Carbon fiber separation positioning method based on machine vision

Abstract

The invention provides a machine vision-based carbon fiber separation positioning method, which comprises the following steps of: controlling the angle and the wavelength of a light source according to obtained interface feature information of a carbon fiber material, and performing time sequence illumination on the carbon fiber material to obtain carbon fiber material images under different illumination conditions; processing the images of the carbon fiber material under different illumination conditions, and separating the images in ultraviolet, visible and near-infrared wavebands through a spectrum splitting method to obtain three-spectrum coaxial acquisition images; performing image segmentation processing on the three-spectrum coaxial acquisition image, extracting contour edge information of the carbon fiber material, and determining a target cutting path of the carbon fiber material in combination with spatial distribution information of the carbon fiber material; on a target cutting path, through multi-sensor cooperative calibration, carbon fiber material information collected by different sensors is subjected to spatial registration and fusion, and a spatial mapping relation of carbon fiber material separation and positioning is established.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a carbon fiber separation and positioning method based on machine vision. Background Art

[0002] The recycling and sorting of carbon fiber composites is extremely important for solving the global problem of the difficulty of efficient reuse of high-performance materials after they are discarded. At the same time, it has irreplaceable value in reducing environmental load and improving industrial economic benefits. With the rapid growth of demand for carbon fiber in industries such as aerospace and automobile manufacturing, how to effectively recycle and reuse these composite materials has become a technical bottleneck that needs to be broken through. In the process of recycling and sorting carbon fiber in a composite environment, there is a technical problem that it is difficult to extract the interface features of carbon fiber materials. The existing recycling methods mostly rely on mechanical crushing or chemical decomposition. Although they can achieve material separation to a certain extent, they generally have problems such as low sorting accuracy, high energy consumption, and serious secondary pollution. Especially for composite materials with complex structures, traditional solutions are difficult to accurately identify and extract the spatial distribution and interface characteristics of carbon fibers, resulting in limited recycling efficiency and material quality. Due to the complex and changeable surface morphology of carbon fiber materials in composite environments and obvious anisotropy, traditional single light source illumination solutions are difficult to fully and accurately extract the surface morphology characteristics of carbon fiber materials. In addition, the color, glossiness, and transparency of carbon fiber materials themselves vary greatly, which brings great challenges to image processing and feature extraction. Therefore, how to build a carbon fiber recycling and sorting system that integrates multi-spectral visual detection, composite lighting solutions and multi-sensor collaborative calibration in a complex environment to achieve precise positioning of the spatial distribution of carbon fibers, efficient extraction of interface features and accurate planning of separation paths has become a key issue that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a carbon fiber separation and positioning method based on machine vision, which mainly includes:

[0004] The image data of the carbon fiber material in the ultraviolet, visible and near-infrared spectral bands are obtained through a multi-spectral visual detection unit, and the distribution position of the carbon fiber material in three-dimensional space is determined according to the image data to obtain the spatial distribution information of the carbon fiber material;

[0005] According to the spatial distribution information of carbon fiber materials, the carbon fiber materials are illuminated by using the composite illumination of orthogonal polarized light and structured light, the angle and intensity of polarized light and structured light are controlled, and the interface characteristic information of carbon fiber materials is obtained;

[0006] According to the obtained interface characteristic information of carbon fiber materials, the angle and wavelength of the light source are controlled to perform sequential illumination on the carbon fiber materials to obtain images of carbon fiber materials under different illumination conditions;

[0007] Process the images of carbon fiber materials under different lighting conditions, separate the images in the ultraviolet, visible, and near-infrared bands through spectral spectroscopy to obtain three-spectrum coaxial acquisition images;

[0008] Perform image segmentation processing on the three-spectrum coaxial acquisition images, extract the contour edge information of the carbon fiber materials, and combine the spatial distribution information of the carbon fiber materials to determine the target cutting path of the carbon fiber materials;

[0009] Through multi-sensor collaborative calibration on the target cutting path, spatially register and fuse the carbon fiber material information collected by different sensors to establish a spatial mapping relationship for carbon fiber material separation and positioning;

[0010] According to the established spatial mapping relationship for carbon fiber material separation and positioning, combined with the target cutting path, use machine vision to guide the robotic arm to cut and separate the carbon fiber materials, and classify and recycle the separated carbon fiber materials.

[0011] Furthermore, image data of the carbon fiber material in three spectral bands of ultraviolet, visible, and near-infrared are obtained through a multispectral vision detection unit. The distribution position of the carbon fiber material in three-dimensional space is determined based on the image data to obtain the spatial distribution information of the carbon fiber material, including: obtaining the reflected spectral image data of the carbon fiber surface in the ultraviolet band of 360 nm to 400 nm, the visible light band of 450 nm to 700 nm, and the near-infrared band of 750 nm to 1400 nm through a multispectral scanner at a preset sampling interval, performing Gaussian filtering and noise reduction processing on the obtained spectral images, and removing noise points according to a preset threshold to obtain spectral image data. A calibration board is arranged around the carbon fiber material, and the corner position coordinates of the checkerboard on the calibration board are extracted using a corner detection algorithm. A mapping matrix between the imaging coordinate system of the multispectral camera and the world coordinate system is established based on the corner position coordinates. The spectral image data is processed by band, the edge line segments of each band image are extracted using the Hough transform method, the initial feature point set is obtained through intersection calculation, and the first feature point set is obtained by screening the feature points according to a preset gray threshold. The region growing algorithm is used to perform region segmentation with the points in the first feature point set as the growth starting points, the contour region of the carbon fiber material is obtained, and the contour boundary points are extracted to form the second feature point set. Three-dimensional coordinate transformation is performed on the second feature point set according to the mapping matrix to obtain the world coordinates of the surface feature points of the carbon fiber material. The spectral feature matching algorithm based on similarity is used to register the feature points in the three bands. By calculating the similarity scores of the gray values and spatial positions of the feature points, the feature point pairs with the highest similarity scores are selected to establish corresponding relationships. Using the registered feature points as control points, the surface of the carbon fiber material is meshed using the cubic spline interpolation algorithm to generate high-density three-dimensional point cloud data. The triangulation algorithm is used to construct an initial mesh model for the point cloud data, and the mesh is optimized through Laplacian smoothing operation to obtain the three-dimensional space distribution model of the carbon fiber material.

[0012] Further, according to the spatial distribution information of the carbon fiber material, the carbon fiber material is illuminated by a combined illumination method of orthogonal polarized light and structured light. The angles and intensities of the polarized light and the structured light are controlled to obtain the interface characteristic information of the carbon fiber material, including: setting the illumination position of the orthogonal polarized light source according to the spatial distribution range of the carbon fiber material, collecting the reflected light intensity signal on the surface of the carbon fiber by a light intensity sensor, adjusting the polarization angle of the polarizer by a closed-loop feedback controller, and obtaining the polarization compensation parameter by comparing the reflected light intensity difference between the two polarization directions. Adjusting the illumination direction of the orthogonal polarized light source with the polarization compensation parameter, obtaining the compensated reflected light intensity distribution through the light intensity sensor, and performing Gaussian filtering on the reflected light intensity signal to obtain the first light intensity image. Generating a structured light fringe pattern by a digital light source controller, calculating the fringe projection density parameter according to the first light intensity image, obtaining the structured light fringe reflection signal by a light intensity sensor, and calculating the fringe deformation parameter by Gaussian fitting. Performing timing control on the orthogonal polarized light and the structured light, generating an alternating illumination sequence by a light source synchronization trigger, collecting the reflected images under the two light sources by a light intensity sensor, and performing denoising filtering on the reflected images to obtain the second light intensity image. Extracting the surface brightness distribution of the carbon fiber according to the second light intensity image, calculating the image gradient by using the Sobel operator to obtain the surface texture feature image. Geometrically correcting the surface texture feature image with the fringe deformation parameter, calculating the fringe distortion amount by a phase demodulation algorithm, and obtaining the surface topography feature image of the carbon fiber material. Fusing the surface texture feature image and the surface topography feature image, and obtaining the interface characteristic information of the carbon fiber material by a weighted superposition method.

[0013] Further, according to the obtained interface characteristic information of the carbon fiber material, control the angle and wavelength of the light source, perform sequential illumination on the carbon fiber material, and obtain images of the carbon fiber material under different illumination conditions, including: according to the texture distribution and morphological characteristics in the interface characteristic information of the carbon fiber material, adjust the illumination angle of the light source through an electric angle controller, use a light intensity sensor to collect the surface reflected light intensity distribution in real time, and obtain the first angle parameter by comparing the reflected light intensity differences at different angles. Use a spectrally tunable filter to scan the wavelength of the light source, select multiple wavelength points according to the surface reflection characteristics of the carbon fiber material, generate a wavelength modulation signal through a wavelength controller, and obtain the first wavelength parameter by using a spectrometer to collect the spectral response data at different wavelengths. Generate an illumination matrix according to the first angle parameter and the first wavelength parameter, generate an illumination pulse sequence through a multi-channel waveform generator, and obtain the timing control parameter by using a timing controller to set the light source switching time interval. Use a light source driver to perform angle and wavelength combined illumination according to the timing control parameter, generate an image acquisition trigger signal through an optoelectronic synchronizer, and obtain an illumination sequence control signal. Drive a high-speed image acquisition device according to the illumination sequence control signal to obtain a sequence of reflected images of the carbon fiber material, and perform denoising processing on the image sequence through a median filter to obtain the first image sequence. Use an image sharpening algorithm to enhance the edges of the first image sequence, and perform gray histogram equalization processing to obtain the second image sequence. Screen the second image sequence according to the signal-to-noise ratio and contrast threshold, and use a brightness correction algorithm to compensate the image brightness distribution to obtain images of the carbon fiber material under different illumination conditions.

[0014] Further, process the carbon fiber material images under different illumination conditions, separate the images in three bands of ultraviolet, visible, and near-infrared through spectral splitting method to obtain three-spectrum coaxial acquisition images, including: calculate the band light intensity distribution according to the carbon fiber material images under different illumination conditions, extract the spectral components in the ultraviolet band of 360 nm to 400 nm, the visible band of 450 nm to 700 nm, and the near-infrared band of 750 nm to 1400 nm through a spectral band-pass filter, and perform Gaussian denoising on the filtered image to obtain the first-band image. Obtain the spectral transmittance curve of a standard whiteboard using a spectrophotometer as a reference standard, perform pixel-by-pixel transmittance compensation on the first-band image according to the reference standard, and obtain the second-band image through brightness normalization processing. Calculate the crosstalk coefficient matrix between bands according to the second-band image, use the non-negative matrix factorization method to separate each band image, and extract the pure spectral components of the band through iterative operation to obtain the third-band image. Perform edge detection on the third-band image, use the Harris corner detection algorithm to extract the set of feature points, and calculate the corresponding relationship of the feature points through the least squares method to obtain the spatial transformation parameters. Establish an image coordinate mapping relationship according to the spatial transformation parameters, perform resampling on the three-band images using the bicubic interpolation algorithm, and obtain the fourth-band image through the sub-pixel registration method. Perform spectral contrast enhancement on the fourth-band image, adjust the gray distribution of the three-band images using the histogram matching method, and obtain the fifth-band image through the superposition of the band images. Perform spectral quality evaluation according to the fifth-band image, quantify the image quality using the signal-to-noise ratio and spectral contrast indicators, and obtain the three-spectrum coaxial acquisition images through quality threshold screening.

[0015] Furthermore, perform image segmentation on the three-spectrum coaxial acquisition image, extract the contour edge information of the carbon fiber material, and combine the spatial distribution information of the carbon fiber material to determine the target cutting path of the carbon fiber material, including: processing the three-spectrum coaxial acquisition image with a multi-scale edge detection operator, decomposing the image into three layers through a Gaussian pyramid, extracting edge candidate points according to the image gradient magnitude and direction information, and screening the candidate points using the double-threshold percentage method to obtain the first contour image. Perform edge tracking on the first contour image, connect the disconnected edges using the eight-neighborhood search method, and perform region growing through the gray similarity criterion to obtain the second contour image. Extract the contour feature point sequence from the second contour image, calculate the curvature of the feature points using the curve fitting method, and obtain the key control points through the extraction of extreme points to get the third contour image. Perform surface mapping on the third contour image, calculate the depth mapping matrix according to the spatial distribution information of the carbon fiber material, and correct the contour edge using the depth compensation method to obtain the fourth contour image. Construct a three-dimensional mesh model based on the fourth contour image, locally encrypt the surface using the mesh subdivision algorithm, and determine the key contour points through curvature calculation to obtain the fifth contour image. Segment the path of the fifth contour image, calculate the path direction according to the surface tangent vector, and connect the segmented paths using the path smoothing algorithm to obtain the first cutting path. Constrain the first cutting path according to the thickness distribution of the carbon fiber material, adjust the cutting depth and angle using the path optimization algorithm, and correct the path using the curvature constraint method to obtain the target cutting path.

[0016] Further, through multi-sensor collaborative calibration on the target cutting path, the carbon fiber material information collected by different sensors is spatially registered and fused to establish a spatial mapping relationship for carbon fiber material separation and positioning, including: arranging a calibration board on the target cutting path, collecting calibration board images through a vision sensor and a laser sensor, extracting calibration feature points using the checkerboard corner detection algorithm, and calculating the sensor focal length, principal point coordinates, and distortion coefficients based on the feature point positions to obtain spatial calibration parameters. Using the target cutting path as a reference, synchronously collect a sequence of carbon fiber material images through multiple sensors, perform Gaussian filtering denoising and histogram equalization enhancement on the image sequence, and obtain the sensor acquisition sequence parameters based on the image frame interval and trigger delay. Calculate the relative position relationship between sensors according to the sensor acquisition sequence parameters, unify the sensor coordinate systems using a spatial coordinate transformation matrix, and optimize the coordinate transformation parameters through the least squares method. Use a matching algorithm based on scale-invariant features to register multi-sensor images, extract corner and edge features in the images, and calculate the feature correspondence relationship through feature descriptors. Calculate the overlapping area of sensor data according to the feature correspondence relationship, perform registration optimization on the overlapping area using the iterative closest point algorithm, and obtain the spatial registration parameters through error feedback. Perform spatial mapping transformation on the registered sensor data, use a three-dimensional mesh reconstruction method to construct the spatial structure of the carbon fiber material, and obtain the material separation and positioning parameters through surface fitting. Establish a spatial mapping relationship for the carbon fiber material according to the material separation and positioning parameters, and perform spatial mapping on the cutting path using a coordinate transformation equation to obtain the spatial mapping relationship for carbon fiber material separation and positioning.

[0017] Further, according to the established spatial mapping relationship for the separation and positioning of carbon fiber materials, in combination with the target cutting path, the robotic arm is guided by machine vision to cut and separate the carbon fiber materials, and the separated carbon fiber materials are classified and recycled, including: establishing a robotic arm coordinate system according to the spatial mapping relationship for the separation and positioning of carbon fiber materials, collecting real-time images of the working space through a vision sensor, extracting a sequence of target feature points using a target tracking method based on template matching, calculating the deviation between the feature points and the planned path to obtain the first position parameter. Calculating the compensation amount of each joint of the robotic arm according to the first position parameter, generating a sequence of joint angles using the quintic polynomial interpolation method, and calculating the speed curve of each joint through a speed planner to obtain the first motion parameter. Controlling the robotic arm to move to the specified position according to the first motion parameter, calculating the deviation between the end effector and the target position using a real-time tracking method based on visual servo, and performing trajectory correction through a proportional-integral controller to obtain the second motion parameter. Adjusting the attitude angle of the cutting tool according to the second motion parameter, collecting the contact force and torque signals during the cutting process in real time through a torque sensor, and adjusting the cutting force through an adaptive impedance controller to obtain the first cutting parameter. Performing the carbon fiber material separation operation according to the first cutting parameter, detecting the contour features of the cutting area in real time through an image processor, and calculating the deviation between the actual cutting trajectory and the target trajectory through an edge extraction algorithm to obtain the second cutting parameter. Controlling the robotic arm to perform local correction on the cutting trajectory using the second cutting parameter, judging the separation integrity according to the visual feedback information, and extracting the shape features of the separated material through an image segmentation algorithm to obtain the separation feature parameter. Classifying the materials according to the separation feature parameter, extracting the surface texture and shape features of the materials using a feature extraction network based on deep learning, and classifying the materials through a support vector machine to obtain the material category parameter. Controlling the classification and recycling device to perform classification actions according to the material category parameter, transporting different types of materials in layers using a multi-layer conveyor belt, and completing the classification and recycling by detecting the material position through a photoelectric sensor.

[0018] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0019] The present invention discloses a carbon fiber separation and positioning method based on machine vision. The spatial distribution information of carbon fiber materials is obtained through multi-spectral vision detection. The interface features are obtained by using composite illumination of orthogonal polarized light and structured light, and images under different illumination conditions are obtained through sequential illumination. The three-spectral coaxial acquisition images are obtained by using spectral splitting technology. The contour edge information is extracted to determine the cutting path, and the spatial mapping relationship for separation and positioning is established, ultimately realizing precise cutting, separation, classification, and recycling under machine vision guidance. The present invention integrates technologies such as multi-spectral imaging, composite illumination, image processing, and multi-sensor fusion, and can efficiently and accurately realize the intelligent separation and recycling of carbon fiber materials, improving the recycling efficiency and quality of carbon fiber materials, and having important economic and environmental benefits. Brief Description of the Drawings

[0020] Figure 1 It is a flowchart of a carbon fiber separation and positioning method based on machine vision according to the present invention.

[0021] Figure 2 It is a schematic diagram of a carbon fiber separation and positioning method based on machine vision according to the present invention. Detailed Embodiments

[0022] To further understand the content of the present invention, the present invention will be described in detail in combination with the drawings and embodiments. The present application will be further described in detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. In addition, it should be noted that only the parts related to the invention are shown in the drawings for the convenience of description.

[0023] As Figure 1-2 , a carbon fiber separation and positioning method based on machine vision in this embodiment may specifically include:

[0024] S101. Collect image data of carbon fiber materials in the ultraviolet, visible, and near-infrared bands through a multispectral imaging unit, and determine the three-dimensional spatial distribution characteristics of the carbon fiber materials according to the image data.

[0025] The built-in separation and positioning function can be triggered through system settings or application programs. The specific triggering method is determined according to the actual scenario. For example, the user can start the function through interface operations, or remote control can be achieved by sending instructions from external devices. After the function is started, first use the multispectral imaging unit to collect image data of carbon fiber materials.

[0026] S1011. Control the multispectral scanner to collect the reflection spectrum data of the carbon fiber surface according to the preset band range, and generate initial image data.

[0027] The multispectral scanner covers the ultraviolet band from 360 nm to 400 nm, the visible band from 450 nm to 700 nm, and the near-infrared band from 750 nm to 1400 nm. The sampling interval is set to 5 nm to ensure data resolution. During the collection process, the band selection is dynamically adjusted according to the material reflection characteristics. The ultraviolet band highlights surface defects, the visible band reflects morphological features, and the near-infrared band reveals internal structures. The initial image data is processed by Gaussian filtering, with a filtering window of 5x5 and a standard deviation set to 1.2 to reduce the influence of spectral noise.

[0028] S1012. Establish a coordinate system mapping relationship based on the calibration plate image, and perform feature extraction on the initial image data to generate a three-dimensional spatial distribution model of the carbon fiber material.

[0029] Arrange a calibration board around the carbon fiber material, using a 9x9 checkerboard pattern with a grid point spacing of 10 mm. Use the corner detection algorithm to extract the corner coordinates of the calibration board with an accuracy better than 0.1 pixel, and then construct the mapping matrix between the multi-spectral camera imaging coordinate system and the world coordinate system. Separate the bands of the initial image data, use the Hough transform to extract edge line segments, set the inclination angle threshold to 5 degrees, and the line segment length threshold to 20 pixels. Through the region growing algorithm, starting from points with a gray level gradient greater than 30, with a neighborhood radius of 3 pixels, segment the carbon fiber contour region and generate a set of feature points.

[0030] S1013. Perform spectral registration and 3D modeling on the set of feature points to generate a high-precision spatial distribution model.

[0031] For the set of feature points, use a similarity-based spectral matching algorithm to calculate the similarity by comprehensively considering the gray value and spatial position, with weights of 0.6 and 0.4 respectively, and set the threshold to 0.8. After registration, select feature point pairs with a spatial deviation less than 2 mm, generate high-density point cloud data through cubic spline interpolation. The point cloud data is triangulated to construct an initial mesh model, which is optimized by Laplacian smoothing, iterated 3 times, with the displacement controlled within 0.5 mm, and the mesh side length ranging from 2 mm to 8 mm. For wrinkled or warped areas, the feature point sampling is encrypted, and the spacing is reduced to 0.5 mm to ensure that the model accuracy is better than 1 mm. The finally generated 3D model can clearly reflect the spatial geometric characteristics of the carbon fiber, providing a reliable basis for separation.

[0032] By combining multi-spectral imaging and coordinate system mapping, the spatial distribution information of carbon fiber can be efficiently obtained. The optimized design of feature extraction and 3D modeling significantly improves the model accuracy and reduces noise interference.

[0033] S102. Use a composite illumination method of orthogonal polarized light and structured light to irradiate the carbon fiber material, dynamically adjust the light source parameters to collect interface feature data, and generate the texture and morphology information of the carbon fiber surface through image processing.

[0034] After obtaining the three-dimensional spatial distribution characteristics of the carbon fiber material, immediately start the composite illumination system, using orthogonal polarized light and structured light as light sources to extract features for the complex morphology of the material surface. The composite illumination can effectively cope with the anisotropic characteristics of carbon fiber, ensuring the comprehensive capture of texture and morphology information. By precisely controlling the light source parameters and image processing algorithms, high-precision interface feature data is generated, providing a reliable basis for subsequent separation path planning.

[0035] S1021. Configure an orthogonal polarized light source, collect the reflected light intensity signal on the carbon fiber surface, and optimize the polarization angle through a closed-loop feedback mechanism to generate the initial light intensity distribution.

[0036] An orthogonal polarization light source with an included angle of 90 degrees between two polarizers is adopted, and the incident light direction is dynamically adjusted according to the spatial distribution characteristics of carbon fibers. When the polarization direction of the incident light is parallel to the fiber orientation of the carbon fiber, the reflected light intensity reaches the peak value, otherwise it drops to the minimum. The reflected signal is collected by a light intensity sensor at a sampling rate of 1000 Hz, and the signal is digitally processed by a 10-bit analog-to-digital converter. The closed-loop feedback controller calculates the polarization compensation parameter according to the light intensity difference between the two polarization directions, and the angle adjustment accuracy is better than 0.1 degree. The compensated light intensity signal is processed by Gaussian filtering, and the filtering window is 5x5, generating an initial light intensity distribution image, and the signal-to-noise ratio is increased to more than 45 dB, laying a foundation for feature extraction.

[0037] S1022. Enable structured light illumination, generate fringe patterns and extract deformation parameters to capture the geometric morphology characteristics of the carbon fiber surface.

[0038] A sinusoidal fringe pattern is projected through a digital light source controller, and the fringe spatial frequency is adaptively adjusted according to the surface curvature of the carbon fiber. The period in the central region is 2 mm, and the period in the high-curvature region with a curvature change rate greater than 0.5 / mm is reduced to 0.2 mm. The light intensity sensor collects the fringe reflection signal, and a Gaussian fitting algorithm is used to calculate the fringe deformation parameters with a fitting window of 7 pixels, and the fitting error is controlled within 0.1 pixel. The fringe deformation parameters reflect the surface height change, providing key data for the extraction of morphology features. A 100 Hz square wave signal is generated by a dual-channel waveform generator to control the alternating illumination of orthogonal polarized light and structured light, and the exposure time is set to 4 ms to ensure sufficient dynamic range of the collected images.

[0039] S1023. Denoise and geometrically correct the reflected image under composite illumination to generate a texture feature image of the carbon fiber surface.

[0040] The reflected image under the alternating illumination of orthogonal polarized light and structured light is processed. A 5x5 mean filtering window is used to remove the ambient light interference, and the gray-scale resolution of the filtered image remains 8 bits. Based on the initial light intensity distribution, the Sobel operator is used to calculate the image gradient, and the window size is 3x3 pixels. The pixel points with a gradient amplitude greater than 25 are marked as texture edges. The set of edge points reflects the fiber weaving direction and local defect characteristics of the carbon fiber surface. Geometric correction is performed according to the fringe deformation parameters, and the image is resampled using the bilinear interpolation algorithm with a grid spacing of 0.1 mm to ensure that the correction accuracy is better than 0.2 mm. The corrected image clearly shows the surface texture details, providing high-quality input for morphology analysis.

[0041] S1024. Extract morphology features through a phase demodulation algorithm and fuse texture information to generate complete interface feature data.

[0042] The four-step phase-shift method is applied to the corrected image for phase demodulation. The phase-shift step is π / 2, and the quantization accuracy of the phase value reaches π / 100. The fringe distortion obtained by demodulation is used to calculate the surface height distribution, generate the morphological feature image of carbon fiber, and the reconstruction accuracy is better than 0.1 mm. The texture feature image and the morphological feature image are fused using the weighted superposition method, with a texture weight of 0.3 and a morphology weight of 0.7. For areas with dense fiber weaving or obvious defects, the texture weight is dynamically increased to 0.5 to highlight details. The fused interface feature data fully reflects the microscopic structure of the carbon fiber surface, including characteristics such as fiber orientation, surface undulation, and local wrinkles, and the boundary positioning accuracy is better than 0.2 mm.

[0043] The orthogonal polarized light and structured light of the compound illumination system cooperate with each other. The former enhances the contrast of texture features, and the latter accurately captures morphological changes. The application of the closed-loop feedback and Gaussian fitting algorithms significantly improves the quality of the light intensity signal and reduces the interference of ambient light and material anisotropy. The fusion of texture and morphology information further enriches the expression ability of interface features and provides solid data support for the precise separation of carbon fibers.

[0044] S103. Collect the reflection image sequences under multiple conditions and enhance the images to generate high-definition carbon fiber surface images.

[0045] Based on the obtained carbon fiber interface feature data, start the sequential illumination system. By precisely controlling the angle and wavelength of the light source, diverse illumination conditions are generated to capture the subtle changes on the carbon fiber surface. Sequential illumination can highlight the texture layers and morphological details of the material surface and provide high-quality input for image segmentation and path planning.

[0046] S1031. Use an electric angle controller to adjust the light source irradiation angle, collect the reflected light intensity distribution, and generate optimized angle parameters.

[0047] Configure an electric angle controller with a scanning range covering 0 degrees to 75 degrees, a stepping interval set to 15 degrees, and a 100-ms pause at each angle to collect stable signals. The light intensity sensor records the reflected light intensity on the carbon fiber surface at a sampling rate of 1000 Hz. The signal is digitized by a 12-bit analog-to-digital converter with sufficient dynamic range. Analyze the light intensity difference at different angles. When the angle between the light source and the fiber orientation is close to 45 degrees, the reflection intensity reaches the peak, and this angle is recorded as the reference. The optimized angle parameters reflect the response characteristics of the surface morphology to light and provide a reference basis for wavelength selection.

[0048] S1032. Scan the wavelength through a spectrally tunable filter, collect the spectral response data, and generate wavelength parameters to optimize the illumination conditions.

[0049] A spectrally tunable filter is adopted, with a wavelength tuning range of 380 nm to 780 nm and a step interval of 50 nm. Continuous wavelength switching is achieved by liquid crystal modulation. The spectrometer acquires the spectral response of the carbon fiber surface at a resolution of 2 nm, focusing on recording the characteristic peaks at 450 nm, 550 nm, and 650 nm. These wavelengths can highlight the fabric structure and surface defects of the material. The wavelength selection is dynamically adjusted according to the light intensity distribution to generate wavelength parameters, ensuring a high degree of matching between the illumination conditions and the interface characteristics. The measurement repeatability error is controlled within 1%, providing a reliable spectral basis for image acquisition.

[0050] S1033. Generate an illumination matrix based on the angle and wavelength parameters, drive sequential illumination, and acquire a sequence of reflected images.

[0051] Combining the angle and wavelength parameters, an illumination matrix containing 5 angles and 3 wavelengths is constructed, forming 15 illumination combinations. A multi-channel waveform generator generates 100 Hz square wave pulses, with a 60-degree phase difference between adjacent channels and a switching interval of 10 ms to ensure the stability of the light source. An optoelectronic synchronizer drives a high-speed image acquisition device with a trigger accuracy of 1 μs, a resolution of 2048x2048 pixels, a frame rate of 200 frames per second, and a single-frame exposure of 2 ms. The acquired sequence of reflected images covers the surface details under different illumination conditions, and the image data for each illumination condition is ensured to be completely recorded through sequential control.

[0052] S1034. Denoise and edge enhance the sequence of reflected images to generate a sequence of high-definition images of the carbon fiber surface.

[0053] Apply a 5x5 median filter to the image sequence to effectively suppress salt-and-pepper noise and maintain the image gray-scale resolution above 8 bits. Subsequently, use the Laplacian operator for edge enhancement, with the enhancement coefficient set to 1.5 to highlight the fiber texture and morphology boundaries. Further, through gray-scale histogram equalization processing, the gray levels are mapped to 256 levels to improve the image contrast. For uneven brightness, the image is divided into 8x8 sub-blocks, the average brightness of each sub-block is calculated, and a brightness compensation coefficient is generated by bilinear interpolation. The selected image sequence needs to meet the thresholds of a signal-to-noise ratio higher than 35 dB and a contrast greater than 0.4. The generated sequence of high-definition images clearly presents the fabric structure and microscopic defects on the carbon fiber surface.

[0054] Sequential illumination, through the combination of multiple angles and wavelengths, significantly enhances the visualization effect of the carbon fiber surface features. For example, the combination of a 30-degree angle and a 550-nm wavelength can highlight the fiber weaving texture, while a 60-degree angle enhances the shadow effect of surface irregularities. The synergistic application of denoising and enhancement processing effectively improves the image quality and reduces the influence of ambient light interference. Through dynamic parameter optimization and efficient image processing, it is ensured that the generated image sequence can provide accurate data support for separation and positioning.

[0055] S104. Decompose the image into components in the ultraviolet, visible, and near-infrared bands through spectral separation and image processing techniques, perform feature enhancement and registration, and generate a three-spectral coaxial acquisition image.

[0056] Use spectral splitting technology to perform multi-band decomposition on the collected carbon fiber image, and perform denoising, compensation, and registration processing for the characteristics of each band to extract multi-level features on the material surface. Spectral separation can effectively distinguish texture and topography information at different depths, while image registration and enhancement techniques ensure spatial consistency and feature clarity across bands. This multi-band processing method provides comprehensive data support for the precise positioning and separation of carbon fibers, significantly improving the material recognition ability during the recycling process.

[0057] S1041. Separate the image bands through a spectral band-pass filter and generate an initial band image using Gaussian denoising.

[0058] Configure a spectral band-pass filter with center wavelengths set at 380 nm, 550 nm, and 850 nm respectively, a full width at half maximum of 40 nm, and a Gaussian distribution of the filter transmittance to ensure pure extraction of band components. For the ultraviolet band from 360 nm to 400 nm, the visible band from 450 nm to 700 nm, and the near-infrared band from 750 nm to 1400 nm, perform spectral decomposition on the image to generate the initial components of each band. Subsequently, apply a 5x5 Gaussian kernel for denoising with a standard deviation set at 1.2 to effectively suppress random noise, and increase the signal-to-noise ratio of the generated initial band image to above 35 dB, providing high-quality input for compensation processing.

[0059] S1042. Calibrate the spectral transmittance using a standard whiteboard, compensate and normalize the band image, and generate an optimized band image.

[0060] Adopt an alumina standard whiteboard with a reflectance of over 95% in the range of 300 nm to 1500 nm as the calibration benchmark for spectral transmittance. Measure the transmittance curve of the whiteboard with a spectrophotometer, calculate the correction coefficient pixel by pixel according to the Lambert-Beer law, and perform transmittance compensation on the initial band image. After compensation, the transmittance flatness of the three bands reaches 98%, and the difference between bands is reduced to within 2%. Further perform brightness normalization on the image, map the gray values to a unified dynamic range, and generate an optimized band image to ensure consistent brightness distribution across bands for subsequent feature analysis.

[0061] S1043. Separate the cross-talk between bands through non-negative matrix factorization, and perform edge detection and feature extraction on the image.

[0062] For the optimized band images, the 3x3-dimensional inter-band crosstalk coefficient matrix is ​​calculated to quantify the energy coupling between the ultraviolet, visible light and near-infrared bands. The non-negative matrix decomposition method is used, and the alternating least squares method is used for iterative optimization. The number of iterations is set to 50, and the convergence threshold is 0.001. Finally, the crosstalk coefficient is reduced to below 0.05 to generate a pure band image. Edge detection is performed on these images, and the Harris corner detection algorithm is used. The response threshold is set to 0.01, combined with a 5x5 pixel non-maximum suppression window, and about 200 feature points are extracted in each band. The feature point set clearly reflects the fiber weaving direction and the location of surface defects, providing a key reference for registration.

[0063] Through bicubic interpolation and sub-pixel registration technology, the band images are spatially corrected and high-precision fused images are generated. Based on the feature point set, the least squares method is used to calculate the spatial transformation parameters, the residual threshold is set to 2 pixels, and the point pairs with large matching errors are eliminated. The bicubic interpolation algorithm calculates the grayscale value with a 4x4 pixel window, and the shape parameter is -0.5 to ensure the smoothness of the interpolation. The registration accuracy is better than 0.1 pixel, and the spatial deviation between bands is controlled within 1 pixel. The grayscale distribution is further adjusted by histogram matching, and the dynamic range of the three bands is consistent by piecewise linear mapping. When fused, the weight ratio of ultraviolet, visible light, and near-infrared bands is 1:2:1, and the dynamic range of the generated three-spectrum coaxial acquisition image is extended to 12 bits, and the spatial resolution is better than 5 microns. The ultraviolet band highlights tiny surface defects, the visible light band shows the fiber weaving texture, and the near-infrared band reveals deep structural information, providing multi-dimensional feature support for carbon fiber separation and positioning.

[0064] S1044. Perform quality assessment on the fused image to ensure the fidelity and consistency of the three-spectrum image.

[0065] The fused images were evaluated by multiple indicators, with the signal-to-noise ratio threshold set to 30 decibels and the spectral contrast threshold set to 0.4. The clarity of the fiber texture and the separation of the band features were quantified by calculating the grayscale gradient and spectral response distribution of the image. Images below the threshold were eliminated, and the retained images were used to generate three-spectrum coaxial acquisition images by band superposition. This strict quality control ensures the excellent performance of the image in terms of spectral fidelity and spatial consistency, providing a reliable data basis for separation path planning.

[0066] The application of multi-band spectral separation technology significantly enhances the ability to express the surface features of carbon fiber. The ultraviolet band is sensitive to surface scratches and microcracks, the visible light band clearly presents the fiber interweaving structure, and the near-infrared band can penetrate shallow materials and reveal internal texture changes. Through the systematic processing of denoising, compensation and registration, the interference of crosstalk between bands and uneven illumination is overcome. The generated image not only retains multi-level features, but also achieves spatial unification across bands, laying a technical foundation for accurate separation.

[0067] S105. Extract the contour edge information of the carbon fiber material through multi-scale image segmentation and spatial mapping processing, and generate an accurate target cutting path in combination with the spatial distribution characteristics.

[0068] Based on the trispectral coaxial acquisition image, adopt multi-scale edge detection and region growing techniques to gradually extract the contour features of the carbon fiber material, ensuring the integrity and accuracy of the edge information. By combining the spatial distribution data, perform three-dimensional mapping and depth correction on the contour to generate a cutting path that conforms to the geometric characteristics and processing requirements of the material. This method can effectively handle the complex undulations and local defects on the carbon fiber surface, providing reliable guidance for efficient separation and recycling.

[0069] S1051. Decompose the image through Gaussian pyramid, extract edge candidate points by gradient detection, and generate an initial contour image.

[0070] Construct a Gaussian pyramid for the trispectral coaxial acquisition image, decompose it into three layers, with the size of each layer of the image halved. Apply Gaussian kernels with standard deviations of 1.0, 2.0, and 4.0 for smoothing respectively to capture edge features at different scales. Calculate the gradient magnitude and direction of each layer of the image, and adopt a double-threshold screening method. The high threshold is set to 75% of the maximum gradient value, and the low threshold is 40% of the high threshold to screen out reliable edge candidate points. These candidate points are fused through multi-scale to generate an initial contour image, retaining the edge details of the fiber texture and defect areas, providing a basis for tracking.

[0071] S1052. Perform edge tracking and region growing on the initial contour image to generate a coherent contour feature image.

[0072] Adopt an eight-neighborhood search algorithm to perform edge tracking on the initial contour image, with a search window of 3x3 pixels. When the gray difference between adjacent pixels is less than 20 and the gradient direction difference is less than 30 degrees, it is determined as a continuation point of the same edge, effectively connecting discontinuous edges. Further, through the region growing algorithm, based on a gray similarity threshold of 0.85, expand the edge region and stop when the growing area exceeds 1000 pixels to avoid over-expansion. The generated contour feature image clearly outlines the fiber weaving boundary and local irregular areas on the carbon fiber surface, laying a foundation for feature point extraction.

[0073] S1053. Extract the key points of the contour and perform surface mapping to generate a three-dimensionally corrected contour image.

[0074] Extract a sequence of feature points from the contour feature image. Calculate the local curvature using a 7-pixel window, and mark the points with a curvature value greater than 0.1 as key control points. The point spacing is maintained between 10 and 30 pixels to ensure description accuracy and conciseness. Perform curve fitting based on the control points, and use the extreme point extraction method to streamline the feature point set to generate a smoother contour image. Subsequently, combine the spatial distribution information of carbon fibers to construct a depth mapping matrix with a resolution of 0.1 mm, and project the contour into three-dimensional space through surface mapping. Apply a depth compensation algorithm to correct the edge position and eliminate the deviation caused by surface undulations to generate a three-dimensional corrected contour image, providing an accurate geometric basis for path planning.

[0075] Further improve the cutting path through mesh modeling and path optimization. Based on the corrected contour image, construct a triangular mesh model with an initial side length of 2 mm, and densify the mesh in the area where the curvature is greater than 0.05, reducing the minimum side length to 0.5 mm to accurately describe complex surfaces. Segment the path according to the equal arc length principle, with each segment length controlled between 50 and 100 mm. Calculate the tangent vector using the five-point difference method to ensure that the path turning angle is less than 45 degrees. Use a circular arc transition at the path connection with a radius of not less than 5 mm to ensure smoothness. Utilize the ultrasonic thickness measurement data with a resolution of 0.01 mm to dynamically adjust the cutting depth, keeping the ratio of the depth to the material thickness between 0.8 and 0.9. The path optimization algorithm ensures that the path curvature radius is not less than 10 mm through curvature constraints to adapt to local variations in material thickness.

[0076] S1054. Smooth and constrain the path to generate the target cutting path.

[0077] Apply a smoothing algorithm to the segmented path with a smoothing radius of 20 mm to eliminate sharp corners and irregular fluctuations. Optimize the cutting angle and depth according to the material thickness distribution and processing technology requirements, and preferentially select the cutting direction along the fiber orientation to reduce fiber breakage. The generated cutting path can accurately bypass surface defect areas such as folds or microcracks, while maintaining path continuity and processing efficiency, providing direct guidance for the precise separation of mechanical devices.

[0078] Multi-scale edge detection effectively captures the fine texture and macroscopic contour of the carbon fiber surface, and region growing and curvature analysis further improve the robustness of edge connection. Depth mapping and mesh densification technologies fully consider the three-dimensional geometric characteristics of the material, ensuring that path planning can adapt to complex surface topographies. Through dynamic constraints and smoothing processing, the generated cutting path not only meets the accuracy requirements but also takes into account the stability and efficiency during the processing, providing technical support for the efficient recycling of carbon fibers.

[0079] S106. Through multi-sensor collaborative calibration and data fusion, register the information collected by visual and laser sensors in space, construct the spatial structure of carbon fiber materials, and generate a spatially mapped relationship for separation and positioning.

[0080] Utilize the complementary characteristics of visual and laser sensors to achieve high-precision data acquisition and registration on the target cutting path. Collaborative calibration ensures the spatial consistency of multi-source data by unifying the sensor coordinate system, while data fusion technology integrates the characteristic information of different sensors into a complete material spatial model. This method can effectively cope with the interference caused by the surface reflection and texture complexity of carbon fiber, providing a reliable spatial positioning basis for precise separation.

[0081] S1061. Extract feature points from the calibration plate image and calculate spatial calibration parameters to unify the multi-sensor coordinate system.

[0082] Arrange a 9x9 checkerboard calibration plate around the target cutting path, with a grid point spacing of 10 mm, to form a closed calibration field to cover the field of view. The visual sensor uses a 2-megapixel industrial camera with a focal length of 8 mm, the principal point located at the center of the image, and the radial distortion coefficient controlled within 0.01; the measurement accuracy of the laser sensor is better than 0.05 mm, and the scanning frequency is 200 Hz. Extract the feature points of the calibration plate through the checkerboard corner detection algorithm, and calculate the internal and external parameters of the sensor based on the corner positions, including the focal length, principal point coordinates, and distortion coefficient. The generated calibration parameters are optimized by least squares, with the position error converging to 0.1 mm and the angle error less than 0.1 degree, providing a unified coordinate reference for subsequent multi-sensor synchronous acquisition.

[0083] S1062. Synchronously acquire multi-sensor image sequences, and perform denoising and enhancement processing to generate acquisition sequence parameters.

[0084] Adopt a hardware trigger mechanism with a trigger frequency of 100 Hz and a delay of 1 ms between adjacent sensors to ensure synchronous acquisition of carbon fiber material image sequences by visual and laser sensors. The acquired image sequences are processed through a 5x5 Gaussian filter window with a standard deviation of 1.2 to effectively remove noise interference. Further apply histogram equalization to enhance the image contrast, with an increase of approximately 30%, making the fiber texture and edge features more prominent. Generate acquisition sequence parameters based on the image frame interval and trigger timestamp, with a timestamp accuracy better than 0.1 ms, providing a reliable time and space reference for subsequent registration.

[0085] Spatial registration of multi-sensor data is achieved through scale-invariant feature matching and iterative optimization. Based on the acquisition sequence parameters, the scale-invariant feature transform algorithm is used to extract approximately 800 feature points from each image, generating 128-dimensional feature descriptors. During feature matching, the nearest neighbor ratio threshold of 0.7 is used to screen the matching point pairs, and the random sample consensus algorithm is used to eliminate the false matches, retaining at least 50 pairs of high-quality matching points. The relative pose between the sensors is calculated, and the 4x4 homogeneous transformation matrix is used to unify the coordinate system. The repeated positioning accuracy of the optimized coordinate transformation parameters is better than 0.2 mm. Further, the iterative closest point algorithm is used to perform fine registration on the overlapping area. After 50 iterations, it converges when the pose change amount is less than 0.05 mm or 0.05 degrees, and the root mean square value of the registration error is controlled within 0.2 mm.

[0086] S1063. Generate the spatial mapping relationship for the separation and positioning of carbon fiber materials through spatial mapping and 3D reconstruction.

[0087] Based on the registered feature correspondence, the multi-sensor data is mapped to a unified spatial coordinate system. The triangular mesh reconstruction method is used to construct the 3D spatial structure of the carbon fiber material. The mesh side length is dynamically adjusted according to the surface curvature, with a range of 1 mm to 5 mm. Cubic spline functions are used for surface fitting, with a control point spacing of 5 mm and a fitting accuracy better than 0.1 mm, generating separation and positioning parameters including position coordinates and normal vectors. According to these parameters, the target cutting path is mapped to the 3D space to generate the spatial mapping relationship, ensuring that the path is highly consistent with the actual geometric features of the material, and the coordinate accuracy is better than 0.2 mm.

[0088] Through targeted parameter optimization and error compensation, the positioning accuracy is significantly improved. During the acquisition process of the vision sensor, the exposure time and gain are dynamically adjusted to suppress overexposure or underexposure caused by surface reflection. The scattering interference of the laser sensor at the material edge is alleviated by taking the average value of multiple scans, and the measurement stability is significantly improved. The installation deviation of the sensor is compensated by the calibration parameters to ensure the minimization of the system error. The application of multi-sensor collaborative calibration and data fusion technology not only provides rich material feature information but also achieves the precise alignment of the path and the material spatial structure, providing a solid guarantee for the efficient execution of subsequent mechanical cutting.

[0089] S107. Through vision guidance and force feedback control, drive the robotic arm to perform precise cutting and separation along the target cutting path, and perform intelligent classification and recycling of the separated materials.

[0090] Based on the spatial mapping relationship, the collaborative feedback of the vision sensor and the torque sensor is utilized to guide the robotic arm to complete the cutting operation of carbon fiber materials in real time. Visual servo technology ensures the accuracy of path tracking, while adaptive force control optimizes the stability during the cutting process. After separation, intelligent classification of the materials is achieved through deep learning and image analysis, and efficient recycling is completed using automated devices. This multi-link collaborative processing method not only improves the cutting accuracy but also significantly enhances the efficiency and classification accuracy of material recycling, providing comprehensive technical support for the circular utilization of carbon fiber.

[0091] S1071. Collect the working space image through the vision sensor, extract the feature points and calculate the path deviation to generate the initial motion parameters.

[0092] Use a 2-million-pixel vision sensor to collect the working space image in real time. Adopt the normalized cross-correlation template matching algorithm, with the template window set to 64x64 pixels, the search area being 2 times the template size, and the matching threshold being 0.85. Select the contour inflection points and edge intersection points as the target feature points, with the point spacing kept above 10 pixels to ensure uniform distribution of the feature points. Calculate the deviation between the feature points and the target cutting path. When the position error is less than 2 pixels, it is determined as effective tracking, and the first position parameter is generated. Based on this parameter, a fifth-order polynomial interpolation is used to generate the robotic arm joint angle sequence, with the acceleration limited to no more than 1.5 times the gravitational acceleration. The velocity curve is planned through a trapezoid, with the acceleration and deceleration segments each accounting for 20%. Calculate the velocity curves of each joint, with the angular velocity upper limit of 90 degrees per second and the motion timing error less than 1 millisecond, generating the first motion parameter, which provides a smooth motion instruction for the precise positioning of the robotic arm.

[0093] S1072. Adjust the cutting posture based on visual servo and force feedback to generate real-time optimized cutting parameters.

[0094] Drive the robotic arm to the specified position according to the first motion parameter. Adopt a visual servo algorithm based on image features, with a control frequency of 200 Hz, a proportional gain of 0.8, and an integral time constant of 0.05 seconds, and calculate the deviation between the end effector and the target position. When the position error converges to 0.2 mm and the posture error is less than 0.2 degrees, the second motion parameter is generated for adjusting the cutting tool posture. Sample the contact force and torque through the torque sensor at 1000 Hz, with a measurement range of 0 to 100 N. Apply adaptive impedance control to dynamically adjust the cutting parameters according to the contact force: when the force exceeds 50 N, reduce the feed speed to 50 mm / s, and when it is lower than 20 N, increase it to 200 mm / s. The cutting tool rotation speed is maintained at 12,000 revolutions per minute, generating the first cutting parameter to ensure a stable and efficient cutting process, reducing fiber breakage and material damage.

[0095] Ensure the precise execution of the cutting path through real-time edge detection and trajectory correction. Use the Canny algorithm for edge detection with a Gaussian smoothing kernel of 5x5 pixels, a low threshold of 50, and a high threshold of 150 to extract the contour features of the cutting area. Calculate the deviation between the actual cutting trajectory and the target trajectory, allowing a deviation of plus or minus 0.5 mm, and trigger local correction when exceeded. Based on visual feedback, evaluate the separation integrity. Judge through contour closure and edge continuity. A separation integrity greater than 0.95 is considered qualified. Further extract the shape features of the separated material and generate separation feature parameters, providing key data for classification.

[0096] S1073. Classify the separated materials using deep learning and image analysis, and drive the recycling device to complete hierarchical transportation.

[0097] Use an improved convolutional neural network to process the separation feature parameters. The input image size is 224x224 pixels, including 5 convolutional layers and 3 fully connected layers to extract the surface texture and shape features of the material. Classify through a support vector machine with a radial basis kernel function penalty factor of 100, and the classification accuracy reaches over 95%, generating material category parameters. Control the classification and recycling device according to the category parameters, configure 4 conveyor belts with a speed of 300 mm / s and a layer spacing of 200 mm. The response time of the photoelectric sensor is less than 1 ms, and the detection accuracy is better than 1 mm to monitor the material position in real time. Through baffle diversion, direct materials of different categories to the corresponding conveyor belts, with a classification accuracy of 98% to achieve efficient hierarchical recycling.

[0098] The combination of visual guidance and force feedback significantly improves the stability and accuracy of cutting. Especially when dealing with wrinkles or local defects on the carbon fiber surface, adaptive control effectively avoids over-cutting or tool damage. The application of deep learning classification technology enables rapid differentiation of different types of carbon fiber materials, such as woven structures or thickness differences, enhancing the pertinence of recycling. The automated recycling device ensures the efficiency and reliability of the classification process through precise diversion and real-time detection, providing a complete technical closed-loop for the recycling of carbon fiber.

[0099] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the technical solutions formed by mutually replacing the above features with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A carbon fiber separation and positioning method based on machine vision, characterized in that: The method comprises: The image data of the carbon fiber material in the ultraviolet, visible and near-infrared spectral bands are obtained through a multi-spectral visual detection unit, and the distribution position of the carbon fiber material in three-dimensional space is determined according to the image data to obtain the spatial distribution information of the carbon fiber material; According to the spatial distribution information of carbon fiber materials, the carbon fiber materials are illuminated by using the composite illumination of orthogonal polarized light and structured light, the angle and intensity of polarized light and structured light are controlled, and the interface characteristic information of carbon fiber materials is obtained; According to the obtained interface characteristic information of carbon fiber materials, the angle and wavelength of the light source are controlled to perform sequential illumination on the carbon fiber materials to obtain images of carbon fiber materials under different illumination conditions; The carbon fiber material images under different lighting conditions are processed, and the images are separated into three bands: ultraviolet, visible, and near-infrared through spectral spectroscopy to obtain three-spectrum coaxial acquisition images; Perform image segmentation processing on the three-spectrum coaxial acquisition image to extract the contour edge information of the carbon fiber material, and determine the target cutting path of the carbon fiber material in combination with the spatial distribution information of the carbon fiber material; Through multi-sensor collaborative calibration on the target cutting path, the carbon fiber material information collected by different sensors is spatially registered and fused to establish the spatial mapping relationship of carbon fiber material separation and positioning; According to the established spatial mapping relationship of carbon fiber material separation and positioning, combined with the target cutting path, the carbon fiber material is cut and separated by guiding the robotic arm through machine vision, and the separated carbon fiber material is classified and recycled.

2. The method according to claim 1, characterized in that The method of obtaining image data of the carbon fiber material in the ultraviolet, visible, and near-infrared spectral bands by a multi-spectral visual detection unit, determining the distribution position of the carbon fiber material in three-dimensional space according to the image data, and obtaining the spatial distribution information of the carbon fiber material includes: According to the carbon fiber surface reflectance spectrum image data acquired by the multi-spectral scanner, the first spectrum image data is obtained by performing Gaussian filtering and noise reduction processing; For the checkerboard image of the calibration plate, a corner point detection algorithm is used to extract the corner point position coordinates, and a mapping matrix between the multispectral camera imaging coordinate system and the world coordinate system is established according to the corner point position coordinates; The first spectral image data is subjected to band separation, edge segments of each band image are extracted using Hough transform, and a second feature point set of the contour area of ​​the carbon fiber material is obtained by a region growing algorithm; For the feature points in the second feature point set, the three bands are aligned using a spectral feature matching algorithm based on similarity, and the spatial distribution information of the carbon fiber material is obtained according to the aligned feature points.

3. The method according to claim 1, characterized in that According to the spatial distribution information of the carbon fiber material, the carbon fiber material is illuminated by a composite illumination method of orthogonal polarized light and structured light, the angle and intensity of the polarized light and the structured light are controlled, and the interface characteristic information of the carbon fiber material is obtained, including: The reflected light intensity signal is collected according to the spatial distribution range of the carbon fiber material, and the polarization angle of the polarizer is adjusted using a closed-loop feedback controller. The polarization compensation parameters are obtained by comparing the difference in reflected light intensity in two polarization directions. Acquire a structured light stripe reflection signal according to the polarization compensation parameter, and perform a Gaussian fitting operation on the structured light stripe reflection signal to obtain a stripe deformation parameter; A light source synchronization trigger is used to generate an alternating illumination sequence, and a second light intensity image is obtained by performing a denoising and filtering process on the reflected image; The second light intensity image is geometrically corrected according to the fringe deformation parameter, and the fringe distortion amount is calculated by a phase demodulation algorithm to obtain a surface morphology characteristic image of the carbon fiber material.

4. The method according to claim 1, characterized in that According to the obtained interface feature information of the carbon fiber material, the angle and wavelength of the light source are controlled to perform sequential illumination on the carbon fiber material to obtain images of the carbon fiber material under different illumination conditions, including: A light intensity sensor is used to receive a reflected light intensity signal from the surface of the carbon fiber material, the reflected light intensity signal is generated by adjusting the irradiation angle of the light source by an electric angle controller, and a first angle parameter is obtained according to the reflected light intensity signal; Controlling the spectrally tunable filter to scan the wavelength of the light source according to the first angle parameter, collecting spectral response data of the surface of the carbon fiber material through the wavelength controller, and obtaining a first wavelength parameter; Generate an illumination matrix according to the first angle parameter and the first wavelength parameter, generate an illumination pulse sequence through a multi-channel waveform generator, and obtain an illumination sequence control signal; According to the illumination sequence control signal, a high-speed image collector is driven to obtain a reflection image sequence of a carbon fiber material, the reflection image sequence is denoised by a median filter to obtain a first image sequence, the first image sequence is edge enhanced by an image sharpening algorithm, and then brightness compensation is performed to obtain carbon fiber material images under different illumination conditions.

5. The method according to claim 1, characterized in that The carbon fiber material images under different lighting conditions are processed, and the images are separated into three bands, namely, ultraviolet, visible, and near-infrared, by a spectral spectrometry method to obtain three-spectrum coaxial acquisition images, including: According to the carbon fiber material image, the spectral components of the ultraviolet band, the visible light band and the near infrared band are extracted by using a spectral bandpass filter, and the first band image is obtained by Gaussian denoising. A spectrophotometer is used to obtain a standard whiteboard spectral transmittance curve, transmittance compensation is performed on the first band image according to the standard whiteboard spectral transmittance curve, and a second band image is obtained by brightness normalization processing; Calculate the inter-band crosstalk coefficient matrix according to the second band image, separate the inter-band crosstalk coefficient matrix by using a non-negative matrix decomposition method, and obtain the third band image by iterative operation; Perform edge detection on the third band image, use Harris corner detection algorithm to extract a set of feature points, calculate spatial transformation parameters based on the feature point set by least squares method, use histogram matching method to adjust the grayscale distribution of the three band images, and obtain a three-spectrum coaxial acquisition image by superimposing the band images.

6. The method according to claim 1, characterized in that The image segmentation process is performed on the three-spectrum coaxial acquisition image to extract the contour edge information of the carbon fiber material, and the target cutting path of the carbon fiber material is determined in combination with the spatial distribution information of the carbon fiber material, including: A Gaussian pyramid is constructed based on the three-spectrum coaxially collected image, the image is decomposed into three layers by the Gaussian pyramid, and the edge candidate points are extracted by a gradient amplitude detection method to obtain a first contour image; The first contour image is edge tracked by an eight-neighborhood search method, and a second contour image is obtained by region growing by a grayscale similarity criterion; Extracting a contour feature point sequence according to the second contour image, calculating the curvature of the feature points by a curve fitting method, and obtaining key control points by extracting extreme points to obtain a third contour image; The third contour image is subjected to surface mapping, a depth mapping matrix is ​​calculated according to the spatial distribution information of the carbon fiber material, and a depth compensation method is used to correct the contour edge to obtain a target cutting path of the carbon fiber material.

7. The method according to claim 1, characterized in that The method of spatially registering and fusing the carbon fiber material information collected by different sensors through multi-sensor collaborative calibration on the target cutting path, and establishing a spatial mapping relationship for separating and positioning the carbon fiber materials, includes: A visual sensor and a laser sensor are used to obtain a calibration plate image, checkerboard corner feature points are extracted according to the calibration plate image, and spatial calibration parameters are calculated through the feature points; Controlling multiple sensors to synchronously collect a sequence of carbon fiber material images according to the spatial calibration parameters, performing Gaussian filtering and histogram equalization processing on the image sequence to obtain sensor acquisition sequence parameters; The sensor acquisition sequence parameters are processed by a matching algorithm based on scale-invariant features, and feature correspondence is obtained by calculating feature descriptors; Performing spatial mapping conversion on the feature correspondence, constructing the spatial structure of the carbon fiber material using a three-dimensional grid reconstruction method, and obtaining material separation positioning parameters; A spatial mapping relationship of the carbon fiber material is established according to the material separation and positioning parameters, and the cutting path is spatially mapped to obtain a spatial mapping relationship of the carbon fiber material separation and positioning.

8. The method according to claim 1, characterized in that The method includes: guiding the mechanical arm to cut and separate the carbon fiber material by machine vision based on the established carbon fiber material separation and positioning spatial mapping relationship and combining the target cutting path, and classifying and recycling the separated carbon fiber material, including: Collecting a workspace image according to a visual sensor, extracting feature points of the workspace image based on a template matching algorithm, and calculating the deviation between the feature points and the planned path to obtain a first position parameter; Generate a joint angle sequence using a quintic polynomial interpolation method according to the first position parameter, and calculate the velocity curve of each joint of the joint angle sequence using a velocity planner to obtain a first motion parameter; Control the robot arm to move to a specified position according to the first motion parameter, and calculate the deviation between the end effector of the robot arm and the target position based on a visual servo algorithm to obtain a second motion parameter; The cutting tool posture angle is adjusted according to the second motion parameter, the contact force and torque signal between the cutting tool and the workpiece are collected by the torque sensor to obtain the first cutting parameter, and after the carbon fiber material separation operation is performed according to the first cutting parameter, the carbon fiber material is classified and recycled.

Images