A Machine Vision-Based Inspection Method for the Production Inspection of Modified Asphalt from Waste Plastics

By employing multispectral and structured light imaging technologies, the real-time and accuracy issues of detecting the dispersion state of plastic particles in the production of waste plastic modified asphalt have been resolved. This enables online detection of the three-dimensional distribution and melting state of plastic particles in the asphalt matrix, supporting the optimization of the production process.

CN123111966APending Publication Date: 2026-11-17HEBEI ROAD & BRIDGE GROUP +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611635580.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies cannot detect the dispersion state of plastic particles in the asphalt matrix during the production of waste plastic modified asphalt in real time and accurately. In particular, it is difficult to identify the three-dimensional distribution and melting and softening degree of the interface between plastic particles and asphalt under high temperature and dynamic flow environment.

Method used

By combining multispectral imaging and structured light imaging, visible light, near-infrared and structured light stripe images of molten mixtures are acquired simultaneously, and temporal and spatial registration is performed to generate a multimodal image set. The exposure height, embedding depth, equivalent particle size and melt state index of plastic particles are calculated to achieve non-contact online detection.

Benefits of technology

It enables precise, multi-dimensional detection of the dispersion state of plastic particles under high-temperature flow conditions, providing a direct and quantitative evaluation of the three-dimensional spatial occupancy and melting degree of plastic particles in asphalt, and supporting real-time adjustment of the production process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123111966A_ABST
    Figure CN123111966A_ABST
Patent Text Reader

Abstract

This invention discloses a machine vision-based inspection method for the production and testing of modified asphalt from waste plastics, belonging to the technical field of quality inspection in the production of modified asphalt from waste plastics. The method simultaneously acquires surface multispectral images and structured light stripe images of the molten mixture at a transparent inspection window in the modified asphalt production pipeline, obtaining visible light image sequences, near-infrared image sequences, and structured light stripe image sequences. Temporal and spatial registration is then performed to generate a multimodal image set. A plastic phase distribution marker map is generated based on the near-infrared image sequence, and a three-dimensional contour map is generated by phase demodulation of the structured light stripe image sequence. The two are then superimposed to calculate the exposure height and embedding depth. Finally, the equivalent particle size distribution and melt state index are calculated based on the visible light image sequence, and the dispersion state detection result is comprehensively generated. This method achieves multi-dimensional, non-destructive, online detection of the dispersion state of plastic particles in the molten mixture.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of production quality testing technology for waste plastic modified asphalt, specifically a production testing method for waste plastic modified asphalt based on machine vision inspection. Background Technology

[0002] In the production process of waste plastic modified asphalt, the dispersion state of waste plastic particles in the high-temperature asphalt matrix directly affects the road performance of the modified asphalt. Existing methods for detecting dispersion state mostly involve manual sampling followed by microscopic cross-section observation or fluorescence microscopy analysis in the laboratory. This method has a long operation cycle, and the test results lag behind the production process, making it impossible to provide real-time feedback for adjusting process parameters.

[0003] Another existing technology uses a single visible light camera to capture images of the surface of the molten mixture, identifies the plastic phase regions through image segmentation, and calculates their area proportion. This approach only acquires two-dimensional planar information of the surface and cannot determine whether plastic particles are floating on the asphalt surface or fully submerged and embedded, leading to distorted assessments of dispersion. Infrared spectroscopy can be used to identify the chemical components of plastics, but conventional methods only provide average spectral information, lacking a comprehensive correlation between spatial distribution and three-dimensional morphology.

[0004] Because both asphalt and plastic are dark, viscous fluids in their molten state, with weak surface texture and low contrast, the aforementioned technologies struggle to simultaneously identify the relative spatial positions of plastic particles and the asphalt matrix, as well as quantify the degree of melting and softening of the plastic particles themselves, under high-temperature, dynamic pipeline flow conditions. Specifically, it is necessary to address how to obtain the three-dimensional distribution morphology of plastic particles relative to the asphalt interface from the surface of the flowing molten mixture, and how to accurately characterize the physical state changes of plastic particles as they transition from a solid to a molten state without interfering with production. A common deficiency in existing detection technologies is the lack of joint analytical capabilities regarding the height of plastic particles exposed outside the matrix, the depth of embedding within the matrix, the degree of melting of the particles themselves, and particle size. Summary of the Invention

[0005] The purpose of this invention is to provide a machine vision-based detection method for the production and testing of waste plastic modified asphalt, so as to achieve non-contact, multi-dimensional online detection of the dispersion state of plastic particles in the molten mixture during the production process of waste plastic modified asphalt.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a machine vision-based detection method for the production and inspection of waste plastic modified asphalt, comprising the following steps: simultaneously acquiring surface multispectral images and structured light stripe images of the molten mixture at a transparent detection window of the modified asphalt production pipeline, obtaining visible light image sequences, near-infrared image sequences, and structured light stripe image sequences; performing temporal and spatial registration on the visible light image sequences, near-infrared image sequences, and structured light stripe image sequences to generate a multimodal image group; generating a plastic phase distribution marker map based on the absorption differences of plastic characteristic bands in the near-infrared image sequence; performing phase demodulation on the structured light stripe image sequence to generate a three-dimensional contour map of the surface of the molten mixture; superimposing the plastic phase distribution marker map with the three-dimensional contour map to calculate the exposure height and embedding depth; calculating the equivalent particle size distribution and melt state index based on the edge contour and color components of the plastic phase region in the visible light image sequence; and generating a dispersion state detection result based on the exposure height, embedding depth, equivalent particle size distribution, and melt state index.

[0007] As a technical solution of this invention, when performing temporal and spatial registration on visible light image sequences, near-infrared image sequences, and structured light fringe image sequences, the acquisition time of the three is controlled by the same hardware trigger signal, ensuring that exposure is completed at the same time point. A calibration board is used to obtain the spatial transformation parameters between the multispectral imaging device and the structured light projection device, mapping each image sequence to the same coordinate space. The three mapped image sets are then aligned according to pixel coordinates to generate a multimodal image group. Through rigorous temporal and spatial registration, the differences and viewing angle deviations between the multi-source images at the moment of acquisition are eliminated, ensuring pixel-level correspondence accuracy for subsequent fusion analysis.

[0008] Preferably, the process of generating the plastic phase distribution marker map is as follows: Intensity values ​​of two characteristic bands are extracted from a near-infrared image sequence, where the first characteristic band corresponds to the carbon-hydrogen bond absorption peak of polyolefins, and the second characteristic band corresponds to the reference wavelength of the asphalt matrix; the ratio of the intensity values ​​of each pixel in the first characteristic band to the second characteristic band is calculated to obtain a ratio image; pixels with a ratio greater than 1 in the ratio image are assigned a plastic marker, and pixels with a ratio not greater than 1 are assigned an asphalt marker, thus generating the plastic phase distribution marker map. This marker method directly utilizes the specific spectral absorption of plastic components, enabling rapid and reliable separation of the plastic phase region from the asphalt matrix under high-temperature molten conditions.

[0009] Preferably, the specific method for generating a three-dimensional contour map of the molten mixture surface is as follows: each frame of the structured light fringe image sequence is decomposed into a fundamental frequency component and an overtone component, and the principal phase value of the fringe is extracted; the principal phase values ​​of two adjacent fringe frames are differentially analyzed to obtain a phase increment; the phase increments are accumulated to generate a continuous phase distribution; and a three-dimensional contour map is generated based on the mapping relationship between the continuous phase distribution and a pre-calibrated phase height. This phase demodulation method effectively overcomes the phase discontinuity problem caused by the molten flow surface and obtains a dense three-dimensional morphology that reflects the true height fluctuations.

[0010] Furthermore, when calculating the exposure height and embedding depth, the connected domain boundaries of each plastic particle are extracted from the plastic phase distribution marker map and mapped onto the 3D contour map. The height values ​​of each pixel within the connected domain boundaries are read, and the maximum value is taken as the particle vertex height. In the plastic phase distribution marker map, the asphalt region adjacent to the connected domain boundary is selected, and the height values ​​of the asphalt region are read. The median height of the asphalt region is taken as the asphalt surface height. The exposure height is obtained by subtracting the asphalt surface height from the particle vertex height. The embedding depth is obtained by subtracting the minimum height value of each pixel within the connected domain boundary from the asphalt surface height. This quantitatively obtains the exposure height of the plastic particle protruding above the asphalt gas-liquid free surface and the apparent embedding depth of the plastic particle on the molten mixture gas-liquid free surface. Due to the limitations of structured light imaging mechanism, structured light cannot penetrate dark asphalt matrix. The embedding depth obtained by this invention is the apparent depression height difference formed by particles at the gas-liquid interface, and is not used as an absolute physical measurement value of the actual embedding depth of plastic particles in the asphalt matrix. This apparent embedding depth is used to compare the sinking and wetting degree of different plastic particles on the surface of molten asphalt, to evaluate the trend of particles being wetted and wrapped by asphalt, and to provide a relative dimensional index for comprehensive evaluation of dispersion state.

[0011] Preferably, the process of calculating the equivalent particle size distribution and melt state index is as follows: Based on the plastic phase distribution marker map, extract the plastic phase pixel set from the visible light image sequence; perform edge detection on the plastic phase pixel set to obtain the closed boundary of each plastic particle; perform ellipse fitting on the closed boundary and calculate the major and minor axis lengths of the fitted ellipse; use the square root of the product of the major and minor axis lengths as the equivalent particle size, and statistically analyze the equivalent particle size of all plastic particles to generate the equivalent particle size distribution; extract the component values ​​of the plastic phase pixel set in the red, green, and blue color channels, and use the ratio of the difference between the maximum and minimum color components to the maximum color component as the melt state index of each plastic particle. The equivalent particle size distribution reflects the degree of refinement of the plastic particle size, while the melt state index uses color uniformity to reflect the melting and mixing process of the plastic.

[0012] In the aforementioned edge detection process, voids are filled in the plastic phase pixels within the plastic phase pixel set to obtain a filled plastic phase region; morphological erosion is then performed on the filled plastic phase region to obtain an eroded plastic phase region; the pixel values ​​of the filled plastic phase region and the eroded plastic phase region are subtracted to obtain the closed boundary. This method can accurately extract the single-pixel-wide continuous boundary of plastic particles, improving the stability of ellipse fitting and the accuracy of particle size calculation.

[0013] As a preferred embodiment of the present invention, the steps for generating dispersion state detection results include: calculating the Euclidean distance between the centroids of any two plastic particles based on the centroid coordinates of each plastic particle, and selecting the distance between each particle and its nearest neighbor as the nearest neighbor distance; statistically analyzing the numerical distribution ranges of the exposure height, embedding depth, melt state index, and nearest neighbor distance of all plastic particles; and correlating the numerical distribution ranges with the equivalent particle size distribution to generate dispersion state detection results. Through comprehensive statistical analysis and correlation of multi-dimensional parameters, the dispersion uniformity, exposure and embedding degree, and melting degree of particles in waste plastic modified asphalt are comprehensively evaluated, providing a quantitative online evaluation basis for production process control.

[0014] Preferably, when simultaneously acquiring images at the transparent detection window, the optical axis of the multispectral imaging device is perpendicular to the outer surface of the transparent detection window, and the projection direction of the structured light projection device forms a set angle with the optical axis of the multispectral imaging device. Before simultaneous acquisition, the inner surface of the transparent detection window is swept with compressed air, and the temperature of the inner surface of the transparent detection window is heated to the same temperature as the molten mixture. This setting can prevent high-temperature asphalt vapor from condensing or coking on the inner surface of the window, eliminate optical distortion caused by temperature gradients, and ensure the clarity and authenticity of multispectral and structured light images.

[0015] Furthermore, the phase-height mapping relationship used to generate the 3D contour map is pre-established through the following steps: a calibration block with multiple known height steps is placed at the transparent detection window; fringe images of the calibration block are acquired in a structured light fringe image sequence, obtaining the corresponding data points of continuous phase values ​​and actual heights for each step; polynomial fitting is performed on the corresponding data points to generate the phase-height mapping relationship. This calibration process directly utilizes the optical path conditions of the detection window, avoiding installation deviations and ensuring high accuracy of the height reconstruction values.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0017] By simultaneously acquiring near-infrared image sequences and structured light fringe image sequences of the molten mixture, and extracting absorption differences in characteristic bands of the plastic from the near-infrared image sequences to generate a plastic phase distribution marker map, while performing phase demodulation on the structured light fringe image sequences to generate a three-dimensional contour map of the molten mixture surface, and then overlaying and analyzing the plastic phase distribution marker map and the three-dimensional contour map, the exposure height and embedding depth of each plastic particle can be calculated. This pixel-level fusion of the spatial distribution of chemical components and the three-dimensional morphology of the surface overcomes the limitation that a single two-dimensional image can only describe planar distribution. It allows for precise differentiation on the surface of a high-temperature flowing molten mixture from whether plastic particles are floating on the asphalt surface or have been effectively wetted and embedded in the matrix, thus providing direct and quantitative geometric parameters for evaluating the three-dimensional spatial occupancy of plastic particles in asphalt.

[0018] By extracting the set of plastic phase pixels from a visible light image sequence using a plastic phase distribution marker map, and calculating the equivalent particle size distribution after fitting the particle boundaries, the component values ​​of the plastic phase pixel set in the red, green, and blue color channels are extracted. The ratio of the difference between the maximum and minimum color components to the maximum component is used to calculate the melt state index. This joint analysis method correlates the physical size characteristics of plastic particles with the degree of chemical and physical phase transition. The equivalent particle size distribution reflects the macroscopic size composition of the particle group, while the ratio calculation based on the color channel differences can capture the color mixing and saturation changes of plastic particles during melting and softening due to surface morphology, refractive index, and partial component analysis, thus directly characterizing the degree of transformation of individual particles from a solid to a viscous flow state. Combining these two pieces of information with the exposure height and embedding depth calculated by overlaying multimodal images, it is possible to three-dimensionally characterize the multi-dimensional dispersion characteristics of particles, such as whether they are "large but unmelted and floating on the surface" or "small but melted and deeply embedded in the matrix." The entire detection process does not require contact with the material and can be completed under pipeline flow conditions. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of a machine vision-based inspection method for the production and testing of modified asphalt from waste plastics.

[0021] Figure 2 This is a flowchart of the plastic phase distribution marker generation process based on dual-band near-infrared ratio images;

[0022] Figure 3 This is a flowchart of the method for extracting the exposure height and embedding depth of plastic particles;

[0023] Figure 4 It is a curve showing the change in pixel grayscale values ​​of a multimodal image sequence;

[0024] Figure 5 This is a distribution map of the exposed height and embedding depth of plastic particles;

[0025] Figure 6 It is the estimated curve of equivalent particle size distribution and kernel density of plastic particles. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] See Figure 1 This invention provides a machine vision-based inspection method for the production and testing of modified asphalt from waste plastics, comprising the following steps:

[0028] At the transparent inspection window of the modified asphalt production pipeline, surface multispectral images and structured light stripe images of the molten mixture are acquired simultaneously to obtain visible light image sequences, near-infrared image sequences, and structured light stripe image sequences.

[0029] The visible light image sequence, near-infrared image sequence, and structured light stripe image sequence are time-registered and spatially registered to generate a multimodal image group.

[0030] Based on the absorption differences in the characteristic bands of plastic in the near-infrared image sequence, a plastic phase distribution marker map is generated.

[0031] Phase demodulation is performed on the structured light stripe image sequence to generate a three-dimensional contour map of the surface of the molten mixture.

[0032] The plastic phase distribution map is overlaid with the three-dimensional contour map to calculate the exposure height and embedding depth.

[0033] The equivalent particle size distribution and melt state index are calculated based on the edge contour and color components of the plastic phase region in the visible light image sequence.

[0034] Based on the exposure height, embedding depth, equivalent particle size distribution, and melt state index, dispersion state detection results are generated.

[0035] In practice, a multispectral imaging device and a structured light projection device are installed outside the transparent inspection window of the modified asphalt production pipeline. The optical axis of the multispectral imaging device is perpendicular to the outer surface of the transparent inspection window. The multispectral imaging device is used to acquire visible light and near-infrared image sequences of the molten mixture. The projection direction of the structured light projection device has a set angle of 30 degrees with the optical axis of the multispectral imaging device. This angle is chosen because a 30-degree angle maximizes the sensitivity of the deformation of the structured light fringes on the surface of the molten mixture to height changes and avoids obstruction between the projection and imaging light paths. The transparent inspection window is made of high-temperature and high-pressure resistant sapphire glass and is embedded in the wall of the modified asphalt production pipeline.

[0036] Before synchronous acquisition begins, the inner surface of the transparent detection window is cleaned by blowing compressed air through it to remove any adhering molten mixture residue. After blowing, the inner surface temperature of the transparent detection window is heated to the same temperature as the molten mixture. The temperature of the molten mixture is measured in real time by a temperature sensor installed upstream of the transparent detection window. The heating element is a resistance heating wire embedded in the frame of the transparent detection window. Closed-loop control ensures that the temperature difference between the inner surface of the transparent detection window and the molten mixture is less than 1 degree Celsius.

[0037] During synchronous acquisition, the acquisition timing of the visible light image sequence, near-infrared image sequence, and structured light fringe image sequence is controlled by the same hardware trigger signal. The hardware trigger signal is issued by the synchronization controller, which outputs three TTL pulse signals. The rising edges of these three TTL pulse signals simultaneously reach the multispectral imaging device and the structured light projection device, ensuring that the visible light camera and near-infrared camera in the multispectral imaging device, and the fringe projection unit in the structured light projection device, complete exposure at the same time point. The visible light camera and near-infrared camera in the multispectral imaging device share the same optical path through a beam splitter prism, while the structured light projection device projects a sinusoidal fringe pattern into the transparent detection window. After synchronous acquisition, time-synchronized visible light image sequences, near-infrared image sequences, and structured light fringe image sequences are obtained.

[0038] To transform visible light image sequences, near-infrared image sequences, and structured light stripe image sequences to the same coordinate space, a calibration plate was used to obtain spatial transformation parameters between the multispectral imaging device and the structured light projection device. The calibration plate was a flat plate with a checkerboard pattern printed on its surface, with each square having a side length of 5 mm. The calibration plate was placed on the inner surface of a transparent detection window, and visible light images and images with structured light stripes were acquired from the calibration plate using both the multispectral imaging device and the structured light projection device. Pixel coordinates of the checkerboard corner points in the visible light image were extracted through corner detection, while the three-dimensional coordinates of the checkerboard corner points in the structured light projection device coordinate system were obtained using phase demodulation and triangulation principles. Spatial transformation parameters, including rotation matrices and translation vectors, were established between the multispectral imaging device image coordinate system and the structured light projection device coordinate system using the pixel coordinates in the visible light image and the three-dimensional coordinates in the structured light projection device coordinate system. Finally, the intrinsic parameter calibration results between the visible light camera and the near-infrared camera in the multispectral imaging device were used to map the near-infrared image sequence to the visible light image coordinate system. Based on spatial transformation parameters, the pixel coordinates of the visible light image sequence and the near-infrared image sequence are mapped to the coordinate system of the structured light projection device. Then, the pixel coordinates of the structured light stripe image sequence are inversely transformed to the same coordinate system of the structured light projection device through the projector. This transforms the three sets of images—visible light image sequence, near-infrared image sequence, and structured light stripe image sequence—to the same coordinate space. The mapped visible light image sequence, the mapped near-infrared image sequence, and the mapped structured light stripe image sequence are then aligned by nearest-neighbor interpolation according to their pixel coordinates, so that the same pixel coordinate position corresponds to the same physical point on the surface of the molten mixture, generating a multimodal image set.

[0039] See Figure 4 The figure shows the curves of pixel grayscale values ​​changing with frame number (image sequence index) for three image sequences simultaneously acquired at the transparent detection window by the multispectral imaging device and structured light projection device used in the above embodiment. The horizontal axis represents the frame number, ranging from 1 to 500 frames, and the vertical axis represents the numerical change of pixel grayscale values ​​in the corresponding image sequence. In the legend, the blue curve corresponds to the visible light image sequence, the orange curve corresponds to the near-infrared image sequence, and the green curve corresponds to the structured light stripe image sequence.

[0040] From the curve trends, the grayscale values ​​of the three image sequences all exhibited periodic fluctuations throughout the acquisition process, indicating that the acquired image data reflected the dynamic changes of the molten mixture at the transparent detection window of the pipeline. The green structured light stripe image sequence had a generally higher grayscale value, typically fluctuating between 160 and 210, with a significantly larger fluctuation amplitude than the other two groups. This reflects the significant deformation produced by the structured light stripes on the surface of the molten mixture, making it suitable for phase demodulation and 3D contour map generation. The blue visible light image sequence had a grayscale value between 130 and 170, with relatively stable fluctuations, reflecting changes in the surface optical characteristics of the mixture. The orange near-infrared image sequence had the lowest grayscale value, fluctuating between approximately 115 and 155. Its fluctuation trend showed a certain synchronicity with the visible light image sequence, reflecting the difference in absorption response between the plastic and asphalt phases in the near-infrared characteristic band.

[0041] In specific implementation, please refer to Figure 2 Intensity values ​​of two characteristic bands were extracted from the near-infrared image sequence. Each frame of the near-infrared image sequence contains data from multiple spectral bands. The first characteristic band corresponds to the carbon-hydrogen bond absorption peak of polyolefins. The center wavelength of the first characteristic band is 1720 nm. At this wavelength, the stretching vibration of carbon-hydrogen bonds in polyolefins produces strong absorption, making the reflection intensity of the plastic phase region significantly lower than that of the asphalt matrix region. The second characteristic band corresponds to the reference wavelength of the asphalt matrix. The center wavelength of the second characteristic band is 1650 nm. At this wavelength, the absorption of polyolefins is weak, while the reflection intensity of the asphalt matrix changes more gradually compared to the reflection intensity at the first characteristic band, thus serving as a reference benchmark.

[0042] For each pixel location in each frame of the near-infrared image sequence, obtain the intensity value under the first feature band. Intensity values ​​under the second characteristic band ,in For the coordinate index of pixels in the near-infrared image, For pixel column coordinates, These are the pixel row coordinates. The intensity value of the first feature band. The intensity value of the second characteristic band was obtained by a near-infrared camera after narrowband filtering with a center wavelength of 1720 nm and a full width at half maximum (FWHM) of 20 nm. The image was captured by the same near-infrared camera after being filtered through a narrowband with a center wavelength of 1650 nm and a full width at half maximum (FWHM) of 20 nm.

[0043] Calculate the ratio of the intensity values ​​of each pixel in the first feature band to the intensity values ​​in the second feature band to obtain a ratio image. The coordinates in the ratio image are... pixel values Defined by the following formula:

[0044]

[0045] in, The ratio image in pixel coordinates The ratio value at that point, This represents the coordinate index of a pixel in a near-infrared image. Pixel column coordinates (dimensionless integers). These are the pixel row coordinates (dimensionless integers). pixel coordinates The intensity value of the first characteristic band, pixel coordinates The intensity value of the second characteristic band. Calculated one by one for each pixel position in the near-infrared image. This yields a ratio image with the same size as a single-frame near-infrared image.

[0046] After the ratio image is generated, pixels with a ratio greater than 1 are assigned a plastic label, and pixels with a ratio less than or equal to 1 are assigned an asphalt label. In the ratio image, when the pixel coordinates... place satisfy When the pixel corresponds to the plastic phase in the molten mixture, a plastic marker is assigned to the pixel, with a value of 1; when At that time, the pixel is determined to correspond to the bitumen phase in the molten mixture, and a bitumen marker is assigned to the pixel with a value of 0. After completing the binary assignment for all pixels, a binary plastic phase distribution marker map is generated. In the plastic phase distribution marker map, the pixel area marked with 1 represents the spatial distribution range of the plastic phase, and the pixel area marked with 0 represents the spatial distribution range of the bitumen phase.

[0047] In the specific implementation, a phase height mapping relationship is established in advance. The process of establishing the phase height mapping relationship is as follows: A calibration block is placed at the transparent detection window. The calibration block has multiple known height steps, and its surface is processed with five steps arranged from low to high, with step heights of 0 mm, 2 mm, 4 mm, 6 mm, and 8 mm respectively. The upper surface of each step is flat and parallel to the outer surface of the transparent detection window. After the calibration block is placed on the inner surface of the transparent detection window, the structured light projection device is activated to project a sinusoidal fringe pattern onto the surface of the calibration block. Simultaneously, the camera in the multispectral imaging device acquires the fringe image of the calibration block in the structured light fringe image sequence. The structured light projection device sequentially projects sinusoidal fringe patterns with a phase shift of four steps and a phase shift increment of π / 2. One frame of fringe image is acquired for each projected pattern to obtain the fringe image sequence of the calibration block.

[0048] Phase demodulation is performed on each pixel position in the stripe image sequence of the calibration block. A four-step phase-shifting algorithm is used to calculate the principal phase value of each pixel. Spatial phase expansion is then performed on the principal phase value to obtain the continuous phase value corresponding to each pixel position on the calibration block surface. For each known height step on the calibration block, pixels in the central region of the step's surface are selected, and the arithmetic mean of the continuous phase values ​​of all pixels in that central region is taken as the representative continuous phase value for that step. The actual height values ​​of the five steps are paired with their corresponding continuous phase value representatives to obtain five corresponding data points. Polynomial fitting is performed on the five corresponding data points. A cubic polynomial is used for the polynomial fitting, and the form of the cubic polynomial is as follows: ,in The height value obtained by fitting, For continuous phase values, , , , The coefficients are the polynomial coefficients obtained from the fitting. The constant coefficient represents the reference height value when the continuous phase value is zero; the value is automatically determined by the fitting algorithm. The coefficient for the first term represents the slope of the linear relationship between continuous phase values ​​and height values; its value is automatically determined by the fitting algorithm. The coefficients of the quadratic term represent the coefficients of the second-order nonlinear component in the phase height mapping relationship, and their values ​​are automatically determined by the fitting algorithm. The coefficients are cubic coefficients, representing the third-order nonlinear component coefficients in the phase-height mapping relationship. Their values ​​are automatically determined by the fitting algorithm. The fitting algorithm uses the least squares method to solve for the polynomial coefficients with the objective of minimizing the sum of squares of the fitting residuals. , , , After fitting, the polynomial coefficients obtained will be... , , , Stored as a phase height mapping relationship.

[0049] It is understandable that after the phase height mapping relationship is established in advance, the stored polynomial coefficients can be directly called when generating a three-dimensional contour map of the molten mixture surface each time. , , , No need for repeated calibration.

[0050] The specific implementation method for generating a three-dimensional contour map of the molten mixture surface is as follows: Each frame of the structured light fringe image sequence is decomposed in the frequency domain. A Fourier transform is used to transform the fringe image from the spatial domain to the frequency domain. In the frequency domain, the fundamental frequency component of the fringe is represented by the fundamental frequency peak, and the harmonic components are represented by higher harmonic peaks. The spectral component corresponding to the fundamental frequency peak is extracted using a bandpass filter. An inverse Fourier transform is performed on the extracted fundamental frequency spectral component to obtain the fundamental frequency complex signal. The phase angle at each pixel position in the fundamental frequency complex signal is the principal phase value of the fringe. The structured light fringe image sequence contains four frames of fringe images acquired by a structured light projection device projecting a four-step phase-shifted sinusoidal fringe pattern sequentially. Frequency domain decomposition and fundamental frequency component extraction are performed on each frame of fringe image to obtain the principal phase value corresponding to each frame. The principal phase values ​​of adjacent frames of fringe images are differentiated. Adjacent frames are fringe images corresponding to two sinusoidal fringe patterns projected sequentially by the structured light projection device. The principal phase value of the subsequent frame of fringe image is subtracted from the principal phase value of the previous frame of fringe image to obtain the phase increment. The phase increment reflects the amount of phase change at each pixel location caused by fringe movement during phase shift. The phase increment is calculated for all three pairs of adjacent frames in the structured light fringe image sequence. All calculated phase increments are accumulated pixel-by-pixel, with the phase increment accumulation operation performed independently at each pixel location during the accumulation process, generating a continuous phase distribution. The continuous phase distribution is a phase matrix with the same size as the single-frame fringe pattern, where each element corresponds to the unwrapped phase value of a detection point on the surface of the molten mixture. Based on the continuous phase values ​​at each pixel location in the continuous phase distribution, the stored polynomial expression for the phase height mapping relationship is substituted. In the process, the height value corresponding to each pixel position is calculated sequentially, and the calculation results are arranged according to pixel position to form a height matrix, generating a 3D contour map. The height value of each pixel position in the 3D contour map represents the vertical height of that point on the surface of the molten mixture relative to the inner surface plane of the transparent detection window.

[0051] In specific implementation, please refer to Figure 3 The connected component boundaries of each plastic particle are extracted from the plastic phase distribution marker map. The plastic phase distribution marker map is a binary image, where the pixel value corresponding to a plastic marker is 1, and the pixel value corresponding to an asphalt marker is 0. An eight-neighbor connected component labeling method is used to analyze the connected components of pixels with a pixel value of 1 in the plastic phase distribution marker map. Pixels that are spatially adjacent and both have a pixel value of 1 are merged into the same connected component, with each connected component corresponding to an independent plastic particle. The outer boundary contour of each connected component is extracted by scanning each pixel within the connected component. Pixels with plastic markers having at least one neighboring pixel with a pixel value of 0 in their four-neighbor or eight-neighbor areas are identified as boundary pixels. The coordinate sequence of these boundary pixels is traced and recorded in a clockwise direction, forming the closed connected component boundaries of the plastic particles.

[0052] The pixel coordinate systems of the plastic phase distribution marker map and the 3D contour map have been unified to the same coordinate space through multimodal image group spatial registration. The plastic phase distribution marker map is located in the coordinate space... The pixels are located at the same coordinates as the 3D contour map. Each pixel corresponds to the same physical point on the surface of the molten mixture. The coordinates of all boundary pixels contained in the connected domain boundary of each plastic particle are extracted one by one and directly mapped onto the 3D contour map according to the coordinate correspondence, thereby transferring the position information of the connected domain boundary of the plastic particle to the 3D contour map.

[0053] For each plastic particle's connected component boundary, the internal region enclosed by the boundary is determined. This internal region consists of all plastic marker pixels located inside the boundary that are not part of the boundary itself. Each pixel coordinate within the internal region is traversed, and the height value at the corresponding coordinate position is read from the 3D contour map to obtain the set of height values ​​for all pixels within the area covered by the plastic particle. Numerical comparisons are performed within this set of height values, and the maximum value is determined as the particle's vertex height.

[0054] In the plastic phase distribution map, asphalt regions adjacent to the boundaries of connected domains of plastic particles are selected. The selection method is as follows: on the plastic phase distribution map, each boundary of a connected domain of plastic particles is morphologically expanded using a circular structuring element, with a radius of 5 pixels. The expanded region is subtracted from the original connected domain region of plastic particles before expansion, resulting in a ring-shaped neighborhood surrounding the boundary of the connected domain of plastic particles. Pixels marked as asphalt (i.e., pixels with a value of 0) are selected from the ring-shaped neighborhood as the asphalt regions adjacent to the boundaries of the connected domains of plastic particles. The height values ​​of the asphalt regions at the corresponding coordinate positions in the 3D contour map are read to obtain a set of asphalt height values. All height values ​​in the asphalt height value set are sorted from smallest to largest, and the median of the sorted sequence is taken as the asphalt surface height. When the number of elements in the asphalt height value set is odd, the median is the single height value located in the middle of the sequence; when the number of elements in the asphalt height value set is even, the median is the arithmetic mean of the two middle height values ​​of the sequence.

[0055] The exposure height is obtained by subtracting the asphalt surface height from the particle apex height. The exposure height characterizes the vertical dimension of the plastic particle protruding above the surrounding asphalt matrix surface.

[0056] To mitigate the adverse effects of image noise and localized minor depressions on minimum point identification, statistical filtering is first applied to the height value set within the connected domain of the plastic particles. Abnormal pixels, which are statistical outliers, are removed from the set. These outliers originate from imaging noise and instantaneous minor pits on the molten surface. After removing these outliers, the height values ​​from the remaining valid height data are selected as the equivalent minimum point height for the plastic particle region. This equivalent minimum point height represents the lowest apparent position of the plastic particle on the free surface of the molten mixture, not the actual physical bottom surface of the plastic particle embedded within the asphalt matrix.

[0057] From the set of effective height values ​​after statistical filtering to remove outlier pixels, the height value is selected as the equivalent minimum point height of the area covered by plastic particles.

[0058] The embedding depth is obtained by subtracting the lowest point height from the asphalt surface height. This embedding depth is the apparent embedding depth. Since structured light vision inspection can only acquire the free surface morphology of the molten mixture's gas-liquid mixture, and light cannot penetrate asphalt material, it is impossible to directly obtain the true bottom position of the plastic particles embedded within the asphalt matrix. This apparent embedding depth reflects the degree of surface depression of the plastic particles at the gas-liquid interface caused by asphalt wetting. The apparent embedding depths of different particles can be compared to distinguish the relative strength of the particles' asphalt wetting and coating. Combining multiple parameters such as exposure height, equivalent particle size distribution, and melt state index, a comprehensive evaluation of the plastic particle dispersion state can be achieved without relying on the absolute measurement value of the particle's actual embedding depth within the asphalt. The embedding depth characterizes the vertical dimension of the portion of the plastic particle embedded in the asphalt matrix.

[0059] See Figure 5 In the figure, the horizontal axis represents the plastic particle number, ranging from 0 to 500, indicating that each of the 500 plastic particles in the molten mixture is numbered individually; the vertical axis represents the height, in millimeters, ranging from 0 to 6 millimeters. Blue dots represent the exposed height of each plastic particle, and orange triangles represent the embedding depth of each plastic particle.

[0060] As shown in the figure, the exposed height of all plastic particles ranges from approximately 0.5 mm to 5.1 mm, with most particles concentrated in the 1 mm to 4 mm range. The exposed height exhibits a relatively discrete distribution, reflecting the diversity of the plastic particles protruding from the asphalt matrix surface. The embedding depth ranges from approximately 2.5 mm to 5.5 mm, significantly higher than the exposed height, indicating that most plastic particles are embedded to a considerable depth in the asphalt matrix. The relatively concentrated embedding depth values ​​reflect the spatial stability of the embedded portion of the particles.

[0061] The distribution trends of exposure height and embedding depth in the figure are relatively independent. The exposure height data points have a wide vertical distribution range and significant differences, while the embedding depth data points are relatively concentrated and are always higher than the exposure height of the same index. This verifies the logic in the above embodiment that the exposure height is calculated by the difference between the particle vertex height and the asphalt surface height, and the embedding depth is calculated by the difference between the asphalt surface height and the lowest point height of the plastic particle.

[0062] In practice, based on the plastic phase distribution marker map, a set of plastic phase pixels is extracted from the visible light image sequence. The plastic phase distribution marker map is a binary image, with a pixel value of 1 corresponding to a plastic marker and a pixel value of 0 corresponding to an asphalt marker. A visible light image frame with the same timestamp as the plastic phase distribution marker map is selected from the visible light image sequence. All pixel coordinates in the plastic phase distribution marker map are traversed, and the pixel coordinates with a pixel value of 1 are mapped one by one to the visible light image. The red, green, and blue color channel component values ​​are extracted from the corresponding coordinate positions in the visible light image. All extracted pixel coordinates and their corresponding red, green, and blue color channel component values ​​are combined to form the plastic phase pixel set.

[0063] Hole filling is performed on the plastic phase pixels in the plastic phase pixel set. In the plastic phase distribution marker map, isolated small regions surrounded by bitumen markers may appear inside the plastic particles due to insignificant local absorption characteristics. These isolated small regions represent cavities inside the plastic particles. The specific implementation of cavity filling is as follows: the plastic phase distribution marker map is inverted, making the original plastic marker area the background and the original bitumen marker area the foreground. Connected component marking is performed in the inverted image. Isolated small connected components with an area smaller than a set threshold are identified as cavities. The threshold is set to 50 pixels, based on the fact that 50 pixels correspond to an actual surface area of ​​the molten mixture of approximately 0.2 square millimeters. Isolated areas smaller than this area value are caused by noise or local detection fluctuations rather than the true bitumen phase. The pixel value of the corresponding position of the identified cavity in the plastic phase distribution marker map is changed from 0 to 1 to obtain the plastic phase filling area.

[0064] Morphological erosion is performed on the plastic phase-filled region using a circular structuring element with a radius of 3 pixels. The circular structuring element is slid across the plastic phase-filled region pixel by pixel. When all pixels within the area covered by the circular structuring element have a value of 1, the value of the central pixel is retained; otherwise, the central pixel is set to 0, resulting in the eroded plastic phase region. The eroded plastic phase region shrinks by one ring of boundary pixels compared to the plastic phase-filled region.

[0065] The pixel values ​​of the filled plastic phase region and the etched plastic phase region are subtracted. This subtraction operation is performed pixel-by-pixel, for coordinates... The pixel at that location will fill the plastic phase region at coordinates The pixel value at that location minus the eroded plastic phase region at coordinates The pixel values ​​at each point are compared. In the difference image, pixels with a value of 1 form the closed boundary of each plastic particle, while pixels with a value of 0 represent the background or the interior of the particle. The closed boundary is spatially continuous and closed, completely surrounding each plastic particle.

[0066] An ellipse fitting is performed on the closed boundary using the least squares ellipse fitting method. For the closed boundary of each plastic particle, the coordinate sequence of all pixels on the closed boundary is extracted. This coordinate sequence is then input into the least squares ellipse fitting algorithm. The goal of the fitting algorithm is to minimize the sum of squared algebraic distances from all pixels on the closed boundary to the fitted ellipse. The fitting algorithm outputs five parameters of the fitted ellipse: the coordinates of the ellipse center. Major axis length minor axis length and rotation angle Major axis length To fit the maximum span along the principal axis of the ellipse, the minor axis length... To fit the maximum span of the ellipse perpendicular to the principal axis, the rotation angle is... To fit the deflection angle of the principal axis of the ellipse relative to the horizontal coordinate axis.

[0067] Major axis length With minor axis length The square root of the product of is used as the equivalent particle size. The equivalent particle size is calculated as follows:

[0068]

[0069] in, For equivalent particle size, To fit the length of the major axis of the ellipse, The length of the minor axis of the fitted ellipse. Equivalent particle size. Irregularly shaped plastic particles are transformed into circles of equal area, with the diameter characterizing the particle size. The equivalent particle size is calculated for each plastic particle detected in the molten mixture. The equivalent particle size values ​​of all plastic particles are arranged in ascending order. A statistical histogram is plotted with the equivalent particle size as the x-axis and the number of particles in the corresponding particle size interval as the y-axis. The group interval of the statistical histogram is set to 25 micrometers to generate the equivalent particle size distribution.

[0070] Extract the component values ​​of the plastic phase pixel set in the red, green, and blue color channels. For each plastic particle, obtain the component values ​​of all plastic phase pixels in the red channel within the connected component corresponding to that plastic particle. Component values ​​of the green channel and the component values ​​of the blue color channel The average red channel component value of the plastic particle is obtained by taking the arithmetic mean of the red channel component values ​​of all pixels within the connected component. The arithmetic mean of the green channel component values ​​of all pixels within the connected component is taken to obtain the average green channel component value of the plastic particle. The arithmetic mean of the blue channel component values ​​of all pixels within the connected component is taken to obtain the average blue channel component value of the plastic particle. In the three groups of average component values , , The numerical values ​​are compared, and the average component value with the largest value is selected as the largest color component. The average component value with the smallest value is selected as the smallest color component. . The largest color component With minimum color component The difference between the maximum color component and the maximum color component The melt state index of each plastic particle is obtained by performing ratio calculation.

[0071] In some implementations, the nearest neighbor distance is calculated based on the centroid coordinates of each plastic particle. The centroid coordinates are obtained by taking the arithmetic mean of the x-coordinates of all pixels within the connected component of each plastic particle. The ordinate of the centroid is obtained by taking the arithmetic mean of the ordinates of all pixels within the connected region of each plastic particle. The coordinates of the centroid of the plastic particle The coordinates of the centroids of all plastic particles are traversed pairwise. The Euclidean distance between the centroids of any two particles is calculated by summing the squares of the differences in their x-coordinates and y-coordinates, and then taking the square root. After calculating the Euclidean distance between each plastic particle and all other plastic particles, for each particle, the smallest value among all its associated Euclidean distances is selected as its nearest neighbor distance.

[0072] The numerical distribution intervals of exposure height, embedding depth, melt state index, and nearest neighbor distance for all plastic particles were statistically analyzed. The minimum, maximum, median, first quartile, and third quartile were calculated for each set of exposure height, embedding depth, melt state index, and nearest neighbor distance. The numerical distribution intervals of exposure height, embedding depth, melt state index, and nearest neighbor distance were correlated with the equivalent particle size distribution. The correlation method was as follows: using the equivalent particle size as the grouping variable, plastic particles within each particle size interval of the equivalent particle size distribution were grouped together. For each group of plastic particles, the numerical distribution intervals of exposure height, embedding depth, melt state index, and nearest neighbor distance within the group were statistically analyzed, forming structured data containing statistical characteristics of various dispersion state indicators under different particle size ranges, and generating dispersion state detection results.

[0073] See Figure 6 This figure shows the equivalent particle size distribution and kernel density estimation curves of plastic particles obtained through joint analysis of visible light image sequences and plastic phase distribution marker maps. The horizontal axis represents the equivalent particle size of the plastic particles in micrometers (µm), and the vertical axis represents the number of particles on the left and the probability density on the right. The blue bars in the figure represent the number distribution of plastic particles within different equivalent particle size intervals, with a statistical class interval of 25 µm; the red curve represents the corresponding kernel density estimation curve, used to smooth the probability density characteristics of the particle size distribution.

[0074] As observed in the figure, the equivalent particle size of the plastic granules is mainly distributed in the range of 50 micrometers to 250 micrometers, with the largest number of particles in the range of 75 micrometers to 100 micrometers, reaching a peak of approximately 100 particles. The number of plastic particles with an equivalent particle size of less than 50 micrometers is relatively small, indicating a scarcity of fine particles in this range. The number of larger particles exceeding 250 micrometers decreases significantly, and the particle number shows a rapid decreasing trend with increasing particle size, indicating a skewed distribution of particle size.

[0075] The nuclear density estimation curve and the histogram show a consistent overall trend. The peak of the curve corresponds to about 100 micrometers. After 100 micrometers, the curve gradually decreases and approaches zero, indicating that the probability distribution of particle size is concentrated in the small to medium particle size range and the distribution has a right-skewed characteristic.

[0076] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for detecting the production of modified asphalt from waste plastics based on machine vision inspection, characterized in that, Includes the following steps: At the transparent inspection window of the modified asphalt production pipeline, surface multispectral images and structured light stripe images of the molten mixture are acquired simultaneously to obtain visible light image sequences, near-infrared image sequences, and structured light stripe image sequences. The visible light image sequence, near-infrared image sequence, and structured light stripe image sequence are time-registered and spatially registered to generate a multimodal image group; Based on the absorption differences of plastic characteristic bands in the near-infrared image sequence, a plastic phase distribution marker map is generated; Phase demodulation is performed on the structured light stripe image sequence to generate a three-dimensional contour map of the surface of the molten mixture; The plastic phase distribution map is overlaid with the three-dimensional contour map to calculate the exposure height and embedding depth. Based on the edge contours and color components of the plastic phase region in the visible light image sequence, the equivalent particle size distribution and melt state index are calculated. Based on the exposure height, embedding depth, equivalent particle size distribution, and melt state index, dispersion state detection results are generated.

2. The method for detecting the production of modified asphalt from waste plastics based on machine vision inspection according to claim 1, characterized in that, The step of performing temporal and spatial registration on the visible light image sequence, near-infrared image sequence, and structured light fringe image sequence to generate a multimodal image group specifically includes: The acquisition times of the visible light image sequence, near-infrared image sequence, and structured light stripe image sequence are controlled by the same hardware trigger signal, so that the three are exposed at the same time point. The spatial transformation parameters between the multispectral imaging device and the structured light projection device are obtained using a calibration plate, and the visible light image sequence, near-infrared image sequence and structured light stripe image sequence are mapped to the same coordinate space. The three mapped images are aligned according to pixel coordinates to generate the multimodal image group.

3. The method for detecting the production of modified asphalt from waste plastics based on machine vision inspection according to claim 1, characterized in that, The step of generating a plastic phase distribution marker map based on the absorption differences of plastic characteristic bands in the near-infrared image sequence specifically includes: Intensity values ​​of two characteristic bands are extracted from the near-infrared image sequence, wherein the first characteristic band corresponds to the carbon-hydrogen bond absorption peak of polyolefin, and the second characteristic band corresponds to the reference wavelength of the asphalt matrix. Calculate the ratio of the intensity values ​​of each pixel in the first feature band to the second feature band to obtain a ratio image; Pixels with a ratio greater than 1 in the ratio image are assigned a plastic label, and pixels with a ratio less than or equal to 1 are assigned an asphalt label, thereby generating the plastic phase distribution label map.

4. The method for detecting the production of modified asphalt from waste plastics based on machine vision inspection according to claim 1, characterized in that, The step of performing phase demodulation on the structured light stripe image sequence to generate a three-dimensional contour map of the surface of the molten mixture specifically includes: Each frame of the structured light stripe image sequence is decomposed into a fundamental frequency component and a harmonic frequency component, and the principal phase value of the stripes is extracted. The phase increment is obtained by differentiating the principal phase values ​​corresponding to two adjacent fringe patterns. The phase increments are accumulated to generate a continuous phase distribution; The three-dimensional contour map is generated based on the continuous phase distribution and the pre-calibrated phase height mapping relationship.

5. The method for detecting the production of modified asphalt from waste plastics based on machine vision inspection according to claim 1, characterized in that, The step of overlaying the plastic phase distribution map with the three-dimensional contour map and calculating the exposure height and embedding depth specifically includes: Extract the connected domain boundaries of each plastic particle from the plastic phase distribution marker map, and map the connected domain boundaries onto the three-dimensional contour map; Read the height value of each pixel within the boundary of the connected component, and take the maximum value among the height values ​​as the particle vertex height; In the plastic phase distribution marking map, select the asphalt region adjacent to the boundary of the connected domain, read the height value of the asphalt region, and take the median of the height values ​​of the asphalt region as the asphalt surface height. The exposure height is obtained by subtracting the asphalt surface height from the particle vertex height. The embedding depth is obtained by subtracting the minimum height value of each pixel within the boundary of the connected domain from the height of the asphalt surface.

6. The method for detecting the production of modified asphalt from waste plastics based on machine vision inspection according to claim 1, characterized in that, The step of calculating the equivalent particle size distribution and melt state index based on the edge contour and color components of the plastic phase region in the visible light image sequence specifically includes: Based on the plastic phase distribution marker map, extract the set of plastic phase pixels from the visible light image sequence; Edge detection is performed on the set of plastic phase pixels to obtain the closed boundary of each plastic particle; An ellipse is fitted to the closed boundary, and the lengths of the major and minor axes of the fitted ellipse are calculated. The square root of the product of the major axis length and the minor axis length is taken as the equivalent particle size. The equivalent particle size of all plastic particles is counted to generate the equivalent particle size distribution. Extract the component values ​​of the plastic phase pixel set in the red, green and blue color channels, and calculate the ratio of the difference between the maximum color component and the minimum color component to the maximum color component to obtain the melting state index of each plastic particle.

7. The method for detecting the production of waste plastic modified asphalt based on machine vision inspection according to claim 6, characterized in that, The step of performing edge detection on the set of plastic phase pixels to obtain the closed boundary of each plastic particle specifically includes: Hole filling is performed on the plastic phase pixels in the plastic phase pixel set to obtain a filled plastic phase region; Morphological etching is performed on the filled plastic phase region to obtain the etched plastic phase region; The closed boundary is obtained by subtracting the pixel value of the filled plastic phase region from that of the corroded plastic phase region.

8. The method for detecting the production of modified asphalt from waste plastics based on machine vision inspection according to claim 1, characterized in that, The generation of dispersion state detection results based on the exposure height, embedding depth, equivalent particle size distribution, and melt state index specifically includes: Based on the centroid coordinates of each plastic particle, calculate the Euclidean distance between the centroids of any two plastic particles, and select the distance between each particle and its nearest neighbor as the nearest neighbor distance. The numerical distribution range of the exposure height, embedding depth, melt state index, and nearest neighbor distance of all plastic particles is statistically analyzed. The numerical distribution range is correlated with the equivalent particle size distribution to generate the dispersion state detection result.

9. The method for detecting the production of modified asphalt from waste plastics based on machine vision inspection according to claim 1, characterized in that, The simultaneous acquisition of surface multispectral images and structured light stripe images of the molten mixture at the transparent detection window of the modified asphalt production pipeline specifically includes: A multispectral imaging device and a structured light projection device are installed on the outside of the transparent detection window, such that the optical axis of the multispectral imaging device is perpendicular to the outer surface of the transparent detection window, and the projection direction of the structured light projection device forms a set angle with the optical axis of the multispectral imaging device. Before synchronous acquisition, the inner surface of the transparent detection window is swept with compressed air, and the temperature of the inner surface of the transparent detection window is heated to the same temperature as the molten mixture. The multispectral imaging device and the structured light projection device are activated to perform synchronous data acquisition.

10. The method for detecting the production of waste plastic modified asphalt based on machine vision inspection according to claim 4, characterized in that, In the step of generating the three-dimensional contour map based on the continuous phase distribution and the pre-calibrated phase height mapping relationship, the pre-calibrated phase height mapping relationship is established through the following steps: A calibration block with multiple known height steps is placed at the transparent detection window; The striated image of the calibration block in the structured light striated image sequence is acquired, and the corresponding data points of the continuous phase value and actual height of each step are obtained. Polynomial fitting is performed on the corresponding data points to generate the phase height mapping relationship.