A coil automatic detection machine defect identification method, system and device

Abstract

The application provides a kind of coil automatic detection machine's defect identification method, system and equipment, it is related to image processing technical field, method includes: when coil reaches target detection area, image acquisition is executed;After distortion and illumination correction are carried out to detection image set, the corrected image set is carried out multispectral feature fusion, and enhanced image is input to window recognition channel to carry out image traversal scanning of enhanced image, and the texture energy and direction consistency index in each sliding window are calculated;Multi-scale image anomaly comparison is carried out to the enhanced image corresponding to the calibration position image and coordinate index;With the pixel point with abnormal value mark is carried out morphological aggregation and boundary continuity analysis, extracts geometric morphological parameter, spectral feature parameter and texture anomaly parameter, and outputs defect mask.Through the present application, the technical problem of low defect recognition accuracy of the coil detection machine in the prior art can be solved, and the technical effect of improving the defect recognition accuracy is achieved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a defect identification method, system, and device for an automatic coil inspection machine. Background Technology

[0002] In modern industrial production, coils are important components of electrical equipment and precision machinery. Their quality directly affects the performance and service life of the equipment. Therefore, defect detection and quality control of coils have become crucial links.

[0003] Currently, existing automatic coil inspection technologies mainly rely on single optical imaging or simple infrared detection methods to identify defects on the coil surface manually or semi-automatically. These technologies typically use single-spectrum image acquisition equipment, and the acquired images are prone to distortion and uneven brightness under different lighting conditions, imaging angles, and environmental interference, thus affecting the accurate extraction of defect features.

[0004] In summary, existing technologies suffer from technical problems such as reliance on single-spectral imaging, lack of multi-feature, multi-scale image fusion analysis, and low defect localization accuracy. These problems make it difficult to accurately identify and quantify minute defects and local anomalies on the coil surface, further affecting real-time monitoring of the production process, defect traceability capabilities, and the stability and reliability of coil product quality. Summary of the Invention

[0005] The purpose of this application is to provide a defect identification method, system, and equipment for an automatic coil inspection machine, in order to solve the technical problems in the prior art, which are that due to reliance on single-spectral imaging, lack of multi-feature, multi-scale image fusion analysis, and low defect positioning accuracy, it is difficult to accurately identify and quantify small defects and local anomalies on the coil surface, which further affects the real-time monitoring of the production process, defect traceability capabilities, and the stability and reliability of coil product quality.

[0006] In view of the above problems, this application provides a defect identification method, system and equipment for an automatic coil inspection machine.

[0007] In a first aspect, this application provides a defect identification method for an automatic coil inspection machine, implemented through a defect identification system for the automatic coil inspection machine, comprising: triggering a detection command of the automatic coil inspection machine when the coil reaches the target detection area; activating a ring-shaped adjustable light source according to the detection command, controlling a visible light camera and an infrared camera positioned directly above the target detection area to perform image acquisition and establish a detection image set; performing distortion and illuminance correction on the detection image set, fusing multispectral features in the corrected image set to construct an enhanced image, inputting the enhanced image into a window recognition channel, performing image traversal scanning of the enhanced image using a sliding window, and calculating the texture energy and orientation consistency index within each sliding window to establish a coordinate index; calling a calibration position image according to the coordinate index, performing multi-scale image anomaly comparison using the calibration position image and the enhanced image corresponding to the coordinate index, and establishing pixels with anomaly identification; performing morphological aggregation and boundary continuity analysis on the pixels with anomaly identification, extracting geometric morphological parameters, spectral feature parameters, and texture anomaly parameters, and outputting a defect mask.

[0008] Preferably, the defect identification method of the automatic coil inspection machine further includes: calling the three-dimensional calibration model of the target inspection area through the calibration database of the inspection machine, and using the three-dimensional calibration model to perform joint correction of radial distortion and tangential distortion on the visible light image and infrared image respectively; acquiring environmental spectral monitoring data of the target inspection area, constructing an illumination distribution surface model based on the environmental spectral monitoring data, and using the illumination distribution surface model to perform illuminance correction under block brightness equalization on the distortion-corrected visible light image and infrared image respectively, as follows: dividing the visible light image or infrared image into N grid blocks, calculating a stable brightness statistic for each grid block; configuring the target brightness constant according to the stable brightness statistic, calculating the gain map of each pixel, and performing multiplicative correction of the original image to complete the illuminance correction.

[0009] Preferably, the defect identification method of the automatic coil inspection machine further includes: performing pixel-level spatial geometric registration between the illuminance-corrected visible light image and the infrared image, performing hierarchical feature decomposition, extracting low-frequency layer features, mid-frequency layer features, and high-frequency layer features, and performing adaptive weight fusion based on the sharpness score of the multi-layer features and the prior probability of defects; and constructing an enhanced image using the adaptive weight fusion result.

[0010] Preferably, the defect identification method of the automatic coil inspection machine further includes: calling the texture extraction layer of the window recognition channel, performing enhanced image texture complexity evaluation, and establishing a texture complexity evaluation identifier; establishing a window shape library for sliding windows, the window shape library including square windows and rectangular windows, and generating position window shape constraints after matching the shape of the window shape library using the texture complexity evaluation identifier; configuring multi-scale position windows using the position window shape constraints, performing texture energy descriptor calculation, establishing texture energy distribution, and establishing a coordinate index based on the texture energy distribution and direction consistency index.

[0011] Preferably, the defect identification method of the automatic coil inspection machine further includes: performing gradient direction clustering on each multi-scale position window and calculating the main direction concentration; using the main direction concentration as a direction consistency index, and establishing a coordinate index based on the texture energy distribution and the direction consistency index.

[0012] Preferably, the defect identification method of the automatic coil inspection machine further includes: calling multi-scale calibration position images in the calibration library according to the coordinate index; performing ensemble correction and spectral normalization processing on the multi-scale calibration position images, performing image registration at different scale layers, and establishing anomaly identification based on spectral differences, texture differences, and ensemble differences.

[0013] Preferably, the defect identification method of the automatic coil inspection machine further includes: performing spatial aggregation of abnormal pixels using morphological connected region analysis, merging neighboring similar abnormal pixels, wherein the connected region analysis includes dilation region analysis, erosion region analysis, and opening / closing operation region analysis; calculating boundary curvature and concavity / convexity indices for each spatial aggregation result to generate boundary continuity analysis results; and extracting geometric morphological parameters, spectral feature parameters, and texture anomaly parameters using the spatial aggregation results and the boundary continuity analysis results.

[0014] Preferably, the defect identification method of the automatic coil inspection machine further includes: obtaining a parameter mapping between coil manufacturing parameters and a defect set; after obtaining a defect mask, performing correlation matching of manufacturing parameters based on the defect mask and the parameter mapping to generate a correlation record; generating production feedback based on the correlation record, and using the production feedback for coil production management.

[0015] Secondly, this application also provides a defect identification system for an automatic coil inspection machine, used to execute a defect identification method for an automatic coil inspection machine as described in the first aspect, comprising: a detection command activation module, used to trigger a detection command of the automatic coil inspection machine when the coil reaches the target detection area; an image acquisition module, used to activate a ring-shaped adjustable light source according to the detection command, and control a visible light camera and an infrared camera set directly above the target detection area to perform image acquisition and establish a detection image set; and an image traversal scanning module, used to perform distortion and illuminance correction on the detection image set, and then perform multispectral feature fusion on the corrected image set to construct an augmented image. The enhanced image is input into the window recognition channel, and a sliding window is used to perform image traversal scanning of the enhanced image. The texture energy and orientation consistency index within each sliding window are calculated to establish a coordinate index. The image anomaly comparison module is used to call the calibration position image according to the coordinate index, and to perform multi-scale image anomaly comparison using the calibration position image and the enhanced image corresponding to the coordinate index to establish pixels with outlier identifiers. The defect mask output module is used to perform morphological aggregation and boundary continuity analysis on the pixels with outlier identifiers, extract geometric morphological parameters, spectral feature parameters and texture anomaly parameters, and output a defect mask.

[0016] Thirdly, this application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of a defect identification method for an automatic coil inspection machine as described in any one of the first aspects above.

[0017] The technical solution provided in this application has at least the following technical effects or advantages: by achieving the technical goal of high-precision, full-spectrum, multi-scale automatic identification and positioning of surface defects in coils, it achieves the technical effects of improving defect detection accuracy, enhancing quantitative defect analysis capabilities, realizing closed-loop feedback of production parameters, and optimizing coil production management.

[0018] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a schematic flowchart of a defect identification method for an automatic coil inspection machine provided in an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of the defect identification system of an automatic coil inspection machine provided in an embodiment of this application.

[0022] Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application.

[0023] Explanation of reference numerals in the attached diagram: Detection instruction activation module 1, image acquisition module 2, image traversal scanning module 3, image anomaly comparison module 4, defect mask output module 5, bus 300, receiver 301, processor 302, transmitter 303, memory 304, bus interface 305. Detailed Implementation

[0024] This application provides a defect identification method, system, and equipment for an automatic coil inspection machine. It addresses the technical problems in existing technologies where reliance on single-spectrum imaging, lack of multi-feature, multi-scale image fusion analysis, and low defect localization accuracy lead to the difficulty in accurately identifying and quantifying minute defects and local anomalies on the coil surface. This, in turn, affects real-time monitoring of the production process, defect traceability capabilities, and the stability and reliability of coil product quality. The application achieves the technical goal of high-precision, full-spectrum, multi-scale automatic identification and localization of coil surface defects, thereby improving defect detection accuracy, enhancing quantitative defect analysis capabilities, enabling closed-loop feedback of production parameters, and optimizing coil production management.

[0025] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0026] Example 1, please refer to the appendix. Figure 1This application provides a defect identification method for an automatic coil inspection machine, applied to a defect identification system for an automatic coil inspection machine, specifically including the following steps: S1: When the coil reaches the target detection area, the detection command of the automatic coil detection machine is triggered.

[0027] Specifically, the target detection area is the workspace used to perform the detection. When the coil enters the pre-set detection position, the position sensor or other triggering device will detect that the coil has reached the target detection area. Once the trigger signal is recognized, a detection command is immediately issued to coordinate the working sequence of each detection device.

[0028] S2: After starting the ring-shaped adjustable light source according to the detection command, control the visible light camera and infrared camera set directly above the target detection area to perform image acquisition and establish a detection image set.

[0029] Specifically, the detection command will control the activation of the ring-shaped adjustable light source. The ring-shaped adjustable light source is distributed in a ring shape, which can uniformly illuminate the coil surface from multiple angles. The brightness or spectral distribution can be adjusted to adapt to different material reflectivities or changes in ambient light, ensuring uniform illumination of the acquired image.

[0030] Simultaneously with the activation of the ring-shaped adjustable light source, two types of cameras installed directly above the target detection area are driven to work synchronously. The visible light camera collects images in the visible wavelength range, which can reflect the color, shape, and obvious defects on the coil surface. The infrared camera collects thermal radiation information, which can detect temperature differences or internal anomalies that are difficult to detect with the naked eye. The images obtained by the visible light camera and the infrared camera together constitute the detection image set, which is used for subsequent defect identification.

[0031] S3: After distortion and illumination correction of the detected image set, multispectral feature fusion is performed on the corrected image set to construct an enhanced image. The enhanced image is input into the window recognition channel, and the image traversal scan of the enhanced image is performed using a sliding window. The texture energy and orientation consistency index within each sliding window are calculated, and a coordinate index is established.

[0032] Specifically, distortion and illumination correction are performed on the detection image set to eliminate deformations caused by the lens or imaging angle. For example, radial distortion can cause barrel or pincushion-shaped deformations at the image edges, while tangential distortion can lead to uneven distortion. Simultaneously, brightness differences caused by uneven illumination are corrected, making the detection image set more realistic and accurate in overall brightness and geometry. The corrected image set undergoes multispectral feature fusion, aligning and extracting features from different spectral bands, such as visible and infrared light, at the pixel level. This preserves more detail and spectral differences, constructing an enhanced image with higher contrast, clearer details, and more complete texture, making subsequent analysis easier to detect defects.

[0033] The enhanced image is input into the window recognition channel, and then sent to the image analysis module for region-by-region scanning and texture feature recognition. A sliding window is used to perform image traversal scanning of the enhanced image; that is, a window of fixed or varying size is gradually slid across the image, analyzing image information region by region. The strength of the texture in a region is measured by statistically analyzing the intensity of pixel grayscale changes, quantifying the stability of the texture direction, recording the position of each window and its texture energy and direction consistency values, and establishing a coordinate index to provide reference data for subsequent precise defect location.

[0034] S4: Call the calibration location image according to the coordinate index, and use the calibration location image and the enhanced image corresponding to the coordinate index to perform multi-scale image anomaly comparison and establish pixel points for anomaly identification.

[0035] Specifically, the calibration location image is retrieved based on the coordinate index. This involves using index information containing the coordinates of specific locations to retrieve the corresponding reference image from the calibration library. The calibration location image serves as the comparison benchmark during detection. Multi-scale image anomaly comparison is performed using the calibration location image and the enhanced image corresponding to the coordinate index. This involves comparing the reference image and the detected enhanced image layer by layer at multiple resolutions or observation scales, analyzing everything from overall contours to subtle textures to comprehensively detect anomalies. Pixels with differences exceeding a set threshold are marked, establishing anomaly markers.

[0036] S5: After performing morphological aggregation and boundary continuity analysis on the pixels with outlier identifiers, extract geometric morphological parameters, spectral feature parameters and texture anomaly parameters, and output the defect mask.

[0037] Specifically, after performing morphological aggregation and boundary continuity analysis on pixels marked as outliers, the individual pixels identified as anomalous are combined according to spatial relationships and adjacency to form more complete anomalous regions. Simultaneously, the continuity, curvature, and concavity / convexity of the region boundaries are analyzed to ensure smooth and complete boundaries and to eliminate isolated noise points. Morphological aggregation can be achieved through methods such as dilation, erosion, and opening / closing operations, while boundary continuity analysis evaluates the shape integrity of the anomalous region by calculating boundary curvature and concavity / convexity indices.

[0038] Quantitative indicators describing the shape, color spectral features, and texture changes of defects are obtained from the aggregated anomalous regions. Geometric morphological parameters, spectral feature parameters, and texture anomaly parameters are extracted. Geometric morphological parameters include area, perimeter, and boundary curvature. Spectral feature parameters represent the difference between the anomalous region and the surrounding region in color or infrared bands. Texture anomaly parameters measure the texture intensity or directional consistency of the region. A binary image of the same size as the original image is generated, and a defect mask is output. Pixels in anomalous regions are marked with 1 or highlighted, while pixels in normal regions are marked with 0 or retain their original color.

[0039] Furthermore, this application also includes: calling the three-dimensional calibration model of the target detection area through the calibration database of the detection machine, and using the three-dimensional calibration model to perform joint correction of radial and tangential distortion on the visible light image and infrared image respectively; acquiring environmental spectral monitoring data of the target detection area, constructing an illumination distribution surface model based on the environmental spectral monitoring data, and using the illumination distribution surface model to perform illuminance correction under block brightness equalization on the distortion-corrected visible light image and infrared image respectively, as follows: dividing the visible light image or infrared image into N grid blocks, calculating a stable brightness statistic for each grid block; configuring the target brightness constant according to the stable brightness statistic, calculating the gain map of each pixel, and performing multiplicative correction of the original image to complete the illuminance correction.

[0040] Specifically, the calibration database of the inspection machine is a pre-stored database of equipment calibration information, containing geometric and optical parameters for various inspection areas. By accessing the calibration database of the inspection machine, the three-dimensional calibration model of the target inspection area can be retrieved, and the corresponding three-dimensional calibration model can be read from the calibration database, thereby accurately describing the spatial relationship between the camera, the light source, and the object being inspected.

[0041] A three-dimensional calibration model is used to perform joint correction of radial and tangential distortion on visible light and infrared images to correct the image bending deformation caused by lens characteristics. Radial distortion refers to the magnification or compression phenomenon that occurs from the center of the image outward, while tangential distortion refers to the tilting deformation caused by the lens not being completely parallel to the imaging plane. Joint correction can ensure the geometric accuracy of the image.

[0042] Environmental spectral monitoring data of the target detection area is acquired. This data, obtained in real-time by a spectral sensor, provides information on the illumination intensity and spectral distribution of the detection area, reflecting the combined influence of the light source and ambient light. An illumination distribution surface model is constructed based on this data to describe the intensity variations of illumination at different locations. This model is then used to perform block-based brightness equalization illuminance correction on the distortion-corrected visible light and infrared images, compensating for uneven brightness in local areas and improving image contrast and detail.

[0043] The visible light or infrared image is divided into N grid blocks to analyze brightness characteristics within a local area. The value of N is set according to the image resolution and detection accuracy requirements. Stable brightness statistics are calculated for each grid block, identifying the relatively stable mean or median brightness under illumination fluctuations to reduce the interference of instantaneous light changes on the results.

[0044] Based on stable brightness statistics, a target brightness constant is configured, and an ideal brightness benchmark is set to calculate the gain map of each pixel, determining the required increase or decrease in brightness for each pixel. Then, multiplicative correction of the original image is performed, multiplying the brightness of each pixel by the corresponding gain coefficient to unify the brightness distribution and achieve illumination correction. Table 1 shows the image correction data for the target detection area.

[0045] Table 1: Image Correction Data Table for Target Detection Region

[0046] Furthermore, this application also includes: performing pixel-level spatial geometric registration between the illuminance-corrected visible light image and the infrared image, performing hierarchical feature decomposition to extract low-frequency layer features, mid-frequency layer features, and high-frequency layer features, and performing adaptive weight fusion based on the sharpness score and prior probability of defects of the multi-layer features; and constructing an enhanced image using the adaptive weight fusion result.

[0047] Specifically, the illuminance-corrected visible light image and the infrared image are subjected to pixel-level spatial geometric registration, precisely aligning the visible light and infrared images at each pixel location so that pixels at the same physical location correspond consistently in both images. This spatial geometric registration needs to consider factors such as imaging viewing angle differences, lens distortion, and positional offsets to ensure the accuracy of subsequent feature fusion.

[0048] After registration, hierarchical feature decomposition is performed to decompose the registered image information into different frequency levels, including low-frequency features, mid-frequency features, and high-frequency features. Low-frequency features include overall shape and brightness distribution, mid-frequency features include local structure and contour information, and high-frequency features include detail and texture information, allowing features at different levels to be analyzed and processed separately.

[0049] After extracting low-frequency, mid-frequency, and high-frequency features, adaptive weight fusion is performed based on the clarity score of the multi-layer features and the prior probability of defects. The clarity score is used to measure the recognizability of the multi-layer features, while the prior probability of defects is based on historical samples or detection models to predict the probability that the multi-layer features may contain defect information. Adaptive weight fusion dynamically allocates the weight ratio of each layer feature in the final result according to the index, making the defect information more prominent.

[0050] An enhanced image is constructed using the adaptive weighted fusion result. The fused multi-layer features are recombined into a complete image, thus preserving key structural and spectral information in both visible and infrared images. This results in higher contrast and detail in defect areas, providing better data input for subsequent detection.

[0051] Furthermore, this application also includes: calling the texture extraction layer of the window recognition channel, performing enhanced image texture complexity evaluation, and establishing a texture complexity evaluation identifier; establishing a window shape library for sliding windows, the window shape library including square windows and rectangular windows, and generating position window shape constraints after matching the shape of the window shape library using the texture complexity evaluation identifier; configuring multi-scale position windows using the position window shape constraints, performing texture energy descriptor calculation, establishing texture energy distribution, and establishing coordinate indexes based on the texture energy distribution and orientation consistency index.

[0052] Specifically, the texture extraction layer of the window recognition channel is invoked, and the processing module for analyzing image texture information is enabled to identify the patterns of grayscale or color changes in the enhanced image. Then, a texture complexity evaluation is performed within the enhanced image, calculating the degree of detail changes in the texture in the enhanced image to measure its complexity. For example, the complexity is quantified by statistically analyzing local contrast, orientation changes, and frequency components. The calculation results are converted into labels or values ​​that can be used in subsequent steps to establish a texture complexity evaluation label.

[0053] A library of sliding window shapes is established for scanning and enhancing images. This library includes different shapes such as square and rectangular windows to accommodate various defect morphologies and sizes. Shape matching from this library is performed using texture complexity evaluation identifiers. This involves selecting the most suitable window shape based on the texture features of the image region. For example, a rectangular window can be selected for regions with regular texture orientations, while a square window can be selected for regions with irregular texture orientations. This process then assigns the window shape to each scanning position, generating positional window shape constraints.

[0054] After configuring multi-scale location windows using location window shape constraints, the window size is varied while maintaining a fixed shape to cover defect areas of different sizes. Texture energy descriptor calculations are performed, calculating texture energy features within each window, and statistically summing the squares of pixel grayscale changes. The texture energy results from all locations are then compiled into a distribution map to measure texture intensity and establish a texture energy distribution that reflects the texture characteristics of different regions in the entire image. A coordinate index is established based on the texture energy distribution and orientation consistency index. Combining the stability of texture intensity and orientation, important detection locations in the enhanced image are recorded for subsequent localization and analysis.

[0055] Furthermore, this application also includes: performing gradient direction clustering on each multi-scale location window and calculating the main direction concentration; using the main direction concentration as a direction consistency index, and establishing a coordinate index based on the texture energy distribution and the direction consistency index.

[0056] Specifically, gradient direction clustering is performed for each multi-scale location window. Within multi-scale location windows of different sizes, the directional information of pixel grayscale changes is analyzed, and gradient vectors with similar directions are grouped into one category to identify the main arrangement direction of textures or edges. The proportion of the most frequent gradient direction among all categories is calculated to obtain the main direction concentration. A higher proportion of main direction concentration indicates a more uniform texture orientation in that region.

[0057] The concentration of principal directions is used as an indicator of directional consistency to measure the stability of texture directions. A higher concentration of principal directions indicates better directional consistency, while a lower concentration indicates more chaotic texture directions. A coordinate index is established based on texture energy distribution and the directional consistency index, combining texture intensity and directional stability to perform feature labeling on each location in the enhanced image, thereby generating an index containing the coordinates of key locations for subsequent defect comparison and localization.

[0058] Furthermore, this application also includes: calling multi-scale calibration location images from the calibration library according to the coordinate index; performing ensemble correction and spectral normalization on the multi-scale calibration location images, performing image registration at different scale layers, and establishing outlier identifiers based on spectral differences, texture differences, and ensemble differences.

[0059] Specifically, the multi-scale calibration location images in the calibration library are retrieved based on the coordinate index. That is, by using the coordinate index containing location and feature information, images of various resolutions or scales at the corresponding locations are retrieved from the standard reference images stored in the calibration library. The multi-scale calibration location images can provide different levels of detail from the whole to the local, which is convenient for subsequent comparison.

[0060] Ensemble correction and spectral normalization are performed on multi-scale calibration location images to unify the geometric and brightness corrections of reference images at different scales, ensuring that the multi-scale calibration location images are consistent with the currently detected image in terms of shape and color spectrum. Ensemble correction refers to unifying the scale, rotation, and positional relationships among multiple images, while spectral normalization eliminates color and brightness deviations caused by differences in imaging equipment or ambient lighting.

[0061] After performing image registration at different scales, the reference image and the detection image are precisely aligned at the pixel level at each scale to ensure that each pixel position in subsequent difference comparisons corresponds correctly. Then, anomaly comparison is performed based on spectral differences, texture differences, and ensemble differences, comparing differences in color spectrum, changes in surface texture, and deviations in overall shape or structure to identify regions that do not conform to the reference standard and establish outlier markers.

[0062] Furthermore, this application also includes: performing spatial aggregation of abnormal pixels using morphological connected region analysis, merging neighboring similar abnormal pixels, wherein the connected region analysis includes dilation region analysis, erosion region analysis, and opening / closing operation region analysis; calculating boundary curvature and concavity / convexity indices for each spatial aggregation result to generate boundary continuity analysis results; and extracting geometric morphological parameters, spectral feature parameters, and texture anomaly parameters using the spatial aggregation results and the boundary continuity analysis results.

[0063] Specifically, spatial aggregation of anomalous pixels is performed using morphological connected component analysis. This involves grouping pixels marked as anomalous during detection according to their spatial connectivity, thereby merging adjacent pixels with similar features into larger anomalous regions. Dilation analysis in connected component analysis involves adding pixels to pixel boundaries, connecting adjacent but not directly contacting anomalous points. Erosion analysis removes pixels on boundaries, eliminating noise or separating irrelevant small regions. Opening and closing operations in connected component analysis combine dilation and erosion; opening removes small, isolated noise, while closing fills in small holes within a region.

[0064] For each spatial aggregation result, boundary curvature and concavity / convexity indices are calculated, that is, the boundary shape of the abnormal region is analyzed. Boundary curvature measures the degree of bending of the boundary, and concavity / convexity indices indicate whether the boundary is convex or concave. Boundary continuity analysis results are generated to evaluate the integrity and regularity of the boundary of the abnormal region.

[0065] By utilizing spatial aggregation results and boundary continuity analysis results, geometric morphological parameters, spectral feature parameters, and texture anomaly parameters are extracted. That is, by combining the overall shape information, boundary characteristics, color spectrum information, and texture features of the anomaly region, a set of parameters that can accurately describe the defect is extracted, providing a basis for subsequent defect classification and judgment.

[0066] Furthermore, this application also includes: obtaining a parameter mapping between coil manufacturing parameters and a defect set; after obtaining a defect mask, performing an association matching of manufacturing parameters based on the defect mask and the parameter mapping to generate an association record; generating production feedback based on the association record, and using the production feedback for coil production management.

[0067] Specifically, this involves obtaining a parameter mapping between coil manufacturing parameters and defect sets, establishing a correspondence between various process parameters, equipment settings, and operating conditions involved in coil production and the detected defect types and their characteristics. Coil manufacturing parameters may include coil thickness, winding tension, current, heating temperature, etc., while defect set parameters include crack length, area, spectral differences, and texture anomaly intensity, etc. Parameter mapping can help analyze how coil production conditions may lead to the occurrence of specific types of defects.

[0068] After obtaining the defect mask, the manufacturing parameters are matched and associated based on the defect mask and parameter mapping to generate associated records. That is, by using the abnormal areas marked in the defect mask, the abnormal features are mapped to the production parameters to find the specific process conditions or equipment status that may cause these defects and form a structured record.

[0069] Production feedback is generated based on correlation records, and this feedback is used for coil production management. This involves transforming correlation records into actionable production improvement suggestions, such as adjusting tension, controlling temperature, or optimizing winding speed, thereby reducing defects and improving product quality in subsequent production. Production management can include adjustment prompts from automated control systems or operational suggestions for manual intervention.

[0070] In summary, the defect identification method for an automatic coil inspection machine provided in this application has the following technical effects: by achieving the technical goal of high-precision, full-spectrum, and multi-scale automatic identification and positioning of surface defects in coils, it achieves the technical effects of improving defect detection accuracy, enhancing quantitative defect analysis capabilities, realizing closed-loop feedback of production parameters, and optimizing coil production management.

[0071] Example 2: Based on the same inventive concept as the defect identification method for an automatic coil inspection machine in the foregoing examples, this application also provides a defect identification system for an automatic coil inspection machine. Please refer to the appendix. Figure 2The system includes: a detection command activation module 1, used to trigger the detection command of the coil automatic detection machine when the coil reaches the target detection area; an image acquisition module 2, used to start the ring adjustable light source according to the detection command, control the visible light camera and infrared camera set directly above the target detection area to perform image acquisition, and establish a detection image set; an image traversal scanning module 3, used to perform distortion and illumination correction on the detection image set, perform multispectral feature fusion on the corrected image set to construct an enhanced image, input the enhanced image to the window recognition channel, use a sliding window to perform image traversal scanning of the enhanced image, and calculate the texture energy and direction consistency index within each sliding window to establish a coordinate index; an image anomaly comparison module 4, used to call the calibration position image according to the coordinate index, use the calibration position image and the enhanced image corresponding to the coordinate index to perform multi-scale image anomaly comparison, and establish pixels with anomaly value identification; and a defect mask output module 5, used to perform morphological aggregation and boundary continuity analysis on the pixels with anomaly value identification, extract geometric morphological parameters, spectral feature parameters and texture anomaly parameters, and output a defect mask.

[0072] Furthermore, the image traversal scanning module 3 is used to perform the following steps: It calls the three-dimensional calibration model of the target detection area through the calibration database of the detector, and uses the three-dimensional calibration model to perform joint correction of radial and tangential distortion on the visible light image and infrared image respectively; it acquires environmental spectral monitoring data of the target detection area, constructs an illumination distribution surface model based on the environmental spectral monitoring data, and uses the illumination distribution surface model to perform illuminance correction under block brightness equalization on the distortion-corrected visible light image and infrared image respectively, performing the following: dividing the visible light image or infrared image into N grid blocks, calculating a stable brightness statistic for each grid block; configuring the target brightness constant according to the stable brightness statistic, calculating the gain map of each pixel, and performing multiplicative correction of the original image to complete the illuminance correction.

[0073] Furthermore, the image traversal scanning module 3 is used to perform the following operation steps: after performing pixel-level spatial geometric registration between the illuminance-corrected visible light image and the infrared image, perform hierarchical feature decomposition to extract low-frequency layer features, mid-frequency layer features, and high-frequency layer features, and perform adaptive weight fusion based on the sharpness score and defect prior probability of the multi-layer features; and construct an enhanced image using the adaptive weight fusion result.

[0074] Furthermore, the image traversal scanning module 3 is used to perform the following steps: call the texture extraction layer of the window recognition channel, perform enhanced image texture complexity evaluation, and establish a texture complexity evaluation identifier; establish a window shape library for sliding windows, the window shape library including square windows and rectangular windows, and generate position window shape constraints after matching the shape of the window shape library using the texture complexity evaluation identifier; configure multi-scale position windows using the position window shape constraints, perform texture energy descriptor calculation, establish texture energy distribution, and establish coordinate indexes based on the texture energy distribution and direction consistency index.

[0075] Furthermore, the image traversal scanning module 3 is used to perform the following operation steps: perform gradient direction clustering for each multi-scale position window and calculate the main direction concentration; use the main direction concentration as the direction consistency index, and establish a coordinate index based on the texture energy distribution and the direction consistency index.

[0076] Furthermore, the image anomaly comparison module 4 is used to perform the following steps: calling the multi-scale calibration location image in the calibration library according to the coordinate index; performing ensemble correction and spectral normalization processing on the multi-scale calibration location image, performing image registration at different scale layers, and performing anomaly comparison based on spectral differences, texture differences, and ensemble differences to establish anomaly identifiers.

[0077] Furthermore, the defect mask output module 5 is used to perform the following steps: spatial aggregation of abnormal pixels using morphological connected component analysis, merging neighboring similar abnormal pixels, wherein the connected component analysis includes dilation region analysis, erosion region analysis, and opening / closing operation region analysis; calculating boundary curvature and concavity / convexity indices for each spatial aggregation result to generate boundary continuity analysis results; and extracting geometric morphological parameters, spectral feature parameters, and texture anomaly parameters using the spatial aggregation results and the boundary continuity analysis results.

[0078] Furthermore, the defect mask output module 5 is used to perform the following steps: obtain the parameter mapping between coil manufacturing parameters and defect set; after obtaining the defect mask, perform the association matching of manufacturing parameters according to the defect mask and the parameter mapping to generate an association record; generate production feedback according to the association record, and use the production feedback for coil production management.

[0079] Through the foregoing detailed description of a defect identification method for an automatic coil inspection machine, those skilled in the art can clearly understand the defect identification system of an automatic coil inspection machine in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0080] In Embodiment 3, based on the inventive concept of the defect identification method of an automatic coil inspection machine in the foregoing embodiments, this application also provides an electronic device, including: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the defect identification method of an automatic coil inspection machine as described in any one of Embodiment 1 above.

[0081] Appendix Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application. Figure 3 In this document, the bus architecture is represented by bus 300. Bus 300 may include any number of interconnected buses and bridges, and bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 can be used to store data used by processor 302 during operation.

[0082] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0083] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A defect identification method for an automatic coil inspection machine, characterized in that, The method includes: When the coil reaches the target detection area, the detection command of the automatic coil detection machine is triggered; After the ring-shaped adjustable light source is activated according to the detection command, the visible light camera and infrared camera set directly above the target detection area are controlled to perform image acquisition and establish a detection image set. After distortion and illumination correction of the detected image set, the corrected image set is fused with multispectral features to construct an enhanced image. The enhanced image is input into the window recognition channel, and the image traversal scan of the enhanced image is performed using a sliding window. The texture energy and orientation consistency index within each sliding window are calculated, and a coordinate index is established. The calibration location image is retrieved based on the coordinate index, and the enhanced image corresponding to the calibration location image and the coordinate index is used to perform multi-scale image anomaly comparison to establish outlier marker pixels. After performing morphological aggregation and boundary continuity analysis on pixels with outlier identifiers, geometric morphological parameters, spectral feature parameters, and texture anomaly parameters are extracted, and a defect mask is output. The step of inputting the enhanced image into the window recognition channel and performing image traversal scanning of the enhanced image using a sliding window includes: The texture extraction layer of the window recognition channel is invoked to perform enhanced image texture complexity evaluation and establish a texture complexity evaluation identifier; A window shape library for sliding windows is established, including square windows and rectangular windows. After matching the shapes in the window shape library using the texture complexity evaluation identifier, position window shape constraints are generated. After configuring multi-scale position windows using position window shape constraints, texture energy descriptor calculation is performed to establish texture energy distribution. Based on the texture energy distribution and orientation consistency index, coordinate index is established.

2. The defect identification method of an automatic coil inspection machine as described in claim 1, characterized in that, The distortion and illumination correction of the detected image set includes: The calibration database of the detection machine is used to call the three-dimensional calibration model of the target detection area, and the three-dimensional calibration model is used to perform joint correction of radial distortion and tangential distortion on the visible light image and the infrared image respectively. Acquire environmental spectral monitoring data of the target detection area, construct an illumination distribution surface model based on the environmental spectral monitoring data, and use the illumination distribution surface model to perform block-based brightness equalization illuminance correction on the distortion-corrected visible light image and infrared image, respectively, as follows: Divide the visible light image or infrared image into N grid blocks, and calculate the stable brightness statistics for each grid block; After configuring the target luminance constant based on the stable luminance statistics, the gain map of each pixel is calculated, and then multiplicative correction of the original image is performed to complete the illuminance correction.

3. The defect identification method of an automatic coil inspection machine as described in claim 2, characterized in that, The step of fusing multispectral features of the corrected image set to construct an enhanced image includes: After performing pixel-level spatial geometric registration between the illuminance-corrected visible light image and the infrared image, hierarchical feature decomposition is performed to extract low-frequency, mid-frequency, and high-frequency features. Adaptive weight fusion is then performed based on the sharpness score and prior probability of defects of the multi-layer features. An enhanced image is constructed using the results of adaptive weight fusion.

4. The defect identification method of an automatic coil inspection machine as described in claim 1, characterized in that, The step of establishing a coordinate index based on the texture energy distribution and orientation consistency index includes: Gradient direction clustering is performed for each multi-scale location window, and the concentration of the main direction is calculated. The concentration of the main direction is used as the directional consistency index, and a coordinate index is established based on the texture energy distribution and the directional consistency index.

5. The defect identification method of an automatic coil inspection machine as described in claim 1, characterized in that, The step of using the calibrated location image and the enhanced image corresponding to the coordinate index to perform multi-scale image anomaly comparison and establish outlier marker pixels includes: The multi-scale calibration location image in the calibration library is retrieved based on the coordinate index; After performing ensemble correction and spectral normalization on the multi-scale calibrated location images, image registration is performed at different scale layers. Anomaly comparison is then conducted based on spectral differences, texture differences, and ensemble differences to establish anomaly identifiers.

6. The defect identification method of an automatic coil inspection machine as described in claim 1, characterized in that, The step of performing morphological aggregation and boundary continuity analysis on pixels with outlier identifiers includes: Spatial aggregation of anomalous pixels is performed using morphological connected region analysis, merging neighboring similar anomalous pixels. The connected region analysis includes dilation region analysis, erosion region analysis, and opening / closing operation region analysis. For each spatial aggregation result, the boundary curvature and concavity / convexity indices are calculated to generate boundary continuity analysis results; Geometric morphological parameters, spectral feature parameters, and texture anomaly parameters are extracted using the spatial aggregation results and the boundary continuity analysis results.

7. The defect identification method of an automatic coil inspection machine as described in claim 1, characterized in that, The output defect mask includes: Obtain the parameter mapping between coil manufacturing parameters and defect sets; After obtaining the defect mask, the manufacturing parameters are matched and associated based on the defect mask and the parameter mapping to generate an associated record; Production feedback is generated based on the associated records, and the production feedback is used for coil production management.

8. A defect identification system for an automatic coil inspection machine, characterized in that, A defect identification method for implementing an automatic coil inspection machine according to any one of claims 1-7, the system comprising: The detection command activation module is used to trigger the detection command of the automatic coil detection machine when the coil reaches the target detection area. The image acquisition module is used to control the visible light camera and infrared camera set directly above the target detection area to perform image acquisition and establish a detection image set after the ring adjustable light source is activated according to the detection command. The image traversal scanning module is used to perform distortion and illumination correction on the detection image set, then perform multispectral feature fusion on the corrected image set to construct an enhanced image, input the enhanced image into the window recognition channel, perform image traversal scanning of the enhanced image using a sliding window, calculate the texture energy and orientation consistency index within each sliding window, and establish a coordinate index. The image anomaly comparison module is used to call the calibration position image according to the coordinate index, and use the calibration position image and the enhanced image corresponding to the coordinate index to perform multi-scale image anomaly comparison and establish pixel points for anomaly identification. The defect mask output module is used to perform morphological aggregation and boundary continuity analysis on pixels with outlier identifiers, extract geometric morphological parameters, spectral feature parameters and texture anomaly parameters, and output a defect mask.

9. An electronic device, characterized in that, include: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the steps of the defect identification method of an automatic coil inspection machine according to any one of claims 1 to 7.

Images