Method and system for analyzing quality of homologous solid waste recycled aggregate based on image processing

Abstract

The invention provides a homologous solid waste recycled aggregate quality analysis method and system based on image processing, and relates to the technical field of image recognition, and the method comprises the steps: obtaining 3D point cloud data of a to-be-analyzed aggregate particle placement area through scanning equipment, and obtaining an aggregate particle point cloud set; clustering the aggregate particle point cloud set to obtain a plurality of independent point cloud clusters and clustering determinacy; acquiring NIR spectral data of each to-be-analyzed aggregate particle, and analyzing to obtain the aggregate type and the type goodness of fit of each to-be-analyzed aggregate particle; extracting point cloud cluster features of the plurality of independent point cloud clusters; obtaining N quality analysis model branches, and outputting N quality parameters; and calculating an initial quality coefficient of the to-be-analyzed aggregate particles, correcting the initial quality coefficient by using the clustering certainty degree and the type goodness of fit of the to-be-analyzed aggregate particles to obtain a quality coefficient, and screening the aggregate particles. The technical problems that in the prior art, homologous solid waste recycled aggregate quality analysis is poor in accuracy and low in efficiency are solved.

Description

Technical Field

[0001] This invention relates to the field of image recognition, and more particularly to a method and system for quality analysis of recycled aggregates from homogeneous solid waste based on image processing. Background Technology

[0002] Homogeneous solid waste recycled aggregates refer to recyclable aggregates formed from the same type of industrial solid waste or construction waste, such as concrete waste and cement slag, after being processed through crushing, screening, and impurity removal. With the advancement of "dual carbon" (carbon diversification and carbon sequestration), the increasing requirements for the resource utilization rate of construction solid waste, and the growing scarcity of natural aggregate resources, the demand for the resource utilization of homogeneous solid waste recycled aggregates in fields such as construction engineering is increasing, and the requirements for precise control of aggregate quality are also increasing accordingly.

[0003] However, traditional aggregate quality analysis methods usually rely on manual sorting to identify aggregate types and physical experiments to test quality. This results in poor accuracy and low efficiency in quality analysis, which seriously restricts the large-scale promotion of the utilization of recycled aggregate resources.

[0004] Therefore, there is an urgent need for a quality analysis method for recycled aggregates from homogeneous solid waste based on image processing, which can solve the pain points of traditional methods and support the efficient resource utilization of recycled aggregates. Summary of the Invention

[0005] This invention addresses the technical problems of poor accuracy and low efficiency in the quality analysis of recycled aggregates from homogeneous solid waste in existing technologies by providing a method and system for quality analysis of recycled aggregates from homogeneous solid waste based on image processing.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing, including: Use a scanning device to acquire 3D point cloud data of the area where the aggregate particles to be analyzed are placed, and remove the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed. Based on a preset clustering distance threshold, aggregate particle point clouds are clustered to obtain multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents an aggregate particle. The NIR spectral data of each aggregate particle to be analyzed is obtained using a scanning device. Based on the NIR spectral data and a preset spectral fingerprint library, the aggregate type and type matching degree of each aggregate particle to be analyzed are obtained. Extract point cloud cluster features from the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume; Obtain N quality analysis model branches trained based on machine learning, input the point cloud cluster features of the aggregate particles to be analyzed and the aggregate type into the N quality analysis model branches respectively, and obtain N quality parameters of the aggregate particles to be analyzed, wherein the N quality parameters include at least crushing value and water absorption rate. Based on preset rules, the initial quality coefficient of the aggregate particles to be analyzed is calculated according to the aggregate type and N quality parameters. The initial quality coefficient is then corrected using the cluster certainty and type fit of the aggregate particles to be analyzed to obtain the quality coefficient. Based on the preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are screened.

[0007] Secondly, the present invention provides a quality analysis system for recycled aggregates from homogeneous solid waste based on image processing, comprising: The point cloud acquisition module is used to acquire 3D point cloud data of the area where the aggregate particles to be analyzed are placed using a scanning device, and remove the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed. The point cloud clustering module is used to cluster aggregate particle point clouds based on a preset clustering distance threshold, to obtain multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents an aggregate particle. The type analysis module is used to acquire NIR spectral data of each aggregate particle to be analyzed using a scanning device. Based on the NIR spectral data and a preset spectral fingerprint library, the aggregate type and type matching degree of each aggregate particle to be analyzed are obtained. The feature extraction module is used to extract the point cloud cluster features of the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume. The quality analysis module is used to obtain N quality analysis model branches trained based on machine learning. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are respectively input into the N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed. The N quality parameters include at least the crushing value and water absorption rate. The output filtering module is used to calculate the initial quality coefficient of the aggregate particles to be analyzed based on preset rules, according to the aggregate type and N quality parameters of the aggregate particles to be analyzed, and to correct the initial quality coefficient using the cluster certainty and type matching degree of the aggregate particles to be analyzed, thereby obtaining the quality coefficient. Based on the preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are filtered.

[0008] The beneficial effects of this invention are: Compared to existing technologies, this application first uses a scanning device to acquire 3D point cloud data of the aggregate particle placement area to be analyzed, obtaining aggregate particle point clouds. This accurately preserves the three-dimensional morphological information of the aggregate to be analyzed, providing a clean and reliable raw data foundation for subsequent point cloud clustering, feature extraction, and other steps. Secondly, the aggregate particle point clouds are clustered to obtain multiple independent point cloud clusters and clustering certainty, providing independent data units for subsequent single aggregate feature extraction and quality analysis, effectively ensuring the accuracy of the overall analysis. Thirdly, the NIR spectral data of each aggregate particle to be analyzed is obtained, and the aggregate type and type conformity of each aggregate particle are analyzed, achieving automated and accurate determination of the type of each aggregate. Furthermore, the point cloud cluster features of multiple independent point cloud clusters are extracted, transforming the discrete point cloud clusters of the aggregate into quantifiable geometric features, providing reliable data support for the subsequent accurate determination of the quality of recycled aggregates from the same source of solid waste. Furthermore, by acquiring N quality analysis model branches and inputting N quality parameters of the aggregate particles to be analyzed, efficient and non-destructive accurate prediction of the quality parameters of the aggregate particles is achieved, providing reliable data support for the automated evaluation of the quality of recycled aggregates from the same source of solid waste. Finally, based on preset rules, quality coefficients are calculated to screen the aggregate particles, achieving accurate and objective screening of the quality of recycled aggregates from the same source of solid waste.

[0009] Through the above technical solutions, this application constructs a fully automated technical chain for quality analysis of recycled aggregates from homogeneous solid waste through 3D point cloud clustering, NIR spectral recognition, machine learning, quality prediction, and error correction. This improves the accuracy and efficiency of quality analysis of recycled aggregates from homogeneous solid waste, and provides reliable technical support for the large-scale resource utilization and industrial application of recycled aggregates from homogeneous solid waste. Attached Figure Description

[0010] Figure 1 A flowchart illustrating the image processing-based method for analyzing the quality of recycled aggregates from homogeneous solid waste provided by this invention. Figure 2 A schematic diagram of the structure of the image processing-based homogeneous solid waste recycled aggregate quality analysis system provided by the present invention.

[0011] In the attached diagram, the components represented by each number are as follows: Point cloud acquisition module 11, point cloud clustering module 12, type analysis module 13, feature extraction module 14, quality analysis module 15, output filtering module 16. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0014] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0015] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing, including: S10: Use a scanning device to acquire 3D point cloud data of the area where the aggregate particles to be analyzed are placed, and remove the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed.

[0016] In the industrial quality analysis of recycled aggregates from the same source of solid waste, traditional methods for obtaining the three-dimensional morphology (such as volume and convexity) of aggregates mostly rely on manual measurement or two-dimensional imaging. However, manual measurement suffers from low efficiency and large errors, while two-dimensional imaging cannot reproduce three-dimensional features, making it difficult to support subsequent accurate quality assessment.

[0017] To address the aforementioned issues, this application uses a scanning device to acquire 3D point cloud data of the area where the aggregate particles to be analyzed are placed, and removes the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed.

[0018] Specifically, step S10 in the method includes: A scanning device is deployed on the aggregate screening facility, the scanning device integrating a 3D scanner and an NIR spectral probe; Trigger the 3D scanner to acquire 3D point cloud data of the entire aggregate particle placement area to be analyzed; Using the RANSAC plane fitting algorithm, planar point clouds representing the background plate are quickly identified and removed from 3D point cloud data, resulting in an aggregate particle point cloud set containing only multiple aggregate particles to be analyzed.

[0019] In this embodiment, a scanning device is first deployed on the aggregate screening facility. This scanning device integrates a 3D scanner and a NIR spectral probe: the 3D scanner collects 3D point cloud data of the area where the aggregate particles to be analyzed are placed, accurately capturing the three-dimensional morphological details of each aggregate particle, including its outline, surface protrusions and depressions, through the spatial coordinate information of millions of points; the NIR spectral probe collects near-infrared spectral data of individual aggregate particles, analyzing the composition information of the aggregate by leveraging the correspondence between spectral characteristics and material composition, providing a basis for subsequent determination of the aggregate type.

[0020] For example, scanning equipment integrating a 3D scanner and an NIR spectral probe is deployed in key locations within the aggregate screening facility. Common deployment locations include beside the detection section of the aggregate conveyor belt or directly above a dedicated detection platform. This deployment method ensures that the scanning range of the equipment completely covers the area where the aggregate particles to be analyzed are placed, avoiding data omissions due to positional deviations, while providing a stable equipment foundation for the accurate acquisition of 3D point cloud data and NIR spectral data.

[0021] Secondly, a 3D scanner is triggered to acquire 3D point cloud data of the entire aggregate placement area to be analyzed. For example, after the aggregate particles are transported to the placement area, the 3D scanner is automatically or manually triggered to perform a complete scan of the entire area, generating 3D point cloud data. This 3D point cloud data contains all the three-dimensional information of the aggregate and the background plate. This 3D scan of the entire aggregate placement area ensures that no aggregate particle is missed, providing a complete data foundation for subsequent point cloud cluster analysis of individual aggregate particles.

[0022] Finally, since 3D point cloud data contains all the three-dimensional information of the aggregate and the background plate, in order to focus on aggregate analysis, the RANSAC plane fitting algorithm can be used to quickly identify and remove the planar point cloud representing the background plate from the 3D point cloud data, obtaining an aggregate particle point cloud set containing only multiple aggregate particles to be analyzed. The RANSAC plane fitting algorithm is used because the background plate of aggregate screening facilities (such as inspection platforms, conveyor belt surfaces, etc.) is usually a regular plane, and its point cloud will exhibit obvious planar features. Conversely, aggregate particles themselves are irregular three-dimensional shapes, and their point cloud distribution is scattered and without a fixed planar pattern, showing significant differences from the planar point cloud of the background plate. The core advantage of the RANSAC plane fitting algorithm is that it can quickly lock planar features between the planar point cloud of the background plate and the non-planar point cloud of the aggregate. Then, through random sampling and verification of point cloud consistency, the planar point cloud belonging to the background plate can be accurately identified. Even if a small number of non-planar points of the aggregate are mixed in, the RANSAC plane fitting algorithm can also eliminate these abnormal points through planar consistency verification. After identification, the planar point cloud representing the background is removed, resulting in an aggregate particle point cloud set containing only multiple aggregate particles to be analyzed. This completely removes background interference and provides clean and pure raw data for subsequent point cloud clustering, feature extraction and other steps.

[0023] In summary, compared to existing technologies, this application uses a scanning device to acquire 3D point cloud data of the aggregate particle placement area to be analyzed, and removes the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed. This accurately preserves the three-dimensional morphological information of the aggregate to be analyzed, providing a clean and reliable raw data foundation for subsequent point cloud clustering, feature extraction, and other steps, effectively avoiding background interference and improving the accuracy and efficiency of the overall quality analysis.

[0024] S20: Based on a preset clustering distance threshold, cluster the aggregate particle point cloud to obtain multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents an aggregate particle.

[0025] In the quality analysis of recycled aggregates from the same source solid waste, the aggregate particle point cloud after background planar point cloud removal is a mixed point cloud data containing multiple aggregates. Subsequent extraction of 3D sphericity, volume and other features of individual aggregates, as well as analysis of quality parameters, all require individual aggregates as independent analysis units. Traditional methods of manually sorting individual aggregates are extremely inefficient and cannot be adapted to industrial continuous production. At the same time, aggregate particles have the characteristics of irregular shape and large size difference. The point clouds of adjacent aggregates in the mixed point cloud are prone to local overlap. If the splitting is based solely on simple distance judgment, it is easy to make misjudgments such as multiple aggregate point clouds clustering into one cluster or a single aggregate point cloud being split into multiple clusters.

[0026] To address the aforementioned issues, this application clusters aggregate particle point clouds based on a preset clustering distance threshold, obtaining multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, wherein each independent point cloud cluster represents an aggregate particle.

[0027] Specifically, step S20 in the method includes: Create an unprocessed point list containing all points that need to be clustered; Randomly select a point from the list of unprocessed points as a seed point. Based on the preset clustering distance threshold, search and aggregate all unprocessed neighboring points within the clustering distance threshold to obtain a new aggregated point. The new aggregation point is used as a new seed point to continue clustering until the clustering can no longer expand, thus obtaining an independent point cloud cluster. The number of clusterings when the independent point cloud cluster is formed is counted. Based on the number of clusterings when the independent point cloud cluster is formed, the clustering certainty of each independent point cloud cluster is calculated, wherein the clustering certainty is inversely proportional to the number of clusterings. Select the next seed point from the unprocessed points and perform clustering until all points are clustered, obtaining multiple independent point cloud clusters, each independent point cloud cluster representing an aggregate particle.

[0028] In this embodiment, an unprocessed point list is first created, which contains all points that need to be clustered. For example, the unprocessed point list is first initialized to an empty set, and then the aggregate particle point cloud containing multiple aggregates to be analyzed is imported into the list. Thus, the unprocessed point list covers all points that need to be clustered, becoming the basic data pool for subsequent clustering operations.

[0029] Secondly, a point is randomly selected from the list of unprocessed points as a seed point. Based on a preset clustering distance threshold, all unprocessed neighboring points within the threshold are searched and aggregated to obtain a new aggregated point. The point cloud aggregation based on the preset clustering distance threshold is based on the fact that the point cloud of the same aggregate exhibits a continuous and dense distribution in space, meaning that there must be an upper limit to the spatial distance between adjacent points on the same aggregate.

[0030] For example, the preset clustering distance threshold can be set according to the common particle size range of the aggregate particles to be analyzed: since the aggregate particle size directly reflects the approximate size of the particles, the distance between adjacent points will not significantly exceed a certain proportion of the particle size. Therefore, using the particle size range as a reference can ensure the rationality of the preset clustering distance threshold. For example, for aggregates with a common particle size of 10-30mm, those skilled in the art can set the preset clustering distance threshold to 5-8mm, and then randomly select a point from the list of unprocessed points as a seed point, and filter out all unprocessed neighboring points whose spatial distance from the seed point is less than the preset clustering distance threshold, thus aggregating them to form a new aggregation point.

[0031] Next, following the same logic and method, the new aggregation point is used as a new seed point to continue clustering, and this iterative process is repeated until the clustering can no longer expand (i.e., no new neighboring points can be included). This yields an independent point cloud cluster, representing one aggregate particle. Simultaneously, the number of clustering operations during the formation of each independent point cloud cluster is counted. Based on this number of clustering operations, the clustering certainty of each independent point cloud cluster is calculated. The clustering certainty is inversely proportional to the number of clustering operations. This is because fewer clustering operations indicate a denser distribution of points within the independent point cloud cluster, easier aggregation, and higher reliability of the clustering results, thus quantifying the credibility of the clustering process for a single aggregate particle. Optionally, the formula for calculating the clustering certainty is: Clustering Certainty = 1 / Number of Clustering Operations. Fewer clustering operations result in higher clustering certainty and higher credibility of the independent point cloud cluster cluster.

[0032] For example, if after 5 clustering iterations no new neighboring points can be included, meaning the clustering cannot be expanded further, then an independent point cloud cluster is obtained, with a clustering certainty of 1 / 5 = 0.2. The clustering certainty reflects the reliability of the clustering process of this independent point cloud cluster.

[0033] Finally, once an independent point cloud cluster is formed, all points contained within it are removed from the list of unprocessed points. Following the same logic and method as generating an independent point cloud cluster, the next seed point is selected from the unprocessed points for clustering, until all points are clustered, resulting in multiple independent point cloud clusters, each representing a single aggregate particle. This achieves precise decomposition of aggregate particle point clouds from multiple aggregates to be analyzed into independent point cloud clusters of a single aggregate, providing a reliable data foundation for subsequent extraction of aggregate features and analysis of quality parameters.

[0034] In summary, compared to existing technologies, this application clusters aggregate particle point clouds based on a preset clustering distance threshold, obtaining multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents one aggregate particle. In this way, the aggregate particle point cloud is clustered and decomposed into multiple independent point cloud clusters with a single aggregate particle as the unit. Simultaneously, the clustering certainty quantifies the reliability of the clustering results, providing independent data units for subsequent single aggregate feature extraction and quality analysis, and also providing a reliable basis for subsequent quality coefficient correction, effectively ensuring the accuracy of the overall analysis.

[0035] S30: Use a scanning device to acquire NIR spectral data for each aggregate particle to be analyzed. Based on the NIR spectral data and a preset spectral fingerprint library, analyze and obtain the aggregate type and type matching degree for each aggregate particle to be analyzed.

[0036] In the quality analysis of recycled aggregates from the same source of solid waste, traditional aggregate type identification relies on manual sorting, which is not only inefficient and highly subjective, making it difficult to adapt to continuous industrial production, but also prone to misclassification due to human judgment errors, thus affecting the accuracy of subsequent analysis of core quality parameters such as crushing value and water absorption rate.

[0037] Meanwhile, aggregates with different compositions (such as concrete containing cement and sand, and red bricks containing clay) have unique material compositions, and their NIR spectral curves have significant differences in absorption peak positions, reflectance intensity, and other characteristics. This characteristic of differences in composition corresponding to differences in spectral features is unique and stable.

[0038] Therefore, the NIR spectral data of the aggregate particles to be analyzed can be obtained to accurately identify the aggregate type.

[0039] To address the aforementioned issues, this application uses a scanning device to acquire NIR spectral data for each aggregate particle to be analyzed. Based on the NIR spectral data and a preset spectral fingerprint library, the aggregate type and type matching degree of each aggregate particle to be analyzed are obtained.

[0040] Specifically, step S30 in the method includes: Collect aggregate particle samples of known types, scan their NIR spectra, and form a standard spectral database. The standard spectral database contains two types of data: spectral characteristics and type labels. The aggregate particles to be analyzed are scanned using a scanning device to obtain their spectral characteristics. The standard spectral feature most similar to the spectral feature to be analyzed is retrieved from the standard spectral database and used as the approximate spectral feature. The type label of the approximate spectral feature is extracted and used as the aggregate type of the aggregate particles to be analyzed. The similarity between the approximate spectral features and the spectral features to be analyzed is used as the type matching degree of the aggregate particles to be analyzed.

[0041] In this embodiment, aggregate particle samples of known types are first collected, and their NIR spectra are scanned to form a standard spectral database. The standard spectral database contains two types of data: spectral features and type labels. For example, the same NIR spectral probe used in step S10 can be used to scan aggregate particle samples of known types (such as concrete blocks, broken red bricks, cement waste, etc.), and the spectral features of each known type of aggregate particle sample, such as NIR spectral curves, can be extracted. The spectral features and type labels of each known type of aggregate particle sample are then correlated to construct the standard spectral database. This provides a unified comparison benchmark for the spectral features of aggregates to be analyzed subsequently.

[0042] Next, a scanning device is used to scan the aggregate particles to be analyzed to obtain the spectral characteristics to be analyzed. For example, the NIR spectral probe in the scanning device deployed on the aggregate screening facility in step S10 is used to scan each aggregate particle to be analyzed, and the spectral characteristics of each aggregate particle to be analyzed, such as the NIR spectral curve, are extracted as the spectral characteristics to be analyzed.

[0043] Next, the standard spectral feature most similar to the spectral feature to be analyzed is retrieved from the standard spectral database. This approximate spectral feature is then extracted as its type label, which is used to determine the aggregate type of the aggregate particles to be analyzed. For example, the spectral feature to be analyzed can be input into the standard spectral database. A similarity algorithm, such as cosine similarity or Euclidean distance, is used to retrieve the most similar standard spectral feature from the database. This approximate spectral feature is then used as the approximate spectral feature. Since the spectral features in the standard spectral database have explicit type labels, the aggregate type of the aggregate particles to be analyzed is obtained by extracting the type label corresponding to the approximate spectral feature.

[0044] For example, the cosine similarity between the spectral feature to be analyzed and all standard spectral features in the standard spectral database is calculated one by one. If the spectral feature to be analyzed has the highest similarity with the "standard spectral feature of concrete sample" among all similarity calculation results, the aggregate type of the aggregate particles to be analyzed can be determined to be "concrete". In this way, the correspondence between the aggregate particles to be analyzed and the aggregate type is determined through NIR spectral similarity analysis.

[0045] Finally, the similarity between the approximate spectral features and the spectral features to be analyzed is used as the type matching degree of the aggregate particles to be analyzed. For example, if the similarity between the approximate spectral features and the spectral features to be analyzed is 0.92 by calculating the cosine similarity, then 0.92 is used as the type matching degree of the aggregate particles to be analyzed. The higher the similarity, the smaller the compositional difference between the aggregate particles to be analyzed and the known type of aggregate particles, and the more reliable the determination of the aggregate type; conversely, the lower the similarity, the greater the compositional difference between the aggregate particles to be analyzed and the known type of aggregate particles, and the less reliable the determination of the aggregate type.

[0046] In summary, compared to existing technologies, this application uses a scanning device to acquire NIR spectral data for each aggregate particle to be analyzed. Based on the NIR spectral data and a pre-set spectral fingerprint library, it analyzes and obtains the aggregate type and type matching degree of each aggregate particle. This replaces traditional manual sorting and identification, achieving automated and accurate determination of the type of each aggregate particle. Furthermore, it quantifies the reliability of the identification results through type matching, providing an accurate type basis and a reliable reference for subsequent aggregate quality coefficient calculation and overall quality analysis.

[0047] S40: Extract the point cloud cluster features of the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume.

[0048] In the quality analysis of recycled aggregates from the same source of solid waste, traditional aggregate geometric feature assessment relies on manual observation of shape, which cannot accurately quantify key geometric features affecting engineering performance. However, the geometric features of recycled aggregates are directly related to aggregate flowability (e.g., the roundness of the shape affects the frictional resistance between aggregates) and structural strength (e.g., uneven shape easily leads to stress concentration). The aforementioned steps obtain independent point cloud clusters of individual aggregate particles through clustering, but discrete point clouds cannot directly reflect the geometric quality of the aggregates; therefore, quantifiable feature parameters need to be extracted from them.

[0049] To address the aforementioned issues, this application extracts point cloud cluster features from the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume.

[0050] Specifically, step S40 in the method includes: Calculate the 3D convex hull of the point cloud cluster, and calculate the volume and surface area of ​​the 3D convex hull. Based on the volume and surface area of ​​the 3D convex hull, calculate the 3D sphericity using the sphericity formula. After triangulating the point cloud cluster to generate a mesh model, the volume of the mesh model is calculated as the actual volume of the aggregate particles corresponding to the point cloud cluster. The convexity of the point cloud cluster is calculated based on the actual volume of the point cloud cluster and the 3D convex hull volume. Find the shortest inertial axis of the point cloud cluster, divide the point cloud cluster into several slices along the direction of the shortest inertial axis, calculate the minimum cross-sectional area and the average cross-sectional area of ​​the slices, and calculate the moment imbalance rate based on the minimum cross-sectional area and the average cross-sectional area.

[0051] In this embodiment, the 3D convex hull of the point cloud cluster is first calculated, and its volume and surface area are calculated. Based on the volume and surface area of ​​the 3D convex hull, the 3D sphericity is calculated using the sphericity formula. The 3D convex hull refers to the smallest convex polyhedron that can completely enclose the point cloud cluster. The volume and surface area of ​​the 3D convex hull accurately reflect the contour characteristics of the aggregate. In practice, mature point cloud processing algorithms such as the Qhull algorithm and a 3D extension of the Graham scan method can be used. By calculating the spatial convex hull of the discrete point cloud, the 3D convex hull of the point cloud cluster is constructed, ensuring that the 3D convex hull fits the outermost contour of the aggregate point cloud, providing a reliable external geometric basis for the accurate calculation of subsequent 3D sphericity and convexity.

[0052] For example, since the 3D convex hull is essentially a closed convex polyhedron composed of multiple triangular facets, the calculation process for its volume and surface area can be as follows: When calculating the surface area, traverse all the triangular facets of the 3D convex hull, and calculate the area of ​​a single facet using the spatial coordinates of the three vertices of each facet, or using the cross product of vectors or Heron's formula. Then, sum the areas of all facets to obtain the surface area of ​​the 3D convex hull. When calculating the volume, the 3D convex hull can be decomposed into multiple small tetrahedrons with the centroid of the convex hull as the vertex and each triangular facet as the base. Calculate the volume of each small tetrahedron using the tetrahedron volume formula, and then sum the volumes of all small tetrahedrons to obtain the volume of the 3D convex hull.

[0053] Among them, 3D sphericity refers to the geometric characteristic index calculated based on the volume and surface area of ​​3D convex hull combined with the sphericity formula. It is used to quantify the similarity between the shape of aggregate and an ideal sphere. The closer the 3D sphericity is to 1, the closer the shape of the aggregate is to an ideal sphere, and the higher the regularity of the shape. In engineering applications, it is more conducive to improving the fluidity of concrete mixtures, while reducing the amount of cementitious materials used, and reducing construction difficulty and cost.

[0054] For example, 3D sphericity is calculated based on the volume and surface area of ​​the 3D convex hull combined with the sphericity formula, wherein the formula for calculating 3D sphericity is: Where V is the volume of the 3D convex hull and S is the surface area of ​​the 3D convex hull. For example, if the volume of the 3D convex hull is 1000 mm²... 3 The surface area of ​​the 3D convex hull is 600 mm². 2 Substituting into the formula, we can calculate the 3D sphericity = =0.68, indicating that the similarity between the aggregate shape and a sphere is 0.68.

[0055] Secondly, after triangulating the point cloud cluster to generate a mesh model, the volume of the mesh model is calculated as the actual volume of the aggregate particles corresponding to the point cloud cluster. Based on the actual volume of the point cloud cluster and the 3D convex hull volume, the convexity of the point cloud cluster is calculated. For example, discrete points in the point cloud cluster can be connected into continuous triangular mesh surfaces using a greedy projection triangulation algorithm to form a mesh model that highly matches the actual shape of the aggregate. Then, based on the spatial coordinates of each vertex of the mesh surface, the mesh model is decomposed into multiple closed tetrahedrons with the mesh vertices as vertices. The volume of each tetrahedron is calculated and summed to obtain the volume of the mesh model, which serves as the actual volume of the aggregate particles corresponding to the point cloud cluster, accurately reflecting the spatial occupancy range of the aggregate point cloud.

[0056] For example, the convexity of a point cloud cluster = the actual volume of the point cloud cluster / the volume of the 3D convex hull. Since the 3D convex hull is the smallest convex body that encloses the aggregate, the volume of the 3D convex hull must be greater than or equal to the actual volume of the aggregate particles. If the aggregate particles are regularly shaped, the actual volume of the point cloud cluster is close to the volume of the 3D convex hull, and the convexity of the point cloud cluster is close to 1. Conversely, if the aggregate particles are irregularly shaped (such as having deep depressions), the actual volume of the point cloud cluster is much smaller than the volume of the 3D convex hull, and the convexity of the point cloud cluster is relatively small.

[0057] For example, if a mesh model is generated by greedy projection triangulation and the volume is calculated, the actual volume of the aggregate particles corresponding to the point cloud cluster is found to be 850 mm. 3 The 3D convex hull volume is 1000 mm. 3 If the convexity of the point cloud cluster is 850 / 1000 = 0.85, it means that the actual volume of the aggregate accounts for 85% of the volume of its smallest circumscribed convex hull. This indicates that although there are some local depressions in the shape of the aggregate, the overall fullness is high and the shape regularity is good.

[0058] Finally, the shortest axis of inertia of the point cloud cluster is found. The cluster is then divided into several slices along this axis. The minimum and average cross-sectional areas of each slice are calculated, and the moment imbalance rate is obtained based on these areas. The axis of inertia is the axis of symmetry reflecting the mass distribution of the point cloud, and the shortest axis of inertia refers to the axis along which the aggregate extends the shortest distance, i.e., the narrowest direction. The moment imbalance rate reflects the spatial uniformity of the aggregate distribution. A higher rate indicates a shape that is more biased towards a particular direction, such as needle-like or strip-like structures, which are prone to stress concentration in engineering applications. The moment imbalance rate is calculated as: 1 - minimum cross-sectional area / average cross-sectional area.

[0059] For example, first, the shortest axis of inertia of the point cloud cluster is determined through point cloud moment of inertia analysis. Then, along the direction of the shortest axis of inertia, the point cloud cluster is uniformly divided into several parallel slices (e.g., slices every 0.5 mm). Subsequently, the area of ​​the region where each slice plane intersects with the point cloud cluster is calculated as the cross-sectional area at the corresponding position, and the minimum cross-sectional area (e.g., 20 mm) is extracted from it. 2 ), and simultaneously calculate the average cross-sectional area (e.g., 30mm²). 2 At this point, the moment imbalance rate = 1 - (20mm) 2 / 30mm 2 The value of ≈0.33 indicates that the cross-sectional area of ​​the aggregate varies greatly along the shortest axis of inertia, the shape is obviously uneven in thickness, and the overall symmetry and balance are weak.

[0060] In summary, compared to existing technologies, this application extracts point cloud cluster features from the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume. In this way, the discrete point cloud clusters of aggregates are transformed into quantifiable geometric features, providing reliable data support for subsequent calculation of the aggregate comprehensive quality coefficient based on multi-dimensional features and for the automated and accurate determination of the quality of recycled aggregates from the same source of solid waste.

[0061] S50: Obtain N quality analysis model branches trained based on machine learning, input the point cloud cluster features of the aggregate particles to be analyzed and the aggregate type into the N quality analysis model branches respectively, and obtain N quality parameters of the aggregate particles to be analyzed, wherein the N quality parameters include at least the crushing value and water absorption rate.

[0062] In the quality analysis of recycled aggregates from the same source of solid waste, the traditional methods of obtaining quality parameters such as crushing value and water absorption rate rely on laboratory physical tests, such as crushing test and saturated weighing test. These methods are not only time-consuming and destructive, but also difficult to adapt to the continuous and batch testing requirements of industrial production lines for aggregates.

[0063] Meanwhile, the point cloud cluster characteristics of aggregate particles (3D sphericity, convexity, moment imbalance rate, volume) are intrinsically related to aggregate type and quality parameters. For example, 3D sphericity affects water absorption rate, and aggregate type affects basic strength, which in turn affects crushing value.

[0064] To address the aforementioned issues, this application obtains N quality analysis model branches trained based on machine learning. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are respectively input into the N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed, wherein the N quality parameters include at least the crushing value and water absorption rate.

[0065] Specifically, step S50 in the method includes: Multiple batches of typical aggregate particle samples were selected, and the point cloud cluster features and aggregate types of multiple batches of typical aggregate particle samples were collected as typical aggregate features. Obtain N quality parameters for each typical aggregate sample, wherein the N quality parameters include at least crushing value and water absorption rate; Using typical aggregate characteristics as training data and N quality parameters as supervision data, N quality analysis model branches are trained respectively. Among them, the N quality parameters used as supervision data are all measured values. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are respectively input into N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed.

[0066] In this embodiment, to ensure that the machine learning model can adapt to the diverse aggregate conditions in actual production, multiple batches of typical aggregate particle samples are first selected, and the point cloud cluster features and aggregate types of these samples are collected as typical aggregate features. The collected batches of typical aggregate particle samples need to cover the main types of recycled aggregates from the same source of solid waste, such as concrete blocks, broken red bricks, and cement waste, while also including different geometric shapes, such as high sphericity and low sphericity, high convexity and low convexity, and differences in size and volume, to avoid insufficient model generalization ability due to a single sample.

[0067] For example, select multiple batches, totaling at least 2000 typical aggregate particle samples covering common types such as concrete blocks, red brick fragments, and cement waste, and containing different geometric shapes. Extract the point cloud cluster features of each typical aggregate particle sample using the same method as in step S40. At the same time, combine the NIR spectral recognition method in step S30 and physical observation, manually label the aggregate type corresponding to each typical aggregate particle sample, and then integrate the point cloud cluster features with the aggregate types one by one to form the typical aggregate features required for model training.

[0068] Secondly, N quality parameters are obtained for each typical aggregate sample. These N quality parameters include at least the crushing value and water absorption rate. The crushing value can be determined through an aggregate crushing test. The crushing value reflects the aggregate's ability to resist crushing by external forces and is a key indicator for determining whether the aggregate can be used in high-strength concrete. The water absorption rate can be determined by the saturated weighing method and is related to the water demand of the concrete mix and the final structural durability.

[0069] Secondly, using typical aggregate characteristics as training data and N quality parameters as supervisory data, N quality analysis model branches are trained respectively. The N quality parameters used as supervisory data are all measured values. For example, if the quality parameters only include crushing value and water absorption rate, then using typical aggregate characteristics as training data and two quality parameters as supervisory data, two quality analysis model branches are trained respectively.

[0070] For example, if the quality parameters only include crushing value and water absorption rate, then typical aggregate characteristics are used as training data, and the measured values ​​of crushing value and water absorption rate are used as supervision data to train two independent quality analysis model branches. One quality analysis model branch focuses on learning the mapping relationship between typical aggregate characteristics and crushing value to achieve crushing value prediction, while the other quality analysis model branch focuses on learning the mapping relationship between typical aggregate characteristics and water absorption rate to achieve water absorption rate prediction. Independent training avoids mutual interference between parameters and improves the prediction accuracy of each parameter.

[0071] For example, the quality analysis model branch can be trained using the following technical path: 1. Data preparation: The typical aggregate characteristics and the corresponding two quality parameters are divided into training set, validation set and test set in a ratio of 7:1.5:1.5.

[0072] 2. Model Construction: Two quality analysis model branches can be constructed using random forests. One quality analysis model branch is used for crushing value prediction, and the other quality analysis model branch is used for water absorption rate prediction. The two have the same structural framework but are optimized for different parameters. A single model branch mainly consists of a basic unit, feature sampling, sample sampling, splitting criteria, and output fusion. The basic unit is composed of 100-500 independent classification and regression trees forming the main body of the forest. The number of trees is determined through validation set performance tuning. Feature sampling randomly selects 3-4 features from point cloud cluster features and aggregate types as splitting candidate sets. Sample sampling uses sampling with replacement to extract a subset of the training set with an amount equivalent to the original sample size, and assigns independent training samples to each tree. The splitting criterion aims to minimize the mean squared error (MSE) of node splitting in each tree. By calculating the MSE of child nodes under different feature thresholds, the threshold that minimizes the total MSE after splitting is selected to complete the node splitting until a preset tree depth (e.g., a maximum depth of 10-20 layers) or a leaf node sample size threshold (e.g., ≤5 samples) is reached. Output fusion is the average of the prediction results of all classification and regression trees.

[0073] 3. Model Training: Taking the training process of a quality analysis model branch as an example, the typical aggregate features in the training set are used as input features, the corresponding quality parameters (crushing value or water absorption rate) are used as supervision labels, and the MSE of the training set is used as the loss function. The splitting threshold and structure of each tree are optimized iteratively. At the same time, the MSE of the validation set is calculated after each iteration to evaluate the model prediction error. When the MSE of the validation set no longer decreases (or the decrease is ≤0.01) in 10 consecutive iterations, and the deviation between the MSE of the test set and the MSE of the validation set is ≤5%, it is considered to have converged. Training is stopped and the model parameters at this time are saved, resulting in a quality analysis model branch that has been trained. The other quality analysis model branch is trained in the same way, resulting in two quality analysis model branches.

[0074] Finally, the point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are input into N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed. For example, the point cloud cluster features of the aggregate particles to be analyzed (such as 3D sphericity 0.68, convexity 0.85, moment imbalance rate 0.33, and volume 850 mm²) are input into the model. 3The data is input into two quality analysis model branches, one for the aggregate type (e.g., concrete) and the other for the aggregate itself. This yields two quality parameters for the aggregate particles to be analyzed, such as a crushing value of 18% and a water absorption rate of 5.2%. Based on point cloud cluster features and the inherent correlation between aggregate type and quality parameters, the quality analysis model branches quickly predict and output accurate quality parameters for the aggregate particles to be analyzed. This provides accurate data for subsequent calculations of the comprehensive quality coefficient, grading of recycled aggregates from the same source of solid waste, and adaptation to engineering application scenarios.

[0075] In summary, compared to existing technologies, this application obtains N quality analysis model branches trained based on machine learning. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are input into these N quality analysis model branches respectively to obtain N quality parameters of the aggregate particles to be analyzed. These N quality parameters include at least the crushing value and water absorption rate. Thus, through the machine learning model, quality parameters such as crushing value and water absorption rate are predicted and output based on the point cloud cluster features and aggregate type of the aggregate particles to be analyzed. This replaces traditional time-consuming and destructive laboratory testing, achieving efficient, non-destructive, and accurate prediction of the quality parameters of the aggregate particles to be analyzed, and providing reliable data support for the automated evaluation of the quality of recycled aggregates from the same source of solid waste.

[0076] S60: Based on preset rules, calculate the initial quality coefficient of the aggregate particles to be analyzed according to the aggregate type and N quality parameters, and correct the initial quality coefficient using the cluster certainty and type matching degree corresponding to the aggregate particles to be analyzed to obtain the quality coefficient. Based on the preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are screened.

[0077] The aforementioned steps obtain the aggregate type and N quality parameters of the aggregate particles to be analyzed, as well as the clustering certainty and type matching degree, which quantify the reliability of the independent point cloud clustering process and the aggregate type identification process. Based on this, a comprehensive quality assessment of the aggregate particles to be analyzed can be carried out, thereby achieving accurate screening of aggregate particles.

[0078] To address the aforementioned issues, this application, based on preset rules, calculates the initial quality coefficient of the aggregate particles to be analyzed according to the aggregate type and N quality parameters, and then corrects the initial quality coefficient using the cluster certainty and type fit of the aggregate particles to be analyzed, thereby obtaining the quality coefficient. Based on the preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are screened.

[0079] Specifically, step S60 in the method includes: Preset crushing value threshold, water absorption threshold, quality coefficient threshold, and aggregate type range; If the aggregate type of the aggregate particles to be analyzed does not meet the aggregate type range, or if the quality parameters of the aggregate particles to be analyzed do not meet either the crushing value threshold or the water absorption threshold, it is judged as unqualified. If the aggregate particles to be analyzed meet the aggregate type range, the crushing value meets the crushing value threshold, and the water absorption rate meets the water absorption rate threshold, then N weights are assigned to the N quality parameters of the aggregate particles to be analyzed, and the initial quality coefficient of the aggregate particles to be analyzed is obtained by weighted calculation. The initial quality coefficient is corrected by using the cluster certainty and type fit of the aggregate particles to be analyzed, and the quality coefficient is obtained. If the quality coefficient of the aggregate particles to be analyzed meets the quality coefficient threshold, it is judged as qualified; otherwise, it is judged as unqualified.

[0080] In this embodiment, the following steps are first preset: a crushing value threshold, a water absorption threshold, a quality coefficient threshold, and a range of aggregate types. For example, these thresholds can be determined based on the actual application scenario of recycled aggregates from the same source solid waste (such as C30 concrete, roadbed filling, etc.) or industry standards. For instance, the preset crushing value threshold can be set to 20%, the water absorption threshold to 6%, the quality coefficient threshold to 60 points, and the aggregate type range to concrete and cement. It should be noted that these thresholds need to be dynamically set according to actual engineering needs and application scenarios to ensure that the judgment results are adapted to the application scenario.

[0081] Secondly, if the aggregate type of the aggregate to be analyzed does not meet the aggregate type range, or if the quality parameters of the aggregate to be analyzed do not meet either the crushing value threshold or the water absorption threshold, it is judged as unqualified. That is, if the aggregate to be analyzed does not meet any of the aggregate type range, crushing value threshold, or water absorption threshold, it is judged as unqualified. For example, if the aggregate type and crushing value of a certain aggregate to be analyzed meet the aggregate type range and crushing value threshold, but the water absorption does not meet the water absorption threshold, then the aggregate to be analyzed is judged as unqualified. In this way, aggregates that clearly do not meet the basic requirements of the project can be quickly filtered out.

[0082] Conversely, if the aggregate particles to be analyzed meet the aggregate type range, the crushing value meets the crushing value threshold, and the water absorption rate meets the water absorption rate threshold, then N weights are assigned to the N quality parameters of the aggregate particles to be analyzed, and the initial quality coefficient of the aggregate particles to be analyzed is obtained by weighted calculation. The sum of the N weights is 1. Each weight can be dynamically set according to the number of quality parameters and the degree of influence of each quality parameter on the overall quality of the aggregate. For example, if the quality parameters only include crushing value and water absorption rate, and the crushing value has a more significant impact on the structural strength of the aggregate (such as in high-strength concrete applications), then the crushing value and water absorption rate can be assigned weights of 0.6 and 0.4 respectively. Those skilled in the art can flexibly adjust the weight allocation according to specific engineering application scenarios and industry standard requirements to ensure that the weight settings are adapted to the actual application needs of the aggregate.

[0083] For example, if the crushing value of the aggregate particles to be analyzed is 18% and the water absorption rate is 5.2%, with weights of 0.6 and 0.4 respectively, a score of 0-100 can be assigned to the crushing value and water absorption rate of the aggregate particles to be analyzed based on preset rules (such as combining parameter thresholds with industry standards). The higher the score, the better the aggregate performance corresponding to that quality parameter. For example, based on the actual usage scenario and preset rules, a crushing value of 18% can be assigned 80 points and a water absorption rate of 5.2% can be assigned 85 points. On this basis, the initial quality coefficient of the aggregate particles to be analyzed can be obtained by weighted calculation = 0.6 × 80 + 0.4 × 85 = 82 points. The initial coefficient reflects the importance of different quality parameters through weights, effectively realizing the comprehensive quantification of multi-dimensional quality parameters.

[0084] It should be noted that the "standards for good and bad" of quality parameters (such as crushing value and water absorption rate) are not fixed and uniform, but will change with the needs of the application scenario. There is no single scoring and calculation model. For example, in high-strength concrete structures (such as building beams and columns), both crushing value and water absorption rate should be as low as possible. A low crushing value ensures that the aggregate is not easily broken under load, ensuring the strength of the concrete structure. A low water absorption rate can prevent the aggregate from competing for moisture in the mixture, preventing structural cracking due to insufficient cement hydration. However, in lightweight aggregate concrete (such as insulated walls), the crushing value still needs to be as low as possible to ensure the stability of the wall foundation structure, while the water absorption rate needs to be moderately high to allow the aggregate to slowly release water during hydration, improving the interfacial bonding performance between the aggregate and cement paste, and enhancing the synergistic effect of insulation and mechanical properties. Therefore, it is necessary to first clarify the "good and bad" direction of each quality parameter according to the specific application scenario, then design the scoring rules accordingly, and finally calculate the initial quality coefficient by combining weights to ensure that the judgment result accurately matches the actual needs of the scenario.

[0085] Furthermore, since the clustering process described above may result in blurred division of independent point cloud clusters due to point cloud noise and particle adhesion, misjudgment may occur in the type identification process. Therefore, it is necessary to correct the initial quality coefficient using the clustering certainty and type fit corresponding to the aggregate particles to be analyzed, thus obtaining the final quality coefficient. The clustering certainty reflects the reliability of the independent point cloud clustering process, while the type fit reflects the reliability of the aggregate type identification process. For example, if the clustering certainty is 0.2, the type fit is 0.92, and the initial quality coefficient is 82, then the corrected quality coefficient = 0.2 × 0.92 × 82 = 15.64. This indicates that although the initial quality coefficient is acceptable, the final quality coefficient is significantly reduced due to the extremely poor reliability of the clustering process, effectively avoiding the risk of misjudgment caused by the initial clustering error and ensuring that the quality judgment result is more consistent with the actual situation of the aggregate.

[0086] Finally, if the quality coefficient of the aggregate particles to be analyzed meets the quality coefficient threshold, it is judged as qualified; otherwise, it is judged as unqualified. For example, if the quality coefficient threshold is 60 points and the quality coefficient of the aggregate particles to be analyzed is 15.64 points, it does not meet the quality coefficient threshold and is judged as unqualified.

[0087] In summary, compared to existing technologies, this application, based on preset rules, calculates the initial quality coefficient of the aggregate particles to be analyzed according to the aggregate type and N quality parameters. The initial quality coefficient is then corrected using the cluster certainty and type fit of the aggregate particles to obtain the final quality coefficient. Based on a preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are screened. This solves the problems of the one-sidedness and poor scenario adaptability of traditional single-parameter judgment, and eliminates the interference of preliminary analysis errors, achieving accurate and objective screening of the quality of recycled aggregates from the same source of solid waste.

[0088] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application first uses a scanning device to acquire 3D point cloud data of the area where the aggregate particles to be analyzed are placed, and then removes the background planar point cloud to obtain a point cloud set of aggregate particles containing multiple aggregate particles to be analyzed. In this way, the three-dimensional morphological information of the aggregate to be analyzed is accurately preserved, providing a clean and reliable raw data foundation for subsequent steps such as point cloud clustering and feature extraction, effectively avoiding background interference, and improving the accuracy and efficiency of overall quality analysis.

[0089] Secondly, this application clusters aggregate particle point clouds based on a preset clustering distance threshold, obtaining multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents one aggregate particle. In this way, the aggregate particle point cloud is clustered and decomposed into multiple independent point cloud clusters with a single aggregate particle as the unit. Simultaneously, the clustering certainty quantifies the reliability of the clustering results, providing independent data units for subsequent single aggregate feature extraction and quality analysis, and also providing a reliable basis for subsequent quality coefficient correction, effectively ensuring the accuracy of the overall analysis.

[0090] Furthermore, this application uses a scanning device to acquire NIR spectral data for each aggregate particle to be analyzed. Based on the NIR spectral data and a pre-set spectral fingerprint database, the aggregate type and type matching degree of each aggregate particle are analyzed. This replaces traditional manual sorting and identification, achieving automated and accurate determination of the type of each aggregate particle. It also quantifies the reliability of the identification results through type matching, providing accurate type basis and reliable reference for subsequent aggregate quality coefficient calculation and overall quality analysis.

[0091] Furthermore, this application extracts the point cloud cluster features of the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume. In this way, the discrete point cloud clusters of aggregate are transformed into quantifiable geometric features, providing reliable data support for subsequent calculation of the aggregate comprehensive quality coefficient based on multi-dimensional features and for the automated and accurate determination of the quality of recycled aggregate from the same source of solid waste.

[0092] Furthermore, this application obtains N quality analysis model branches trained based on machine learning. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are input into these N quality analysis model branches respectively to obtain N quality parameters of the aggregate particles to be analyzed. These N quality parameters include at least the crushing value and water absorption rate. Thus, through the machine learning model, quality parameters such as crushing value and water absorption rate are predicted and output based on the point cloud cluster features and aggregate type of the aggregate particles to be analyzed. This replaces traditional time-consuming and destructive laboratory testing, achieving efficient, non-destructive, and accurate prediction of the quality parameters of the aggregate particles to be analyzed, providing reliable data support for the automated evaluation of the quality of recycled aggregates from the same source of solid waste.

[0093] Finally, based on preset rules, this application calculates the initial quality coefficient of the aggregate particles to be analyzed according to the aggregate type and N quality parameters. The initial quality coefficient is then corrected using the clustering certainty and type fit of the aggregate particles to obtain the final quality coefficient. Based on a preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are screened. This solves the problems of the one-sidedness and poor scenario adaptability of traditional single-parameter judgment, and eliminates the interference of errors in the initial analysis, achieving accurate and objective screening of the quality of recycled aggregates from the same source of solid waste.

[0094] Through the above technical solutions, this application constructs a fully automated technical chain for quality analysis of recycled aggregates from homogeneous solid waste through 3D point cloud clustering, NIR spectral recognition, machine learning, quality prediction, and error correction. This improves the accuracy and efficiency of quality analysis of recycled aggregates from homogeneous solid waste, and provides reliable technical support for the large-scale resource utilization and industrial application of recycled aggregates from homogeneous solid waste.

[0095] Example 2, as Figure 2 As shown, based on the same inventive concept as the image processing-based method for analyzing the quality of recycled aggregates from homogeneous solid waste provided in Embodiment 1, this embodiment of the invention also provides an image processing-based system for analyzing the quality of recycled aggregates from homogeneous solid waste, including: The point cloud acquisition module 11 is used to acquire 3D point cloud data of the aggregate particle placement area to be analyzed using a scanning device, and remove the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed. The point cloud clustering module 12 is used to cluster aggregate particle point clouds based on a preset clustering distance threshold to obtain multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, wherein each independent point cloud cluster represents an aggregate particle. The type analysis module 13 is used to acquire the NIR spectral data of each aggregate particle to be analyzed using a scanning device, and to analyze and obtain the aggregate type and type matching degree of each aggregate particle to be analyzed based on the NIR spectral data and the preset spectral fingerprint library. Feature extraction module 14 is used to extract point cloud cluster features of the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume. The quality analysis module 15 is used to obtain N quality analysis model branches based on machine learning training. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are respectively input into the N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed. The N quality parameters include at least the crushing value and water absorption rate. The output filtering module 16 is used to calculate the initial quality coefficient of the aggregate particles to be analyzed based on preset rules, according to the aggregate type and N quality parameters of the aggregate particles to be analyzed, and to correct the initial quality coefficient using the clustering certainty and type matching degree corresponding to the aggregate particles to be analyzed, thereby obtaining the quality coefficient. Based on the preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are filtered.

[0096] Specifically, the point cloud acquisition module 11 is used for: A scanning device is deployed on the aggregate screening facility, the scanning device integrating a 3D scanner and an NIR spectral probe; Trigger the 3D scanner to acquire 3D point cloud data of the entire aggregate particle placement area to be analyzed; Using the RANSAC plane fitting algorithm, planar point clouds representing the background plate are quickly identified and removed from 3D point cloud data, resulting in an aggregate particle point cloud set containing only multiple aggregate particles to be analyzed.

[0097] Specifically, the point cloud clustering module 12 is used for: Create an unprocessed point list containing all points that need to be clustered; Randomly select a point from the list of unprocessed points as a seed point. Based on the preset clustering distance threshold, search and aggregate all unprocessed neighboring points within the clustering distance threshold to obtain a new aggregated point. The new aggregation point is used as a new seed point to continue clustering until the clustering can no longer expand, thus obtaining an independent point cloud cluster. The number of clusterings when the independent point cloud cluster is formed is counted. Based on the number of clusterings when the independent point cloud cluster is formed, the clustering certainty of each independent point cloud cluster is calculated, wherein the clustering certainty is inversely proportional to the number of clusterings. Select the next seed point from the unprocessed points and perform clustering until all points are clustered, obtaining multiple independent point cloud clusters, each independent point cloud cluster representing an aggregate particle.

[0098] Specifically, the type analysis module 13 is used for: Collect aggregate particle samples of known types, scan their NIR spectra, and form a standard spectral database. The standard spectral database contains two types of data: spectral characteristics and type labels. The aggregate particles to be analyzed are scanned using a scanning device to obtain their spectral characteristics. The standard spectral feature most similar to the spectral feature to be analyzed is retrieved from the standard spectral database and used as the approximate spectral feature. The type label of the approximate spectral feature is extracted and used as the aggregate type of the aggregate particles to be analyzed. The similarity between the approximate spectral features and the spectral features to be analyzed is used as the type matching degree of the aggregate particles to be analyzed.

[0099] Specifically, the feature extraction module 14 is used for: Calculate the 3D convex hull of the point cloud cluster, and calculate the volume and surface area of ​​the 3D convex hull. Based on the volume and surface area of ​​the 3D convex hull, calculate the 3D sphericity using the sphericity formula. After triangulating the point cloud cluster to generate a mesh model, the volume of the mesh model is calculated as the actual volume of the aggregate particles corresponding to the point cloud cluster. The convexity of the point cloud cluster is calculated based on the actual volume of the point cloud cluster and the 3D convex hull volume. Find the shortest inertial axis of the point cloud cluster, divide the point cloud cluster into several slices along the direction of the shortest inertial axis, calculate the minimum cross-sectional area and the average cross-sectional area of ​​the slices, and calculate the moment imbalance rate based on the minimum cross-sectional area and the average cross-sectional area.

[0100] Specifically, the quality analysis module 15 is used for: Multiple batches of typical aggregate particle samples were selected, and the point cloud cluster features and aggregate types of multiple batches of typical aggregate particle samples were collected as typical aggregate features. Obtain N quality parameters for each typical aggregate sample, wherein the N quality parameters include at least crushing value and water absorption rate; Using typical aggregate characteristics as training data and N quality parameters as supervision data, N quality analysis model branches are trained respectively. Among them, the N quality parameters used as supervision data are all measured values. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are respectively input into N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed.

[0101] Specifically, the output filtering module 16 is used for: Preset crushing value threshold, water absorption threshold, quality coefficient threshold, and aggregate type range; If the aggregate type of the aggregate particles to be analyzed does not meet the aggregate type range, or if the quality parameters of the aggregate particles to be analyzed do not meet either the crushing value threshold or the water absorption threshold, it is judged as unqualified. If the aggregate particles to be analyzed meet the aggregate type range, the crushing value meets the crushing value threshold, and the water absorption rate meets the water absorption rate threshold, then N weights are assigned to the N quality parameters of the aggregate particles to be analyzed, and the initial quality coefficient of the aggregate particles to be analyzed is obtained by weighted calculation. The initial quality coefficient is corrected by using the cluster certainty and type fit of the aggregate particles to be analyzed, and the quality coefficient is obtained. If the quality coefficient of the aggregate particles to be analyzed meets the quality coefficient threshold, it is judged as qualified; otherwise, it is judged as unqualified.

[0102] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application first uses a point cloud acquisition module to acquire 3D point cloud data of the aggregate particle placement area using a scanning device, obtaining aggregate particle point cloud sets and accurately preserving the three-dimensional morphological information of the aggregate to be analyzed. Secondly, a point cloud clustering module clusters the aggregate particle point cloud sets, obtaining multiple independent point cloud clusters and clustering certainty, providing independent data units for subsequent single aggregate feature extraction and quality analysis. Thirdly, a type analysis module acquires the NIR spectral data of each aggregate particle to be analyzed, obtaining the aggregate type and type conformity of each particle, achieving automated and accurate determination of each aggregate type. Furthermore, a feature extraction module extracts the point cloud cluster features of multiple independent point cloud clusters, transforming the discrete point cloud clusters of the aggregate into quantifiable geometric features, providing reliable data support for the subsequent accurate determination of the quality of recycled aggregates from the same source of solid waste. Furthermore, through the quality analysis module, N quality analysis model branches are obtained, and N quality parameters of the aggregate particles to be analyzed are input, achieving efficient and non-destructive accurate prediction of the quality parameters of the aggregate particles to be analyzed, providing reliable data support for the automated evaluation of the quality of recycled aggregates from the same source of solid waste. Finally, through the output screening module, quality coefficients are calculated based on preset rules to screen aggregate particles, achieving accurate and objective screening of the quality of recycled aggregates from the same source of solid waste. In this way, the accuracy and efficiency of the quality analysis of recycled aggregates from the same source of solid waste are improved, providing reliable technical support for the large-scale resource utilization and industrial application of recycled aggregates from the same source of solid waste.

[0103] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0104] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0109] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing, characterized in that, The method includes: Use a scanning device to acquire 3D point cloud data of the area where the aggregate particles to be analyzed are placed, and remove the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed. Based on a preset clustering distance threshold, aggregate particle point clouds are clustered to obtain multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents an aggregate particle. The NIR spectral data of each aggregate particle to be analyzed is obtained using a scanning device. Based on the NIR spectral data and a preset spectral fingerprint library, the aggregate type and type matching degree of each aggregate particle to be analyzed are obtained. Extract point cloud cluster features from the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume; Obtain N quality analysis model branches trained based on machine learning, input the point cloud cluster features of the aggregate particles to be analyzed and the aggregate type into the N quality analysis model branches respectively, and obtain N quality parameters of the aggregate particles to be analyzed, wherein the N quality parameters include at least crushing value and water absorption rate. Based on preset rules, the initial quality coefficient of the aggregate particles to be analyzed is calculated according to the aggregate type and N quality parameters. The initial quality coefficient is then corrected using the cluster certainty and type fit of the aggregate particles to be analyzed to obtain the quality coefficient. Based on the preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are screened.

2. The method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing according to claim 1, characterized in that, The 3D point cloud data of the aggregate particle placement area to be analyzed is acquired using a scanning device, and the background planar point cloud is removed to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed, including: A scanning device is deployed on the aggregate screening facility, the scanning device integrating a 3D scanner and an NIR spectral probe; Trigger the 3D scanner to acquire 3D point cloud data of the entire aggregate particle placement area to be analyzed; Using the RANSAC plane fitting algorithm, planar point clouds representing the background plate are quickly identified and removed from 3D point cloud data, resulting in an aggregate particle point cloud set containing only multiple aggregate particles to be analyzed.

3. The method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing according to claim 1, characterized in that, Based on a preset clustering distance threshold, the aggregate particle point cloud is clustered to obtain multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents an aggregate particle, including: Create an unprocessed point list containing all points that need to be clustered; Randomly select a point from the list of unprocessed points as a seed point. Based on the preset clustering distance threshold, search and aggregate all unprocessed neighboring points within the clustering distance threshold to obtain a new aggregated point. The new aggregation point is used as a new seed point to continue clustering until the clustering can no longer expand, thus obtaining an independent point cloud cluster. The number of clusterings when the independent point cloud cluster is formed is counted. Based on the number of clusterings when the independent point cloud cluster is formed, the clustering certainty of each independent point cloud cluster is calculated, wherein the clustering certainty is inversely proportional to the number of clusterings. Select the next seed point from the unprocessed points and perform clustering until all points are clustered, obtaining multiple independent point cloud clusters, each independent point cloud cluster representing an aggregate particle.

4. The method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing according to claim 1, characterized in that, The NIR spectral data of each aggregate particle to be analyzed is acquired using a scanning device. Based on the NIR spectral data and a pre-set spectral fingerprint database, the aggregate type and type conformity of each aggregate particle to be analyzed are obtained, including: Collect aggregate particle samples of known types, scan their NIR spectra, and form a standard spectral database. The standard spectral database contains two types of data: spectral characteristics and type labels. The aggregate particles to be analyzed are scanned using a scanning device to obtain their spectral characteristics. The standard spectral features most similar to the spectral features to be analyzed are retrieved from the standard spectral database and used as approximate spectral features. The type labels of the approximate spectral features are extracted and used as the aggregate type of the aggregate particles to be analyzed. The similarity between the approximate spectral features and the spectral features to be analyzed is used as the type matching degree of the aggregate particles to be analyzed.

5. The method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing according to claim 1, characterized in that, Extract the point cloud cluster features of the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume, including: Calculate the 3D convex hull of the point cloud cluster, and calculate the volume and surface area of ​​the 3D convex hull. Based on the volume and surface area of ​​the 3D convex hull, calculate the 3D sphericity using the sphericity formula. After triangulating the point cloud cluster to generate a mesh model, the volume of the mesh model is calculated as the actual volume of the aggregate particles corresponding to the point cloud cluster. The convexity of the point cloud cluster is calculated based on the actual volume of the point cloud cluster and the 3D convex hull volume. Find the shortest inertial axis of the point cloud cluster, divide the point cloud cluster into several slices along the direction of the shortest inertial axis, calculate the minimum cross-sectional area and the average cross-sectional area of ​​the slices, and calculate the moment imbalance rate based on the minimum cross-sectional area and the average cross-sectional area.

6. The method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing according to claim 5, characterized in that, Obtain N quality analysis model branches trained based on machine learning. Input the point cloud cluster features of the aggregate particles to be analyzed and the aggregate type into the N quality analysis model branches respectively to obtain N quality parameters of the aggregate particles to be analyzed. The N quality parameters include at least the crushing value and water absorption rate, including: Multiple batches of typical aggregate particle samples were selected, and the point cloud cluster features and aggregate types of multiple batches of typical aggregate particle samples were collected as typical aggregate features. Obtain N quality parameters for each typical aggregate sample, wherein the N quality parameters include at least crushing value and water absorption rate; Using typical aggregate characteristics as training data and N quality parameters as supervision data, N quality analysis model branches are trained respectively. Among them, the N quality parameters used as supervision data are all measured values. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are respectively input into N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed.

7. The method for quality analysis of recycled aggregates from homogeneous solid waste based on image processing according to claim 1, characterized in that, Based on preset rules, the initial quality coefficient of the aggregate particles to be analyzed is calculated according to the aggregate type and N quality parameters. The initial quality coefficient is then corrected using the clustering certainty and type fit of the aggregate particles to be analyzed, yielding the final quality coefficient. Based on a preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are screened, including: Preset crushing value threshold, water absorption threshold, quality coefficient threshold, and aggregate type range; If the aggregate type of the aggregate particles to be analyzed does not meet the aggregate type range, or if the quality parameters of the aggregate particles to be analyzed do not meet either the crushing value threshold or the water absorption threshold, it is judged as unqualified. If the aggregate particles to be analyzed meet the aggregate type range, the crushing value meets the crushing value threshold, and the water absorption rate meets the water absorption rate threshold, then N weights are assigned to the N quality parameters of the aggregate particles to be analyzed, and the initial quality coefficient of the aggregate particles to be analyzed is obtained by weighted calculation. The initial quality coefficient is corrected by using the cluster certainty and type fit of the aggregate particles to be analyzed, and the quality coefficient is obtained. If the quality coefficient of the aggregate particles to be analyzed meets the quality coefficient threshold, it is judged as qualified; otherwise, it is judged as unqualified.

8. A quality analysis system for recycled aggregates from homogeneous solid waste based on image processing, characterized in that, For performing the method according to any one of claims 1-7, comprising: The point cloud acquisition module is used to acquire 3D point cloud data of the area where the aggregate particles to be analyzed are placed using a scanning device, and remove the background planar point cloud to obtain an aggregate particle point cloud set containing multiple aggregate particles to be analyzed. The point cloud clustering module is used to cluster aggregate particle point clouds based on a preset clustering distance threshold, to obtain multiple independent point cloud clusters and the clustering certainty of each independent point cloud cluster, where each independent point cloud cluster represents an aggregate particle. The type analysis module is used to acquire NIR spectral data of each aggregate particle to be analyzed using a scanning device. Based on the NIR spectral data and a preset spectral fingerprint library, the aggregate type and type matching degree of each aggregate particle to be analyzed are obtained. The feature extraction module is used to extract the point cloud cluster features of the multiple independent point cloud clusters, wherein the point cloud cluster features include at least 3D sphericity, convexity, moment imbalance rate, and volume. The quality analysis module is used to obtain N quality analysis model branches trained based on machine learning. The point cloud cluster features of the aggregate particles to be analyzed and the aggregate type are respectively input into the N quality analysis model branches to obtain N quality parameters of the aggregate particles to be analyzed. The N quality parameters include at least the crushing value and water absorption rate. The output filtering module is used to calculate the initial quality coefficient of the aggregate particles to be analyzed based on preset rules, according to the aggregate type and N quality parameters of the aggregate particles to be analyzed, and to correct the initial quality coefficient using the cluster certainty and type matching degree of the aggregate particles to be analyzed, thereby obtaining the quality coefficient. Based on the preset quality coefficient threshold and the quality coefficient of the aggregate particles to be analyzed, the aggregate particles are filtered.

Images