Metal surface weld feature positioning method and system based on polarization imaging

Abstract

The invention relates to a metal surface weld feature positioning method and system based on polarization imaging, and the method comprises the steps: collecting a plurality of polarization information images of a metal surface in different polarization directions, and calculating the Stokes parameter images of the polarization information images; processing the plurality of polarization information images to obtain strong reflection light component images; segmenting the strong reflection light component image to obtain an initial target laser line image; segmenting the second Stokes parameter image to obtain an initial background laser line image; performing morphological operation on the initial background laser line image to obtain a final background laser line image; performing logical operation on the initial target laser line image and the final background laser line image to obtain a final target laser line image; and reconstructing a laser line track and positioning welding seam feature points. According to the method, the dependence of the traditional technology on a large number of manual parameters is avoided, and the robustness, the adaptive capability and the generalization performance in the complex illumination and reflection environment are greatly improved through the adaptive image processing flow.

Description

Technical Field

[0001] This invention relates to the field of polarization imaging measurement technology, and in particular to a method and system for locating weld features on metal surfaces based on polarization imaging. Background Technology

[0002] Industrial welding, as a core link in manufacturing, is accelerating its paradigm shift from "experience-driven" to "data-driven." As the foundation for scene perception and feature reconstruction, the high-precision measurement capabilities of visual sensors are not only a key prerequisite for achieving intelligent welding but also a crucial technological support for promoting "agile and precise" flexible manufacturing and the implementation of "lights-out factories." Among these technologies, weld seam recognition and tracking technology based on "laser + vision" sensors has become the most promising sensing method for robotic automation and intelligent welding due to its advantages such as non-contact operation, high precision, and anti-interference capabilities. This technology integrates a laser emitter and an image sensor to achieve real-time recognition and feedback of weld seam features, guiding the robot to dynamically adjust its welding trajectory, thereby ensuring high precision and consistency in the welding process.

[0003] However, visual sensors are extremely sensitive to background light, and their image quality is easily affected by the reflective properties of the object being measured, especially in scenarios with multiple reflections. For example, when welding highly reflective metal materials such as aluminum sheets, the multiple reflected rays (false laser lines) and the directly projected rays (real laser lines) exhibit high similarity in brightness and shape, severely interfering with the accurate identification of the real laser lines and the precise positioning of weld feature points. This leads to defects such as weld misalignment and incomplete penetration, significantly impacting welding quality and efficiency. Furthermore, due to the strong specular reflection of high-power lasers on metal surfaces, image sensors are prone to oversaturation, resulting in bright, overexposed areas in the acquired images. This oversaturation masks the detailed features of the real laser lines, severely affecting subsequent critical operations such as laser line centerline extraction and weld positioning.

[0004] To address this issue, invention patent CN119941834A discloses an anti-interference target laser line extraction method. By integrating algorithms such as brightness abrupt change detection, linewidth constraint, grayscale centroid extraction, and line-by-line tracking, it possesses a certain ability to identify and eliminate false laser lines. However, this method primarily relies on manually set parameters and lacks adaptive adjustment capabilities for complex scenes. Especially when facing high reflectivity interference such as strong specular reflection from metal surfaces and overexposure saturation, its suppression ability remains limited, making it difficult to effectively extract genuine laser lines. Furthermore, its line-by-line traversal strategy has low computational efficiency when processing high-resolution images, making it unsuitable for high-speed online applications. Therefore, accurately extracting genuine laser lines under strong reflection interference conditions is a key issue in achieving high-precision, automated welding.

[0005] In recent years, the development of polarization imaging technology has provided a new solution for the efficient and accurate extraction of laser lines in environments with strong reflection interference. By utilizing the polarization physics of metal surfaces and the response differences of different reflected light in various polarization directions, polarization imaging can enhance the ability to distinguish between real and false laser lines, thereby significantly improving the detection accuracy and stability of real laser lines.

[0006] Lei et al. (Opt. Lasers Eng., 186, 108820, 2025) proposed a method for extracting noise-reducing features of aluminum alloy fillet welds based on polarized laser vision. This method utilizes the reflection suppression advantage of polarized imaging, combined with minimum pixel value calculation, region erosion, and discrete noise removal, to effectively remove various reflection noises. Furthermore, through multiple set constraints, it achieves high-precision extraction of the laser centerline. However, while this method effectively suppresses some reflection noise, its adaptability to complex lighting environments and oversaturated laser lines is limited, and its generalization ability is insufficient. Moreover, this method involves multiple manually set parameters, making adaptive extraction difficult. In addition, invention patent CN118195933A discloses a method and system for denoising aluminum alloy fillet welds based on polarized vision. It extracts real laser lines through image difference combined with pixel segmentation and fitting algorithms. Although this method performs well under weak reflection interference conditions, it still struggles to effectively suppress false laser lines in strong reflection scenarios, exhibiting limited robustness and adaptability. Chinese patent application CN119934968A discloses a method and apparatus for structured light polarization visual inspection of fillet welds under strong reflective conditions, utilizing multi-angle polarized images for reflective removal and weld position extraction. However, this technical solution does not fully exploit the polarization characteristics of reflected light, resulting in limited line extraction performance under high reflective backgrounds. Furthermore, the method's processing flow relies on multiple image acquisitions and step-by-step filtering, leading to significant computational overhead, and the parameter settings lack adaptability, making automated industrial deployment difficult.

[0007] In summary, existing methods largely rely on traditional image processing techniques, failing to fully exploit the polarization characteristics of reflected laser lines. Their adaptability and robustness in complex interference scenarios such as strong reflection and overexposure remain insufficient, making it difficult to guarantee laser line extraction accuracy. Therefore, to overcome these technical bottlenecks, it is urgent to construct a polarization-based visual perception and laser line extraction scheme for highly reflective metal welding scenarios. This requires developing a polarization-based laser visual perception method with rapid response, accurate detection, and automatic adaptation capabilities to achieve efficient and stable laser line recognition and weld feature extraction in complex reflection environments. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a method and system for locating weld features on metal surfaces based on polarization imaging, which solves the technical problems of low laser line recognition accuracy, poor robustness, insufficient generalization ability, and poor weld feature point positioning effect in the existing technology under complex interference environments such as strong reflection and overexposure.

[0009] To achieve the above objectives, in a first aspect, the technical solution adopted by the present invention is a method for locating weld features on a metal surface based on polarization imaging, the method comprising the following steps: S1. Projecting a linear laser beam with a specific polarization state onto the metal surface; S2. Acquire several polarization information images of the metal surface in different polarization directions, and calculate the Stokes parameter image based on the several polarization information images; S3. Based on the decoupling and separation model of the reflected laser line, the polarization information images are processed to obtain the strong reflected light component image of the reflected laser line reflected from the metal surface. S4. Using a threshold segmentation algorithm, the image of the strong reflected light component is segmented into the initial target laser line region to obtain a binarized initial target laser line image. S5. Using a threshold segmentation algorithm, the Stokes second parameter image in the Stokes parameter image is segmented into the initial background laser line region to obtain a binarized initial background laser line image. S6. Perform morphological operations on the initial background laser line image to obtain a binarized final background laser line image. S7. Perform a logical NOT operation on the final background laser line image, and then perform a logical AND operation with the initial target laser line image to obtain the final target laser line image. S8. Extract the center feature points of the laser line in the final target laser line image; S9. Fit the center feature point to obtain the laser line trajectory curve, and locate the weld feature point according to the laser line trajectory curve.

[0010] The beneficial effects of this invention are as follows: Firstly, by utilizing the polarization-based imaging-based method for locating weld features on metal surfaces, the non-polarized strong reflected light component is accurately separated using a decoupling and separation model based on reflected laser lines, significantly suppressing background stray light and diffuse reflection interference, and enhancing the signal-to-noise ratio of the real laser line. Secondly, by innovatively combining Stokes' second parameter image segmentation and morphological processing, false laser line interference formed by multiple reflections can be effectively distinguished and eliminated, overcoming the adverse effects of oversaturated regions on feature extraction. This method avoids the reliance on numerous manual parameters found in traditional techniques and significantly improves robustness, adaptability, and generalization performance under complex lighting and reflection environments through an adaptive image processing workflow. Finally, it achieves high-precision extraction of the center point of the real laser line and stable and accurate positioning of weld feature points, providing reliable technical support for automated welding and significantly improving welding quality and efficiency.

[0011] Preferably, step S2 includes the following steps: S2.1. Use a polarization camera to acquire N polarization information images at equal intervals in different polarization directions, where N≥3; S2.2. Based on the principle of polarization imaging, obtain the polarization intensity value of each polarization information image. Specifically, it is expressed as follows: ;in, Indicates the first The polarization angle corresponding to the amplitude polarization information image; , , These represent the Stokes first parameter image, the Stokes second parameter image, and the Stokes third parameter image, respectively. S2.3. Based on the polarization intensity values ​​of the polarization information image obtained in step S2.2, obtain the Stokes parameter image, which is specifically represented as follows: .

[0012] Preferably, step S3 includes the following steps: S3.1 Arrange the N polarization information images obtained in step S2 in different polarization directions in order to construct an image matrix. The image matrix Specifically, it is expressed as follows: ;in, , Indicates the polarization angle as The non-polarized component at time, and Independent of polarization angle It changes with the changes. , Represents the image of the strongly reflected light component; Indicates the polarization angle as polarization components at time, , This represents the linear polarization angle of the incident laser. Represents the image of the weakly reflected light component; S3.2, Based on the image matrix obtained in step S3.1 Construct a measurement matrix The measurement matrix The measurement matrix is ​​used to describe the linear mapping relationship between incident laser light and the acquired polarization information image at different polarization angles. Specifically, it is expressed as follows: ; S3.3 Calculate the measurement matrix generalized inverse matrix And solve for the vector The vector Specifically, it is expressed as follows: ;in, , , , ; S3.4, Based on the vector obtained in step S3.3 The image of the weakly reflected light component in the reflected laser line was calculated. Image with strong reflected light component Specifically, it is expressed as: , .

[0013] Preferably, in step S4, the specific process of performing initial target laser line region segmentation on the strongly reflected light component image is as follows: setting a standard threshold. The image of strongly reflected light components that are greater than the standard threshold The pixels that are in the target laser line area are assigned to the target laser line area, while those that are not are assigned to the background area. After processing, a binarized initial target laser line image is obtained.

[0014] Preferably, in step S5, the specific process of segmenting the initial background laser line region of the Stokes second parameter image in the Stokes parametric image is as follows: setting a standard threshold. The Stokes second parameter image containing values ​​greater than the standard threshold Pixels with good morphology are assigned to the background area, while those with poor morphology are assigned to the target laser line area. After processing, the initial background laser line image is obtained.

[0015] Preferably, the threshold segmentation algorithm is a manually selected threshold segmentation method, or the OTSU maximum inter-class variance threshold segmentation method, or the maximum entropy-based threshold segmentation method, or the iterative threshold segmentation method.

[0016] Preferably, step S6 includes the following steps: S6.1. Perform dilation processing on the initial background laser line image obtained in step S5 to obtain the first processed image. Specifically, it is expressed as: ; where ⊕ represents the dilation operator, Indicates an expansion structural element. , Image representing the initial background laser lines pixel position coordinates, Represents expansion structural element The pixel position coordinates in the text; S6.2, Process the first processed image obtained in step S6.1. The second processed image is obtained by performing erosion processing in morphological operations. Specifically, it is expressed as: ;in, Represents the erosion operator. Indicates the corrosive structural element. , For the first image processing The pixel position coordinates in Represents structural element The pixel position coordinates in the text; S6.3, Process the second processed image obtained in step S6.2. The convex hull operation is performed on each connected region in the image to obtain the final background laser line image. Specifically, it is expressed as: ;in, Indicates the first The set of pixel coordinates of a connected region; Conv(.) represents the convex hull operation; U represents the union operation.

[0017] Preferably, in step S8, the method used to extract the center feature points of the laser line in the final target laser line image is the geometric center method, or the extreme value method, or the skeleton method, or the gray-scale centroid method, or the normal centroid method, or the Steger algorithm based on the Hessian matrix.

[0018] Preferably, step S9 includes the following steps: S9.1 Based on the central feature points obtained in step S8, the line trajectory curves of two independent laser lines formed on the left and right sides of the weld are constructed using a fitting method. S9.2 Calculate the fitting function for the two line trajectory curves respectively; S9.3. Based on the obtained fitting function, calculate the coordinates of the intersection point of the two line trajectory curves. The coordinates of the intersection point are the characteristic points of the weld.

[0019] Secondly, the technical solution adopted by the present invention is a metal surface weld feature localization system based on polarization imaging, comprising: case; A laser emission window and a polarized light incident window are provided on the front side of the housing; Laser emitting module, used to generate a line-structured laser beam; The rotating base is mounted on the bottom plate of the housing, and the laser emitting module is mounted on the rotating base. The spatial projection angle of the laser emitting module can be changed by adjusting the rotating base. A polarization camera mounted on the bottom plate of the housing; The polarization modulation module installed in the housing has its optical axis coincident with the optical axis of the laser emission module. It is used to polarize the line structure laser beam generated by the laser emission module to generate a line structure laser beam with a specific polarization state. The line structure laser beam with a specific polarization state is projected onto the metal surface to be tested through the laser emission window. The filter module is installed inside the housing; The imaging lens installed in the housing has optical axes that coincide with the optical axis of the polarization camera. The reflected laser line with a specific polarization state reflected from the metal surface under test passes through the polarization light incident window, then passes through the filter module and the imaging lens in sequence, and is focused onto the polarization camera, which generates a polarization information image. The image acquisition and processing module is connected to the polarization camera and is used to acquire polarization information images and execute image processing algorithms to extract real laser lines and identify weld positions.

[0020] The aforementioned polarization imaging-based metal surface weld feature localization system integrates a polarization modulation module, a filter module, and a polarization camera into a single package. Active polarization modulation effectively enhances the difference between the laser line and the complex reflective background, improving the signal-to-noise ratio from the source. The combination of the filter module and the adjustable rotating base allows the system to flexibly adapt to different working conditions and ambient light interference, exhibiting excellent industrial field adaptability. This integrated structure enables efficient and stable acquisition of polarization information, providing a high-quality data foundation for backend processing algorithms. This supports accurate laser line extraction and stable weld location even under harsh conditions such as strong reflection and overexposure. The system has a compact structure, high reliability, and is easy to integrate and deploy in the field. Attached Figure Description

[0021] Figure 1This is a schematic diagram of the overall structure of a metal surface weld feature localization system based on polarization imaging according to the present invention. Figure 2 This is a schematic diagram of the internal structure of a metal surface weld feature localization system based on polarization imaging according to the present invention. Figure 3 This is a schematic diagram illustrating the system principle of a metal surface weld feature localization system based on polarization imaging according to the present invention. Figure 4 This is a schematic diagram of the welding scenario of V-shaped aluminum sheet metal in this invention; Figure 5 This is a flowchart of a method for locating weld features on a metal surface based on polarization imaging, as described in this invention. Figure 6 A schematic diagram of the polarization information images acquired in different directions and the calculated Stokes parameter images in this embodiment of the invention; Figure 7 A flowchart of step S3 in this invention; Figure 8 The polarization component image calculated in the embodiments of the present invention With non-polarized component image A schematic diagram; Figure 9 In Figure (a), the Stokes second parameter image and the Stokes third parameter image in an embodiment of the present invention are as a function of the incident laser angle. A schematic diagram of the changing curve; Figure 9 (b) shows the Stokes second parameter image and the Stokes third parameter image in an embodiment of the present invention as a function of the first reflection angle. φ 1. A schematic diagram of the changing curve; Figure 10 A schematic diagram of the initial target laser line image, the initial background laser line image, and the final background laser line image in an embodiment of the present invention; Figure 11 A schematic diagram of the final target laser line image in an embodiment of the present invention; Figure 12 A schematic diagram showing the results of laser line center point extraction and weld feature point positioning on the surface of an aluminum plate in this embodiment of the invention. Figure 13 Schematic diagram of the results of laser line extraction and weld feature point localization using different methods in this embodiment of the invention; As shown in the figure: 1. Housing; 2. Rotating base; 3. Filter module; 4. Imaging lens; 5. Polarization camera; 6. Laser emission module; 7. Polarization modulation module; 8. Image acquisition and processing module; 9. Laser emission window; 10. Polarized light incident window; 11. V-shaped aluminum sheet; 12. Working platform; 13. Controller. Detailed Implementation

[0022] The invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can implement it based on the description. The scope of protection of the invention is not limited to these specific embodiments.

[0023] Those skilled in the art should understand that, in the disclosure of this invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting this invention.

[0024] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0025] In the description of the embodiments of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0026] This invention provides a metal surface weld feature localization system based on polarization imaging. A schematic diagram of the overall system structure is shown below. Figure 1 As shown in the diagram, please refer to the internal structure diagram. Figure 2 As shown, the system's structural schematic diagram is available for reference. Figure 3As shown. The system specifically includes: a housing 1, a rotating base 2, a filter module 3, an imaging lens 4, a polarization camera 5, a laser emission module 6, a polarization modulation module 7, and an image acquisition and processing module 8. The laser emission module 6 generates a linear laser beam, which is modulated by the polarization modulation module 7 to form a laser line with a specific polarization state. This laser line with a specific polarization state passes through the laser emission window 9 on the front side of the housing and is projected onto the metal surface to be tested. The reflected laser line with a specific polarization state reflected from the metal surface passes through the polarization light incident window 10 on the front side of the housing 1, and is focused onto the polarization camera 5 after passing through the filter module 3 and the imaging lens 4, forming an image containing polarization information in different polarization directions. The image acquisition and processing module 8 controls the polarization camera to acquire polarization information images containing different polarization directions through host computer software, and combines this with backend processing algorithms to extract the actual laser line and accurately identify the weld position.

[0027] like Figure 1 and Figure 2 As shown, the housing 1 is used to structurally fix and encapsulate the various components. The rotating base 2 and the polarization camera 5 are mounted on the bottom plate of the housing 1. A laser emission window 9 and a polarization light incident window 10 are provided on the front side of the housing 1.

[0028] like Figure 2 As shown, the rotating base 2 is used to install the laser emitting module 6 and to make the spatial attitude of the laser emitting module 6 adjustable to adapt to the laser projection angle under different working conditions.

[0029] like Figure 1 and Figure 2 As shown, the axis of the filter module 3 coincides with the optical axis of the imaging lens 4 and the polarizing camera 5. The filter module 3 is used to suppress ambient stray light.

[0030] like Figure 2 As shown, the imaging lens 4 includes an optical lens group for imaging. The imaging lens 4 is used to focus the light field information reflected from the scene onto the polarization camera 5, and its axis coincides with the optical axis of the polarization camera 5.

[0031] like Figure 2 As shown, the polarization camera 5 is used to receive light signals passing through the filter module 3 and the imaging lens 4, and convert them into electrical signals to achieve the acquisition of polarization information images of the scene.

[0032] like Figure 1 and Figure 2 As shown, the laser emitting module 6 is used to project a line-structured laser line onto the metal surface, and works with the polarization modulation module 7 to generate a line-structured laser beam with a specific polarization state.

[0033] like Figure 2 As shown, the axis of the polarization modulation module 7 coincides with the optical axis of the laser emission module 6, and is used to actively modulate the polarization direction of the laser to enhance the polarization difference between the laser line and the background.

[0034] like Figure 3 As shown, the image acquisition and processing module 8 is connected to the polarization camera 5 and is used to acquire the polarization information image generated by the polarization camera 5, and to solve the polarization information image to extract the real laser lines on the metal surface and extract the weld feature points.

[0035] Specifically, the rotating base 2 is a manually adjustable structure used to realize the adjustable configuration of the projection direction of the laser emitting module 6; the rotating base 2 is fixedly connected to the bottom plate of the housing 1 by bolts, and supports continuous rotation in the horizontal direction from 0 to 360°, wherein the angle between the emitted light of the laser emitting module 6 and the optical axis of the polarization camera 5 is set to 25°.

[0036] Specifically, the filter module 3 includes a filter, which is detachably connected to the housing 1; the filter is a narrowband spectral filter with a working wavelength of 660nm±15nm.

[0037] Specifically, the operating band of the imaging lens 4 is matched according to the polarization camera 5 connected to it, and the interface can be a C-type interface; the lens type can be selected as a fixed-focus lens according to the application requirements, with a focal length of 16mm and a field of view of 31.4°×23.7°.

[0038] Specifically, the polarization camera 5 is used to convert light signals into electrical signals, capture the light beam passing through the imaging lens 4, and generate digital image data. According to the usage scenario and system design requirements, the polarization camera 5 is a focal plane type structure. It uses a Sony IMX264MZR CMOS image sensor, transmits image data through a USB 3.0 data interface, has a resolution of 2448×2048, and a pixel size of 3.45μm.

[0039] Specifically, the laser emitting module 6 can be a red laser, whose emission wavelength matches that of the filter module 3 and the polarization camera 5; depending on the application scenario, the thickness and brightness of the laser beam emitted by the laser emitting module 6 can be adjusted; the output wavelength of the laser emitting module 6 is 655nm, the rated voltage is 2.8~5.2V, and the power is 200mW.

[0040] Specifically, the polarization modulation module 7 includes a polarization device, which is detachably connected to the housing 1. The polarization device is a metal wire grid polarizer, which has a TM wave transmittance of better than 83% and an extinction ratio of better than 800:1 (29dB) in the 420~700nm band.

[0041] Specifically, the image acquisition and processing module 8 executes image processing algorithms to extract the real laser line and identify the weld location, specifically employing steps S2 to S9 of the polarization imaging-based metal surface weld feature localization method proposed in this invention. The image acquisition and processing module 8 is responsible for image acquisition, data processing, storage, and display.

[0042] Based on the aforementioned polarization imaging-based metal surface weld feature localization system, such as Figure 4 As shown, this embodiment of the invention also addresses the welding scenario of V-shaped aluminum sheet 11, providing a method for locating weld features on the metal surface based on polarization imaging. The flowchart of the method is as follows. Figure 5 As shown, the specific process includes the following steps: S1. Projecting a linear laser beam with a specific polarization state onto the metal surface; In step S1, a metal surface weld feature positioning system based on polarization imaging is set up, and a linear laser beam with a specific polarization state is projected onto the metal surface by the laser emission module 6 in the system. S2. Acquire several polarization information images of the metal surface in different polarization directions, and calculate the Stokes parameter image based on the several polarization information images; In step S2, polarization information images with different polarization directions are acquired, and polarization feature parameter images are calculated according to the polarization imaging principle and Stokes theory. Specifically, the polarization feature parameter images are linear Stokes parameter images. S3. Based on the decoupling and separation model of the reflected laser line, process the several polarization information images to obtain the strong reflected light component image of the reflected laser line; In step S3, the polarization information images of different polarization directions obtained in step S2 are used in combination with the decoupling and separation model of the reflected laser line on the metal surface to calculate the strong reflected light component image of the reflected laser line, so as to suppress the interference of background stray weak reflected light and enhance the target laser line. S4. Use a threshold segmentation algorithm to segment the background region and the initial target laser line region of the strong reflection light component image to obtain a binarized initial target laser line image. S5. Using a threshold segmentation algorithm, the Stokes second parameter image in the Stokes parameter image is segmented into the background region and the strong reflection interference line region in the background to obtain a binarized initial background laser line image. In step S5, based on the polarization characteristics of the primary and secondary reflected light from the metal surface, a threshold segmentation algorithm is applied to the Stokes second parameter image obtained in step S2 to extract strong reflection interference lines in the background, resulting in a binarized initial background laser line image. S6. Perform morphological operations on the initial background laser line image to obtain a binarized final background laser line image. In step S6, the binarized initial background laser line image obtained in step S5 is processed by morphological filtering algorithms such as dilation and erosion to remove isolated pixels, connect broken areas, and fill internal holes, thereby obtaining a structurally complete final background laser line image. S7. Perform a logical NOT operation on the final background laser line image calculated in step S6, and then perform a logical AND operation on the initial target laser line image calculated in step S4 to obtain the final target laser line image. S8. Extract the center feature points of the laser line in the final target laser line image; In step S8, based on the geometric shape and light intensity distribution characteristics of the final target laser line image obtained in step S7, the center position coordinates of the laser line are calculated for each column of pixels of the final target laser line, and the center feature points of the laser line are extracted. S9. Fit the central feature point to reconstruct the laser line trajectory and locate the weld feature point; In step S9, the complete laser line center trajectory and weld feature points are reconstructed using a discrete point fitting method, which is used for subsequent guidance, positioning, or tracking control tasks.

[0043] Furthermore, the specific implementation process of step S2 includes the following steps: S2.1. Acquire four polarization information images using the polarization camera 5. I 0° , I 45° , I 90° as well as I 135° The interval angle between the four polarization information images is 45°; S2.2. Based on the principle of polarization imaging, obtain the polarization intensity value of each polarization information image. Specifically, it is expressed as: ;in, Indicates the first The polarization angle corresponding to the amplitude polarization information image; , , , ; , , These represent the Stokes first parameter image, the Stokes second parameter image, and the Stokes third parameter image, respectively. S2.2. Based on the polarization intensity values ​​of the polarization information image obtained in step S2.2, obtain the Stokes parameter image, which is specifically represented as follows: In a specific embodiment, N =4.

[0044] Furthermore, to more clearly illustrate the technical solution of the present invention, please refer to... Figure 6 Stokes first parametric image S 0. The composition and optical properties of the reflected light after the laser line is projected onto a highly reflective metal surface (such as aluminum alloy) during laser-guided welding are described as follows: When a laser beam strikes a metal surface, the reflected light consists mainly of specular reflection and diffuse reflection components. The specular reflection component originates primarily from the microscopically smooth regions of the metal surface, exhibits high directionality, and is usually accompanied by strong polarization characteristics, resulting in a high degree of polarization. In contrast, the diffuse reflection component is mainly generated by scattering from the rough regions of the surface, has a relatively uniform directional distribution, weaker polarization characteristics, and a relatively lower degree of polarization.

[0045] However, highly reflective metal surfaces exhibit significant specular reflection characteristics. When a line laser beam irradiates a metal surface, the reflected light mainly consists of the following three components: The first specular reflection component of the laser: The first directional specular reflection produced by the incident laser beam on the metal surface presents a bright, fine line with a high degree of directionality, which is an effective source of information for weld feature extraction and positioning; Secondary specular reflection component of laser: formed by multiple specular reflections caused by the weld structure, it appears as a high-intensity radial line or bright area, which is easily confused with the primary reflection signal and is the main source of interference in weld tracking. Laser diffuse reflection component: Due to the micro-rough structure and scattering effect of the metal surface, some laser energy is converted into non-directional diffuse reflection after specular reflection or multiple reflections, which appears as a smoother, low-intensity scattered background area in the image. Background light reflection component: The reflection component from non-laser light sources such as ambient light and auxiliary lighting, which usually has randomness and spatial non-uniformity.

[0046] The four polarization information images obtained in the specific embodiments of the present invention in different polarization directions are as follows: Figure 6 As shown, from Figure 6As can be seen, with the change in the direction angle of the polarizing optical element, the light intensity in some areas of the aluminum plate surface shows a trend of changing from bright to dark. This indicates that the polarization imaging process can effectively modulate the reflected light intensity, and the light intensity response of different areas has significant differences with the polarization direction. The more obvious the change in light intensity, the more significant the polarization characteristics. However, due to the camera overexposure caused by excessively high reflected light intensity, the light intensity in some areas of primary and secondary specular reflection of the laser did not change significantly with the polarization angle, which poses a significant challenge to the subsequent extraction of the real laser line.

[0047] Furthermore, the flowchart for step S3 is as follows: Figure 7 As shown, the specific implementation process includes the following steps: S3.1 Arrange the four polarization information images obtained in step S2 in different polarization directions in sequence to construct an image matrix. The image matrix Specifically, it is expressed as follows: ;in, , Indicates the polarization angle as The non-polarized component at time, and Independent of polarization angle It changes with the changes. , This represents the image of the strongly reflected light component (the image of the intensity of the unpolarized component). Indicates the polarization angle as polarization components at time, , This represents the linear polarization angle of the incident laser. Image representing weakly reflected light component (polarization component intensity image); S3.2, Based on the image matrix obtained in step S3.1 Construct a measurement matrix The measurement matrix The measurement matrix is ​​used to describe the linear mapping relationship between incident laser light and the acquired polarization information image at different polarization angles. Specifically, it is expressed as follows: ; S3.3 Calculate the measurement matrix The generalized inverse matrix (pseudo-inverse matrix) And solve for the vector The vector Specifically, it is expressed as follows: ; in, , , , ; S3.4, Based on the vector obtained in step S3.3 The image of the weak reflected light component in the reflected light was calculated. Image with strong reflected light component Specifically, it is expressed as: , .

[0048] According to step S3, the polarization component image of this embodiment of the invention is obtained. Non-polarization component images like Figure 8 As shown, from Figure 8 As can be seen, due to the inherent polarization characteristics and the interaction of the light field between the interface and the environment, the two images exhibit significant differences in different regions. Specifically, this is reflected in the polarization component image. In the image, large-area reflection areas are more obvious, and reflection interference characteristics are prominent; while in the non-polarization component image... This effectively suppressed most of the background reflected light, retaining only a small number of significant reflective edge features, thus significantly enhancing the real laser line region. Based on this difference, the effective separation of reflected light components in the laser line image was initially achieved, providing a reliable data foundation for subsequent extraction of the real laser line.

[0049] Therefore, the specific implementation process of step S4 is as follows: for the strong reflection light component image calculated in step S3.4... Since the target laser line pixel brightness is significantly higher than the background pixel brightness, the two can be distinguished using a threshold segmentation method. Specifically, a standard threshold is set. The image of strongly reflected light components exceeding this standard threshold will be... Pixels with high resolution are assigned to the target laser line region, while those with low resolution are assigned to the background region. The resulting binarized image is the initial target laser line image. In other words, a threshold segmentation algorithm is used to segment the background region and the initial target laser line region to obtain the binarized initial target laser line image. Specifically, it is expressed as: In the binarized initial target laser line image, the region with a logic value of 1 is the initial target laser line region.

[0050] In a specific embodiment, the threshold segmentation algorithm used in step S4 is the OTSU maximum inter-class variance method.

[0051] Furthermore, to more clearly illustrate the technical solution of the present invention, the polarization characteristics of the primary and secondary reflected light from the metal surface described in step S5 are explained as follows: With the incident plane as a reference, the electric field vector of the incident laser can be decomposed into two orthogonal components: perpendicular (s-component) and parallel (p-component). For an air-metal interface, according to Fresnel's law of reflection, the reflectivity coefficient... and They are represented as follows: , ; in, Indicates the angle of incidence. Indicates the angle of refraction, and ; This represents the refractive index of air (usually taken as 1). It represents the complex refractive index of a metal surface.

[0052] For a single reflection from a metal surface, its Mueller matrix can be expressed as: ; in, Indicates the amplitude ratio; It represents the phase difference, reflecting the phase change of light; It represents the reflectivity of a metal surface during its first reflection.

[0053] Therefore, when the polarization state of the incident laser is S i After one reflection from the metal surface, the Stokes vector of the reflected light is: .

[0054] Secondary reflection from a metal surface can be viewed as a cascaded process of two separate reflections, each described by a Mueller matrix. Therefore, the Mueller matrix for secondary reflection from a metal surface can be expressed as: ; in, , These represent the amplitude ratios of the first and second reflections, respectively; , These represent the phase difference between the first and second reflections, respectively. It represents the reflectivity of a metal surface during a second reflection.

[0055] Therefore, when the polarization state of the incident laser is S i After secondary reflection from the metal surface, the Stokes vector of the reflected light is: .

[0056] Based on the above theoretical analysis of the polarization characteristics of primary and secondary reflected light, assuming the reflecting surface is a metallic aluminum surface with a complex refractive index of , Considering that the incident laser is linearly polarized, with the polarization direction set at 45°, its Stokes vector is then... When the metal surface undergoes primary reflection, such as Figure 9 As shown in (a), Figure 9 (a) Stokes second parameter image and Stokes third parameter image of the reflected light as the incident laser angle changes. The changing curve, from Figure 9 (a) It can be seen that: within a large angular range of 0° to 80°, the Stokes third parameter image It is a negative value, and it increases with the increase of the incident angle. It increases rapidly, approaching 1.

[0057] When the metal surface exhibits secondary reflection, such as Figure 9 As shown in (b), Figure 9 (b) Stokes second parameter image and Stokes third parameter image of the reflected light as the first reflection angle changes. The changing curve, from Figure 9 (b) It can be seen that: With the first reflection angle The changes exhibit clear periodic fluctuations, with values ​​alternating between positive and negative across the entire angular range. This is particularly evident in the angular region of approximately 22.5° to 80°. The value is positive. This difference provides a valid basis for distinguishing between primary and secondary reflected light.

[0058] Therefore, step S5 targets the second Stokes parameter image in the Stokes parameter image. Since the background pixel brightness is significantly higher than the target laser line, the two can be distinguished using a threshold segmentation method. Specifically, a standard threshold is set. The Stokes second parameter image containing values ​​greater than the standard threshold Pixels with good or bad performance are assigned to the background region, while those with good or bad performance are assigned to the target laser line region. The resulting binarized image is the initial background laser line image; that is, step S5 uses the Stokes second parameter image in the Stokes parameter image. The background region and the region of strong reflection interference lines in the background are segmented to obtain a binarized initial background laser line image. , can be represented as: In the binarized initial background laser line image, the regions with a logic value of 1 are the initial background laser line regions.

[0059] In a specific embodiment, the threshold segmentation algorithm used in step S5 is the OTSU maximum inter-class variance method.

[0060] Furthermore, the specific implementation steps of step S6 are as follows: S6.1, Based on the binarized initial background laser line image obtained in step S5 The first processed image is obtained by performing dilation, a morphological operation. Specifically, it is expressed as: ; Where ⊕ represents the dilation operator, Indicates an expansion structural element. , Image representing the initial background laser lines pixel position coordinates, Represents expansion structural element The pixel position coordinates in the image; after dilation processing, the boundary of the target area can be expanded, bridging the broken areas of the same region. S6.2, Process the first processed image obtained in step S6.1. The second processed image is obtained by performing erosion processing in morphological operations. Specifically, it is expressed as: ;in, Represents the erosion operator. Indicates the corrosive structural element. , For the first image processing The pixel position coordinates in Represents structural element The pixel position coordinates in the image; residual noise points can be eliminated through erosion processing; S6.3, Process the second processed image obtained in step S6.2. The convex hull operation is performed on each connected region in the image to obtain the final background laser line image. Specifically, it is expressed as: ;in, Indicates the first The set of pixel coordinates of a connected region; Conv(.) represents the convex hull operation; U represents the union operation. The connected region refers to the set of pixels in the binarized final background laser line image that have the same pixel value (e.g., all pixels with 1), are spatially adjacent, and can be connected.

[0061] like Figure 10 As can be seen, there is significant large-area reflection interference in the initial target laser line image, making it difficult to distinguish the real laser lines. By utilizing the difference in polarization characteristics between the primary and secondary reflected light, most of the background laser lines in the initial background laser line image have been successfully extracted, but due to camera overexposure, some areas are still missed. The final background laser line image, based on this, uses the edge contour information of the initial background laser lines to achieve complete and continuous background laser line extraction, significantly improving the region segmentation effect.

[0062] Further, the specific implementation method of step S7 is as follows: First, a logical NOT operation is performed on the final background laser line image, that is, white pixels (value 1) are converted to black pixels (value 0), and black pixels (value 0) are converted to white pixels (value 1), thus obtaining the inverted image. Next, a logical AND operation is performed pixel by pixel on the inverted image and the initial target laser line image. The output is 1 only when pixels at the same position in both images are simultaneously 1; otherwise, it is 0. Through this process, background interference can be effectively removed, and a binarized final target laser line image is obtained. ,Right now: In the binarized final target laser line image, the region with a logic value of 1 is the final target region.

[0063] like Figure 11 As shown, Figure 11 The image shown is of the final target laser line in this embodiment of the invention. It can be seen that the metal surface weld feature localization method based on polarization imaging proposed in this invention successfully eliminates the interference of secondary specular reflection light from the laser, effectively overcomes the problem of inaccurate measurement of the real laser line caused by camera oversaturation, and the extracted laser line is clear and continuous as a whole, without obvious false areas. This verifies the ability of this method to effectively extract the target laser line in complex reflection scenarios.

[0064] Furthermore, the specific implementation method of step S8 is as follows: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] The center feature points are extracted, and the discrete center feature point sequence of the laser line can be represented as: .

[0065] Furthermore, the method used for extracting the center feature point in step S8 is the gray-level centroid method. This method determines the coordinates of the center point by weighting the gray-level values ​​of the pixels in the target region of the image, i.e.: , ;in, Represents pixels grayscale value, This represents the corresponding pixel coordinates.

[0066] During weld inspection, when the laser irradiates the weld area, the laser stripes will form two independent trajectory curves on the left and right sides of the weld due to the influence of the weld groove or bevel structure. Therefore, the specific implementation method of step S9 includes the following steps: S9.1 Based on the central feature points obtained in step S8, the line trajectory curves of two independent laser lines formed on the left and right sides of the weld are constructed using a fitting method. S9.2 Calculate the fitting function for each of the two described line trajectory curves; the fitting function is specifically expressed as: The fitting function for laser line #1 is: The fitting function for laser line #2 is: ; S9.3. Based on the obtained fitting function, calculate the coordinates of the intersection point of the two line trajectory curves. , The coordinates of the intersection point are the weld feature points, thus realizing the positioning of the weld feature points.

[0067] In a specific embodiment, the fitting method described in step S9 is a first-order polynomial fitting based on the least squares method.

[0068] like Figure 12 As shown, Figure 12 The results of laser line extraction and weld feature point localization on the surface of aluminum plate are compared with the coordinates of the actual weld feature points. , Compared to the previous method, the weld feature points of the proposed method in this invention are... x The axial error is 2.70 pixels (considering lens parameters and pixel size, the corresponding actual measurement error is approximately 0.26 mm); weld feature points are... y The on-axis error is 1.58 pixels (corresponding to an actual measurement error of approximately 0.15 mm). That is, the average error in weld feature point positioning is 3.13 pixels, corresponding to an actual measurement average error of 0.30 mm. These results demonstrate that the method proposed in this invention can effectively suppress interference from secondary specular reflection of the laser, significantly alleviate the camera oversaturation problem caused by strong reflected light, thereby achieving high-quality extraction of the real laser line and accurate positioning of weld feature points. This method provides solid technical support for subsequent high-quality welding process control and high-precision weld tracking.

[0069] Furthermore, to verify the advancement of the method proposed in this invention, existing laser line extraction techniques are used as comparative methods. Specifically, Existing Method 1 is a method for extracting noise-reducing features of aluminum alloy fillet welds based on polarized laser vision, proposed by Lei et al. (Opt. Lasers Eng., 186, 108820, 2025). Existing Method 2 is an invention patent with publication number CN118195933A, which discloses a method and system for denoising aluminum alloy fillet welds based on polarized vision. Existing Method 3 is an invention patent with publication number CN119934968A, which discloses a method and device for structured light polarized vision detection of fillet welds under strong reflective light.

[0070] against Figure 6 Tests were conducted in the scenarios shown, where the coordinates of the actual feature points of the weld are ( , ).like Figure 13As shown, this paper presents a comparison of the laser line centerline extraction results and weld feature point positioning effects obtained by different methods. Figure 13 As can be seen, the method proposed in this invention can still stably extract the center line of the laser line even when it is subjected to strong reflection interference and large-area scattering noise. The fitted straight line is highly consistent with the main direction of the laser line, and the intersection point is accurately located, which fully demonstrates the superiority and robustness of the method. Existing methods 1 and 3 are severely affected by strong reflection interference, making it difficult to accurately extract the real laser line, resulting in a certain deviation between the fitting results and the feature point positioning results (the average error of weld feature point positioning is about 3.76 pixels and 5.33 pixels, respectively). Existing method 2, due to problems such as laser line breakage and blurred line edges, has a fitted straight line that deviates significantly from the real main direction of the line, resulting in a significant drift in the intersection point position and a large error (the average error of weld feature point positioning is about 245.94 pixels). In summary, the comparison results intuitively demonstrate the differences in robustness and adaptability of various methods in complex laser scenes, and verify the superiority of the method proposed in this invention in laser line extraction and weld feature point positioning.

Claims

1. A method for locating weld features on a metal surface based on polarization imaging, characterized in that: Includes the following steps: S1. Projecting a linear laser beam with a specific polarization state onto the metal surface; S2. Acquire several polarization information images of the metal surface in different polarization directions, and calculate the Stokes parameter image based on the several polarization information images; S3. Based on the decoupling and separation model of the reflected laser line, the polarization information images are processed to obtain the strong reflected light component image of the reflected laser line reflected from the metal surface. S4. Using a threshold segmentation algorithm, the image of the strong reflected light component is segmented into the initial target laser line region to obtain a binarized initial target laser line image. S5. Using a threshold segmentation algorithm, the Stokes second parameter image in the Stokes parameter image is segmented into the initial background laser line region to obtain a binarized initial background laser line image. S6. Perform morphological operations on the initial background laser line image to obtain a binarized final background laser line image. S7. Perform a logical NOT operation on the final background laser line image, and then perform a logical AND operation with the initial target laser line image to obtain the final target laser line image; S8. Extract the center feature points of the laser line in the final target laser line image; S9. Fit the center feature point to obtain the laser line trajectory curve, and locate the weld feature point according to the laser line trajectory curve.

2. The method for locating weld features on a metal surface based on polarization imaging according to claim 1, characterized in that: Step S2 includes the following steps: S2.

1. Use a polarization camera to acquire N polarization information images at equal intervals in different polarization directions, where N≥3; S2.

2. Based on the principle of polarization imaging, obtain the polarization intensity value of each polarization information image. Specifically, it is expressed as: ； in, Indicates the first The polarization angle corresponding to the amplitude polarization information image; , , These represent the Stokes first parameter image, the Stokes second parameter image, and the Stokes third parameter image, respectively. S2.

3. Based on the polarization intensity values ​​of the polarization information image obtained in step S2.2, obtain the Stokes parameter image, which is specifically represented as follows: 。 3. The method for locating weld features on a metal surface based on polarization imaging according to claim 2, characterized in that: Step S3 includes the following steps: S3.1 Arrange the N polarization information images obtained in step S2 in different polarization directions in order to construct an image matrix. The image matrix Specifically, it is expressed as follows: ； in, , Indicates the polarization angle as The non-polarized component at time, and Independent of polarization angle It changes with the changes. , Image representing the strongly reflected light component; Indicates the polarization angle as polarization components at time, , This represents the linear polarization angle of the incident laser. Represents the image of the weakly reflected light component; S3.2, Based on the image matrix obtained in step S3.1 Construct a measurement matrix The measurement matrix The measurement matrix is ​​used to describe the linear mapping relationship between incident laser light and the acquired polarization information image at different polarization angles. Specifically, it is expressed as follows: ； S3.3 Calculate the measurement matrix generalized inverse matrix And solve for the vector The vector Specifically, it is expressed as follows: ； in, , , , ; S3.4, Based on the vector obtained in step S3.3 The image of the weakly reflected light component in the reflected laser line was calculated. Image with strong reflected light component Specifically, it is expressed as: , 。 4. The method for locating weld features on a metal surface based on polarization imaging according to claim 3, characterized in that: In step S4, the specific process of performing initial target laser line region segmentation on the strongly reflected light component image is as follows: setting a standard threshold. The image of strongly reflected light components that are greater than the standard threshold The pixels that are in the target laser line area are assigned to the target laser line area, while those that are not are assigned to the background area. After processing, a binarized initial target laser line image is obtained.

5. The method for locating weld features on a metal surface based on polarization imaging according to claim 4, characterized in that: In step S5, the specific process of segmenting the initial background laser line region of the Stokes second parameter image in the Stokes parametric image is as follows: setting a standard threshold. The Stokes second parameter image containing values ​​greater than the standard threshold Pixels with good morphology are assigned to the background area, while those with poor morphology are assigned to the target laser line area. After processing, the initial background laser line image is obtained.

6. The method for locating weld features on a metal surface based on polarization imaging according to claim 5, characterized in that: The threshold segmentation algorithm can be a manually selected threshold segmentation method, or the OTSU maximum inter-class variance threshold segmentation method, or the maximum entropy-based threshold segmentation method, or the iterative threshold segmentation method.

7. The method for locating weld features on a metal surface based on polarization imaging according to claim 5, characterized in that: Step S6 includes the following steps: S6.

1. Perform dilation processing on the initial background laser line image obtained in step S5 to obtain the first processed image. Specifically, it is expressed as: ； Where ⊕ represents the dilation operator, Indicates an expansion structural element. , Image representing the initial background laser lines The pixel position coordinates in Represents expansion structural element The pixel position coordinates in the text; S6.2, Process the first processed image obtained in step S6.

1. The second processed image is obtained by performing erosion in morphological operations. Specifically, it is expressed as: ； in, Represents the erosion operator. Indicates the corrosive structural element. , For the first image processing The pixel position coordinates in the image. Represents structural element The pixel position coordinates in the text; S6.3, Process the second processed image obtained in step S6.

2. The convex hull operation is performed on each connected region in the image to obtain the final background laser line image. Specifically, it is expressed as: ； in, Indicates the first The set of pixel coordinates of a connected region; Conv(.) represents the convex hull operation; U represents the union operation.

8. The method for locating weld features on a metal surface based on polarization imaging according to claim 7, characterized in that: In step S8, the method used to extract the center feature points of the laser line in the final target laser line image is the geometric center method, or the extreme value method, or the skeleton method, or the gray-level centroid method, or the normal centroid method, or the Steger algorithm based on the Hessian matrix.

9. A method for locating weld features on a metal surface based on polarization imaging according to claim 7 or 8, characterized in that: Step S9 includes the following steps: S9.1 Based on the central feature points obtained in step S8, the line trajectory curves of two independent laser lines formed on the left and right sides of the weld are constructed using a fitting method. S9.2 Calculate the fitting function for the two line trajectory curves respectively; S9.

3. Based on the obtained fitting function, calculate the coordinates of the intersection point of the two line trajectory curves. The coordinates of the intersection point are the characteristic points of the weld.

10. A metal surface weld feature localization system based on polarization imaging, used to implement the method as described in any one of claims 1 to 9, characterized in that: include: Shell (1); A laser emission window (9) and a polarized light incident window (10) are provided on the front side of the housing (1); A laser emitting module (6) is used to generate a line-structured laser beam; The rotating base (2) is installed on the bottom plate of the housing (1), and the laser emitting module (6) is installed on the rotating base (2). The spatial projection angle of the laser emitting module (6) can be changed by adjusting the rotating base (2). A polarization camera (5) is mounted on the base plate of the housing (1); The polarization modulation module (7) installed in the housing (1) has its optical axis coincident with the optical axis of the laser emission module (6). It is used to polarize the line structure laser beam generated by the laser emission module (6) to generate a line structure laser beam with a specific polarization state. The line structure laser beam with a specific polarization state passes through the laser emission window (9) and is projected onto the metal surface to be tested. The filter module (3) is installed inside the housing (1); The imaging lens (4) installed in the housing (1) has the optical axes of the filter module (3) and the imaging lens (4) coincident with the optical axis of the polarization camera (5). The reflected laser line with a specific polarization state reflected from the metal surface to be tested passes through the polarization light incident window (10), and then passes through the filter module (3) and the imaging lens (4) in sequence, and is focused onto the polarization camera (5). The polarization camera (5) generates a polarization information image. The image acquisition and processing module (8) is connected to the polarization camera (5) and is used to acquire polarization information images and execute image processing algorithms to extract real laser lines and identify weld positions.

Images