Vision analysis based adhesive tape surface defect detection system

Abstract

The application discloses a visual analysis-based adhesive tape surface defect detection system, and relates to the technical field of computer vision and adhesive tape quality inspection.The system comprises a double-light-path acquisition module, a synchronous registration module, a pixel difference operation module, an interface feature extraction module and a defect classification output module.The system acquires adhesive tape surface layer reflection and interlayer transmission optical signals through a double-light-path synchronous acquisition and spatial registration framework, realizes two-way image pixel-level time sequence and spatial alignment, completely covers full-width quality information, and eliminates the internal interface defect detection blind area of single imaging.Meanwhile, the system filters normal coating texture interference, accurately identifies hidden interface defects, uniformly maps multiple defects to a production line global coordinate system, realizes defect classification output and position correlation, can be connected to a production line control and data storage, and effectively improves the accuracy and operation efficiency of adhesive tape quality control.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and tape quality inspection technology, specifically a tape surface defect detection system based on visual analysis. Background Technology

[0002] Adhesive tape, as a core material for adhesive encapsulation and structural bonding, is widely used in industrial fields such as consumer electronics, new energy power batteries, and food and pharmaceutical packaging. The quality stability of its surface and internal adhesive layers directly affects the sealing performance, bonding strength, and reliability of the final product. During the high-speed coating and slitting production of adhesive tape, surface defects such as scratches, pits, and foreign matter adhesion are easily generated due to factors such as raw material ratio, coating process, and rolling precision, as well as internal interface defects such as uneven adhesive layer thickness, interlayer bubbles, and substrate delamination. Currently, machine vision inspection has become the mainstream method for surface quality control of adhesive tape, but the multi-layer composite structure of adhesive tape places higher demands on the dimensional coverage and recognition accuracy of defect detection.

[0003] Existing visual inspection solutions for adhesive tapes mostly employ a single-path reflective light imaging architecture, which can only acquire reflected optical signals from the surface morphology of the tape. While it can identify visible surface defects such as scratches and breaks, it cannot penetrate the adhesive layer and substrate to capture abnormal information at the interlayer interface. This results in blind spots for hidden defects such as insufficient adhesive, interlayer bubbles, and substrate delamination, leading to a high rate of missed detections. In some solutions employing dual-path acquisition, the two imaging systems are triggered independently. Affected by production line vibrations and speed fluctuations, the two images exhibit significant temporal discrepancies and spatial coordinate misalignments, failing to achieve pixel-level precise correspondence and directly limiting the effectiveness of subsequent multi-path fusion operations. Furthermore, existing dual-path solutions often use simple image overlay or fixed threshold segmentation, failing to effectively separate surface texture signals from interface abnormal signals. This easily leads to misjudging normal coating textures as defects and insufficient feature extraction capabilities for minute interface defects, making it difficult to accurately distinguish defect types and achieve unified mapping of production line locations.

[0004] In summary, existing tape surface defect detection technologies consistently suffer from core challenges such as limited detection dimensions, insufficient dual-optical-path registration accuracy, and low interface defect feature recognition. They cannot simultaneously detect both surface morphology defects and interlayer interface defects comprehensively, and the global coordinate mapping accuracy of defect locations is insufficient, making it difficult to meet the online detection needs of high-speed production lines for full-area, multi-type defects. To address these technical challenges, there is an urgent need to develop a detection system based on simultaneous dual-optical-path acquisition and pixel differential analysis. This system achieves precise alignment of the two images through temporal synchronization and spatial registration, and extracts interface anomaly information by combining differential feature separation. Ultimately, it enables high-precision detection and classification of tape defects across all dimensions, meeting the stable quality control requirements of industrial mass production scenarios. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a visual analysis-based tape surface defect detection system. This system can acquire surface images of reflected light and transmitted light from the tape, synchronize them temporally and register them spatially, and then perform pixel-level differential calculations to generate differential feature maps. Through connected component filtering and adaptive threshold determination, it separates surface texture interference and interface anomaly signals, extracting defect geometric and grayscale features. Finally, it completes defect classification and maps surface and interface defects to the production line's global coordinate system, outputting structured detection results. This system can interface with the production line control unit and archive data, accommodating both surface and interlayer interface defects, thus improving detection comprehensiveness and recognition accuracy.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a visual analysis-based tape surface defect detection system, the system comprising: Dual-path acquisition module: Acquires the reflected light surface image and the transmitted light transmission image of the tape under test. The two acquisition fields of view overlap and cover the full width of the tape. Obtain the optical signals corresponding to the surface morphology and interlayer structure of the tape, and output two raw image data. Synchronous registration module: Receives two channels of raw image data, controls the acquisition triggering sequence of the two channels of images to maintain synchronization, performs spatial coordinate alignment on the two acquired images, and outputs the registered two aligned images; Pixel difference calculation module: Receives two aligned images after registration, performs grayscale preprocessing and pixel-level difference calculation, generates a difference feature map reflecting the difference between the surface layer and interlayer signals of the tape, and outputs the difference feature map data; Interface feature extraction module: Receives differential feature map data, separates the surface texture signal and interface layer abnormal signal in the differential feature map, extracts the geometric shape and gray-scale distribution features of the interface abnormal area, and outputs defect feature data. Defect classification output module: Receives defect feature data, matches the corresponding defect type, associates the tape position coordinates corresponding to the defect, and outputs complete defect detection results.

[0007] Furthermore, in the dual-optical-path acquisition module, the reflected optical path and the transmitted optical path are respectively located on both sides of the tape and their acquisition directions are coaxial and opposite. The image acquisition direction of the reflected optical path is perpendicular to the tape surface and is used to receive the optical signal reflected from the surface. The image acquisition direction of the transmitted optical path is opposite to the reflected light direction and is used to receive the optical signal that penetrates the entire tape. The field of view of the two acquisition paths are completely corresponding and together cover the entire width of the tape. During the acquisition process, the illumination unit is driven by a constant current source to ensure that the light intensity fluctuation does not exceed ±5%, and the acquisition parameters of the two images are kept uniform. The two output original images carry corresponding acquisition timing identifiers and position identifiers, respectively.

[0008] Furthermore, the synchronous registration module uses the production line running speed pulse signal as the common trigger source for the acquisition of the two images, so that the start time of the acquisition of the two images is consistent. After the acquisition is completed, the contour lines of the two sides of the tape in the two images are extracted, and the two images are transformed by translation and scaling based on the contour lines until the pixel coordinates of the tape edges in the two images are completely coincident, and the overall structure of the image is kept intact during the transformation process, generating the two aligned images after registration.

[0009] Furthermore, after receiving the two aligned images after registration, the pixel difference operation module first performs grayscale normalization processing on the two images respectively, mapping the grayscale value range of the two images to the same numerical range, and then performs smoothing filtering processing on the two normalized images respectively to filter out random noise in the images. After filtering, the preprocessed reflected light image and transmitted light image are obtained. The two preprocessed images maintain the same resolution and pixel coordinate system.

[0010] Furthermore, the pixel difference calculation module performs pixel-by-pixel difference calculation on the two preprocessed images. It traverses all pixel coordinate positions in a point-by-point scanning manner, performs a difference calculation on the gray values ​​of reflected light and transmitted light at the same coordinate, and takes the absolute value. This absolute value is used as the gray value of the difference feature map at that coordinate. The specific formula is as follows: ,in, This represents the grayscale value of the difference feature map at coordinates (x, y). The grayscale value of the reflected light image; The grayscale values ​​are the transmitted light image values; after traversing all pixel coordinates, a complete differential feature map is generated.

[0011] Furthermore, when the interface feature extraction module separates the surface texture signal and the interface layer abnormal signal, it performs connected component analysis on the differential feature map, extracts all connected regions whose gray levels differ from the background, statistically analyzes the size attributes of each connected region, filters out tiny noise regions whose area and shape do not conform to the interface defect characteristics, retains candidate abnormal regions that conform to the size range, completes the preliminary screening of interface abnormal signals, and eliminates the influence of residual interference signals.

[0012] Furthermore, the interface feature extraction module performs grayscale threshold determination on the filtered candidate abnormal regions, counts the grayscale values ​​of all pixels in the current frame's differential feature map, and generates an adaptive differential grayscale threshold by combining it with a preset sensitivity coefficient. The specific formula is as follows: Where Th is the differential grayscale threshold; μ is the global grayscale mean of the differential feature map of the current frame; σ is the global grayscale standard deviation of the differential feature map of the current frame; k is a preset sensitivity coefficient, which is pre-calibrated according to the sensitivity requirements of the detection scene, and the value range is 1.8-2.8; the average grayscale value in the candidate area is compared with Th, and the area higher than Th is determined as the interface abnormal area; the geometric shape features and grayscale distribution features of the interface abnormal area are extracted. The geometric shape features include the area, aspect ratio and shape regularity of the area, and the grayscale distribution features include the grayscale mean and grayscale uniformity within the area; a built-in defect feature matching rule is used to compare the actual features of the candidate abnormal area with the defect feature matching rule one by one to determine the defect type of the corresponding interface abnormal area, and to generate a dataset containing the feature parameters of each interface abnormal area; the defect feature matching rule is a pre-set judgment logic based on the defect geometric shape and grayscale distribution features, which includes the correspondence between feature parameter thresholds and defect types. The actual features of the candidate abnormal area are compared with the rule one by one to determine the defect type.

[0013] Furthermore, after receiving the defect feature data, the defect classification output module converts the pixel coordinates of the abnormal area into the physical coordinates of the tape, and matches the longitudinal position of the tape corresponding to the defect with the production line running sequence; at the same time, it obtains the surface defect coordinate information identified by the reflected light image, and maps the coordinates of the surface defect and the interface defect to the same global coordinate system of the production line, so that multiple types of defects at the same physical location correspond to the same location identifier.

[0014] Furthermore, the defect classification output module integrates all defect type information, location coordinate information, and distribution range information to generate structured defect detection results. Each defect record in the defect detection results corresponds to a unique location identifier. The defect detection results are output to the outside through a communication interface, which can be connected to the production line's execution control unit to output corresponding trigger signals based on the defect detection results. At the same time, all defect records are written to the local storage unit to form a complete defect detection result archive.

[0015] Compared with existing technologies, this vision analysis-based tape surface defect detection system has the following advantages: I. This invention utilizes an integrated architecture of dual-path acquisition and synchronous registration to simultaneously acquire images of the reflected light surface and transmitted light penetration of the tape. A unified trigger source controls the timing of the two acquisition paths to maintain synchronization, and precise alignment of spatial coordinates is achieved based on the tape edge contour, enabling pixel-level correspondence between the two images in both temporal and spatial dimensions. This architecture can simultaneously acquire optical signals of the tape surface morphology and interlayer structure, fully covering the quality information of the entire tape area, solving the blind spot problem of internal interface defect detection in single-path imaging schemes, and improving the comprehensiveness of defect detection. The two images use unified acquisition parameters and stable illumination conditions to ensure the consistency and comparability of optical signals, providing an accurate and reliable data source for subsequent differential feature calculations, and adapting to the online inspection needs of high-speed continuous production scenarios.

[0016] II. This invention utilizes pixel-level differential operations and interface feature layering extraction mechanisms to perform grayscale preprocessing and pixel-by-pixel difference calculations on the registered two-channel images, generating feature maps reflecting the differences between surface and interlayer signals. Then, through connected component filtering and feature judgment, surface texture interference and interface abnormal signals are separated, accurately extracting the morphological and distribution features of defects. This mechanism effectively filters out misjudgment interference caused by normal coating textures on the tape, improving the accuracy of identifying hidden interface defects. Simultaneously, it maps surface and interface defects to a unified global production line coordinate system, enabling location association and classification output for multiple types of defects. The detection results can be directly connected to the production line control unit, while also completing local data archiving, providing complete data support for production process optimization and quality traceability, thus improving the accuracy and operational efficiency of tape quality control.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

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

[0019] Figure 1 This is a block diagram showing the modular components of a vision-based tape surface defect detection system. Figure 2 The flowchart shows the dual-optical-path acquisition and synchronous registration process of a vision-based tape surface defect detection system. Figure 3This is a flowchart of the pixel difference operation execution of a visual analysis-based tape surface defect detection system. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example 1 This system can be applied to various scenarios, such as the pre-winding section after coating and curing in high-speed tape coating production lines, the offline rewinding and inspection process of finished tape rolls, and the feeding section in the precision electronic tape slitting process. It can perform full-width, uninterrupted online or offline defect detection on continuously running tape products and promptly identify various surface and interlayer quality problems generated during the production process.

[0022] In high-speed coating production scenarios, the dual-optical-path acquisition module acquires reflected light surface images and transmitted light transmission images of the tape under test. The reflected light path and the transmitted light path are located on both sides of the tape and their acquisition directions are coaxial and opposite. The image acquisition direction of the reflected light path is perpendicular to the tape surface and receives the optical signals reflected from the surface, which can clearly present surface defects such as scratches, indentations, foreign matter adhesion, and wrinkles. The image acquisition direction of the transmitted light path is opposite to the reflected light direction and receives the optical signals that penetrate the entire tape, which can penetrate the tape substrate and adhesive layer structure, revealing internal defects such as bubbles, impurities, uneven coating, and interlayer separation that are difficult to observe directly with the naked eye. The two optical paths work together to simultaneously cover the needs of surface and interlayer defect detection, avoiding the limitation of a single optical path being able to identify only a single layer of defects. The two acquisition channels have completely corresponding fields of view, jointly covering the entire width of the tape. The overlapping fields of view enable full-range, blind-spot-free detection of the tape's lateral surface. Simultaneously, optical signals corresponding to the surface morphology and interlayer structure at the same location on the tape are acquired, ensuring effective capture of various types of defects. During acquisition, a constant current source drives the illumination unit. Stable power supply current avoids brightness variations caused by voltage fluctuations, keeping illumination intensity fluctuations stable within ±3.9%, consistently within the allowable range of ±5%. This results in high consistency in imaging brightness and stable grayscale references between different frames, effectively reducing background noise interference in subsequent differential calculations. Furthermore, the acquisition parameters for both images are kept consistent, with core imaging parameters such as exposure and gain being completely identical. This ensures that the basic imaging conditions for both images are the same, avoiding systematic grayscale differences caused by differences in imaging parameters. The two output raw images carry corresponding acquisition time and location identifiers, which can correspond one-to-one with the production line operation data, providing a matching benchmark for subsequent time synchronization and spatial registration, and also providing location basis for subsequent defect tracing, ultimately outputting two raw image data.

[0023] The synchronous registration module receives two channels of raw image data. Using the production line speed pulse signal as the common trigger source for both image acquisitions, it controls the acquisition timing of the two images to remain synchronized. This ensures that the start times of the two image acquisitions are consistent, guaranteeing that the corresponding longitudinal positions of the tape in both images are the same. This avoids mismatches in the image content caused by misaligned acquisition times and eliminates spatial misalignment caused by timing deviations during tape transport. After acquisition, the contour lines of the tape's double edges are extracted from both images. The tape edges have clear grayscale boundary characteristics in both images, serving as a natural benchmark for spatial alignment. No additional markers are needed on the tape, and this does not interfere with the tape production process or the product itself. Using the contour line as a reference, spatial transformations of translation and scaling are performed on the two images to adjust their lateral position and overall proportions. This corrects positional differences caused by installation deviations and tape transport offsets in the two optical paths, until the pixel coordinates of the tape edges in the two images are completely aligned. The transformation process maintains the integrity of the overall image structure, preventing stretching, distortion, or local loss of image content, and ensuring that the morphology and relative position of defects within the tape remain unchanged. After spatial coordinate alignment of the two acquired images, their pixel coordinate systems are completely unified. The same physical location corresponds to identical pixel coordinates in both images, providing a precise spatial basis for subsequent pixel-level difference calculations and avoiding false difference signals due to positional misalignment. Finally, the registered aligned images are output.

[0024] The pixel difference operation module receives the two registered aligned images and first performs grayscale normalization on both images, mapping their grayscale values ​​to the same range. This eliminates overall grayscale deviations caused by differences in imaging optical paths and light attenuation, ensuring a consistent background grayscale baseline and preventing systematic shifts in the difference results due to varying overall brightness. Next, smoothing filtering is applied to the normalized images to remove random noise, reduce point noise from imaging device thermal noise and environmental dust interference, while preserving the main structure and boundaries of abnormal areas. This prevents noise from being amplified in subsequent difference operations, forming false defect signals and improving the signal-to-noise ratio. After filtering, preprocessed reflected and transmitted light images are obtained. These two preprocessed images maintain the same resolution and pixel coordinate system, with completely corresponding spatial positions and a unified grayscale baseline, allowing for direct subsequent difference operations. Subsequently, pixel-by-pixel difference calculation is performed on the two preprocessed images. All pixel coordinates are traversed using a point-by-point scanning method. The absolute value of the difference between the reflected light gray value and the transmitted light gray value at the same coordinate is taken as the gray value of the difference feature map at that coordinate. The formula is: ,in, This represents the grayscale value of the difference feature map at coordinates (x, y). The grayscale value of the reflected light image; This represents the grayscale value of the transmitted light image. The uniform surface layer and interlayer regions of the tape exhibit consistent grayscale trends in both images, resulting in lower grayscale values ​​after differential analysis, forming a uniform low-grayscale background area. Regions with interface anomalies will produce different grayscale responses in the two images, exhibiting higher grayscale values ​​after differential analysis. This effectively highlights the abnormal signals at the interlayer interfaces, suppresses the uniform surface texture signal, and makes the difference between interlayer defects and the background more pronounced. After traversing all pixel coordinates, a differential feature map reflecting the signal differences between the tape surface layer and interlayer regions is generated. This visually presents the distribution of interface anomalies within the tape, significantly reducing the difficulty of subsequent defect identification. The final output is the differential feature map data.

[0025] The interface feature extraction module receives differential feature map data and separates surface texture signals from interface layer abnormal signals in the differential feature map. This distinguishes genuine interface defects from normal surface textures and background noise, preventing normal production textures from being misjudged as defects. First, connected component analysis is performed on the differential feature map to extract all connected regions with gray levels different from the background. Scattered independent pixels are integrated into complete region units, facilitating subsequent feature statistics and filtering. The size attributes of each connected region are statistically analyzed, filtering out tiny noise regions whose area and shape do not conform to interface defect characteristics. Candidate abnormal regions within the specified size range are retained, completing the initial screening of interface abnormal signals, eliminating the influence of residual interference signals, reducing the amount of data processed in subsequent operations, and adapting to the inspection cycle of high-speed production lines. Then, gray-level threshold determination is performed on the filtered candidate abnormal regions. The gray-level values ​​of all pixels in the current frame's differential feature map are statistically analyzed, and an adaptive differential gray-level threshold is generated based on a preset sensitivity coefficient. The formula is: Where Th is the differential grayscale threshold; μ is the global grayscale mean of the differential feature map of the current frame; σ is the global grayscale standard deviation of the differential feature map of the current frame; and k is a preset sensitivity coefficient, pre-calibrated according to the sensitivity requirements of the detection scenario, with a value range of 1.8-2.8. In high-speed coating production scenarios, due to the high production line speed, the false negative rate of defects is extremely sensitive, and k can be calibrated to a larger value, such as 2.2, to prioritize the full detection of defects. The adaptive threshold can be dynamically adjusted according to the overall grayscale level of the current frame image, adapting to the background grayscale changes under different batches of tape substrates and different adhesive layer thicknesses, avoiding detection fluctuations caused by a fixed threshold, and improving the stability of detection. By comparing the average grayscale value in the candidate area with the differential grayscale threshold, areas higher than the threshold are determined as interface abnormal areas, which can accurately distinguish between normal texture fluctuations and real interface defects. The system extracts geometric and grayscale distribution features of abnormal interface regions. Geometric features include region area, aspect ratio, and shape regularity, while grayscale distribution features include the mean and uniformity of grayscale within the region. These multi-dimensional feature parameters comprehensively describe the attributes of abnormal regions, providing sufficient basis for defect classification. Built-in defect feature matching rules compare the actual features of candidate abnormal regions with the specified rules to determine the defect type of the corresponding interface abnormal region. It can accurately identify different types of interlayer defects such as bubbles, delamination, impurities, and uneven coating, generating a dataset containing feature parameters for each interface abnormal region, and finally outputting defect feature data.

[0026] The defect classification output module receives defect feature data and performs surface defect identification on the registered reflected light surface image to obtain surface defect information. It converts the pixel coordinates of abnormal areas into physical coordinates of the tape and matches the longitudinal position of the tape corresponding to the defect with the production line's runtime sequence, achieving precise positioning of the defect on the actual tape. This facilitates targeted processing of defective areas in subsequent processes. Simultaneously, it acquires the surface defect coordinate information identified from the reflected light image and maps the coordinates of surface defects and interface defects to the same global coordinate system of the production line. Multiple types of defects at the same physical location correspond to the same location identifier, enabling associated management of defects at different levels at the same location. This avoids duplicate recording of multiple defects at the same location and improves the regularity of defect data. Finally, it integrates all defect type information, location coordinate information, and distribution range information to generate structured defect detection results. Each defect record in the defect detection results corresponds to a unique location identifier, and the data format is standardized and uniform, facilitating reading and processing by various subsequent systems. Defect detection results are output via a communication interface to the production line's execution control unit. Based on the defect detection results, corresponding trigger signals are output to drive subsequent marking and rejection devices, promptly separating defective segments to prevent them from flowing into the winding process and reducing subsequent rework costs. Simultaneously, all defect records are written to a local storage unit, forming a complete defect detection result archive. The archived data contains detailed information on each defect and its corresponding production sequence, which can be used for subsequent quality traceability and production process analysis. This helps production personnel pinpoint the specific causes of defects, optimize coating production parameters, and gradually improve overall product yield. Figure 1 As shown.

[0027] Example 2 When this inspection system is deployed on an offline rewinding and inspection equipment for finished tape rolls, the dual-optical-path acquisition module acquires reflected light surface images and transmitted light transmission images of the tape under test during the tape rewinding process. The reflected light path and the transmitted light path are located on both sides of the tape along the tape path and their acquisition directions are coaxial and opposite. The image acquisition direction of the reflected light path is perpendicular to the tape surface and receives the optical signals reflected from the surface, which can clearly capture surface abnormalities such as minor scratches, pressure points, excess adhesive, and dust adhesion on the tape surface. The image acquisition direction of the transmitted light path is opposite to the reflected light direction and receives the optical signals that penetrate the entire tape, which can penetrate the substrate and adhesive layer of the tape, revealing interlayer abnormalities such as tiny bubbles, small impurities, uneven adhesive layer thickness, and slight delamination between layers. The two optical paths work together to comprehensively cover the various defect detection needs of the tape from the surface to the interior, without any layer-level missed detections. The two acquisition channels have completely corresponding and overlapping fields of view, covering the entire width of the tape. This allows for comprehensive, blind-spot-free inspection of the tape across its entire lateral width, accurately capturing even minute defects at the tape's edges. During acquisition, a constant current source drives the illumination unit. In offline re-inspection environments with minimal external interference, the constant current drive further ensures stable light source brightness, with intensity fluctuations not exceeding ±5%. This results in uniform and consistent image brightness, maintaining a stable grayscale baseline across different frames, preventing brightness fluctuations from affecting the inspection results. Furthermore, the acquisition parameters for both images are kept consistent, with all imaging settings identical, ensuring the same basic imaging conditions and avoiding systematic grayscale deviations caused by parameter differences. The two output raw images each carry corresponding acquisition timing and position identifiers, corresponding one-to-one with the tape length data of the rewinding equipment. This provides a matching benchmark for subsequent timing synchronization and spatial registration, as well as a length basis for subsequent defect localization, ultimately outputting two raw image data channels.

[0028] The synchronous registration module receives two channels of raw image data. Using the operating speed pulse signal of the rewinding equipment as a common trigger source for the acquisition of both images, it controls the acquisition timing of the two images to remain synchronized, ensuring that the start time of the acquisition is consistent. In offline rewinding scenarios, the tape speed is stable and uniform, resulting in higher timing synchronization accuracy. This ensures that the two images strictly correspond to the same longitudinal position of the tape, preventing image misalignment caused by tape movement fluctuations. After acquisition, the contour lines of the tape's double-sided edges are extracted from the two images. The tape edges have clear grayscale boundaries in both images. During rewinding, the tape moves smoothly with minimal edge position fluctuations, using this as a spatial alignment benchmark for higher alignment accuracy. Using the contour lines as a benchmark, spatial transformations of translation and scaling are performed on the two images to adjust their lateral position and overall proportions. This corrects positional differences caused by installation errors and slight tape offsets in the two optical paths until the pixel coordinates of the tape edges in the two images completely overlap. During the transformation process, the overall image structure remains intact, without distortion or loss of image content, ensuring that the original shape and relative position of defects are unaffected. After spatial coordinate alignment of the two acquired images, the pixel coordinate systems of the two images are completely unified. The same physical location corresponds to the exact same pixel coordinates in both images, providing a precise spatial basis for subsequent pixel-level difference operations. This effectively avoids false difference signals caused by positional misalignment, reduces the probability of false detection, and finally outputs the registered aligned images, such as... Figure 2 As shown.

[0029] The pixel difference operation module receives the two aligned images after registration. First, it performs grayscale normalization on both images, mapping their grayscale values ​​to the same range. This eliminates overall grayscale deviations caused by differences in optical path attenuation and imaging characteristics, ensuring a consistent background grayscale baseline for the two images and guaranteeing a unified baseline for the difference operation. Next, it performs smoothing filtering on the normalized images to remove random noise. In offline scenarios with stable imaging environments, filtering can further remove background noise from the imaging device while preserving defect details, preventing noise from being amplified and forming false defects during the difference operation. After filtering, preprocessed reflected and transmitted light images are obtained. These two preprocessed images maintain the same resolution and pixel coordinate system, with completely corresponding spatial positions and a highly unified grayscale baseline, allowing for direct subsequent difference operations. Subsequently, pixel-by-pixel differential calculation is performed on the two preprocessed images. All pixel coordinates are traversed using a point-by-point scanning method. The absolute value of the difference between the reflected and transmitted light gray values ​​at the same coordinate is used as the gray value of the differential feature map at that coordinate. Uniform, defect-free tape areas show consistent gray-scale responses in both images, exhibiting lower gray-scale values ​​after differential calculation, forming a uniform background. Areas with interlayer anomalies show significant differences in gray-scale responses in both images, exhibiting higher gray-scale values ​​after differential calculation. This effectively amplifies the abnormal signals at the interlayer interfaces, suppresses uniform surface texture signals, and allows even minute interlayer defects to be clearly displayed. After traversing all pixel coordinates, a differential feature map reflecting the differences between the tape surface and interlayer signals is generated. This visually presents the distribution of interface anomalies within the tape, significantly improving the identification of minute defects. The final output is the differential feature map data.

[0030] The interface feature extraction module receives differential feature map data and separates surface texture signals from interface layer anomaly signals in the differential feature map, distinguishing genuine interface defects from normal substrate textures and background noise to ensure the accuracy of the re-inspection results. First, connected component analysis is performed on the differential feature map to extract all connected regions with gray levels different from the background, integrating scattered pixels into complete region units for subsequent feature statistics and filtering. The size attributes of each connected region are statistically analyzed, filtering out tiny noise areas whose area and shape do not conform to interface defect characteristics, retaining candidate anomaly regions within the size range, completing the initial screening of interface anomaly signals, eliminating the influence of residual interference signals, and reducing the amount of data processed subsequently. Offline re-inspection requires higher detection sensitivity. Then, gray-level threshold judgment is performed on the filtered candidate anomaly regions. The gray-level values ​​of all pixels in the current frame's differential feature map are statistically analyzed, and an adaptive differential gray-level threshold is generated by combining it with a preset sensitivity coefficient. The sensitivity coefficient k is pre-calibrated according to the strict detection standards of the re-inspection scenario; in this scenario, 2.0 can be selected to balance the detection rate and false detection rate. The adaptive threshold dynamically adjusts according to the overall grayscale level of the current frame image, adapting to background grayscale differences in different batches and materials of tape, avoiding detection bias caused by a fixed threshold. The average grayscale value within the candidate region is compared with the differential grayscale threshold; regions exceeding the threshold are identified as interface anomaly regions, accurately distinguishing normal texture fluctuations from genuine interface defects. Geometric morphological features and grayscale distribution features of interface anomaly regions are extracted. Geometric morphological features include region area, aspect ratio, and shape regularity; grayscale distribution features include the region's internal grayscale mean and uniformity. These multi-dimensional feature parameters comprehensively characterize the attributes of the anomaly region, providing sufficient basis for defect classification. Built-in defect feature matching rules compare the actual features of candidate anomaly regions with the defect feature matching rules one by one to determine the defect type of the corresponding interface anomaly region. This accurately distinguishes different defect types such as bubbles, impurities, delamination, and uneven coating, generating a dataset containing feature parameters for each interface anomaly region, and finally outputting defect feature data.

[0031] The defect classification output module receives defect feature data and performs surface defect identification on the registered reflected light surface image. It converts the pixel coordinates of abnormal areas into physical coordinates of the tape and matches the longitudinal position of the tape corresponding to the defect with the rewinding sequence, achieving precise length positioning of the defect on the entire roll of tape. This facilitates subsequent manual review and defect marking. Simultaneously, it acquires the surface defect coordinate information identified from the reflected light image and maps the coordinates of surface and interface defects to the same global coordinate system of the re-inspection equipment. Multiple types of defects at the same physical location correspond to the same location identifier, enabling the association of different levels of defects at the same location. This avoids duplicate counting of multiple defects at the same location, improving the completeness and regularity of defect data. Finally, it integrates all defect type information, location coordinate information, and distribution range information to generate structured defect detection results. Each defect record in the defect detection results corresponds to a unique location identifier, and the data format is clear and standardized, facilitating reading and statistics. The defect detection results are output externally through a communication interface, connecting to the marking control unit of the re-inspection equipment. Based on the defect detection results, the unit outputs a corresponding marking trigger signal, marking the tape position corresponding to the defect, facilitating subsequent defect removal and manual review. At the same time, all defect records are written to the local storage unit to form a complete defect detection report archive for a single roll of finished product. The report contains information such as the total number of defects in the whole roll of tape, the distribution of defect types, and the distribution of defect locations. It can be used as a quality certificate for finished product shipment and can also be used for subsequent quality statistics and process improvement analysis.

[0032] Example 3 In the feeding section of the precision electronic tape slitting process, this system uses a dual-optical-path acquisition module to capture reflected light surface images and transmitted light transmission images of the tape under test during the tape feeding and feeding process. The reflected light path and the transmitted light path are located on both sides of the tape along the feeding path, and their acquisition directions are coaxial and opposite. The image acquisition direction of the reflected light path is perpendicular to the tape surface and receives the optical signals reflected from the surface. It can accurately capture minute surface defects such as micro-scratches, indentations, excess adhesive, and micro-dust adhesion on the electronic tape surface. The image acquisition direction of the transmitted light path is opposite to the reflected light direction and receives the optical signals that penetrate the entire tape. It can penetrate the ultra-thin substrate and precision adhesive layer of the electronic tape and reveal internal defects such as micro-bubbles, minute impurities, uneven adhesive layer thickness, and interlayer micro-detachment. The two optical paths work together to comprehensively cover the high-precision inspection requirements of electronic tape from the surface to the interior, meeting the stringent quality requirements of the electronics industry for tape. The two acquisition channels have completely corresponding and overlapping fields of view, covering the entire width of the tape. This allows for comprehensive, blind-spot-free detection of the tape across its entire lateral width, ensuring that even minor defects at the edges are not missed. During acquisition, a constant current source drives the illumination unit, guaranteeing stable power supply and ensuring that light intensity fluctuations do not exceed ±5%, resulting in uniform and consistent image brightness. This prevents brightness fluctuations from affecting the identification of minor defects. Furthermore, the acquisition parameters for both images are kept consistent, with all imaging settings identical. This ensures that the basic imaging conditions for both images are the same, avoiding grayscale deviations caused by parameter differences and guaranteeing differential contrast for minor defects. The two output raw images each carry corresponding acquisition timing and location identifiers, which correspond one-to-one with the material feeding length data of the slitting equipment. This provides a matching benchmark for subsequent timing synchronization and spatial registration, as well as a positional basis for subsequent defect location and slitting removal. The final output consists of two raw image data channels.

[0033] The synchronous registration module receives two channels of raw image data. Using the feeding speed pulse signal of the slitting equipment as the common trigger source for the two image acquisitions, it controls the acquisition trigger timing of the two images to remain synchronized, ensuring that the start time of the two image acquisitions is consistent. This results in a stable and controllable conveyor belt speed during the slitting and feeding process, with high timing synchronization accuracy. It ensures that the two images strictly correspond to the same longitudinal position of the tape, preventing image content deviations caused by timing misalignment. After acquisition, the contour lines of the tape's double-sided edges are extracted from the two images. The electronic tape edges are neat, and the grayscale boundaries are clear in both images. This serves as a spatial alignment benchmark, achieving extremely high alignment accuracy. Using the contour lines as a benchmark, spatial transformations of translation and scaling are performed on the two images to adjust their lateral position and overall proportion. This corrects positional differences caused by installation accuracy and slight tape misalignment in the two optical paths, until the pixel coordinates of the tape edges in the two images completely overlap. During the transformation process, the overall image structure remains intact, without causing image content deformation or loss of detail, ensuring that the original shape and relative position of minor defects are unaffected. After spatial coordinate alignment of the two acquired images, the pixel coordinate system of the two images is completely unified. The same physical position corresponds to the exact same pixel coordinate in the two images, providing a precise spatial basis for subsequent pixel-level differential operations, effectively avoiding false differential signals caused by position misalignment, ensuring the accuracy of micro-defect detection, and finally outputting the registered two aligned images.

[0034] The pixel difference operation module receives the two aligned images after registration. First, it performs grayscale normalization on both images, mapping their grayscale values ​​to the same range. This eliminates overall grayscale deviations caused by differences in optical path characteristics and light attenuation, ensuring a consistent background grayscale baseline for the two images and guaranteeing a unified baseline for the difference operation. Next, it performs smoothing filtering on the normalized images to remove random noise. While preserving details of minor defects, it removes background noise from the imaging device and environment, preventing noise from being confused with minor defects and reducing the probability of false positives and false negatives. After filtering, the preprocessed reflected light and transmitted light images are obtained. These two preprocessed images maintain the same resolution and pixel coordinate system, with completely corresponding spatial positions and a highly unified grayscale baseline, allowing for direct subsequent difference operations. Subsequently, pixel-by-pixel differential calculation is performed on the two preprocessed images. All pixel coordinates are traversed using a point-by-point scanning method. The absolute value of the difference between the reflected and transmitted light gray values ​​at the same coordinate is used as the gray value of the differential feature map at that coordinate. Uniform, defect-free electronic tape regions exhibit consistent gray-scale responses in both images, resulting in lower gray-scale values ​​after differential calculation, forming a uniform, low-noise background. Regions with interlayer anomalies show different gray-scale responses in the two images, resulting in higher gray-scale values ​​after differential calculation. This effectively amplifies weak anomalous signals at the interlayer interface, suppresses uniform surface texture signals, and allows even micron-level interlayer defects to be clearly displayed. After traversing all pixel coordinates, a differential feature map reflecting the differences between the tape surface and interlayer signals is generated. This visually presents the distribution of interface anomalies within the tape, significantly improving the recognition rate of minute defects. The final output is the differential feature map data, such as... Figure 3 As shown.

[0035] The interface feature extraction module receives differential feature map data and separates surface texture signals from interface layer abnormal signals in the differential feature map, distinguishing genuine interface defects from normal substrate textures and background noise to meet the accuracy requirements of high-precision electronic tape detection. First, connected component analysis is performed on the differential feature map to extract all connected regions with gray levels different from the background, integrating scattered pixels into complete region units for subsequent feature statistics and filtering. The size attributes of each connected region are statistically analyzed, filtering out tiny noise areas whose area and shape do not conform to interface defect characteristics, retaining candidate abnormal regions within the size range, completing the initial screening of interface abnormal signals, eliminating the influence of residual interference signals, and reducing the amount of data processed subsequently. For the high-precision detection requirements of electronic tape, gray-level threshold judgment is then performed on the filtered candidate abnormal regions. The gray-level values ​​of all pixels in the current frame's differential feature map are statistically analyzed, and an adaptive differential gray-level threshold is generated by combining it with a preset sensitivity coefficient. The sensitivity coefficient k is pre-calibrated according to the high-precision detection requirements of electronic tape; in this scenario, 2.4 can be selected to prioritize the full detection of extremely minor defects. The adaptive threshold dynamically adjusts according to the overall grayscale level of the current frame image, adapting to background grayscale differences in different batches and materials of electronic tapes. This avoids detection fluctuations caused by fixed thresholds and ensures the stability of detection results. The average grayscale value within the candidate region is compared with the differential grayscale threshold; regions exceeding the threshold are identified as interface anomaly regions, accurately distinguishing normal substrate texture fluctuations from genuine interface defects. Geometric morphological features and grayscale distribution features of interface anomaly regions are extracted. Geometric morphological features include region area, aspect ratio, and shape regularity; grayscale distribution features include the mean grayscale value and grayscale uniformity within the region. These multi-dimensional feature parameters comprehensively characterize the attributes of the anomaly region, providing sufficient basis for defect classification. Built-in defect feature matching rules compare the actual features of candidate anomaly regions with the defect feature matching rules one by one to determine the defect type of the corresponding interface anomaly region. This accurately distinguishes different defect types such as microbubbles, minor impurities, micro-delamination, and uneven adhesive layers, generating a dataset containing feature parameters for each interface anomaly region, and finally outputting defect feature data.

[0036] The defect classification output module receives defect feature data and performs surface defect identification on the registered reflected light surface image. It converts the pixel coordinates of abnormal areas into physical coordinates of the tape and matches the longitudinal position of the tape corresponding to the defect with the material feeding sequence, achieving precise length positioning of the defect on the raw material roll. This provides accurate location information for subsequent slitting and removal. Simultaneously, it acquires the surface defect coordinate information identified from the reflected light image and maps the coordinates of surface defects and interface defects to the same global coordinate system of the slitting equipment. Multiple types of defects at the same physical location correspond to the same location identifier, enabling associated management of defects at different levels at the same location. This avoids duplicate processing of multiple defects at the same location, improving the efficiency of the slitting process. Finally, it integrates all defect type information, location coordinate information, and distribution range information to generate structured defect detection results. Each defect record in the defect detection results corresponds to a unique location identifier, and the data format is standardized and clear, facilitating reading and processing by the slitting control system. Defect detection results are output via a communication interface to the rejection control unit of the slitting equipment. Based on the defect detection results, the unit outputs a corresponding slitting rejection trigger signal to accurately reject defective segments during the slitting process, preventing defective materials from entering the finished product, improving the yield rate, and reducing ineffective processing waste of raw materials. Simultaneously, all defect records are written to the local storage unit, forming a complete defect detection result archive for the slitting batch. The archived data includes defect statistics for raw materials within the batch, which can be used for raw material quality evaluation and supplier management, and also provides data support for subsequent process optimization.

[0037] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A visual analysis-based tape surface defect detection system, characterized in that, The system includes: Dual-path acquisition module: Acquires the reflected light surface image and the transmitted light transmission image of the tape under test. The two acquisition fields of view overlap and cover the full width of the tape. Obtain the optical signals corresponding to the surface morphology and interlayer structure of the tape, and output two raw image data. Synchronous registration module: Receives two channels of raw image data, controls the acquisition triggering sequence of the two channels of images to maintain synchronization, performs spatial coordinate alignment on the two acquired images, and outputs the registered two aligned images; Pixel difference calculation module: Receives two aligned images after registration, performs grayscale preprocessing and pixel-level difference calculation, generates a difference feature map reflecting the difference between the surface layer and interlayer signals of the tape, and outputs the difference feature map data; Interface feature extraction module: Receives differential feature map data, separates the surface texture signal and interface layer abnormal signal in the differential feature map, extracts the geometric shape and gray-scale distribution features of the interface abnormal area, and outputs defect feature data. Defect classification output module: Receives defect feature data, matches the corresponding defect type, associates the tape position coordinates corresponding to the defect, and outputs complete defect detection results.

2. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, In the dual-optical-path acquisition module, the reflected optical path and the transmitted optical path are located on opposite sides of the tape and their acquisition directions are coaxial and opposite. The image acquisition direction of the reflected optical path is perpendicular to the tape surface and is used to receive the optical signal reflected from the surface. The image acquisition direction of the transmitted optical path is opposite to the reflected light direction and is used to receive the optical signal that penetrates the entire tape. The field of view of the two acquisition paths are completely corresponding and together cover the entire width of the tape. During the acquisition process, the illumination unit is driven by a constant current source to ensure that the light intensity fluctuation does not exceed ±5%, and the acquisition parameters of the two images are kept consistent. The two output original images carry corresponding acquisition timing and position identifiers, respectively.

3. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, The synchronous registration module uses the production line running speed pulse signal as the common trigger source for the acquisition of two images, so that the start time of the acquisition of the two images is consistent. After the acquisition is completed, the contour lines of the two sides of the tape in the two images are extracted. The two images are then transformed by translation and scaling based on the contour lines until the pixel coordinates of the tape edges in the two images are completely coincident. The overall structure of the image is kept intact during the transformation process, and the two aligned images are generated after registration.

4. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, After receiving the two aligned images after registration, the pixel difference operation module first performs grayscale normalization processing on the two images respectively, mapping the grayscale value range of the two images to the same numerical range. Then, it performs smoothing filtering processing on the two normalized images respectively to filter out random noise in the images. After filtering, the preprocessed reflected light image and transmitted light image are obtained. The two preprocessed images maintain the same resolution and pixel coordinate system.

5. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, The pixel difference calculation module performs pixel-by-pixel difference calculation on the two preprocessed images. It traverses all pixel coordinates using a point-by-point scanning method, calculates the difference between the reflected light gray value and the transmitted light gray value at the same coordinate, and takes the absolute value. This absolute value is used as the gray value of the difference feature map at that coordinate. The specific formula is as follows: ,in, This represents the grayscale value of the difference feature map at coordinates (x, y). The grayscale value of the reflected light image; The grayscale values ​​are the transmitted light image values; after traversing all pixel coordinates, a complete differential feature map is generated.

6. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, When the interface feature extraction module separates the surface texture signal and the interface layer abnormal signal, it performs connected component analysis on the differential feature map, extracts all connected regions whose gray levels differ from the background, counts the size attributes of each connected region, filters out tiny noise regions whose area and shape do not conform to the interface defect characteristics, retains candidate abnormal regions that conform to the size range, completes the preliminary screening of interface abnormal signals, and eliminates the influence of residual interference signals.

7. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, The interface feature extraction module performs grayscale threshold determination on the filtered candidate abnormal regions, counts the grayscale values ​​of all pixels in the current frame's differential feature map, and generates an adaptive differential grayscale threshold by combining it with a preset sensitivity coefficient. The specific formula is as follows: Where Th is the differential grayscale threshold; μ is the global grayscale mean of the differential feature map of the current frame; σ is the global grayscale standard deviation of the differential feature map of the current frame; k is the preset sensitivity coefficient; the average grayscale value in the candidate region is compared with Th, and the region with a value higher than Th is determined as an interface abnormal region; the geometric morphological features and grayscale distribution features of the interface abnormal region are extracted, and the built-in defect feature matching rules are used to compare the actual features of the candidate abnormal region with the defect feature matching rules one by one to determine the defect type of the corresponding interface abnormal region, and generate a dataset containing the feature parameters of each interface abnormal region.

8. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, After receiving the defect feature data, the defect classification output module converts the pixel coordinates of the abnormal area into the physical coordinates of the tape and matches the longitudinal position of the tape corresponding to the defect in combination with the production line operation sequence; at the same time, it obtains the surface defect coordinate information identified by the reflected light image and maps the coordinates of the surface defect and the interface defect to the same global coordinate system of the production line, so that multiple types of defects at the same physical location correspond to the same location identifier.

9. The tape surface defect detection system based on visual analysis according to claim 1, characterized in that, The defect classification output module integrates all defect type information, location coordinate information, and distribution range information to generate structured defect detection results. Each defect record in the defect detection results corresponds to a unique location identifier. Furthermore, the defect detection results are output to the outside via the communication interface, which can be connected to the production line's execution control unit. Based on the defect detection results, corresponding trigger signals are output, and all defect records are written to the local storage unit to form a complete defect detection result archive.

Images