Metal scrap recovery treatment method based on intelligent control

Abstract

The invention provides a metal scrap recovery treatment method based on intelligent control, and relates to the technical field of metal scrap recovery. Comprising the following steps: S1, collecting a multi-angle metal scrap image and metal mass; s2, extracting color features and texture features of the metal scrap from the metal scrap image; s3, according to the colors and textures of the metal scraps, a gradient lifting tree model is used for conducting preliminary classification on the metal scraps; s4, calculating the metal volume by using a 3D modeling technology and a hexahedral mesh method; and S5, the metal density is calculated according to the metal volume and the metal mass, and classification verification is conducted on the metal waste again according to the metal density. According to the method, the metal density is indirectly calculated through the 3D modeling technology and the hexahedral mesh method, the metal waste is classified and verified according to the density, and efficient and accurate metal waste recycling treatment is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of metal waste recycling, and in particular to a metal waste recycling processing method based on intelligent control. Background Art

[0002] In modern machining, the manufacturing of various metal parts involves a variety of processes, including turning, milling, drilling, and grinding. These operations inevitably generate significant amounts of metal waste. For example, turning produces long, continuous chips, while milling often produces lumpy or flaky scraps, drilling debris, and fine metal powder from grinding. Furthermore, with the continuous development of the machining industry and the expansion of production scale, the amount of metal waste generated is also showing a sustained growth trend.

[0003] Taking the automotive manufacturing industry as an example, a large auto plant generates tons, even tens of tons, of metal scrap every day during the machining of components like engines and body frames. Similarly, in aerospace parts manufacturing, where precision machining is crucial, while the amount of scrap generated by machining a single component is relatively small, the high value of the components, the complex machining process, and the large number of processing batches contribute significantly to the accumulated amount of scrap. If this scrap is not properly recycled and disposed of, it not only takes up significant workshop space, impacting the cleanliness and orderliness of the production environment, but also results in a significant waste of resources.

[0004] In terms of classification accuracy, it relies heavily on manual judgment and lacks precise quantitative classification criteria, making it prone to classification errors. Manual operation is inefficient, lacks data integration and intelligent decision-making, and is difficult to handle in complex scenarios. In terms of comprehensive recycling and processing, it uses a single perspective and cannot fully analyze and process metal scrap. This can easily lead to the omission of valuable scrap or the inability to accurately determine its value and recycling methods. Intelligent control-based methods can address these shortcomings. Summary of the Invention

[0005] The present invention provides a metal waste recycling and processing method based on intelligent control, which is used to solve the defects of low manual operation efficiency and inability to comprehensively analyze and process metal waste in the prior art.

[0006] In one aspect, the present invention provides a metal waste recycling method based on intelligent control, comprising:

[0007] S1: Collect multi-angle scrap metal images and metal quality.

[0008] S2: Extract the color features and texture features of metal scrap from the metal scrap image.

[0009] S3: Use the gradient boosting tree model to perform preliminary classification of metal scrap based on its color and texture.

[0010] S4: Calculate the metal volume using 3D modeling technology and hexahedral mesh method.

[0011] S5: Calculate the metal density based on the metal volume and metal mass, and classify and verify the metal scrap again based on the metal density.

[0012] According to a metal scrap recycling method based on intelligent control provided by the present invention, in step S1, images of the metal scrap are captured from the front, side, top, bottom, and oblique angles.

[0013] According to a metal scrap recycling method based on intelligent control provided by the present invention, in step S2, the color features include: HSV color space, color channel features, color histogram, color moment, and color clustering.

[0014] Texture features include: grayscale, image filtering, gray-level co-occurrence matrix, local binary, and directional gradient histogram.

[0015] According to a metal waste recycling method based on intelligent control provided by the present invention, in the calculation of the color moment:

[0016] The calculation formula for the third-order color moment of the red channel is:

[0017]

[0018] Where, the total number of pixels in the image is N, and the pixel value in the red channel is B i , μ R For each pixel point in the red channel, the pixel value and the first-order color moment of the red channel.

[0019] According to a metal scrap recycling method based on intelligent control provided by the present invention, in step S3, the specific steps of using the gradient boosting tree model to preliminarily classify the metal scrap are as follows:

[0020] S31: Use principal component analysis to reduce the dimension of color features and texture features to obtain a feature subset that contributes most to classification.

[0021] S32: Divide the feature subset into training set, validation set and test set, and label the metal samples according to the metal type.

[0022] S33: Initialize the GBDT model and set the hyperparameters of the GBDT model.

[0023] S34: Use the training set to iteratively train the GBDT model to obtain a trained GBDT model.

[0024] S35: Analyze the importance scores of color features and texture features based on the trained GBDT model, verify the contribution ratio of color and texture features to classification, and determine whether the importance of texture features is lower than the threshold. If so, adjust the feature extraction strategy.

[0025] S35: Evaluate the performance of the GBDT model using the test set.

[0026] According to a metal scrap recycling method based on intelligent control provided by the present invention, in step S4, the specific steps of calculating the metal volume using 3D modeling technology are as follows:

[0027] S41: Obtaining discrete point cloud data from the spatial coordinate information of each point in the multi-angle metal scrap image.

[0028] S42: Importing point cloud data sets into 3D modeling.

[0029] S43: Use surface reconstruction in 3D modeling to convert discrete point cloud data into a continuous metal 3D model with topological structure.

[0030] S44: Calculate the volume V of the metal object using the hexahedral mesh method based on the metal three-dimensional model.

[0031] According to a metal scrap recycling method based on intelligent control provided by the present invention, in step S43, the specific steps of converting discrete point cloud data into a continuous metal three-dimensional model with a topological structure are:

[0032] S431: 3D modeling reconstructs surfaces based on the distribution patterns and spatial relationships of discrete point cloud data.

[0033] S432: Divide the surface into mesh units according to the mesh generation function to obtain a continuous metal three-dimensional model with a topological structure.

[0034] According to a metal scrap recycling method based on intelligent control provided by the present invention, in step S44, the specific steps of calculating the volume V of the metal object using the hexahedral grid method are as follows:

[0035] S441: Perform hexahedral meshing on the metal three-dimensional model according to the meshing module to obtain N hexahedral units.

[0036] S442: Obtain the vertex coordinates of N hexahedral units.

[0037] S443: Calculate the volumes of N hexahedral units using a hexahedral volume calculation formula according to the vertex coordinates.

[0038] S444: Sum the volumes of the N hexahedral units to obtain the metal volume.

[0039] According to a metal waste recycling method based on intelligent control provided by the present invention, in step S443, the volume V of the hexahedral unit i The calculation formula is:

[0040]

[0041] In the formula, (x i1 ,y i1 , z i1 ), (x i2 ,y i2 , z i2 ), (x i3 ,y i3 , z i3 ), (x i4 ,y i4 , z i4 ) are vertex coordinates.

[0042] According to a metal waste recycling method based on intelligent control provided by the present invention, in step S5, the density value of the metal waste is calculated using a density calculation formula, and the average value is taken by multiple measurements. The metal type is obtained by comparing the measured metal density with the standard metal density.

[0043] The present invention provides a metal scrap recycling method based on intelligent control. It uses 3D modeling technology and a hexahedral mesh method to indirectly calculate metal density and classify and verify metal scrap based on density. The beneficial effects achieved are:

[0044] A gradient boosting tree model is used for preliminary classification. This model, through iterative learning, continuously optimizes the relationship between features and classification results. It effectively handles complex nonlinear relationships and uncovers underlying patterns between color and texture features and different metal types. Furthermore, during model training, principal component analysis (PCA) dimensionality reduction, rational data set partitioning, meticulous parameter tuning, and timely strategy adjustments based on feature importance analysis enable the model to efficiently complete classification tasks even with large amounts of complex feature data, while also improving classification accuracy and reducing the complexity and errors associated with manual classification.

[0045] Utilizing 3D modeling technology combined with a hexahedral mesh method to calculate metal volume, this method first converts multi-angle images into point cloud data, then constructs a continuous three-dimensional model. Finally, the volume is precisely calculated using the hexahedral mesh method. This method can accurately obtain the volume of irregularly shaped scrap metal. Compared to traditional rough estimates or calculations based solely on regular shape assumptions, this method produces more realistic volume results, providing reliable data support for subsequent accurate density calculations based on volume and mass, and ultimately, more precise classification and verification.

[0046] A preliminary classification is performed based on color and texture features, and then the density is calculated based on the calculated metal volume and mass. The metal scrap is then classified and verified again based on the comparison of the density with the standard metal density, forming a multi-verification mechanism. Even if there are certain deviations in the initial classification, the classification results can be further corrected through the verification of the physical property of density, improving the accuracy of the overall classification, ensuring the quality of metal scrap classification during the recycling process, and facilitating subsequent targeted recycling operations according to accurate categories. Compared with traditional recycling methods that rely on manual observation, manual measurement, and empirical judgment, this greatly reduces manpower input, increases processing speed, and can more efficiently handle the classification and recycling of large amounts of metal scrap, thereby improving the operational efficiency and benefits of the entire metal scrap recycling industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 This is a flow chart of a metal waste recycling method based on intelligent control provided by an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of a process for calculating metal volume using 3D modeling technology provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0051] The following combination Figure 1-Figure 2 The present invention describes a metal waste recycling method based on intelligent control.

[0052] like Figure 1-Figure 2 As shown, an embodiment of the present invention provides a metal waste recycling method based on intelligent control, comprising:

[0053] S1: Collect multi-angle metal scrap image data and quality data.

[0054] Use high-definition digital cameras, industrial cameras, and other cameras to capture clear, high-quality images. These cameras have sufficient resolution and can flexibly adjust parameters like shooting angle and focal length. These cameras can be used to capture images under various lighting conditions, such as natural light and even artificial light. This is because varying lighting conditions can affect the color and texture of the scrap metal's surface, allowing for a comprehensive display of its appearance. Photograph the scrap metal from multiple angles, such as the front, side, top, bottom, and even at oblique angles. For irregularly shaped scrap metal or scrap metal with distinctive textures and patterns, further angles should be used to ensure every detail is captured. The same scrap metal can also be photographed at different distances to capture both the overall appearance and the details of the individual pieces. Organize and save the captured images in a unified format, annotating them with relevant information such as the corresponding shooting angle and lighting conditions for easy subsequent analysis.

[0055] To collect scrap metal quality data, you need to prepare appropriate measuring tools, such as high-precision electronic scales for accurate weighing of scrap metal. For larger scrap, you may also need to use equipment such as a floor scale to ensure accurate weighing. Use measuring tools such as calipers and micrometers to measure the dimensions of the scrap metal, including basic geometric dimensions such as length, width, and thickness. For complex shapes, you also need to measure the local dimensions of different parts to determine its volume or shape characteristics.

[0056] S2: Extract the color features and texture features of metal scrap from the metal scrap image.

[0057] S21: HSV color space, color channel features, color histogram, color moment, color clustering.

[0058] S211: RGB color space is the most basic representation method, but HSV is more in line with human intuitive perception of color and is often more convenient when performing color analysis, so images are usually converted from RGB color space to HSV color space first.

[0059] S212: After converting to the target color space, the color space is split into different channels. For example, in the HSV color space, the hue (H) channel, saturation (S) channel, and value (V) channel can be extracted separately, so that subsequent operations such as statistics and analysis can be performed on each channel separately.

[0060] S213: Color Histogram Calculation: A color histogram is a statistical representation of the color distribution in an image, showing the percentage of pixels within each color interval. By plotting histograms for different channels (such as the H, S, and V channels in HSV), we can intuitively understand the color distribution characteristics of the scrap metal image.

[0061] S214: Color Moment Calculation: Color moments are low-order moment statistics based on the image color distribution. They primarily include mean, variance, and skewness, and can concisely describe the overall characteristics of color. First-order color moments reflect the average intensity of color, second-order color moments reflect the dispersion of the color distribution, and third-order color moments describe the asymmetry of the color distribution.

[0062] First-order color moments, for a color image, are usually calculated in the RGB color space. The pixel value in the G (green) channel is Gi.

[0063] The calculation formula for the first-order color moment (mean) of the red channel is:

[0064]

[0065] Where, the total number of pixels in the image is N, and the pixel value of the i-th pixel in the R (red) channel is R i .

[0066] This formula means that the pixel values ​​of all pixels in the red channel are added and divided by the total number of pixels to obtain the mean color of the red channel, which reflects the average intensity of the red hue of the entire image.

[0067] The calculation formula for the first-order color moment (mean) of the green channel is:

[0068]

[0069] Where, the total number of pixels in the image is N, and the pixel value in the G (blue) channel is G i .

[0070] The average intensity of the green channel color is calculated through the same summing and averaging method, reflecting the average green tone of the image.

[0071] The calculation formula for the first-order color moment (mean) of the blue channel is:

[0072]

[0073] Where, the total number of pixels in the image is N, and the pixel value in the B (blue) channel is B i The average intensity of the blue channel color represents the average level of the overall blue tone of the image.

[0074] The calculation formula for the second-order color moment (variance) of the red channel is:

[0075]

[0076] Where μ RThe pixel value of each pixel in the red channel and the first-order color moment of the red channel. The variance of the red channel color reflects the degree of dispersion of the color distribution of the red tones in the image. The larger the variance, the more dispersed the color distribution.

[0077] The calculation formula for the second-order color moment (variance) of the green channel is:

[0078]

[0079] Where μ G For each pixel point in the green channel, the pixel value and the first-order color moment of the green channel are G(blue) and G(green) respectively. i .

[0080] The calculation formula for the second-order color moment (variance) of the blue channel is:

[0081]

[0082] Where μ B The pixel value of each pixel in the blue channel and the first-order color moment of the blue channel. The variance of the blue channel color represents the degree of dispersion of the blue tone color distribution of the image.

[0083] Third-order color moment (skewness)

[0084] The calculation formula for the third-order color moment (skewness) of the red channel is:

[0085]

[0086] Where, the total number of pixels in the image is N, and the pixel value in the red channel is B i , μ R For each pixel point in the red channel, the pixel value and the first-order color moment of the red channel.

[0087] It describes the characteristics of the asymmetry of the red color distribution of the image. Different skewness values ​​reflect different morphological characteristics of the color distribution.

[0088] The calculation formula for the third-order color moment (skewness) of the green channel is:

[0089]

[0090] It is used to measure the asymmetry of the green channel color distribution. The skewness is calculated based on the cube of the difference between the green channel pixel value and the mean, following the same calculation logic as the green channel.

[0091] The calculation formula for the third-order color moment (skewness) of the blue channel is:

[0092]

[0093] Where, the total number of pixels in the image is N, and the pixel value in the B (blue) channel is B i , μ R For each pixel point in the blue channel, the pixel value and the first-order color moment of the blue channel.

[0094] The skewness of the blue channel color is calculated by this formula, which reflects the asymmetry of the blue color distribution of the image.

[0095] S215: Color clustering: When you want to classify the colors in the image into several main categories, a clustering algorithm is used to divide the pixels in the image into a specified number of clusters based on color similarity. Each cluster represents a main color category, which facilitates further analysis and differentiation of metal scrap in different color areas.

[0096] S22: Texture features include: grayscale, image filtering, gray-level co-occurrence matrix, local binary, and directional gradient histogram.

[0097] S221: Grayscale: Texture feature extraction is often more convenient and effective when performed on grayscale images. This is because color information may not be critical for texture analysis, and grayscale simplifies subsequent calculations. Grayscale conversion typically involves converting the RGB channels of a color image into a single grayscale channel through a weighted average.

[0098] S222: Image filtering: Filtering is performed to reduce the interference of image noise on texture feature extraction. Common filtering methods include mean filtering, Gaussian filtering, and median filtering. Mean filtering replaces the center pixel value with the average value of the pixels in the neighborhood, Gaussian filtering performs a weighted average of the neighborhood pixels according to a Gaussian distribution, and median filtering takes the median of the neighborhood pixel values ​​as the center pixel value. Each filtering method has its own characteristics, and the appropriate filtering method should be selected based on the noise level of the image.

[0099] S223: Gray Level Co-occurrence Matrix: The gray level co-occurrence matrix describes the probability of gray levels occurring in an image at specific directions and distances. It is used to calculate multiple statistics that reflect texture characteristics. For example, it can reflect texture characteristics such as coarseness, directionality, and periodicity. Different textures will have significantly different gray level co-occurrence matrices and related statistics.

[0100] Determine the parameters for calculating the grayscale co-occurrence matrix, including direction and distance. The choice of these parameters will affect the final extracted texture features. Generally, experiments and adjustments are required based on the characteristics of the metal scrap texture and the needs of subsequent analysis. Use image library functions to calculate the grayscale co-occurrence matrix. For example, in Python, use the greycomatrix() function in the skimage.feature library to pass in parameters such as grayscale image data, distance, and direction to obtain the grayscale co-occurrence matrix at the specified direction and distance. Calculate texture feature statistics based on the grayscale co-occurrence matrix. Commonly used ones include contrast, correlation, energy, and entropy. For example, contrast can measure the clarity of the texture by calculating the sum of the squares of the differences between the matrix elements and the center element. Correlation reflects the linear relationship between row or column elements in the grayscale co-occurrence matrix, reflecting the directionality of the texture.

[0101] S224: Local Binary Pattern: The LBP operator compares each pixel in an image with its neighboring pixels, converts the grayscale values ​​of the neighboring pixels to the grayscale value of the central pixel into binary codes, and then calculates the distribution of these codes to describe texture features. It is robust to changes in illumination and can effectively extract local texture information from images.

[0102] S225: Histogram of Directed Gradients: HOG features are mainly used to describe the distribution of gradient directions in local areas of an image. By dividing the image into multiple small areas, calculating the gradient direction histogram of the pixels in each cell unit, and then combining these histograms, it can well capture the edges, contours and other texture features of objects in the image. It is especially suitable for objects with obvious shapes and texture structures, and can also better extract features for metal scraps with different shapes and textures.

[0103] To calculate the gradient of each pixel in the image, the gradient magnitude and gradient direction can be obtained by calculating the first-order partial derivatives in the horizontal and vertical directions. The image is divided into multiple small cell units. For each cell unit, the gradient direction histogram of the pixels in it is statistically calculated. The gradient direction is usually divided into a certain number of intervals. The amplitude of the pixel gradient direction is accumulated into the corresponding interval according to its gradient direction to form the gradient direction histogram of the cell unit. To enhance the robustness of the feature, several adjacent cell units are combined into a larger block, and the histograms of all cell units in the block are normalized. Finally, the normalized histograms of all blocks are combined to obtain the directional gradient histogram feature vector of the entire image, which is used to describe the texture characteristics of the image.

[0104] S3: Preliminary classification of metal scraps based on their color and texture by using a gradient boosting tree model.

[0105] The Gradient Boosted Tree (GBDT) model is a powerful machine learning classification model that relies on iterative learning through decision tree optimization. Specifically, it first constructs an initial decision tree, using the color and texture features of the metal scrap as input (for example, color is quantized into specific numerical indicators such as RGB values, and texture features such as roughness and directionality are extracted using texture analysis algorithms) to make a preliminary classification decision. However, this decision tree's judgment is often inaccurate and can result in a certain degree of error, meaning there is a residual difference between the predicted result and the true category.

[0106] S31: Use principal component analysis to reduce the dimensionality of color features and texture features, and retain the feature subset that contributes most to classification.

[0107] S32: Divide the preprocessed dataset into training, validation, and test sets. Manually label samples according to metal type to ensure label accuracy. If the sample size is insufficient, expand the training set through data augmentation techniques such as rotation, flipping, and brightness adjustment.

[0108] S33: GBDT Model Training and Parameter Tuning: Initialize the GBDT model and set hyperparameters such as the number of decision trees, depth, and learning rate. Evaluate model performance on a validation set using cross-validation, and tune hyperparameters using grid search or Bayesian optimization to maximize classification accuracy or F1 score.

[0109] S34: Iteratively train the GBDT model using the training set, optimizing the loss function through gradient descent during each training round. After training, analyze the feature importance scores to verify the contribution of color and texture features to classification. If texture features are less important, adjust the feature extraction strategy.

[0110] S35: Evaluate model performance on the test set, calculating metrics such as classification accuracy, confusion matrix, and macro-average / micro-average F1 scores. If the model is overfitting, optimize it by adding regularization terms, reducing tree depth, or increasing the sample size. If the model is underfitting, increase the number of trees or reduce the learning rate.

[0111] S35: Deploy the trained GBDT model to an industrial edge device or cloud server to perform real-time feature extraction and classification prediction on newly collected scrap metal images. Output the classification results and record the classification log for subsequent model iterations.

[0112] S4: Calculate the metal volume using 3D modeling technology and hexahedral mesh method.

[0113] S41: The spatial coordinate information of each point in the multi-angle scrap metal image is aggregated. This spatial coordinate information covers the specific location of the scrap metal in three-dimensional space. This coordinate information is aggregated to form a point cloud data set. This set contains a total of n points. For ease of description and subsequent calculations, the coordinates of each point are represented as (xi, yi, zi), where i ranges from 1 to n.

[0114] S42: Import the point cloud data set into the 3D modeling process. During the import process, the software will read the point cloud data and initially present the approximate shape corresponding to the discrete data in the software's working interface. At this time, what you see is just a pile of unprocessed, scattered point clouds.

[0115] S43: Use its built-in surface reconstruction, mesh generation and other functions to process the imported point cloud data.

[0116] S431: Based on the distribution patterns of these discrete point cloud data and the spatial relationships between them, the software will first attempt to construct continuous surfaces through the surface reconstruction function. During this process, the neighborhood relationships of each point and the overall spatial trend will be analyzed, allowing the originally isolated points to gradually connect to form surfaces.

[0117] S432: Using the mesh generation function, these surfaces are further divided into small grid cells, transforming all the discrete point cloud data into a continuous, topologically structured 3D metal model. In this model, the various parts have clear geometric and spatial connections. Instead of the previous scattered point cloud state, a complete 3D metal object can be intuitively seen, including its shape and surface features, laying the foundation for subsequent volume calculations.

[0118] S44: Calculate the volume V of the metal object using the hexahedral mesh method based on the metal three-dimensional model.

[0119] S441: The metal 3D model is meshed into a hexahedral grid using the meshing module to obtain N hexahedral units. Meshing involves determining the density of the grid, which is controlled by setting the global grid size or the size parameters of the local refinement area, and setting the average side length. A smaller side length means a finer grid, which increases the computational effort and memory usage. For the surface areas of the model that come into contact with the outside world, boundary constraints are set according to the actual physical conditions to ensure that the meshing conforms to the geometric and physical properties of the object. After completing the settings, the meshing function is activated, and the metal object model is automatically divided into numerous small hexahedral units according to the set parameters.

[0120] S442: Obtain the vertex coordinates of N hexahedral units. Obtain the 8 vertex coordinates of each unit. For the i-th hexahedral unit (i ranges from 1 to N), its 8 vertex coordinates are recorded as (x i1 ,y i1 , z i1 )、(x i2 ,y i2 , z i2 )、(x i3 ,y i3 , z i3 )、(x i4 ,y i4 , z i4 )、(x i5 ,y i5 , z i5 )、(x i6 ,y i6 , z i6 )、(x i7 ,y i7 , z i7 )、(x i8 ,y i8 , z i8 ). When the number of cells is small, use the visual query function to view and record the coordinates of each cell one by one.

[0121] S443: Calculate the volume of N hexahedral units using the hexahedral volume calculation formula based on the vertex coordinates. Taking the special and commonly used hexahedron as an example (complex hexahedrons can be converted into parallelepipeds by appropriate coordinate transformation and other methods to simplify the relevant calculations), first determine the three adjacent edge vectors of the hexahedral unit, such as vector a from the vertex (x i1 ,y i1 , z i1 ) points to (x i2 ,y i2 , z i2 ), its coordinates are expressed as a=(x i2 -x i1 ,y i2 -y i1 , z i2 -z i2 ). The vector b is given by the vertex (x i2 ,y i2 , z i2 ) points to (x i3 ,y i3 , z i3 ), the coordinate is expressed as b=(x i3 -x i2 ,y i3 -y i2 , z i3 -zi2 ). The vector c is given by the vertex (x i3 ,y i3 , z i3 ) points to (x i4 ,y i4 , z i4 ), the coordinate is expressed as c=(x i4 -x i3 ,y i4 -y i3 , z i4 -z i3 ).

[0122] According to the mixed product formula, the volume V of the hexahedral unit is i The calculation formula is:

[0123]

[0124] Expanded into determinant:

[0125]

[0126] The volume V of the i-th hexahedral unit is obtained by determinant operation i .

[0127] S444: Sum the volumes of the N hexahedral units to obtain the metal volume.

[0128] The calculation formula for the summation is:

[0129]

[0130] Where V i is the volume of the hexahedral unit.

[0131] S5: Calculate the metal density based on the color characteristics and texture characteristics, and classify and verify the metal scrap again based on the metal density.

[0132] According to the density calculation formula "density = mass ÷ volume", the mass and volume values ​​of each metal scrap sample measured are substituted into the formula to calculate the corresponding density value. The unit is usually grams / cubic centimeter, kilograms / cubic meter, etc. During the calculation process, the accuracy of the data must be ensured, and the calculation must be performed in the correct order of operations. If necessary, multiple measurements should be taken to obtain the average value to reduce the error.

[0133] The calculated density values ​​of each scrap metal sample are compared with known standard density values ​​for different metals. Common metals, such as iron, have a density of approximately 7.87 g / cm³, copper, 8.96 g / cm³, and aluminum, 2.7 g / cm³. A certain margin of error must be considered when making these comparisons, as errors are inevitable in actual measurements, such as the precision limitations of measuring tools and slight deviations during operation. Generally speaking, if the calculated density value is within a reasonable error range of a known standard density value (for example, a set error range of ±5%, which can be determined based on actual measurement accuracy requirements), the scrap metal sample is likely to belong to the corresponding metal category, thereby verifying the accuracy of the initial classification.

[0134] Analyze and compile the validation classification results for all scrap metal samples, examining the proportion of initially correctly classified samples to the total number of samples—this is known as the classification accuracy rate. A high accuracy rate indicates that the initial classification method was reliable. A low accuracy rate indicates that the initial classification process and basis for classification need to be re-examined to identify any issues and refine and optimize the classification method to further improve classification accuracy.

[0135] In summary, this embodiment provides a metal scrap recycling method based on intelligent control. It uses 3D modeling technology and a hexahedral mesh method to indirectly calculate metal density and classify and verify metal scrap based on density. The beneficial effects achieved are:

[0136] A gradient boosting tree model is used for preliminary classification. This model, through iterative learning, continuously optimizes the relationship between features and classification results. It effectively handles complex nonlinear relationships and uncovers underlying patterns between color and texture features and different metal types. Furthermore, during model training, principal component analysis (PCA) dimensionality reduction, rational data set partitioning, meticulous parameter tuning, and timely strategy adjustments based on feature importance analysis enable the model to efficiently complete classification tasks even with large amounts of complex feature data, while also improving classification accuracy and reducing the complexity and errors associated with manual classification.

[0137] Utilizing 3D modeling technology combined with a hexahedral mesh method to calculate metal volume, this method first converts multi-angle images into point cloud data, then constructs a continuous three-dimensional model. Finally, the volume is precisely calculated using the hexahedral mesh method. This method can accurately obtain the volume of irregularly shaped scrap metal. Compared to traditional rough estimates or calculations based solely on regular shape assumptions, this method produces more realistic volume results, providing reliable data support for subsequent accurate density calculations based on volume and mass, and ultimately, more precise classification and verification.

[0138] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the above technical solution is essentially embodied in the form of a software product, or the portion that contributes to the prior art. This software product is stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device to execute the methods of each embodiment or certain portions of the embodiments.

[0139] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will appreciate that modifications may be made to the technical solutions described in the aforementioned embodiments, or that some of the technical features therein may be replaced with equivalents. However, such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A metal waste recycling method based on intelligent control, characterized in that: include: S1: Collect multi-angle metal scrap images and metal quality; S2: extracting color features and texture features of the metal scrap from the metal scrap image; S3: Preliminarily classifying the metal scraps using a gradient boosting tree model according to the color and texture of the metal scraps; S4: Calculate metal volume using 3D modeling technology and hexahedral mesh method; S5: Calculate the metal density according to the metal volume and the metal mass, and classify and verify the metal waste again according to the metal density.

2. The metal waste recycling method based on intelligent control according to claim 1 is characterized in that: In step S1 , images of the metal scrap are captured from the front, side, top, bottom, and oblique angles.

3. The metal waste recycling method based on intelligent control according to claim 1 is characterized in that: In step S2, Color features include: HSV color space, color channel features, color histogram, color moment, and color clustering; Texture features include: grayscale, image filtering, gray-level co-occurrence matrix, local binary, and directional gradient histogram.

4. The metal waste recycling method based on intelligent control according to claim 3 is characterized in that: In the calculation of color moments: The calculation formula for the third-order color moment of the red channel is: Where, the total number of pixels in the image is N, and the pixel value in the red channel is B i , μ R For each pixel point in the red channel, the pixel value and the first-order color moment of the red channel.

5. The metal waste recycling method based on intelligent control according to claim 1 is characterized in that: In step S3, the specific steps of using the gradient boosting tree model to preliminarily classify metal scrap are as follows: S31: Use principal component analysis to reduce the dimension of color features and texture features to obtain a feature subset that contributes most to classification; S32: Dividing the feature subset into a training set, a validation set, and a test set, and labeling the metal samples according to the metal type; S33: Initialize the GBDT model and set the hyperparameters of the GBDT model; S34: Iteratively train the GBDT model using the training set to obtain a trained GBDT model; S35: Analyze the importance scores of the color features and texture features according to the trained GBDT model, verify the contribution ratio of the color and texture features to the classification, and determine whether the importance of the texture feature is lower than a threshold. If so, adjust the feature extraction strategy; S35: Evaluate the performance of the GBDT model using the test set.

6. The metal waste recycling method based on intelligent control according to claim 1 is characterized in that: In step S4, the specific steps of calculating the metal volume using 3D modeling technology are as follows: S41: Obtaining discrete point cloud data from the spatial coordinate information of each point in the multi-angle metal scrap image; S42: Importing the point cloud data set into 3D modeling; S43: using surface reconstruction in 3D modeling to convert the discrete point cloud data into a continuous metal three-dimensional model with a topological structure; S44: Calculate the volume V of the metal object using a hexahedral mesh method based on the metal three-dimensional model.

7. The metal waste recycling method based on intelligent control according to claim 6 is characterized in that: In step S43, the specific steps of converting the discrete point cloud data into a continuous metal three-dimensional model with a topological structure are: S431: 3D modeling to reconstruct a surface based on the distribution pattern and spatial relationship of the discrete point cloud data; S432: Divide the surface into mesh units according to the mesh generation function to obtain a continuous metal three-dimensional model with a topological structure.

8. The metal waste recycling method based on intelligent control according to claim 6 is characterized in that: In step S44, the specific steps of calculating the volume V of the metal object using the hexahedral mesh method are as follows: S441: performing hexahedral mesh division on the metal three-dimensional model according to a mesh division module to obtain N hexahedral units; S442: Obtain vertex coordinates of N hexahedral units; S443: Calculating the volumes of the N hexahedral units using a hexahedral volume calculation formula according to the vertex coordinates; S444: Sum the volumes of the N hexahedral units to obtain the metal volume.

9. The metal waste recycling method based on intelligent control according to claim 8 is characterized in that: In step S443, the volume V of the hexahedral unit i The calculation formula is: In the formula, (x i1 ,y i1 , z i1 ), (x i2 ,y i2 , z i2 ), (x i3 ,y i3 , z i3 ), (x i4 ,y i4 , z i4 ) are vertex coordinates.

10. The metal waste recycling method based on intelligent control according to claim 1, characterized in that: In step S5, the density of the metal scrap is calculated using a density calculation formula, and the average value is obtained by multiple measurements. The metal type is obtained by comparing the measured metal density with the standard metal density.

Images