samples with known defect types and parameters; An automatic loading and unloading unit that transports the standard sample to the detection station; A calibration evaluation unit that calculates the defect recognition accuracy, missed detection rate, and positioning error by comparing the system detection results with the labeled data of the standard sample, and triggers system parameter correction if the error exceeds the threshold. 9.The vision recognition based product quality detection system for high-speed curtain coating production line according to claim 1, wherein, The multi-spectral image acquisition method includes: A time-sharing exposure strategy that uses short exposure 5μs to capture surface details in the visible light band, medium exposure 20μs to penetrate the film surface layer to detect internal structures in the near-infrared band, and long exposure 50μs to excite fluorescent defects in the ultraviolet band; A band registration unit that aligns multi-spectral images at the pixel level through a feature point matching algorithm to eliminate spatial shifts caused by lens distortion or lighting differences in different bands. 10.The vision recognition based product quality detection system for high-speed curtain coating production line according to claim 1, wherein, It also includes a data storage and traceability module for: Storing detection results and corresponding production line process parameters in a local database in real time; Uploading data to the cloud platform through OPCUA protocol to support defect distribution visualization analysis based on time axis or product batch, and process parameter and defect type association rule mining.

Images