A method and system for identifying and classifying components of recycled aggregate from waste concrete

Abstract

This invention discloses a method and system for identifying and classifying the components of recycled aggregates from waste concrete. The method includes: acquiring multispectral image sequences under different spectral bands for the recycled orthopedic sample to be tested; calculating the local texture entropy in four directions for each pixel in the near-infrared band image according to a set of scale parameters to form a family of entropy change curves; dividing the family of entropy change curves into hardened cement stone region, natural aggregate region, and impurity region according to a preset scale parameter threshold; classifying the hardened cement stone region into closed-pore hardened cement stone and open-pore hardened cement stone by combining the fluorescence intensity value of the ultraviolet-excited fluorescence band image; classifying the natural aggregate region into siliceous natural aggregate and calcareous natural aggregate using reflectance values; and forming a visualized aggregate distribution map using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity region.

Description

Technical Field

[0001] This invention relates to the field of concrete aggregate testing, and in particular to a method and system for identifying and classifying the components of recycled aggregates from waste concrete. Background Technology

[0002] With the continuous development of the construction industry and the increasing scarcity of resources and the environment, the recycling and reuse of construction waste has become an important research topic. Concrete is one of the most common materials in building structures, and a large amount of waste concrete is generated during building demolition or reconstruction. How to effectively recycle this waste concrete and convert it into recycled aggregate has become a key technology for improving resource utilization efficiency and reducing environmental pollution. However, the properties of recycled aggregate from waste concrete are usually affected by a variety of factors, including its composition, structure, and porosity, which directly affect the performance and effectiveness of the recycled aggregate. In order to improve the utilization efficiency of recycled aggregate, accurate component identification and classification are essential.

[0003] Currently, the compositional analysis of recycled concrete aggregates mainly relies on traditional experimental methods, such as chemical analysis and microscopic observation. These methods are not only time-consuming and labor-intensive, but also complex to operate and difficult to process large batches of samples efficiently. Therefore, developing a novel, rapid, and efficient method for component identification and classification has become an urgent technical problem to be solved.

[0004] In recent years, with the continuous advancement of spectral imaging technology, component analysis methods based on multispectral images have gradually gained widespread attention. Multispectral imaging technology can acquire image information in different spectral bands, and by analyzing the spectral features in the images, it is possible to effectively distinguish the components of different materials. However, existing multispectral imaging technologies are usually limited to a single spectral band or lack sufficient feature analysis, making it difficult to achieve accurate identification and classification of multiple components in recycled concrete aggregates. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for identifying and classifying the components of recycled aggregates from waste concrete, which solves the aforementioned technical problems pointed out in the prior art.

[0006] This invention provides a method for identifying and classifying the components of recycled aggregates from waste concrete, the specific steps of which include:

[0007] Multispectral image sequences under different spectral bands were obtained for the regenerative orthopedic samples to be tested.

[0008] The multispectral image sequence includes visible light band images, near-infrared band images, and ultraviolet excited fluorescence band images;

[0009] For each pixel in the near-infrared image, the local texture entropy in four directions is calculated according to a set of scale parameters to form a family of entropy change curves. Based on a preset scale parameter threshold, the family of entropy change curves is divided into hardened cement stone regions, natural aggregate regions, and impurity regions. The hardened cement stone regions are further divided into closed-pore hardened cement stone and open-pore hardened cement stone by combining the fluorescence intensity values ​​of the ultraviolet-excited fluorescence image. The visible light image and the near-infrared image are combined using preset feature band combinations to obtain reflectance values. The natural aggregate regions are then divided into siliceous natural aggregate and calcareous natural aggregate. Finally, a heterogeneous component distribution vector is formed using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity regions.

[0010] A visual aggregate distribution map is constructed using the heterogeneous component distribution vector to perform visual classification of the regenerative orthopedic samples to be tested.

[0011] Preferably, the local texture entropy in four directions is calculated for each pixel in the near-infrared image according to a set of scale parameters, forming a family of entropy change curves; the family of entropy change curves is then divided into hardened cement stone regions, natural aggregate regions, and impurity regions according to a preset scale parameter threshold. The specific operation steps are as follows:

[0012] A set of scale parameters is set for the near-infrared band images in the multispectral image sequence; a local window is obtained for each pixel in the near-infrared band image as the center, according to each scale parameter in the scale parameter set and four directions; the gray-level co-occurrence matrix of the pixel in each direction within the local window is calculated; the gray-level co-occurrence matrix is ​​normalized to obtain the probability matrix;

[0013] Calculate the local texture entropy of the local window at the given scale parameter and orientation using each unit element in the probability matrix;

[0014] Sort the local texture entropy of all scale parameters of the current pixel in ascending order according to the current direction to obtain the entropy change curve of that direction; obtain all entropy change curves of the current pixel in the four directions respectively to obtain the entropy change curve family.

[0015] For each entropy change curve, calculate the first-order difference of the scale parameter arranged in ascending order to obtain a numerical derivative sequence; select the extreme points of the maximum and minimum numerical derivatives in the numerical derivative sequence; select the scale parameter corresponding to the first non-zero extreme point of all extreme points as the characteristic scale parameter in that direction;

[0016] The comprehensive feature scale parameters are obtained by taking a weighted average of the feature scale parameters in all directions.

[0017] Preset the first scale parameter threshold and the second scale parameter threshold; determine the relationship between the comprehensive feature scale parameter of the current pixel and the first scale parameter threshold and the second scale parameter threshold;

[0018] If the comprehensive feature scale parameter is less than the first scale parameter threshold, the pixel is determined to be a high-porosity texture region and marked as a hardened cement stone region; if the comprehensive feature scale parameter is greater than the second scale parameter threshold, the pixel is determined to be a large-scale high-entropy region and marked as a natural aggregate region; if the comprehensive feature scale parameter is greater than or equal to the first scale parameter threshold and less than or equal to the second scale parameter threshold, the pixel is determined to be an impurity region.

[0019] Preferably, the hardened cement stone region is classified into closed-pore type hardened cement stone and open-pore type hardened cement stone by combining the fluorescence intensity values ​​of the ultraviolet-excited fluorescence band image. The specific operation steps are as follows:

[0020] Obtain the comprehensive feature scale parameters of the pixels in the high-porosity texture region, find the closest scale parameter in the set of scale parameters, and extract the local texture entropy in the four directions of the closest scale parameter to form a direction entropy vector;

[0021] The maximum and minimum values ​​in the directional entropy vector are selected, and the extreme value ratio of directional entropy is calculated. The fluorescence intensity value of the pixel in the high-porosity texture region at the corresponding coordinate position in the ultraviolet-excited fluorescence band image is obtained. The reflectance value of the pixel in the high-porosity texture region at a preset characteristic wavelength in the near-infrared band image is obtained. The ratio of the fluorescence intensity value to the reflectance value of the pixel in the high-porosity texture region is calculated as the fluorescence excitation efficiency index.

[0022] An anisotropy threshold and an efficiency threshold are preset. If the extreme value ratio of the directional entropy of a pixel in the high-porosity texture region is greater than the anisotropy threshold and the fluorescence excitation efficiency index is less than the efficiency threshold, then the pixel is determined to be closed-pore hardened cement stone. If the extreme value ratio of the directional entropy of a pixel in the high-porosity texture region is less than or equal to the anisotropy threshold and the fluorescence excitation efficiency index is greater than or equal to the efficiency threshold, then the pixel is determined to be open-porosity hardened cement stone. If neither of the above determination conditions is met, then the pixel is determined to be ordinary hardened cement stone.

[0023] Preferably, the reflectance is obtained by combining the visible light band image and the near-infrared band image using a preset feature band combination; the reflectance value is used to divide the natural aggregate region into siliceous natural aggregate and calcareous natural aggregate; a heterogeneous component distribution vector is formed using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity region. The specific operation steps are as follows:

[0024] For the visible light and near-infrared images, reflectance values ​​are extracted using preset feature band combinations. Based on the reflectance values ​​of different feature band combinations, the siliceous and calcareous natural aggregates to which the pixels belong are identified. A sliding window process is applied to the recycled aggregate sample to be tested, and the reflectance change rate and texture entropy gradient of the pixels within the sliding window are calculated. The spatial phase difference is calculated based on the reflectance change rate and texture entropy gradient. The mean and standard deviation of the spatial phase difference are calculated, and a dynamic threshold baseline value is obtained. The recycled aggregate sample to be tested is sieved to obtain its nominal maximum particle size, and the dynamic threshold baseline value is corrected based on the nominal maximum particle size to obtain the final dynamic threshold. A heterogeneous interface response intensity map is generated for each pixel based on the reflectance change rate and texture entropy gradient. The heterogeneous interface response intensity map is binarized using the final dynamic threshold to obtain the boundary line. Based on the boundary line division results, the heterogeneous component distribution vectors of the natural aggregate region, hardened cement stone region, and impurity region are obtained.

[0025] Preferably, for the visible light band image and the near-infrared band image, reflectance values ​​are extracted through preset feature band combinations, and the siliceous natural aggregate or calcareous natural aggregate to which the pixel belongs is identified based on the reflectance values ​​of different feature band combinations. The specific operation steps are as follows:

[0026] The visible light band image and the near-infrared band image are combined using a preset feature band combination pair;

[0027] For each characteristic band combination pair, reflectance values ​​are extracted from the corresponding visible light band image and near-infrared band image respectively; the ratio between the reflectance values ​​of the visible light band image and the near-infrared band image is calculated, and the ratios between all calculated reflectance values ​​are formed into a band ratio vector; the band ratio vector is projected onto the CIELAB color space to obtain the chromaticity coordinates in the CIELAB color space;

[0028] Obtain the centers of the standard chromaticity clusters of siliceous minerals and calcareous minerals from the pre-constructed standard database; calculate the first Euclidean distance between the chromaticity coordinates and the centers of the standard chromaticity clusters of siliceous minerals; calculate the second Euclidean distance between the chromaticity coordinates and the centers of the standard chromaticity clusters of calcareous minerals;

[0029] Multiple visible light characteristic wavelengths are preset, and the reflectance value of the pixel is obtained according to the preset multiple visible light characteristic wavelengths; the ratio between the reflectance values ​​of different visible light characteristic wavelengths is calculated as the feature ratio; all feature ratios are combined to obtain the visible light feature ratio vector.

[0030] The fluorescence excitation efficiency index, the band ratio vector, and the visible light characteristic ratio vector are standardized and then combined to obtain the multimodal spectral characteristic vector.

[0031] Preset deviation threshold and fluorescence threshold; if the first Euclidean distance of the pixel is less than the deviation threshold and the fluorescence excitation efficiency index is less than the fluorescence threshold, then the pixel is determined to be siliceous natural aggregate; if the second Euclidean distance of the pixel is less than the deviation threshold and the fluorescence excitation efficiency index is greater than or equal to the fluorescence threshold, then the pixel is determined to be calcareous natural aggregate.

[0032] Preferably, a sliding window process is applied to the recycled aggregate sample to be tested, and the reflectance change rate and texture entropy gradient of the pixels in the sliding window are calculated; the spatial phase difference is calculated based on the reflectance change rate and texture entropy gradient; the mean and standard deviation of the spatial phase difference are calculated, and then the dynamic threshold reference value is calculated; the recycled aggregate sample to be tested is sieved to obtain its nominal maximum particle size, and the dynamic threshold reference value is corrected based on the nominal maximum particle size to obtain the final dynamic threshold. The specific operation steps are as follows:

[0033] A sliding window is applied to the regenerative orthopedic sample to be tested. Each sliding window is uniformly divided into grids to generate grid points. N grid points are randomly selected from each sliding window as the center grid points, and their coordinate positions are determined. The first-order gradient vector of reflectance values ​​is calculated for all pixels in each sliding window as the reflectance change rate. The second-order gradient tensor of local texture entropy is calculated for all pixels in each sliding window as the texture entropy gradient. The spatial phase difference is calculated based on the reflectance change rate and the texture entropy gradient. The mean and standard deviation of the spatial phase difference of pixels within each sliding window are calculated.

[0034] For each sliding window, obtain the coordinates of the window center point and the coordinates of the image center point of the regenerative orthopedic sample to be tested; calculate the central Euclidean distance between the coordinates of the window center point and the coordinates of the image center point.

[0035] The weighting coefficients are calculated using the central Euclidean distance;

[0036] The weighted average and the weighted average standard deviation are obtained by using the weighting coefficients to calculate the mean and standard deviation of the spatial phase difference of all sliding windows, respectively.

[0037] The dynamic threshold benchmark value is calculated using the weighted average and the weighted average standard deviation.

[0038] The nominal maximum particle size of the recycled aggregate sample to be tested was determined by sieving.

[0039] A preset monotonically decreasing function is used; the particle size adjustment coefficient is calculated using the nominal maximum particle size and the monotonically decreasing function; and the surface roughness parameters of the recycled aggregate sample to be tested are obtained.

[0040] A preset monotonically increasing function is used; the surface roughness parameter and the monotonically increasing function are used to calculate the roughness correction coefficient; the final dynamic threshold is calculated using the dynamic threshold reference value, the particle size adjustment coefficient, and the roughness correction coefficient.

[0041] Preferably, the recycled aggregate sample to be tested is sieved to obtain its nominal maximum particle size, and the dynamic threshold reference value is corrected according to the nominal maximum particle size to obtain the final dynamic threshold; a heterogeneous interface response intensity map is generated for each pixel based on the reflectance change rate and texture entropy gradient; the heterogeneous interface response intensity map is binarized using the final dynamic threshold to obtain the boundary line; based on the boundary line division result, the heterogeneous component distribution vectors of the natural aggregate region, hardened cement stone region, and impurity region are obtained. The specific operation steps are as follows:

[0042] The heterogeneous interface response intensity is calculated for each pixel based on the reflectance change rate and texture entropy gradient, generating a heterogeneous interface response intensity map. The map is then binarized and subjected to morphological operations using the final dynamic threshold to obtain the boundary line. The number of pixels within the natural aggregate region, hardened cement stone region, and impurity region is counted based on the boundary line's range, and the cosine value of the included angle is calculated. The area fractions of the natural aggregate region, hardened cement stone region, and impurity region are obtained based on the cosine value of the included angle. A heterogeneous component distribution vector is constructed based on each area fraction.

[0043] The cosine value of the included angle includes the cosine value of the silicon included angle, the cosine value of the calcium included angle, and the cosine value of the sealing and hardening included angle;

[0044] The area fraction includes the area fraction of siliceous natural aggregate, the area fraction of calcareous natural aggregate, the area fraction of sealed hardened cement stone, the area fraction of impurities, and the area fraction of the high water absorption risk subclass.

[0045] The area fraction of the hardened cement stone region is divided into the area fraction of the closed hardened cement stone and the area fraction of the high water absorption risk subclass. In this scheme, the area fraction is not equal to the area. The area fraction represents the proportion of each material.

[0046] Preferably, the heterogeneous interface response intensity is calculated for each pixel based on the reflectance change rate and texture entropy gradient to generate a heterogeneous interface response intensity map; the heterogeneous interface response intensity map is then binarized and subjected to morphological operations using the final dynamic threshold to obtain the boundary line. The specific operation steps are as follows:

[0047] The multimodal feature response value is obtained by multiplying the reflectance change rate with the texture entropy gradient.

[0048] The pre-defined principal direction angle of the gradient vector of reflectivity change rate is The principal direction angle of the texture entropy gradient vector is Calculate the directional difference; multiply the multimodal characteristic response value with the same directional coefficient to obtain the heterogeneous interface response intensity and generate a heterogeneous interface response intensity map;

[0049] The heterogeneous interface response intensity map is binarized using the final dynamic threshold to obtain a candidate heterogeneous interface pixel binary map. Morphological opening operations are performed on the pixels in the candidate heterogeneous interface pixel binary map according to a preset pixel radius to obtain a morphologically opened candidate heterogeneous interface pixel binary map. Then, a morphological closing operation is performed on the morphologically opened candidate heterogeneous interface pixel binary map using a preset cross-shaped structure window to obtain a repaired boundary binary map.

[0050] Pixels are extracted from the binary boundary map of the repaired area to form the boundary lines between the natural aggregate area, the hardened cement stone area, and the impurity area.

[0051] Preferably, based on the extent of the boundary line, the number of pixels within the natural aggregate region, the hardened cement stone region, and the impurity region is counted, and the cosine value of the included angle is calculated; based on the cosine value of the included angle, the area fractions of the natural aggregate region, the hardened cement stone region, and the impurity region are obtained; and a heterogeneous component distribution vector is constructed based on each area fraction. The specific operation steps are as follows:

[0052] Based on the boundary line division, the number of pixels inside the natural aggregate region, the hardened cement stone region, and the impurity region is counted; the standard siliceous aggregate spectral vector and the standard calcareous aggregate spectral vector of each pixel are obtained from the standard database.

[0053] The cosine value of the silica angle is calculated using the multimodal spectral feature vector of the pixel and the standard siliceous aggregate spectral vector; the cosine value of the calcium angle is calculated using the multimodal spectral feature vector of the pixel and the standard calcareous aggregate spectral vector.

[0054] Obtain the standard hardened cement stone spectral vector for each pixel from the standard database;

[0055] The number of pixels in the hardened cement stone region with closed pores is counted; the closed pore correction factor is calculated by using the extreme ratio of the directional entropy of the pixels in the hardened cement stone with closed pores and a preset correction intensity coefficient; the closed pore correction factor is used to correct the multimodal spectral feature vector to obtain the corrected multimodal spectral feature vector.

[0056] The cosine value of the closed hardening angle is calculated by using the corrected multimodal spectral feature vector and the standard hardened cement stone spectral vector of each pixel in the hardened cement stone region; the standard impurity spectral vector of each pixel is obtained from the standard database; and the cosine value of the impurity angle is calculated by using the standard impurity spectral vector and the multimodal spectral feature vector.

[0057] Preferably, the area fractions of the natural aggregate region, the hardened cement stone region, and the impurity region are obtained based on the cosine value of the included angle; a heterogeneous component distribution vector is constructed based on each area fraction. The specific operation steps are as follows:

[0058] A preset matching threshold is set; the area fractions of siliceous natural aggregate, calcareous natural aggregate, sealed hardened cement stone, and impurities are calculated by using the cosine values ​​of all the included angles mentioned above with the matching threshold.

[0059] Obtain the number of pixels in the open-pore type hardened cement stone;

[0060] The ratio between the number of pixels in the open-pore type hardened cement stone and the number of pixels in the hardened cement stone region is calculated and used as the area fraction of the high water absorption risk subclass.

[0061] The area fractions of siliceous natural aggregate, calcareous natural aggregate, sealed hardened cement stone, impurities, and high water absorption risk subclasses were constructed as heterogeneous component distribution vectors.

[0062] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0063] Analysis of the above-mentioned method and system for identifying and classifying recycled aggregates from waste concrete provided by this invention reveals that, in practical applications, this method first acquires multispectral images in the visible, near-infrared, and ultraviolet-excited fluorescence bands, and then performs a comprehensive analysis by combining information from different bands. The scheme effectively distinguishes hardened cement stone, natural aggregates, and impurity areas by calculating the local texture entropy in four directions through texture analysis of the near-infrared band images. Simultaneously, it further classifies closed-pore hardened cement stone and open-pore hardened cement stone using the fluorescence intensity values ​​of the ultraviolet fluorescence images, and classifies natural aggregates by combining the reflectance characteristics of the visible and near-infrared bands.

[0064] Furthermore, the advantage of this scheme lies in its ability to efficiently distinguish different material regions, including the pore type of hardened cement stone and the type of natural aggregate, through comprehensive analysis of multi-band images. The distinction between closed-pore and open-pore hardened cement stone relies on the response of ultraviolet excitation light to different pore structures, significantly improving the accuracy of identification. The classification of natural aggregates is achieved through the absorption characteristics in the near-infrared band and the reflectivity characteristics in the visible light band, effectively differentiating between siliceous and calcareous aggregates.

[0065] Furthermore, this scheme not only improves classification accuracy but also provides detailed quantitative information for the microstructure analysis of recycled aggregates, such as pore type and the ratio of siliceous to calcareous aggregates, providing a scientific basis for durability assessment and quality control of recycled aggregates. By constructing a visualized aggregate distribution map, users can intuitively see the spatial distribution of different material regions, providing a clear image for subsequent microstructure analysis and greatly improving the efficiency and accuracy of concrete recycled aggregate analysis. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the main process of identifying and classifying recycled aggregates from waste concrete, as described in Example 1.

[0067] Figure 2 This is a flowchart illustrating the classification of hardened cement stone in a waste concrete recycled aggregate component identification and classification method, as described in Example 1.

[0068] Figure 3 This is a schematic diagram of the entropy change curve family of a waste concrete recycled aggregate composition identification and classification method in Example 1;

[0069] Figure 4 This is a schematic diagram illustrating the classification of hardened cement stone in a waste concrete recycled aggregate component identification and classification method according to Example 1.

[0070] Figure 5 This is a flowchart of the dynamic threshold for a waste concrete recycled aggregate composition identification and classification method in Example 1;

[0071] Figure 6 The band diagram is shown in Example 1, illustrating a method for identifying and classifying recycled aggregates from waste concrete.

[0072] Figure 7 This is a flowchart of the heterogeneous component distribution vector of a waste concrete recycled aggregate component identification and classification method in Example 1.

[0073] Figure 8 This is a flowchart of the heterogeneous component distribution vector of a waste concrete recycled aggregate component identification and classification system according to Example 2.

[0074] Labels: Acquisition module 10; Analysis module 20; Identification module 30. Detailed Implementation

[0075] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0076] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0077] Example 1

[0078] like Figure 1 As shown, this invention provides a method for identifying and classifying the components of recycled aggregates from waste concrete. The specific steps include:

[0079] S10: Obtain multispectral image sequences under different spectral bands for the regenerative orthopedic sample to be tested;

[0080] The multispectral image sequence includes visible light band images, near-infrared band images, and ultraviolet excited fluorescence band images;

[0081] It should be noted that the recycled aggregate sample to be tested is crushed and reduced to obtain a representative particle sample (that is, the recycled aggregate sample to be tested is composed of countless independent and complete aggregate particles); the sample is evenly and flatly laid on the stage to ensure that the sample surface is relatively flat and covers the field of view of the imaging system; in the image acquisition and analysis control software, three acquisition sequences are set sequentially: the visible light band set (i.e., setting the wavelength range (e.g., 400-700 nm) and sampling interval (e.g., acquiring one image every 10 nm, for a total of 31 images), the near-infrared band set (i.e., setting the wavelength range (e.g., 900-1700 nm) and sampling interval (e.g., acquiring one image every 20 nm, for a total of 41 images), and the ultraviolet excitation band set (i.e., setting the wavelength of the ultraviolet excitation source (e.g., 365 nm) and setting the range of fluorescence emission bands to be acquired (e.g., 400-700 nm)). (nm), which can be selected to acquire a single wideband fluorescence image or multiple narrowband fluorescence images, and the automatic acquisition program is started; the system first switches to the visible light source, and sequentially traverses the preset visible light bands. Each time a band is switched, the camera is exposed once, the image is recorded and saved as raw data; after the visible light acquisition is completed, the system switches to the near-infrared source and repeats the above traversal acquisition process; finally, the system turns on the ultraviolet excitation source and turns off other light sources, and performs acquisition under the set fluorescence emission band. All acquired images are preprocessed such as noise reduction to obtain a multispectral image sequence;

[0082] S20: Calculate the local texture entropy in four directions for each pixel in the near-infrared image according to a set of scale parameters, forming a family of entropy change curves; divide the family of entropy change curves into hardened cement stone region, natural aggregate region, and impurity region according to a preset scale parameter threshold; divide the hardened cement stone region into closed-pore hardened cement stone and open-pore hardened cement stone by combining the fluorescence intensity value of the ultraviolet-excited fluorescence band image; obtain reflectance value by combining the visible light band image and the near-infrared band image with preset feature bands; divide the natural aggregate region into siliceous natural aggregate and calcareous natural aggregate using the reflectance value; form a heterogeneous component distribution vector using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity region.

[0083] It should be noted that local texture entropy is calculated for each pixel in four directions (usually horizontal, vertical, and diagonal). A family of entropy change curves is formed using set scale parameters (such as different window sizes) to record the changes in texture complexity at different scales. Based on this family of entropy change curves, material regions are divided. According to a preset scale threshold, the image is divided into hardened cement stone regions (relatively uniform texture, small entropy changes), natural aggregate regions (more complex texture, large entropy changes), and impurity regions (textures different from both aggregate and cement stone, possibly random). The hardened cement stone pore type classification combines ultraviolet-excited fluorescence images, using fluorescence intensity values ​​to distinguish between closed-pore hardened cement stone (weak or uniform fluorescence) and open-pore hardened cement stone (strong fluorescence or bright spots). The natural aggregate type classification utilizes the reflectance characteristics of the visible and near-infrared bands to further divide the natural aggregate region into siliceous and calcareous natural aggregates. This system integrates the distribution information of each material type and pore type into a vector representation to describe the proportion and distribution of each heterogeneous component within the sample. Different material regions exhibit varying texture complexity. There are differences. Closed-pore hardened cement stone has closed pores, making fluorescent substances inaccessible, resulting in weak or uniform fluorescence intensity. Open-pore hardened cement stone has interconnected pores, allowing fluorescent substances to be excited, resulting in strong fluorescence intensity or obvious bright spots. This distinction is based on the principle that open-pore structures allow ultraviolet excitation light to enter the pores, exciting fluorescent substances (such as unhydrated cement particles, admixtures, etc.) on the pore walls or within the pores, producing a strong fluorescence signal; while closed pores block the entry of excitation light, resulting in a weaker fluorescence signal. For those already identified... In the natural aggregate region, the reflectance characteristics of the visible and near-infrared bands are used to further distinguish aggregate types. Siliceous natural aggregates have specific absorption characteristics in the near-infrared band, and their fluorescence signals are usually weak. Calcareous natural aggregates have different reflectance combinations in the visible and near-infrared bands, and may produce weak fluorescence due to impurities or surface adsorption. This differentiation method is based on the following principle: the difference in mineral composition between siliceous and calcareous aggregates leads to different spectral reflectance characteristics. By pre-setting the reflectance combination of characteristic bands, the two types of aggregates can be effectively distinguished.

[0084] This invention vectorizes information about different components, providing a unified data structure for quantitative analysis. It accurately divides hardened cement stone, natural aggregate, and impurity regions, further refining the pore type and aggregate type of hardened cement stone. This provides detailed quantitative information for the microstructure analysis of recycled aggregate, supporting durability assessment (pore type) and component area fraction calculation (siliceous / calcareous aggregate ratio), forming a data structure that can be directly used for visualization and subsequent analysis.

[0085] S30: Construct a visual aggregate distribution map using the heterogeneous component distribution vector to perform visual classification of the regenerative orthopedic sample to be tested.

[0086] It should be noted that by analyzing the distribution of closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity areas, a visual image is formed. This allows for a visual view of which areas are closed-pore hardened cement stone and which areas are calcareous natural aggregate, etc.

[0087] Each material type is assigned a different color or label: closed-pore cement stone is light gray, open-pore cement stone is dark gray or blue, siliceous aggregate is red, calcareous aggregate is yellow, and impurities are green. A color image is output, which can be used to visually view the spatial distribution of materials. The solution uses images to view the distribution of different material regions, pore structure, and aggregate type combinations, which is convenient for evaluating the uniformity of the sample's microstructure.

[0088] like Figure 2 As shown, specifically, in step S20, the local texture entropy in four directions is calculated for each pixel in the near-infrared band image according to a set of scale parameters, forming a family of entropy change curves; the family of entropy change curves is divided into hardened cement stone region, natural aggregate region, and impurity region according to a preset scale parameter threshold; the hardened cement stone region is divided into closed-pore hardened cement stone and open-pore hardened cement stone by combining the fluorescence intensity value of the ultraviolet-excited fluorescence band image; the visible light band image and the near-infrared band image are combined with a preset feature band pair to obtain reflectance values; the natural aggregate region is divided into siliceous natural aggregate and calcareous natural aggregate using the reflectance values; a heterogeneous component distribution vector is formed using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity region. The specific operation steps are as follows:

[0089] S21: Set a set of scale parameters for the near-infrared band images in the multispectral image sequence (i.e., S={3,5,7,9,11,13,15} (i.e., unit pixels at different scales), corresponding to the window half-width or step size of the gray-level co-occurrence matrix (GLCM).

[0090] Centered on each pixel in the near-infrared band image, a local window is obtained for each pixel according to each scale parameter in the scale parameter set and four directions (i.e., θ∈{0°,45°,90°,135°}). (i.e., the scale parameter s represents the half-width of the window (i.e., the window side length is 2s+1 pixels). When s=3, the window size is 7×7 pixels; when s=15, the window size is 31×31 pixels.)

[0091] Calculate the gray-level co-occurrence matrix of the pixel in each direction within the local window. (i.e., each unit element in the matrix) This represents the number of times the pixel pair with gray levels i and j appears in the specified direction (i.e., the four directions mentioned above) and at a distance (distance being the scale parameter S); that is, within this window, with a stride of d = 1, the gray-level co-occurrence matrix is ​​calculated. ; Each element in This represents the number of occurrences of a pair of pixels, separated by one pixel in direction θ, with gray values ​​i and j respectively.

[0092] Normalize the gray-level co-occurrence matrix to obtain the probability matrix (i.e., each unit element in the probability matrix). This is represented by the element value in the i-th row and j-th column of the normalized gray-level co-occurrence matrix; that is, the joint probability of two pixels with gray level i and gray level j appearing simultaneously within a specified image window, under a given direction (such as 0°, 45°, etc.) and a given distance (determined by the scale parameter s).

[0093] Calculate the local texture entropy of the local window at that scale parameter and in that direction using each unit element in the probability matrix. (i.e., the formula is) Among them, when hour, );

[0094] For the current pixel, sort the local texture entropy of all scale parameters in ascending order according to the current direction to obtain the entropy change curve in that direction; obtain all entropy change curves of the current pixel in the four directions respectively to obtain the entropy change curve family (that is, for each pixel, calculate the local texture entropy according to the four directions and multiple scale parameters, sort the entropy values ​​of all scale parameters in each direction to obtain a curve, and the four curves are called the entropy change curve family).

[0095] It should be noted that the above steps, by defining a set of scale parameters (e.g., S={3,5,7,9,11,13,15}), and based on the expected particle size distribution range of the recycled aggregate sample (typically 5mm-30mm) and the spatial resolution of the image acquisition system (e.g., each pixel corresponds to an actual physical size of 0.5mm after calibration), convert the pixel units of the scale parameters into physical sizes. That is, s=3 corresponds to a local window size of approximately 3.5mm ((2×3+1)×0.5mm=3.5mm), which can capture the texture features of fine particles (such as fine sand, cement stone micropores); s=15 corresponds to approximately 15.5mm. The local window size can capture the texture features of coarse aggregate particles; this scale range covers the typical particle size range of various components in recycled aggregates, ensuring the integrity and effectiveness of multi-scale texture analysis. The gray-level co-occurrence matrix (GLCM) is calculated to obtain the local texture entropy of each pixel. The scale parameter reflects the fineness of the texture; a smaller scale parameter reflects a finer texture structure, while a larger scale parameter reflects a coarser texture structure. Multiple scales (different window sizes) are used to capture texture information of the image at different scales. The gray-level co-occurrence matrix is ​​a commonly used tool in texture feature extraction, capable of describing the spatial relationship between pixel pairs. The gray-level co-occurrence matrix is ​​normalized so that each element represents the joint probability distribution of gray-level pairs. Local texture entropy reflects the complexity and disorder of local regions of the image; the higher the entropy value, the more complex the texture and the greater the information content in that region. The step also uses the entropy change curves of the pixel in four directions (0°, 45°, 90°, 135°), such as... Figure 3 As shown, the texture directionality of an image can be analyzed, which further helps to distinguish different texture features; the above steps, by calculating local texture entropy, are used to reveal the microstructural features of the image, especially the complexity of the surface texture, which is helpful for subsequent identification and classification.

[0096] S22: Calculate the first-order difference of each scale parameter in ascending order for each entropy change curve to obtain a numerical derivative sequence (i.e., calculate the first-order difference for each scale parameter to obtain a numerical derivative, forming a numerical derivative sequence).

[0097] Filter the extreme points of the maximum and minimum numerical derivatives in the numerical derivative sequence (that is, find local maxima and local minima).

[0098] The scale parameter corresponding to the first non-zero extreme point (that is, the extreme point whose numerical derivative is closest to 0) is selected from all extreme points and used as the feature scale parameter in that direction.

[0099] The feature scale parameters in all directions (i.e., the four directions of the current pixel) are weighted and averaged to obtain the comprehensive feature scale parameters.

[0100] The system presets a first scale parameter threshold and a second scale parameter threshold. The first scale parameter threshold is set to 5, corresponding to a window size of 11×11 pixels, which, after resolution calibration, corresponds to a physical size of approximately 5.5 mm. This threshold is used to distinguish between cement stone matrix (regions with feature scales smaller than this value) and aggregate particles (regions with feature scales larger than this value). The second scale parameter threshold is set to 11, corresponding to a window size of 23×23 pixels, which corresponds to a physical size of approximately 11.5 mm. This threshold is used to distinguish between fine aggregate (feature scales between the two thresholds) and coarse aggregate (feature scales larger than the second threshold). These thresholds are obtained through experimental calibration. Recycled aggregate samples with known compositions are collected, their comprehensive feature scale parameters are calculated, and the scale parameter distributions of the cement stone region, fine aggregate region, and coarse aggregate region are statistically analyzed. The critical value that maximizes the classification accuracy is selected as the preset threshold.

[0101] Determine the relationship between the comprehensive feature scale parameter of the current pixel and the threshold values ​​of the first and second scale parameters;

[0102] If the comprehensive feature scale parameter is less than the first scale parameter threshold, the pixel is determined to be a high-porosity texture region and marked as a hardened cement stone region (that is, it indicates that the texture of the pixel exhibits a high entropy value at a very small scale parameter, indicating a complex microstructure and abundant pores, so the pixel is classified as a high-porosity texture region and marked as a candidate for hardened cement stone region; here, high-porosity texture region is not directly equivalent to high porosity, but refers to the region exhibiting complex texture changes at the microscale. This texture feature usually corresponds to hardened cement stone (especially hardened cement stone with a porous structure) in recycled aggregate; for dense hardened cement stone, although its texture complexity is low, it can be further distinguished by the fluorescence excitation efficiency index in the subsequent step S24; this step only completes the preliminary region candidate division and does not constitute the final classification).

[0103] If the comprehensive feature scale parameter is greater than the second scale parameter threshold, the pixel is determined to be a large-scale high-entropy region and marked as a natural aggregate region (that is, it indicates that the pixel requires a large scale parameter window to detect texture changes, indicating that the surface is uniform and dense, so the pixel is classified as a large-scale high-entropy region and marked as a candidate for natural aggregate region; the large-scale high-entropy region indicates that the texture complexity of the pixel requires a large scale parameter window to be detected, reflecting that the surface grayscale change in this region is relatively gentle and the texture unit scale is large. In actual recycled aggregate, natural aggregate (especially relatively uniform siliceous or calcareous aggregate) usually exhibits this type of texture feature, so this region is marked as a candidate for natural aggregate region).

[0104] If the comprehensive feature scale parameter is greater than or equal to the first scale parameter threshold and less than or equal to the second scale parameter threshold, then the pixel is determined to be an impurity region.

[0105] It should be noted that the above steps find local extrema by calculating the first difference of the entropy change curve and determine the feature scale parameters in each direction; calculating the numerical derivative of the entropy change curve can reveal the rate of change of texture entropy, further helping to capture the changing characteristics of texture structure; this step is used in this way to more accurately capture texture transformation at the regional scale; the above steps also determine the most representative scale parameters by finding the local maxima and minima of the entropy change curve, which reflect the typical texture features in that direction; combining the feature scale parameters in four directions, a global texture feature scale parameter is obtained, which helps to form a comprehensive analysis of the pixel, effectively reduces the influence of boundary effects, and accurately identifies the feature scale parameters in the image;

[0106] S23: Obtain the comprehensive feature scale parameters of the pixels in the high-porosity texture region, find the closest scale parameter in the set of scale parameters, and extract the local texture entropy in the four directions of the closest scale parameter to form a direction entropy vector;

[0107] The maximum and minimum values ​​in the directional entropy vector are selected, and the extreme value ratio of directional entropy is calculated (i.e., the extreme value ratio of directional entropy = the ratio between the maximum and minimum values ​​in the directional entropy vector; the larger the ratio, the stronger the directionality of the pore structure (e.g., cracks, directional channels); a ratio close to 1 indicates that the pore structure is isotropic (e.g., uniformly distributed micropores); at the same time, this ratio also reflects the degree of difference in texture complexity of pixels in different directions: the larger the ratio, the more complex the texture of the region is in one direction than in other directions, suggesting that the pore structure may be directional (e.g., cracks, layered pores); a ratio close to 1 indicates that the texture complexity in each direction is similar, and the pore structure tends to be isotropic (e.g., uniformly distributed micropores)).

[0108] It should be noted that the above steps extract the orientation entropy vector that is closest to the scale parameter, which can more accurately describe the directionality of high-porosity texture regions. The maximum and minimum values ​​of the orientation entropy vector reflect the strength and directionality of the texture. The larger the ratio of orientation entropy extrema, the stronger the directionality of the pore structure (such as cracks and oriented channels). When the ratio is close to 1, it indicates that the pore structure is isotropic and may be uniformly distributed micropores. This step can further analyze the orientation or uniformity of pores, providing quantitative characteristics of pore structure for subsequent classification and helping to distinguish different types of pores.

[0109] S24: Obtain the fluorescence intensity value of the pixel in the high-porosity texture region at the corresponding coordinate position in the ultraviolet-excited fluorescence band image (i.e., since each pixel in the ultraviolet-excited fluorescence band image has an excitation fluorescence intensity, which is the fluorescence intensity value, the connectivity of the pores and the fluid accessibility are evaluated in combination with the fluorescence intensity).

[0110] Obtain the reflectance value of the pixel in the high-porosity texture region at a preset characteristic wavelength in the near-infrared image;

[0111] The ratio of fluorescence intensity to reflectance of pixels in the high-porosity texture region is calculated as the fluorescence excitation efficiency index. A high fluorescence excitation efficiency index reflects strong fluorescence intensity and weak moisture absorption (high reflectance), indicating that fluorescent materials (such as unhydrated cement particles or admixtures) are exposed on the surface, or that the pores are open, allowing excitation light to easily penetrate. A low fluorescence excitation efficiency index reflects weak fluorescence intensity or strong moisture absorption (low reflectance), possibly due to a water film covering the surface or closed pores, making fluorescent materials inaccessible (which may also indirectly reflect ordinary aggregate or a damp surface). In other words, in open-pore hardened cement stone, the pores are interconnected. Ultraviolet excitation light can penetrate into the pores to excite fluorescent substances (such as unhydrated cement particles, additives, etc.), producing a strong fluorescence intensity. Simultaneously, water can be adsorbed within the interconnected pores, and water molecules have strong absorption in the preset characteristic wavelength band, leading to a decrease in the reflectance value of that preset characteristic wavelength. Therefore, the fluorescence excitation efficiency index of open-pore hardened cement stone exhibits a high value. Conversely, the pores of closed-pore hardened cement stone are not interconnected, making it difficult for fluorescent substances to be excited (low fluorescence intensity value), and there is no water or water is inaccessible within the pores (the reflectance value of the preset characteristic wavelength is relatively high), thus exhibiting a low fluorescence excitation efficiency index.

[0112] Preset anisotropy threshold and efficiency threshold;

[0113] If the extreme ratio of directional entropy of a pixel in the high-porosity texture region is greater than the anisotropy threshold and the fluorescence excitation efficiency index is less than the efficiency threshold, then the pixel is determined to be a closed-pore type hardened cement stone (that is, the pore structure is directional (possibly layered or closed cracks), and the fluorescence excitation efficiency is low (fluorescent material is inaccessible or water is trapped in the pores), indicating that the pores are mostly in a closed state).

[0114] If the extreme value ratio of the directional entropy of a pixel in the high-porosity texture region is less than or equal to the anisotropy threshold, and the fluorescence excitation efficiency index is greater than or equal to the efficiency threshold, then the pixel is determined to be an open-pore type hardened cement stone (that is, the pore structure is isotropic (uniformly distributed), and the fluorescence excitation efficiency is high (fluorescent substances can be excited, or moisture in the pores can be detected), indicating that the pores are interconnected and form an open network).

[0115] If any of the above determination conditions are not met, the pixel is determined to be ordinary hardened cement stone.

[0116] It should be noted that at this point, the pixels are classified into the natural aggregate region identified in step S22, the closed-pore hardened cement stone identified in step S24, the open-pore hardened cement stone, and ordinary hardened cement (that is, if two conditions are met simultaneously or neither is met in step S24, it can be marked as mixed / transitional hardened cement stone, which is ordinary hardened cement). Figure 4 As shown;

[0117] The above steps calculate the fluorescence excitation efficiency index by combining ultraviolet-excited fluorescence and near-infrared reflectance, and determine the type of pores by combining the directional entropy extreme ratio. Fluorescence intensity reflects the connectivity and fluid accessibility of pores, while reflectance reveals surface characteristics. The combination of the two can comprehensively evaluate the properties of pores. A high fluorescence excitation efficiency index usually means that there are exposed fluorescent substances on the surface or that the pores are open, while low efficiency may indicate that the pores are closed. The directional entropy extreme ratio and fluorescence excitation efficiency index are combined to determine whether the pores are closed (directional cracks, closed pores) or open (connected pores, microporous networks). This step provides a more detailed classification of pores in the image by combining multiple feature analyses, thereby providing a scientific basis for the performance evaluation of hardened cement stone.

[0118] S25: For the visible light and near-infrared images, extract reflectance values ​​using preset feature band combinations, and identify the siliceous or calcareous natural aggregate to which the pixels belong based on the reflectance values ​​of different feature band combinations; perform sliding window processing on the recycled aggregate sample to be tested, and calculate the reflectance change rate and texture entropy gradient of the pixels in the sliding window; calculate the spatial phase difference based on the reflectance change rate and texture entropy gradient; calculate the mean and standard deviation of the spatial phase difference, and then calculate the dynamic threshold benchmark value; sieve the recycled aggregate sample to be tested to obtain its nominal maximum particle size, and correct the dynamic threshold benchmark value based on the nominal maximum particle size to obtain the final dynamic threshold; generate a heterogeneous interface response intensity map for each pixel based on the reflectance change rate and texture entropy gradient; perform binarization processing on the heterogeneous interface response intensity map using the final dynamic threshold to obtain the boundary line; obtain the heterogeneous component distribution vector of the natural aggregate region, hardened cement stone region, and impurity region based on the boundary line division result;

[0119] It should be noted that the above steps analyze the rate of change of reflectance and the texture entropy gradient through a sliding window, calculate the spatial phase difference, and then screen out the heterogeneous interface features of recycled aggregates. The rate of change of reflectance reflects the surface changes in the image, while the texture entropy gradient reveals the rate of change of texture. Combining the two can effectively analyze the features of heterogeneous interfaces. This step, through the calculation of spatial phase difference, can accurately identify the boundaries between different regions, thereby assisting in subsequent classification. Dynamic thresholds help to adaptively adjust and reduce manual intervention. After generating the heterogeneous interface response intensity map, binarization is performed to identify the aggregate region and the hardened cement stone region, further completing accurate classification. This step can automatically process complex recycled aggregate samples, identify their heterogeneous structures, and then perform accurate classification, assisting in material property analysis and evaluation.

[0120] like Figure 5 As shown, specifically, in step S25, reflectance values ​​are extracted from the visible light and near-infrared images using preset feature band combinations. Based on the reflectance values ​​of different feature band combinations, the siliceous and calcareous natural aggregates to which the pixels belong are identified. A sliding window process is applied to the sample of the recycled aggregate to be tested, and the reflectance change rate and texture entropy gradient of the pixels within the sliding window are calculated. The spatial phase difference is calculated based on the reflectance change rate and texture entropy gradient. The mean and standard deviation of the spatial phase difference are then calculated to obtain the dynamic threshold. The following steps are taken: A baseline value is set; the recycled aggregate sample to be tested is sieved to obtain its nominal maximum particle size, and the dynamic threshold baseline value is corrected according to the nominal maximum particle size to obtain the final dynamic threshold; a heterogeneous interface response intensity map is generated for each pixel based on the reflectance change rate and texture entropy gradient; the heterogeneous interface response intensity map is binarized using the final dynamic threshold to obtain the boundary line; based on the boundary line division result, the heterogeneous component distribution vectors of the natural aggregate region, hardened cement stone region, and impurity region are obtained.

[0121] S251: Perform preset feature band combination pairs on the visible light band image and the near-infrared band image (such as combination 1 visible light band image λVIS, 550nm / λVIS, 670nm (used to distinguish iron minerals); combination 2 near-infrared band image λNIR, 900nm / λNIR, 1100nm (used to distinguish silicate minerals), etc., to identify and distinguish water minerals and calcareous minerals).

[0122] For each characteristic band combination pair, reflectance values ​​are extracted from the corresponding visible light band image and near-infrared band image, respectively;

[0123] Calculate the ratio between the reflectance value of the visible light band image and the reflectance value of the near-infrared band image, and form a band ratio vector from all the calculated reflectance values.

[0124] Project the band ratio vector onto the CIELAB chromaticity space to obtain the chromaticity coordinates in the CIELAB chromaticity space;

[0125] Obtain the standard chromaticity cluster centers of siliceous minerals and calcareous minerals from the pre-built standard database;

[0126] Calculate the first Euclidean distance between the chromaticity coordinates and the center of the standard chromaticity cluster of the siliceous mineral;

[0127] Calculate the second Euclidean distance between the chromaticity coordinates and the center of the standard chromaticity cluster of calcareous minerals (i.e., the smaller the first and second distances are, the closer the spectral characteristics of the pixel are to the corresponding mineral type).

[0128] It should be noted that the visible light (VIS) and near-infrared (NIR) band images are combined, for example: VIS: 550 nm / 670 nm, used to distinguish ferrous minerals; NIR: 900 nm / 1100 nm, used to distinguish silicate minerals. Pixel reflectance values ​​are extracted for each band combination, and band ratios (e.g., R550 / R670) are calculated to form a band ratio vector. This vector is then projected onto the CIELAB chromaticity space to obtain chromaticity coordinates. The Euclidean distance between these coordinates and the pre-constructed mineral standard cluster centers (siliceous and calcareous minerals) is calculated. Multispectral reflectance characteristics are used for preliminary mineral type identification. The ratio eliminates the influence of light intensity and is more sensitive to mineral color differences. Different minerals have specific spectral responses, and the band ratio can enhance these differences. A smaller Euclidean distance indicates that the pixel's spectral characteristics are closer to a particular mineral; a larger Euclidean distance indicates lower discrimination.

[0129] Projecting onto the CIELAB chromaticity space and calculating the Euclidean distance is to convert spectral reflectance data into a color difference measure that conforms to human visual perception, thereby enabling accurate identification of mineral types; simply put, the band ratio vector is the spectral data, the CIELAB chromaticity coordinates are the colors perceived by the human eye, and the Euclidean distance is the magnitude of the color difference.

[0130] S252: Preset multiple visible light characteristic wavelengths, and obtain the reflectivity value of the pixel point according to the multiple preset visible light characteristic wavelengths;

[0131] The reflectance values ​​of different visible light characteristic wavelengths are calculated to form a ratio, which is then used as a characteristic ratio. All characteristic ratios are combined to obtain a visible light characteristic ratio vector. (Since the preset visible light characteristic wavelengths are different, the reflectances are also different (c1, c2, c3, cn). The reflectances c1c2, c1c4, c3cn, etc. are calculated to form a ratio, which captures multi-dimensional material properties, such as silicon and calcium.)

[0132] The fluorescence excitation efficiency index, the band ratio vector, and the visible light feature ratio vector are standardized and then merged to obtain a multimodal spectral feature vector (that is, features from different spectral domains (visible light ratio, near-infrared ratio, fluorescence index) are fused to form a comprehensive spectral feature vector).

[0133] Preset deviation threshold and fluorescence threshold;

[0134] If the first Euclidean distance of the pixel is less than the deviation threshold and the fluorescence excitation efficiency index is less than the fluorescence threshold, then the pixel is determined to be siliceous natural aggregate (i.e., the spectral color of the pixel is close to the standard cluster of siliceous minerals and the fluorescence signal is weak (siliceous aggregates usually do not contain fluorescent substances)).

[0135] If the second Euclidean distance of the pixel is less than the deviation threshold, and the fluorescence excitation efficiency index is greater than or equal to the fluorescence threshold, then the pixel is determined to be calcareous natural aggregate (i.e., the spectral chromaticity of the pixel is close to the standard cluster of calcareous minerals, and the fluorescence signal is strong (calcareous aggregates such as limestone may produce weak fluorescence due to impurities or surface adsorption)). Figure 6 As shown;

[0136] It should be noted that the above steps preset multiple visible light characteristic wavelengths, for example, blue band λ1=450nm, green band λ2=550nm, red band λ3=670nm, and red-edge band λ4=700nm; obtain the reflectance values ​​corresponding to each visible light characteristic wavelength, and calculate the ratios between them, such as a ratio of 1. (Green / Red ratio, reflecting iron mineral content); Ratio 2 (Blue / Green ratio, reflecting organic matter or clay minerals); Ratio 3 (Red border index, reflecting vegetation or iron oxides); ratio 4 (Normalized difference index); The above steps utilize multispectral information in the visible light band to construct a ratio feature sensitive to mineral color, thereby enhancing the ability to distinguish between siliceous and calcareous minerals.

[0137] If none of the above judgment conditions are met, then a soft classification is performed based on the relative size of the first Euclidean distance and the second Euclidean distance. If the first Euclidean distance is less than the second Euclidean distance, it tends to be siliceous; otherwise, it tends to be calcareous.

[0138] Calculate ratios for multiple visible light characteristic wavelengths (e.g., blue / green, green / red, red-edge index); combine these ratios to obtain a visible light characteristic ratio vector; standardize and merge the fluorescence excitation efficiency index, visible light ratio vector, and band ratio vector to obtain a multimodal spectral feature vector; set deviation thresholds and fluorescence thresholds: if the first Euclidean distance < threshold and the fluorescence index < threshold, it is classified as siliceous natural aggregate; if the second Euclidean distance < threshold and the fluorescence index ≥ threshold, it is classified as calcareous natural aggregate; if the conditions are not met, soft classification is performed based on the relative magnitude of the first and second Euclidean distances; fuse features from different spectral domains (visible light ratio, near-infrared ratio, fluorescence information) to form a comprehensive spectral feature vector; siliceous aggregates typically do not produce fluorescence, while calcareous aggregates may exhibit weak fluorescence; multimodal fusion can reduce misclassification of single bands and improve classification accuracy.

[0139] S253: Apply a sliding window (each window can be 50 pixels) to the regenerative orthopedic sample to be tested, and uniformly divide each sliding window into a grid to generate grid points;

[0140] For each sliding window, randomly select N grid points as the center grid point and determine their coordinate positions;

[0141] For all pixels in each sliding window, calculate the first-order gradient vector of reflectance value, which is used as the reflectance change rate.

[0142] For all pixels in each sliding window, calculate the second-order gradient tensor of the local texture entropy, which serves as the texture entropy gradient (i.e., the curvature information representing the change in texture entropy).

[0143] The spatial phase difference is calculated based on the reflectivity change rate and texture entropy gradient (i.e., firstly, the pixel is defined). The gradient vector of the rate of change of reflectance at point is Its gradient magnitude is Define the texture entropy gradient vector as Its gradient magnitude is ;in, and It is expressed as the horizontal partial derivative (i.e., the rate of change of reflectivity and texture entropy in the horizontal direction (x direction) of the image, i.e., the horizontal gradient component). and Expressed as the vertical partial derivative (i.e., the rate of change of reflectivity and texture entropy in the vertical direction (y-direction) of the image, i.e., the vertical gradient component); the spatial phase difference is calculated as follows: ;in, Represented as a very small positive number; (represented as the dot product of two gradient vectors).

[0144] Calculate the mean and standard deviation of the spatial phase difference of pixels within each sliding window;

[0145] For each sliding window, obtain the coordinates of the window center point, and obtain the coordinates of the image center point of the regenerative orthopedic sample to be tested;

[0146] Calculate the Euclidean distance between the center point coordinates of the window and the center point coordinates of the image;

[0147] The weighting coefficients are calculated using the central Euclidean distance (i.e., the calculation formula is...). ;in, It is a very small positive number (such as 1) to prevent division by zero; this formula makes the weight of the window smaller the farther away from the center.

[0148] The weighted average and the weighted average standard deviation are obtained by using the weighting coefficients to calculate the mean and standard deviation of the spatial phase difference of all sliding windows, respectively.

[0149] The dynamic threshold benchmark value is calculated by using the weighted average value and the weighted standard deviation average value (i.e., summation calculation; this benchmark value comprehensively reflects the average intensity and variability of the heterogeneous interface on the sample surface).

[0150] It should be noted that a sliding window (e.g., 50 pixels per window) is applied to the image, and the window is uniformly divided into a grid. N grid points are randomly selected as centers, and the spatial phase difference is calculated for the pixels within the window. The mean and standard deviation of the spatial phase difference for each window are calculated. The weighted mean and standard deviation are obtained based on the distance between the window and the image center, and a dynamic threshold benchmark value is further calculated. This process identifies heterogeneous interfaces (natural aggregate and hardened cement stone boundaries) on the surface of recycled aggregate samples. The reflectivity gradient reflects compositional changes, and the texture entropy gradient reflects microstructural changes. The spatial phase difference integrates the information from both, improving the sensitivity to minute boundaries. The weighting factor takes into account the distance from the center to reduce edge interference, providing basic data for heterogeneous component distribution analysis and ensuring accurate boundary detection.

[0151] S254: The recycled aggregate sample to be tested is sieved to determine the nominal maximum particle size (i.e., different sieve diameters can be set to sieve the sample, such as a maximum of 30mm, 25mm, and a minimum of 20mm; the sieve aperture size (maximum 30mm) corresponding to a 90% sample pass rate is obtained through sieving and used as the nominal maximum particle size).

[0152] A pre-defined monotonically decreasing function is used (i.e., the function takes the nominal maximum particle size as the independent variable, and the output value decreases monotonically as the particle size increases, which is used to dynamically adjust the sensitivity of the dynamic threshold reference value; the larger the particle size, the wider the spatial scale of the interface reflectivity change, and the weak gradient response needs to be appropriately suppressed to avoid over-detection; conversely, the smaller the particle size, the richer the interface details, the higher the function output, and the enhanced ability to identify subtle heterogeneous boundaries; the function form can adopt an inverse proportional or exponential decay structure, and different coefficients can be set according to the particle size range).

[0153] The nominal maximum particle size and the monotonically decreasing function are used to calculate the particle size adjustment coefficient (i.e., the larger the particle size, the more continuous the particle boundary and the gentler the gradient change, the lower the threshold should be (i.e., the dynamic threshold reference value) to avoid missed detection; the smaller the particle size, the denser the particle boundary and the more drastic the gradient change, the higher the threshold should be (i.e., the dynamic threshold reference value) to avoid over-detection).

[0154] Obtain the surface roughness parameters of the recycled aggregate sample to be tested (i.e., use a surface roughness meter (such as a contact profilometer) to measure the roughness at multiple points on the sample surface and take the average value; the specific operation is common knowledge and will not be described in detail).

[0155] A pre-defined monotonically increasing function is used; the surface roughness parameter is used to calculate the roughness correction coefficient using the monotonically increasing function (i.e., the rougher the surface, the more severe the unevenness of shadows and illumination, and the greater the phase difference fluctuation, so the threshold should be appropriately increased (i.e., dynamic threshold reference value) to avoid noise being misjudged as a boundary; the smoother the surface, the phase difference change is mainly caused by compositional differences, so the threshold can be appropriately decreased (i.e., dynamic threshold reference value) to improve the boundary detection sensitivity).

[0156] The final dynamic threshold is calculated using the dynamic threshold reference value, the particle size adjustment coefficient, and the roughness correction coefficient (i.e., multiplication calculation).

[0157] It should be noted that the sample to be tested is sieved to obtain the nominal maximum particle size. For example, by setting different sieve aperture sizes (30mm, 25mm, 20mm, etc.), the sieve aperture size with a 90% sample throughput is obtained as the nominal maximum particle size. Based on the nominal maximum particle size, the particle size adjustment coefficient is calculated. The output of this monotonically decreasing function decreases monotonically as the particle size increases, usually in an inverse proportional or exponential decay form. The larger the particle size, the wider the spatial scale of the interface reflectivity change, and the gentler the gradient change. Therefore, the dynamic threshold should be appropriately lowered to avoid missed detections. The particle boundaries are finer, and the interface changes drastically. The threshold should be appropriately increased to avoid over-detection.

[0158] The surface roughness of the sample under test is measured at multiple points using a surface roughness meter (such as a contact profilometer) to obtain the average value; the roughness correction coefficient is calculated based on the roughness parameters; the rougher the surface, the more likely the shading and uneven illumination will be aggravated, and the phase difference fluctuation will increase, so the threshold should be increased to reduce noise interference; the smoother the surface, the phase difference is mainly caused by compositional differences, so the threshold should be appropriately reduced to improve the boundary detection sensitivity.

[0159] The dynamic threshold baseline is multiplied by the particle size adjustment coefficient and the roughness correction coefficient to obtain the final dynamic threshold. The threshold is adjusted according to the different characteristics of particle size and surface roughness to ensure adaptability to different types of samples (large particles, smooth surfaces, rough surfaces). The particle size and surface roughness of the sample affect the degree of gradient change, which directly affects the sensitivity of heterogeneous interface detection. When the particle size is large, the interface change is relatively gentle, so the threshold needs to be lowered. However, when the surface is rough, it will generate more noise, so the threshold needs to be increased to reduce false detections and ensure accurate identification of heterogeneous interfaces under different particle sizes and surface conditions.

[0160] S255: Calculate the heterogeneous interface response intensity for each pixel based on the reflectivity change rate and texture entropy gradient to generate a heterogeneous interface response intensity map; perform binarization and morphological operations on the heterogeneous interface response intensity map using the final dynamic threshold to obtain the boundary line; count the number of pixels inside the natural aggregate region, hardened cement stone region, and impurity region based on the range of the boundary line, and calculate the cosine value of the included angle; obtain the area fractions of the natural aggregate region, hardened cement stone region, and impurity region based on the cosine value of the included angle; construct a heterogeneous component distribution vector based on the area fractions.

[0161] The cosine value of the included angle includes the cosine value of the silicon included angle, the cosine value of the calcium included angle, and the cosine value of the sealing and hardening included angle;

[0162] The area fraction includes the area fraction of siliceous natural aggregate, the area fraction of calcareous natural aggregate, the area fraction of sealed hardened cement stone, the area fraction of impurities, and the area fraction of the high water absorption risk subclass.

[0163] The area fraction of the hardened cement stone region is divided into the area fraction of the closed hardened cement stone and the area fraction of the high water absorption risk subclass. In this scheme, the area fraction is not equal to the area. The area fraction represents the proportion between various materials (such as the proportion between silica and calcium in natural aggregate, which reflects the essential characteristics of the composition, and will not change with the sample size like the area; at the same time, the areas of samples from different batches and of different sizes cannot be directly compared; while the area fraction can be directly compared regardless of the sample size).

[0164] It should be noted that the heterogeneous interface response intensity map is obtained by calculating the reflectance change rate and texture entropy gradient for each pixel, which can quantify the interface characteristics of different regions in the sample (such as the difference in interface strength between natural aggregate and hardened cement stone regions). The heterogeneous interface response intensity is usually closely related to the differences in material composition and surface texture, reflecting the heterogeneity between regions. The calculated final dynamic threshold is used to binarize the heterogeneous interface response intensity map, that is, to divide the response intensity map into boundary and non-boundary categories. Morphological operations (such as erosion and dilation) are performed on the binarized image to further refine the shape and position of the boundaries, remove noise, and enhance the real boundary information. Based on the binary image after morphological operations, all boundary lines in the sample are extracted. These boundary lines represent the boundary between natural aggregate and hardened cement stone regions. Based on the boundary lines, the natural aggregate region and the hardened cement stone region are statistically analyzed. The number of pixels in the cement stone region and impurity region is calculated; then, the area fraction of each region is calculated using the cosine of the included angle, the cosine value of which is related to the area of ​​the region; based on the calculated area fraction of each region, a heterogeneous component distribution vector is constructed; this vector provides quantitative information on the component ratio of different regions in the sample; this vector can be used for sample quality control, material performance evaluation, and optimized proportioning; images and data reflecting the heterogeneity and mineral component distribution of the sample surface are generated, providing a basis for mineral region division and component analysis; the heterogeneous interface response intensity map can effectively reflect the mineral differences on the sample surface and help to further quantify the area fraction of each region in the sample; it provides the accurate area fraction of each region (such as natural aggregate, hardened cement stone, impurities); by generating the heterogeneous component distribution vector, the component analysis, classification, and evaluation of the sample are realized, which can be further used for quality control and optimization;

[0165] like Figure 7 As shown, specifically, in step S255, the heterogeneous interface response intensity is calculated for each pixel based on the reflectivity change rate and texture entropy gradient, generating a heterogeneous interface response intensity map; the heterogeneous interface response intensity map is binarized and subjected to morphological operations using the final dynamic threshold to obtain the boundary line; the number of pixels inside the natural aggregate region, hardened cement stone region, and impurity region is counted based on the range of the boundary line, and the cosine value of the included angle is calculated; the area fractions of the natural aggregate region, hardened cement stone region, and impurity region are obtained based on the cosine value of the included angle; a heterogeneous component distribution vector is constructed based on each area fraction. The specific operation steps are as follows:

[0166] S2551: Multiply the reflectivity change rate with the texture entropy gradient to obtain the multimodal feature response value (i.e., this step is different from step S253 above, which calculates the spatial phase difference, and the formula is different; this step is a direct product calculation).

[0167] The pre-defined principal direction angle of the gradient vector of reflectivity change rate is The principal direction angle of the texture entropy gradient vector is The directional difference (i.e., the included angle) is calculated as the same direction coefficient (i.e., same direction coefficient = 1, the directions are completely consistent (the included angle is 0° or 180°); same direction coefficient = 0, the directions are orthogonal (the included angle is 45° or 135°); same direction coefficient = −1, the directions are completely opposite (the included angle is 90°); this coefficient is used to enhance pixels whose spectral change direction is consistent with the texture change direction, these pixels are more likely to be true heterogeneous interfaces).

[0168] The multimodal characteristic response value is multiplied by the same-direction coefficient to obtain the heterogeneous interface response intensity (i.e., the range of values). When the directions are completely consistent hour, The response intensity doubles; when the directions are orthogonal hour, The response intensity remains unchanged when the directions are completely opposite; hour, The response intensity is suppressed to zero; this modulation mechanism ensures that only those regions whose spectral changes are consistent with the direction of texture changes are enhanced into heterogeneous interfaces, effectively suppressing noise and false boundaries), generating heterogeneous interface response intensity maps;

[0169] It should be noted that the multimodal feature response value is obtained by directly multiplying the reflectance change rate and the texture entropy gradient, thus determining the principal directions of the reflectance change rate and the texture entropy gradient. A co-directional coefficient (1: completely consistent, 0: orthogonal, -1: completely opposite) is calculated based on the difference in their principal directions. The multimodal feature response value is then multiplied by this co-directional coefficient to generate a heterogeneous interface response intensity map. This direct multiplication captures the synergistic information of spectral (reflectance change) and texture (entropy gradient) changes; only pixels showing changes in both spectrum and texture are significantly amplified. The principal direction alignment coefficient is used to enhance the real heterogeneous interface while suppressing noise or false boundaries. Consistent directions indicate that the pixel is likely located at the interface, while inconsistent directions may indicate texture noise or random variations. This improves the accuracy of heterogeneous interface detection, ensuring that only truly significant interface regions are marked, while reducing false responses. Pixels with doubled response intensity represent high-confidence interfaces; pixels with unchanged or suppressed responses represent low-confidence or non-interface regions. The generated heterogeneous interface response intensity map is a boundary probability map, providing a foundation for subsequent binarization and region extraction.

[0170] S2552: The heterogeneous interface response intensity map is binarized using a final dynamic threshold to obtain a candidate heterogeneous interface pixel binary map (i.e., the heterogeneous interface response intensity map is equivalent to a boundary probability map, where each point has a boundary probability score). Binarizing the heterogeneous interface response intensity map using a final dynamic threshold ensures that without threshold filtering, all pixels with non-zero response intensities would be considered, potentially leading to minor texture changes within the material being misjudged as boundaries. Dynamic threshold filtering ensures that only truly significant interface changes are retained. The calculation formula is... (x, y)= ;in, Represented as the response intensity of a heterogeneous interface; (represented as the final dynamic threshold).

[0171] Morphological opening operations (i.e., erosion followed by dilation) are performed on the pixels in the binary image of the candidate heterogeneous interface according to a preset pixel radius. ;in, Represented as a binary image after morphological opening operation; Represented as a smooth image; Represented as a circular structure with a preset pixel radius; the erosion operation eliminates isolated burrs and small protrusions, and the dilation operation restores the width of the main line segment), resulting in a binary image of the candidate heterogeneous interface pixels after morphological opening operation processing;

[0172] Then, using a preset cross-shaped structure window, a morphological closing operation (i.e., dilation followed by erosion) is performed on the binary image of the candidate heterogeneous interface pixels after the morphological opening operation. ;in, Represented as a cross-shaped structure window; the expansion operation connects adjacent broken line segments, and the erosion operation restores the line width to prevent excessive expansion), resulting in a binary map of the repaired boundary;

[0173] Pixels are extracted from the binary boundary map of the repaired area to form the boundary lines of the natural aggregate area, the hardened cement stone area, and the impurity area (the three different areas are distinguished by the boundary lines, and they do not overlap).

[0174] It should be noted that a dynamic threshold is used to binarize the heterogeneous interface response intensity map to obtain a candidate heterogeneous interface pixel map. A circular morphological closing operation (dilation followed by erosion) is performed on the candidate map to connect broken line segments and restore the line width. Then, a cross-shaped closing operation (erosion followed by dilation) is used to eliminate isolated points and small protrusions, resulting in a repaired boundary binary map. The boundary is extracted to form the boundary lines of the natural aggregate, hardened cement stone, and impurity regions. Binarization transforms the continuous response intensity into a clear candidate boundary, eliminating weak noise. The two morphological processes respectively solve the line segment breakage problem (circular closing operation) and the isolated noise problem (cross-shaped closing operation). After extracting the boundary lines, three different regions can be clearly divided, ensuring that they do not overlap, obtaining a complete, continuous, and accurate heterogeneous interface boundary, eliminating false boundaries, and enhancing the reliability of interface detection. The output boundary lines can be used as the basis for dividing the natural aggregate, hardened cement stone, and impurity regions.

[0175] S2553: Based on the boundary line division, count the number of pixels inside the natural aggregate area, the hardened cement stone area, and the impurity area respectively (that is, the hardened cement stone area is divided into two types of pixels, namely open-pore hardened cement stone and closed-pore hardened cement stone; in the extended step S255 of this scheme, the number of pixels, that is, the total number, can be used as the area of ​​the region).

[0176] Obtain the standard siliceous aggregate spectral vector and the standard calcareous aggregate spectral vector for each pixel from the standard database;

[0177] The cosine value of the silica angle (i.e.,) is calculated using the multimodal spectral feature vector of this pixel and the spectral vector of the standard silica aggregate. ;in, Represented as a multimodal spectral eigenvector; (represented as the standard siliceous aggregate spectral vector).

[0178] The cosine value of the angle between the calcium aggregate and the standard calcium aggregate is calculated using the multimodal spectral feature vector of the pixel and the standard spectral vector (i.e., the calculation formula is the same as the formula for calculating the cosine value of the angle between the silica aggregate and the standard spectral vector, and will not be repeated here).

[0179] It should be noted that, based on the boundary lines, each region is divided, the number of pixels within each region is counted, and the spectral vectors of standard siliceous and calcareous aggregates are obtained. The cosine of the angle between the multimodal spectral vector of each pixel and the standard vector is calculated to assess the degree of spectral matching. The cosine of the angle is used to quantify the similarity between the sample spectrum and the standard spectrum, accurately determining which aggregate type each pixel belongs to. Pixel counting provides more refined aggregate distribution information while avoiding interference from mixed regions. Quantitative analysis of the type ratio of natural aggregates facilitates the calculation of area fractions and further material analysis. The output cosine matrix of the angle between siliceous and calcareous natural aggregates provides basic data for area fraction calculation and can be used to evaluate the composition and structure of recycled aggregates, providing a basis for quality control.

[0180] S2554: Obtain the standard hardened cement stone spectral vector for each pixel from the standard database;

[0181] The number of pixels in the hardened cement stone region with closed pores is counted; the closed pore correction factor is calculated by using the extreme ratio of the directional entropy of the pixels in the hardened cement stone with closed pores and the preset correction strength coefficient.

[0182] The multimodal spectral feature vector is corrected using the closed pore correction factor to obtain the corrected multimodal spectral feature vector (i.e., the correction aims to reduce the interference of closed pores (non-connected pores) on spectral features, because the substances in these pores may not have fully participated in the hydration reaction, and their spectral features differ from those of normal cement stone).

[0183] The cosine value of the closed hardening angle is calculated by using the corrected multimodal spectral feature vector and the standard hardened cement stone spectral vector of each pixel in the hardened cement stone region (i.e., the calculation formula is the same as the formula for calculating the cosine value of the silica angle, and will not be repeated here).

[0184] It should be noted that the process involves obtaining the standard hardened cement stone spectral vector, statistically analyzing the pixels of the closed-pore type hardened cement stone, calculating the closed-pore correction factor, and using the correction factor to correct the multimodal spectral feature vector to obtain the corrected feature vector. The cosine of the angle between the corrected feature vector and the standard hardened cement stone spectral vector is then calculated to obtain the cosine of the closed-pore hardening angle. Closed pores may cause spectral anomalies because the material is not fully hydrated, which can easily interfere with spectral matching. The corrected feature vector is used to reduce the interference of these abnormal pixels on the overall matching, improve the accuracy of hardened cement stone identification, and ensure more accurate calculation of the hardened cement stone area fraction. In particular, non-connected pores will not mislead the area calculation, thus improving the reliability of the hardened cement stone area fraction and spectral matching.

[0185] S2555: Obtain the standard impurity spectral vector for each pixel from the standard database;

[0186] The cosine of the impurity angle is calculated using the standard impurity spectral vector and the multimodal spectral eigenvector (i.e., the multimodal spectral eigenvector is not corrected here).

[0187] A preset matching threshold (which can be set to 0.85; when the cosine of the angle is greater than or equal to 0.85, the spectral match is considered successful).

[0188] By using the cosine values ​​of all the included angles mentioned above and calculating with the matching thresholds, we obtain the area fractions of siliceous natural aggregate (i.e., the proportion of pixels in the natural aggregate region whose spectral features match the siliceous standard), calcareous natural aggregate (i.e., the proportion of pixels in the natural aggregate region whose spectral features match the calcareous standard), sealed hardened cement stone (i.e., the proportion of pixels in the hardened cement stone region whose spectral features match the cement stone standard), and impurity area fraction (i.e., the proportion of pixels in the impurity region whose spectral features match the impurity standard; the calculation formula is...). ;in, Expressed as the cosine of the impurity angle; This is represented as the matching threshold; The total number of pixels in the impurity region is represented as the area; that is, all area fractions are substituted using the same formula, which will not be elaborated further).

[0189] Obtain the number of pixels in the open-pore type hardened cement stone;

[0190] The ratio between the number of pixels in the open-pore hardened cement stone and the number of pixels in the hardened cement stone region is calculated and used as the area fraction of the high water absorption risk subclass (i.e., open-pore hardened cement stone has a high water absorption rate and permeability; this index is output separately to assess the durability risk of recycled aggregates).

[0191] The area fractions of siliceous natural aggregate, calcareous natural aggregate, sealed hardened cement stone, impurities, and high water absorption risk subclasses are constructed as heterogeneous component distribution vectors.

[0192] It should be noted that the process involves obtaining the standard impurity spectral vector, calculating the cosine value of the impurity angle (uncorrected), setting a matching threshold (e.g., 0.85), filtering the cosine values, and calculating the area fraction of each region: siliceous aggregate, calcareous aggregate, closed hardened cement stone, and impurities. The proportion of pixels in open-pore hardened cement stone is statistically analyzed and used as the area fraction of the high water absorption risk subclass. All area fractions are combined into a heterogeneous component distribution vector. The matching threshold ensures that only pixels with consistent spectral height are included in the area fraction, eliminating abnormal impurities. Open-pore hardened cement stone has high water absorption, and its separate output index is used for durability assessment. This generates complete heterogeneous component distribution information, including the main aggregate type, the proportion of hardened cement stone and impurities, as well as potential risk indicators, providing quantitative component analysis results.

[0193] Example 2

[0194] like Figure 8 As shown, the present invention also provides a waste concrete recycled aggregate composition identification and classification system, including: a data acquisition module 10; an analysis module 20; and an identification module 30.

[0195] The acquisition module 10 is used to acquire multispectral image sequences under different spectral bands of the regenerative orthopedic sample to be tested.

[0196] The analysis module 20 is used to calculate the local texture entropy in four directions for each pixel in the near-infrared band image according to a set of scale parameters, forming a family of entropy change curves; to divide the family of entropy change curves into hardened cement stone region, natural aggregate region, and impurity region according to a preset scale parameter threshold; to divide the hardened cement stone region into closed-pore hardened cement stone and open-pore hardened cement stone by combining the fluorescence intensity value of the ultraviolet-excited fluorescence band image; to obtain reflectance value by combining the visible light band image and the near-infrared band image with preset feature bands; to divide the natural aggregate region into siliceous natural aggregate and calcareous natural aggregate using the reflectance value; and to form a heterogeneous component distribution vector using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity region.

[0197] The identification module 30 is used to construct a visual aggregate distribution map from the heterogeneous component distribution vector and to perform visual classification of the regenerative orthopedic sample to be tested.

[0198] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying and classifying the components of recycled aggregate from waste concrete, characterized in that, The specific operating steps include: Multispectral image sequences under different spectral bands were obtained for the regenerative orthopedic samples to be tested. The multispectral image sequence includes visible light band images, near-infrared band images, and ultraviolet excited fluorescence band images; For each pixel in the near-infrared image, the local texture entropy in four directions is calculated according to a set of scale parameters to form a family of entropy change curves. Based on a preset scale parameter threshold, the family of entropy change curves is divided into hardened cement stone regions, natural aggregate regions, and impurity regions. The hardened cement stone regions are further divided into closed-pore hardened cement stone and open-pore hardened cement stone by combining the fluorescence intensity values ​​of the ultraviolet-excited fluorescence image. The visible light image and the near-infrared image are combined using preset feature band combinations to obtain reflectance values. The natural aggregate regions are then divided into siliceous natural aggregate and calcareous natural aggregate. Finally, a heterogeneous component distribution vector is formed using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity regions. A visual aggregate distribution map is constructed using the heterogeneous component distribution vector to perform visual classification of the regenerative orthopedic samples to be tested.

2. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 1, characterized in that, For each pixel in the near-infrared image, the local texture entropy in four directions is calculated according to a set of scale parameters to form a family of entropy change curves. Based on a preset scale parameter threshold, the family of entropy change curves is divided into hardened cement stone regions, natural aggregate regions, and impurity regions. The specific operation steps are as follows: A set of scale parameters is set for the near-infrared band images in the multispectral image sequence; a local window is obtained for each pixel in the near-infrared band image as the center, according to each scale parameter in the scale parameter set and four directions; the gray-level co-occurrence matrix of the pixel in each direction within the local window is calculated; the gray-level co-occurrence matrix is ​​normalized to obtain the probability matrix; Calculate the local texture entropy of the local window at the given scale parameter and orientation using each unit element in the probability matrix; Sort the local texture entropy of all scale parameters of the current pixel in ascending order according to the current direction to obtain the entropy change curve of that direction; obtain all entropy change curves of the current pixel in the four directions respectively to obtain the entropy change curve family. For each entropy change curve, calculate the first-order difference of the scale parameter arranged in ascending order to obtain a numerical derivative sequence; select the extreme points of the maximum and minimum numerical derivatives in the numerical derivative sequence; select the scale parameter corresponding to the first non-zero extreme point of all extreme points as the characteristic scale parameter in that direction; The comprehensive feature scale parameters are obtained by taking a weighted average of the feature scale parameters in all directions. Preset the first scale parameter threshold and the second scale parameter threshold; determine the relationship between the comprehensive feature scale parameter of the current pixel and the first scale parameter threshold and the second scale parameter threshold; If the comprehensive feature scale parameter is less than the first scale parameter threshold, then the pixel is determined to be a high-porosity texture area and marked as a hardened cement stone area. If the comprehensive feature scale parameter is greater than the second scale parameter threshold, then the pixel is determined to be a large-scale high-entropy region and marked as a natural aggregate region. If the comprehensive feature scale parameter is greater than or equal to the first scale parameter threshold and less than or equal to the second scale parameter threshold, then the pixel is determined to be an impurity region.

3. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 2, characterized in that, The hardened cement stone region was classified into closed-pore type hardened cement stone and open-pore type hardened cement stone by combining the fluorescence intensity values ​​of the ultraviolet excitation fluorescence band image. The specific operation steps are as follows: Obtain the comprehensive feature scale parameters of the pixels in the high-porosity texture region, find the closest scale parameter in the set of scale parameters, and extract the local texture entropy in the four directions of the closest scale parameter to form a direction entropy vector; The maximum and minimum values ​​in the directional entropy vector are selected, and the extreme value ratio of directional entropy is calculated. The fluorescence intensity value of the pixel in the high-porosity texture region at the corresponding coordinate position in the ultraviolet-excited fluorescence band image is obtained. The reflectance value of the pixel in the high-porosity texture region at a preset characteristic wavelength in the near-infrared band image is obtained. The ratio of the fluorescence intensity value to the reflectance value of the pixel in the high-porosity texture region is calculated as the fluorescence excitation efficiency index. An anisotropy threshold and an efficiency threshold are preset. If the extreme value ratio of the directional entropy of a pixel in the high-porosity texture region is greater than the anisotropy threshold and the fluorescence excitation efficiency index is less than the efficiency threshold, then the pixel is determined to be closed-pore hardened cement stone. If the extreme value ratio of the directional entropy of a pixel in the high-porosity texture region is less than or equal to the anisotropy threshold and the fluorescence excitation efficiency index is greater than or equal to the efficiency threshold, then the pixel is determined to be open-porosity hardened cement stone. If neither of the above determination conditions is met, then the pixel is determined to be ordinary hardened cement stone.

4. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 3, characterized in that, The reflectance is obtained by combining the visible light and near-infrared images using preset feature bands; the reflectance values ​​are used to divide the natural aggregate region into siliceous natural aggregate and calcareous natural aggregate; a heterogeneous component distribution vector is formed using the closed-pore hardened cement stone, open-pore hardened cement stone, siliceous natural aggregate, calcareous natural aggregate, and impurity regions. The specific operation steps are as follows: For the visible light and near-infrared images, reflectance values ​​are extracted using preset feature band combinations. Based on the reflectance values ​​of different feature band combinations, the siliceous natural aggregate and calcareous natural aggregate to which the pixels belong are identified. The recycled aggregate sample to be tested is processed using a sliding window, and the reflectance change rate and texture entropy gradient of the pixels in the sliding window are calculated. Based on the reflectivity change rate and texture entropy gradient, the spatial phase difference is calculated; the mean and standard deviation of the spatial phase difference are calculated, and then the dynamic threshold reference value is calculated; the recycled aggregate sample to be tested is screened to obtain its nominal maximum particle size, and the dynamic threshold reference value is corrected according to the nominal maximum particle size to obtain the final dynamic threshold. Based on the reflectivity change rate and texture entropy gradient, a heterogeneous interface response intensity map is generated for each pixel; the heterogeneous interface response intensity map is binarized using the final dynamic threshold to obtain the boundary line; based on the boundary line division result, the heterogeneous component distribution vectors of the natural aggregate region, hardened cement stone region, and impurity region are obtained.

5. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 4, characterized in that, For the visible light and near-infrared images, reflectance values ​​are extracted using preset feature band combinations. Based on the reflectance values ​​of different feature band combinations, the siliceous or calcareous natural aggregates to which the pixels belong are identified. The specific operation steps are as follows: The visible light band image and the near-infrared band image are combined using a preset feature band combination. For each characteristic band combination pair, reflectance values ​​are extracted from the corresponding visible light band image and near-infrared band image respectively; the ratio between the reflectance values ​​of the visible light band image and the near-infrared band image is calculated, and the ratios between all calculated reflectance values ​​are formed into a band ratio vector; the band ratio vector is projected onto the CIELAB color space to obtain the chromaticity coordinates in the CIELAB color space; Obtain the centers of the standard chromaticity clusters of siliceous minerals and calcareous minerals from the pre-constructed standard database; calculate the first Euclidean distance between the chromaticity coordinates and the centers of the standard chromaticity clusters of siliceous minerals; calculate the second Euclidean distance between the chromaticity coordinates and the centers of the standard chromaticity clusters of calcareous minerals; Multiple visible light characteristic wavelengths are preset, and the reflectance value of the pixel is obtained according to the preset multiple visible light characteristic wavelengths; the ratio between the reflectance values ​​of different visible light characteristic wavelengths is calculated as the feature ratio; all feature ratios are combined to obtain the visible light feature ratio vector. The fluorescence excitation efficiency index, the band ratio vector, and the visible light characteristic ratio vector are standardized and then combined to obtain the multimodal spectral characteristic vector. Preset Deviation threshold and fluorescence threshold; if the first Euclidean distance of the pixel is less than the deviation threshold and the fluorescence excitation efficiency index is less than the fluorescence threshold, then the pixel is determined to be siliceous natural aggregate; if the second Euclidean distance of the pixel is less than the deviation threshold and the fluorescence excitation efficiency index is greater than or equal to the fluorescence threshold, then the pixel is determined to be calcareous natural aggregate.

6. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 5, characterized in that, The recycled aggregate sample to be tested is subjected to sliding window processing, and the reflectance change rate and texture entropy gradient of the pixels in the sliding window are calculated. Based on the reflectance change rate and texture entropy gradient, the spatial phase difference is calculated. The mean and standard deviation of the spatial phase difference are then calculated to obtain the dynamic threshold baseline value. The recycled aggregate sample to be tested is sieved to obtain its nominal maximum particle size, and the dynamic threshold baseline value is corrected based on the nominal maximum particle size to obtain the final dynamic threshold. The specific operation steps are as follows: A sliding window is applied to the regenerative orthopedic sample to be tested. Each sliding window is uniformly divided into grids to generate grid points. N grid points are randomly selected from each sliding window as the center grid points, and their coordinate positions are determined. For all pixels in each sliding window, calculate the first-order gradient vector of reflectance value as the reflectance change rate; calculate the second-order gradient tensor of local texture entropy for all pixels in each sliding window as the texture entropy gradient; calculate the spatial phase difference based on the reflectance change rate and texture entropy gradient. Calculate the mean and standard deviation of the spatial phase difference of pixels within each sliding window; For each sliding window, obtain the coordinates of the window center point and the coordinates of the image center point of the regenerative orthopedic sample to be tested; calculate the central Euclidean distance between the coordinates of the window center point and the coordinates of the image center point. The weighting coefficients are calculated using the central Euclidean distance; the weighting coefficients are then used to calculate the mean and standard deviation of the spatial phase difference of all sliding windows, respectively, to obtain the weighted average and the weighted average standard deviation; the dynamic threshold benchmark value is then calculated using the weighted average and the weighted average standard deviation. The nominal maximum particle size of the recycled aggregate sample to be tested is sieved; a preset monotonically decreasing function is used; the particle size adjustment coefficient is calculated using the nominal maximum particle size and the monotonically decreasing function; the surface roughness parameter of the recycled aggregate sample to be tested is obtained; a preset monotonically increasing function is used; the roughness correction coefficient is calculated using the surface roughness parameter and the monotonically increasing function; the final dynamic threshold is calculated using the dynamic threshold reference value, the particle size adjustment coefficient, and the roughness correction coefficient.

7. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 6, characterized in that, The recycled aggregate sample to be tested is sieved to obtain its nominal maximum particle size, and the dynamic threshold benchmark value is corrected according to the nominal maximum particle size to obtain the final dynamic threshold. Based on the reflectivity change rate and texture entropy gradient, a heterogeneous interface response intensity map is generated for each pixel; the heterogeneous interface response intensity map is binarized using the final dynamic threshold to obtain the boundary line; based on the boundary line division result, the heterogeneous component distribution vectors of the natural aggregate region, hardened cement stone region, and impurity region are obtained. The specific operation steps are as follows: The heterogeneous interface response intensity is calculated for each pixel based on the reflectance change rate and texture entropy gradient, generating a heterogeneous interface response intensity map; the heterogeneous interface response intensity map is binarized and subjected to morphological operations using the final dynamic threshold to obtain the boundary line; the number of pixels inside the natural aggregate region, hardened cement stone region, and impurity region is counted based on the range of the boundary line, and the cosine value of the included angle is calculated. Based on the cosine value of the included angle, obtain the area fractions of the natural aggregate region, the hardened cement stone region, and the impurity region; construct the heterogeneous component distribution vector based on the area fractions. The cosine value of the included angle includes the cosine value of the siliceous included angle, the cosine value of the calcareous included angle, and the cosine value of the sealed hardening included angle; the area fraction includes the area fraction of siliceous natural aggregate, the area fraction of calcareous natural aggregate, the area fraction of sealed hardening cement stone, the area fraction of impurities, and the area fraction of the high water absorption risk subclass; the area fraction of the hardened cement stone region is divided into the area fraction of sealed hardening cement stone and the area fraction of the high water absorption risk subclass; in this scheme, the area fraction is not equal to the area, and the area fraction represents the proportion between each material.

8. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 7, characterized in that, The heterogeneous interface response intensity is calculated for each pixel based on the reflectance change rate and texture entropy gradient, generating a heterogeneous interface response intensity map. The map is then binarized and subjected to morphological operations using the final dynamic threshold to obtain the boundary line. The specific steps are as follows: The multimodal feature response value is obtained by multiplying the reflectance change rate with the texture entropy gradient. The pre-defined principal direction angle of the gradient vector of reflectivity change rate is The principal direction angle of the texture entropy gradient vector is Calculate the directional difference; multiply the multimodal characteristic response value with the same directional coefficient to obtain the heterogeneous interface response intensity and generate a heterogeneous interface response intensity map; The heterogeneous interface response intensity map is binarized using the final dynamic threshold to obtain a candidate heterogeneous interface pixel binary map. A morphological opening operation is performed on the pixels in the candidate heterogeneous interface pixel binary map according to a preset pixel radius to obtain a morphologically opened candidate heterogeneous interface pixel binary map. A morphological closing operation is then performed on the morphologically opened candidate heterogeneous interface pixel binary map using a preset cross-shaped window structure to obtain a repaired boundary binary map. Pixels in the repaired boundary binary map are extracted to form the boundary lines between the natural aggregate region, the hardened cement stone region, and the impurity region.

9. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 8, characterized in that, The number of pixels within the natural aggregate region, hardened cement stone region, and impurity region is counted based on the boundary line range, and the cosine value of the included angle is calculated. Based on the cosine value of the included angle, the area fractions of the natural aggregate region, hardened cement stone region, and impurity region are obtained. A heterogeneous component distribution vector is constructed based on each area fraction. The specific operation steps are as follows: Based on the boundary line division, the number of pixels within the natural aggregate region, hardened cement stone region, and impurity region is counted. The standard siliceous aggregate spectral vector and standard calcareous aggregate spectral vector of each pixel are obtained from the standard database. The cosine value of the siliceous angle is calculated using the multimodal spectral feature vector of the pixel and the standard siliceous aggregate spectral vector. The cosine value of the calcareous angle is also calculated using the multimodal spectral feature vector of the pixel and the standard calcareous aggregate spectral vector. Obtain the standard hardened cement stone spectral vector for each pixel from the standard database; The number of pixels in the hardened cement stone region with closed pores is counted; the closed pore correction factor is calculated by using the extreme ratio of the directional entropy of the pixels in the hardened cement stone with closed pores and the preset correction strength coefficient. The multimodal spectral feature vector is corrected using the closed pore correction factor to obtain the corrected multimodal spectral feature vector; The cosine value of the closed hardening angle is calculated by using the corrected multimodal spectral feature vector and the standard hardened cement stone spectral vector of each pixel in the hardened cement stone region; the standard impurity spectral vector of each pixel is obtained from the standard database; and the cosine value of the impurity angle is calculated by using the standard impurity spectral vector and the multimodal spectral feature vector.

10. The method for identifying and classifying the components of recycled aggregate from waste concrete according to claim 9, characterized in that, Based on the cosine value of the included angle, obtain the area fractions of the natural aggregate region, the hardened cement stone region, and the impurity region; construct a heterogeneous component distribution vector based on each area fraction, the specific operation steps are as follows: A preset matching threshold is set; the area fractions of siliceous natural aggregate, calcareous natural aggregate, closed hardened cement stone, and impurities are calculated by using the cosine values ​​of all the included angles mentioned above with the matching threshold; the number of pixels of the open-pore type hardened cement stone is obtained. The ratio between the number of pixels in the open-pore type hardened cement stone and the number of pixels in the hardened cement stone region is calculated and used as the area fraction of the high water absorption risk subclass. The area fractions of siliceous natural aggregate, calcareous natural aggregate, sealed hardened cement stone, impurities, and high water absorption risk subclasses were constructed as heterogeneous component distribution vectors.

Images