Ore granularity detection method

Abstract

The invention relates to the technical field of ore processing detection, and discloses an ore particle size detection method which comprises the following steps: step 1, data acquisition: acquiring a visible light image Iv of ore particles through an industrial camera, acquiring an infrared image Iir through an infrared camera, and acquiring a depth image Id through a laser structured light module to obtain multi-modal data; 2, performing multi-modal fusion, namely fusing the multi-modal data acquired in the step 1 to generate an enhanced image Ifuse; and step 3, image preprocessing: Gaussian filtering is carried out on the enhanced image Ifuse generated in the step 2 to remove noise. Through the multi-mode sensing technology, the significance and boundary definition of ore particles are remarkably improved, especially in severe environments such as common dust interference, low illumination intensity and complex background in mine production, high-quality image data can be collected, a reliable basis is provided for subsequent segmentation and particle size calculation, and the method is suitable for popularization and application. The problem that single-mode imaging is limited is solved.

Description

Technical Field

[0001] The invention relates to the technical field of ore processing detection, in particular to an ore particle size detection method. Background Art

[0002] The automation and intelligentization process of the mining industry is accelerating, and the accurate detection of ore particle size has become an indispensable part of the ore processing process. The particle size detection results directly affect the efficiency and effect of ore crushing, screening and other processing technologies, and are also an important basis for the optimization of subsequent mineral processing processes. Accurate analysis of ore particle size distribution can significantly improve ore processing efficiency, reduce energy consumption, and provide basic data for parameter optimization of production processes. However, existing technologies still have technical bottlenecks in ore particle size detection.

[0003] The traditional screening method is a classic particle size detection method. It physically separates ore particles through sieves of different mesh sizes and counts the proportion of particles in each particle size range. However, this method has the disadvantages of long detection cycle, low efficiency, and susceptibility to ore moisture and shape. It is not suitable for real-time detection of dynamic production lines. In addition, the screening process will cause particle breakage, affecting the true accuracy of the detection.

[0004] Laser particle size analyzer is based on the principle of light scattering. It calculates the particle size range by analyzing the light scattering distribution of particle groups. This method has high accuracy and is particularly suitable for particle size detection of fine particles. However, its equipment is expensive and has strict requirements on environmental conditions such as dust and vibration in industrial sites. It is not suitable for complex working conditions in mining production. In addition, laser particle size analyzer has errors in detecting irregularly shaped particles and cannot fully reflect the true characteristics of ore particles.

[0005] The particle size detection method based on two-dimensional images relies on high-resolution cameras and image processing technology to calculate the particle size distribution by extracting the boundaries and sizes of particles. This method has the advantages of non-contact and high degree of automation, and is suitable for online detection. However, in environments such as particle stacking, irregular shapes, insufficient lighting or complex backgrounds, the accuracy of two-dimensional image processing technology will be significantly reduced. In addition, traditional image segmentation methods often rely on fixed rules and have limited ability to segment particles with adhesion or complex shapes, affecting the accuracy of detection.

[0006] The mining production environment is complex, accompanied by dust interference, changes in light intensity, and the stacking and adhesion of particles. This places higher demands on existing particle size detection technologies. In particular, traditional methods have poor adaptability in harsh environments and cannot guarantee the stability and reliability of the detection results. In addition, most methods can only provide two-dimensional particle size information, but cannot accurately reflect the three-dimensional characteristics of non-spherical particles.

[0007] Therefore, those skilled in the art provide a method for detecting ore particle size to solve the above-mentioned problems. Summary of the invention

[0008] In view of the deficiencies of the prior art, the present invention provides a method for detecting ore particle size to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for detecting ore particle size, comprising:

[0010] Step 1: Data collection: collect visible light images of ore particles using industrial cameras. v , collect infrared images through infrared cameras I ir , collect depth map I through laser structured light module d , multimodal data;

[0011] Step 2: Multimodal fusion: fuse the multimodal data collected in step 1 to generate an enhanced image I fused ;

[0012] Step 3: Image preprocessing: the enhanced image I generated in step 2 fused Gaussian filtering is used to remove noise. The filtering formula is:

[0013]

[0014] Where σ is the filter standard deviation, I smooth (x, y) represents the smoothing result of the fused image. represents the Gaussian kernel function, and * represents the convolution operation;

[0015] Step 4: Particle segmentation: Use the deep learning segmentation model to perform particle segmentation on the image preprocessed in step 3, extract the particle region mask M(x, y), and combine morphological post-processing to optimize the particle segmentation result. The morphological post-processing formula is:

[0016]

[0017] in, represents the corrosion operation, represents the dilation operation, and B is the morphological structure element;

[0018] Step 5: Particle size calculation: for the particle area segmented and optimized in step 4, calculate the particle equivalent diameter D according to the particle area F. eq :

[0019] Among them, D eq represents the equivalent diameter, F represents the area of ​​the target area, and π represents pi;

[0020] And the particle volume V is calculated by the laser depth map collected in step 1:

[0021]

[0022] Among them, A i is the area of ​​the triangular mesh unit, h i is the corresponding depth value, N is the number of layers;

[0023] Step 6: Transmit the particle size distribution parameters D10, D50, and D90 calculated in step 5 to the control system for dynamic adjustment of the crusher, and upload to the cloud for storage.

[0024] Preferably, the calculation formulas of the modal weights α1, α2, and α3 are:

[0025]

[0026] Among them, σ i represents the noise standard deviation of the i-th modality image, σ j represents the noise standard deviation of the j-th modality image.

[0027] Preferably, the particle segmentation adopts the Mask R-CNN deep learning model, and the loss function after the segmentation result is optimized is:

[0028] G=G cls +G box +G mask ,

[0029] Among them, G cls is the classification loss, G box is the bounding box loss, G mask is the mask loss.

[0030] Preferably, the particle size distribution parameter is calculated by the cumulative distribution function F(D), and the formula is:

[0031]

[0032] Where D is the particle size, N i is the number of particles in the i-th particle size range, M is the total particle size range, F(D is the cumulative distribution function, D i is a threshold or a specific value.

[0033] Preferably, the particle size distribution parameter is calculated by the cumulative distribution function F(D), and the formula is:

[0034]

[0035] Where D is the particle size, N iis the number of particles in the i-th particle size range, M is the total particle size range, F(D) is the cumulative distribution function, and D i is a threshold or a specific value.

[0036] Preferably, the real-time feedback process adjusts the rotation speed ω of the crusher according to the particle size distribution result, and the adjustment formula is:

[0037] ω′=ω+k·(D target -D measured ),

[0038] Among them, k is the adjustment coefficient, D target is the target particle size, D measured is the current detected particle size.

[0039] Preferably, the standard deviation σ of the Gaussian filter is dynamically adjusted according to the global noise level σ of the image. noise Calculate and satisfy the following relationship:

[0040] σ=c·σ noise ,

[0041] Where c is the filter adjustment coefficient.

[0042] Preferably, the particle volume V calculated from the depth map uses a point-by-point integration method based on a triangular mesh, and the area A of the triangular mesh unit in the depth map is i Calculated by the following formula:

[0043]

[0044] Among them, (x1, y1), (x2, y2), and (x3, y3) are the vertex coordinates of the triangle mesh.

[0045] Preferably, the multimodal fusion generates an enhanced image I by fusing the multimodal data collected in step 1 according to the following formula: fused

[0046] I fused (x, y) = α1I v (x,y)+α2I ir (x,y)+α3I d (x, y),

[0047] Among them, α1, α2, α3 are modal weights, I fused (x, y) represents the intensity value of the fused image, I fused (x, y) represents the intensity value of the visible light image, I ir (x, y) represents the intensity value of the infrared light image, I d (x, y) represents other types of intensity values.

[0048] Preferably, the enhanced image I fused The calculation of (x, y) takes into account the grayscale contrast of the infrared image and compensates it through the following formula:

[0049]

[0050] Among them, G ir and G v is the grayscale average of the infrared image and the visible light image,

[0051] α1, α2, α3 are modal weights,

[0052] I fused (x, y) represents the intensity value of the fused image,

[0053] I fused (x, y) represents the intensity value of the visible light image,

[0054] I ir (x, y) represents the intensity value of the infrared light image,

[0055] I d (x, y) represents other types of intensity values.

[0056] The present invention provides a method for detecting ore particle size, which has the following beneficial effects:

[0057] 1. The present invention significantly improves the prominence and boundary clarity of ore particles through multimodal sensing technology, especially in harsh environments such as dust interference, low light intensity, and complex background commonly seen in mining production. It can collect high-quality image data, provide a reliable basis for subsequent segmentation and granularity calculation, and solve the problem of limited single-modality imaging.

[0058] 2. The present invention adopts a deep learning segmentation model combined with morphological post-processing technology to accurately separate ore particles, especially when the particles are adhered or irregularly shaped, the segmentation accuracy is significantly improved. In addition, the deep learning model has good generalization ability through a large number of sample training, and is suitable for the segmentation needs of different ore types and shapes. The intelligent segmentation method of image data overcomes the limitations of traditional image processing technology that relies on rules and is easily affected by shape.

[0059] 3. The present invention calculates the particle size distribution parameters D10, D50, and D90, and feeds back to the ore processing equipment in real time, dynamically adjusts the equipment operating parameters, and realizes intelligent control of particle size, effectively solving the drawbacks of delayed parameter adjustment and reliance on experience in traditional ore processing, thereby improving processing efficiency and reducing resource waste.

[0060] 4. The present invention constructs a three-dimensional point cloud of particles through a laser depth map, calculates the volume of particles, and realizes high-precision measurement of non-spherical particles. Compared with the traditional two-dimensional method, the three-dimensional particle size calculation can fully reflect the true characteristics of the particles and is particularly suitable for the particle size distribution statistics of ore particles with complex shapes. The application of three-dimensional measurement technology is a key innovation point in the development of ore particle size detection from two dimensions to multiple dimensions. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0062] In order to make the technical personnel in the technical field understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in combination with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a partial embodiment of the present invention, not a complete embodiment. Based on the embodiment of the present invention, other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0063] The present invention is described in detail below in conjunction with the accompanying drawings:

[0064] Example:

[0065] Please see attached Figure 1 The embodiment of the present invention provides a method for detecting ore particle size, comprising:

[0066] Step 1: Data collection: collect visible light images of ore particles using industrial cameras. v , collect infrared images through infrared cameras I ir , collect depth map I through laser structured light module d , multimodal data;

[0067] Step 2: Multimodal fusion: fuse the multimodal data collected in step 1 to generate an enhanced image I fused ;

[0068] Step 3: Image preprocessing: the enhanced image I generated in step 2 fused Gaussian filtering is used to remove noise. The filtering formula is:

[0069]

[0070] Where σ is the filter standard deviation, I smooth (x, y) represents the smoothing result of the fused image. represents the Gaussian kernel function, and * represents the convolution operation;

[0071] Step 4: Particle segmentation: Use the deep learning segmentation model to perform particle segmentation on the image preprocessed in step 3, extract the particle region mask M(x, y), and combine morphological post-processing to optimize the particle segmentation result. The morphological post-processing formula is:

[0072]

[0073] in, represents the corrosion operation, represents the dilation operation, and B is the morphological structure element;

[0074] Step 5: Particle size calculation: for the particle area segmented and optimized in step 4, calculate the particle equivalent diameter D according to the particle area F. eq :

[0075] Among them, D eq represents the equivalent diameter, F represents the area of ​​the target area, and π represents pi;

[0076] And the particle volume V is calculated by the laser depth map collected in step 1:

[0077]

[0078] Among them, A i is the area of ​​the triangular mesh unit, h i is the corresponding depth value, N is the number of layers;

[0079] Step 6: Transmit the particle size distribution parameters D10, D50, and D90 calculated in step 5 to the control system for dynamic adjustment of the crusher, and upload to the cloud for storage.

[0080] Step 1: Collect multimodal data of ore particles through industrial cameras, infrared cameras and laser structured light modules to overcome the limitations of single-modal imaging in dust interference, low light intensity or complex background environments. The collaborative collection of multimodal data improves image quality and the integrity of particle information, provides high-quality basic data for subsequent processing, and adapts to the needs of complex mining environments;

[0081] Step 2: Fuse visible light, infrared and laser depth data to generate enhanced images, combining the advantages of each modality to significantly improve the saliency and boundary clarity of the particle area. Multimodal fusion technology solves the problem of blurred boundaries or insufficient depth information in single-modality data, providing clearer and more reliable image input for subsequent image preprocessing and segmentation;

[0082] Step 3: Smooth the enhanced image through Gaussian filtering to effectively remove noise and texture interference in the image while maintaining the edge details of the particles. The preprocessed image has better smoothness and contrast, laying the foundation for the accuracy of particle segmentation. Especially in an environment with severe dust interference, preprocessing significantly improves the image quality;

[0083] Step 4: Use the deep learning segmentation model to segment the preprocessed image into particles, and optimize the segmentation results by combining morphological post-processing technology, which can effectively separate the adhering particles and accurately extract the particle area. This step solves the problem that the traditional segmentation method has a strong dependence on regularity and insufficient segmentation ability, especially in the case of irregular shapes and stacked particles.

[0084] Step 5: Calculate the equivalent diameter by the area of ​​the particle region, and calculate the three-dimensional volume of the particle in combination with the laser depth map to achieve high-precision measurement of the two-dimensional particle size, which can accurately reflect the true three-dimensional characteristics of the particle. Especially in the particle size measurement of non-spherical particles, this step can comprehensively count the particle distribution characteristics and provide reliable support for scientific decision-making in ore processing;

[0085] Step 6: The calculated particle size distribution parameters are transmitted to the control system in real time to dynamically adjust the operating parameters of the crusher, realize intelligent control of particle size, improve ore processing efficiency and reduce resource waste. In addition, by uploading data to cloud storage, long-term trend analysis and process optimization are supported, providing important data support for intelligent mine management;

[0086] The benefits of the Gaussian filter formula are that the enhanced image is Gaussian smoothed through the convolution operation, effectively removing random noise interference while maintaining the significant features of the particle edge. The standard deviation of the Gaussian kernel function is dynamically adjusted according to the global noise level, enhancing the filter's adaptability to complex mining environments, ensuring the quality of the preprocessed image, and providing clear input for particle segmentation;

[0087] Benefits of the morphological post-processing formula The noise and edge defects in the particle segmentation results are repaired through morphological processing of first corrosion and then expansion. The corrosion operation removes the interference points on the segmentation edge, and the expansion operation restores the actual boundary of the particle, making the segmentation result smoother and more accurate. Combined with the preliminary segmentation results of the deep learning model, this formula greatly improves the quality and stability of particle area extraction;

[0088] Benefits of the equivalent diameter calculation formula: By calculating the equivalent diameter based on the particle area, the particle size information of irregular particles is standardized to the same size parameter as that of round particles, simplifying the particle size statistical analysis of non-spherical particles. This formula provides an accurate and intuitive indicator for the calculation of two-dimensional particle size distribution characteristics, which is suitable for rapid statistics of large-scale ore particles;

[0089] The benefits of the particle volume calculation formula are that the three-dimensional volume of the particle is calculated using the laser depth map through point-by-point integration, achieving high-precision measurement of the true volume of the particle. The combination of the triangular grid unit area and the depth value can fully reflect the three-dimensional geometric characteristics of the particle surface, solving the limitation of the traditional method that can only analyze the two-dimensional particle size, and is particularly suitable for the true volume statistics of irregular particles;

[0090] The benefits of the particle size distribution parameter transmission formula are that it can quickly adapt to changes in particle size distribution by calculating the deviation between the target particle size and the current detected particle size in real time, dynamically adjusting the crusher's rotation speed, and adjusting the coefficient to control the response speed, thereby ensuring the operating efficiency and stability of the processing equipment, avoiding waste of particle size resources, and realizing intelligent closed-loop control.

[0091] The formula plays a key role in different links, from data preprocessing, segmentation optimization, to particle size calculation and dynamic control, to build a complete ore particle size detection technology chain. The accuracy and flexibility of the formula significantly improve the environmental adaptability, real-time and measurement accuracy of the system, providing reliable support for intelligent mine production.

[0092] The calculation formula of modal weights α1, α2, and α3 is:

[0093]

[0094] Among them, σ i represents the noise standard deviation of the i-th modality image, σ j represents the noise standard deviation of the j-th modality image.

[0095] Through the calculation formula of modal weight, the noise level of multimodal data is included in the basis of weight distribution, and the contribution ratio of each modality in the fusion process is dynamically adjusted, so that the modality with lower noise occupies a greater weight. This method effectively reduces the impact of high-noise modalities on the quality of fused images, enhances the clarity and stability of enhanced images, and improves the extraction effect of significant areas of ore particles in complex environments, providing better data input for subsequent processing steps;

[0096] The weight distribution makes the low-noise mode contribute more and maximizes the signal-to-noise ratio of the fused image, especially in complex environments such as dust and low light, showing a superior clarity improvement effect. The infrared mode accurately extracts contours in complex backgrounds, and the laser depth mode compensates for boundary details, further enhancing the overall quality of the image. In the multimodal fusion process, the proportion of high-noise mode data is significantly reduced, reducing the impact of single mode failure on the final fused image, providing high-quality input for subsequent processing steps, and ensuring the robustness and reliability of the system. The dynamically adjusted modal weights enable the fusion method to adapt to a variety of complex environments, including strong dust, high humidity, low light and other harsh conditions in mines. In scenes with large environmental changes, such as ore yards and underground mining environments, real-time adjustment of modal weights ensures the continuity and accuracy of data acquisition and processing.

[0097] The Mask R-CNN deep learning model is used for particle segmentation, and the loss function after segmentation optimization is:

[0098] G=G cls +G box +G mask ,

[0099] Among them, G cls is the classification loss, G box is the bounding box loss, G mask is the mask loss.

[0100] By adopting the Mask R-CNN deep learning model and optimizing the loss function of the segmentation results, the classification loss, bounding box loss and mask loss are organically combined to effectively improve the accuracy and robustness of ore particle segmentation. The classification loss ensures that the model can accurately identify particles, the bounding box loss optimizes the positioning accuracy of the particle area, and the mask loss further refines the shape contour of the particle. The overall optimized loss function enables the model to maintain efficient segmentation performance in environments with particle adhesion, irregular shapes or complex environments, providing high-quality particle area information for particle size calculation.

[0101] The particle size distribution parameters are calculated by the cumulative distribution function F(D), the formula is:

[0102]

[0103] Where D is the particle size, N i is the number of particles in the i-th particle size range, M is the total particle size range, F(D) is the cumulative distribution function, and D i is a threshold or a specific value;

[0104] The key characteristic values ​​D10, D50, and D90 of the particle size distribution represent the percentage points corresponding to the particle size in the cumulative distribution, and are the particle size values ​​when the cumulative distribution of particle size reaches 10%, 50%, and 90%, respectively.

[0105] The particle size distribution parameters are calculated through the cumulative distribution function to fully reflect the distribution of ore particles in different particle size ranges, ensuring the scientificity and accuracy of the statistical results. The cumulative distribution characteristic value clearly calibrates the key percentage points in the particle distribution, provides an intuitive basis for evaluating the overall trend of particle size, can meet the dynamic adjustment needs of ore processing equipment, and provide important support for the scientific analysis and long-term trend prediction of particle size distribution, making process optimization more accurate and efficient.

[0106] The real-time feedback process adjusts the crusher speed ω according to the particle size distribution results. The adjustment formula is:

[0107] ω′=ω+k·(D target -D measured ),

[0108] Among them, k is the adjustment coefficient, D target is the target particle size, D measured is the current detected particle size.

[0109] Through the real-time feedback process, the adjustment formula is used to dynamically adjust the crusher speed according to the particle size distribution results, so that the actual particle size gradually approaches the target particle size, ensuring the precise control of the particle size during ore processing. The adjustment coefficient k is flexibly adapted to different working conditions, can quickly respond to particle size deviations, and avoid excessive or insufficient adjustments. This method significantly improves processing efficiency, reduces resource waste, and reduces the operating load of equipment and unnecessary energy consumption, providing a strong technical guarantee for the realization of intelligent and stable ore processing.

[0110] The standard deviation σ of the Gaussian filter is adjusted dynamically according to the global noise level σ of the image. noise Calculate and satisfy the following relationship:

[0111] σ=c·σ noise ,

[0112] Where c is the filter adjustment coefficient.

[0113] By dynamically adjusting the standard deviation of the Gaussian filter according to the global noise level of the image, the smoothing effect of the filter can adapt to the noise characteristics of the image. The introduction of the adjustment coefficient c ensures that the filtering ability is enhanced in high noise conditions, while the edge and detail features of the image are maintained in low noise conditions. This method improves the flexibility and adaptability of the filter, effectively reduces the random noise interference in the ore particle image, and retains the particle boundary information, providing higher quality image data for subsequent segmentation and particle size calculation.

[0114] The particle volume V calculated from the depth map uses a point-by-point integration method based on a triangular mesh. The area of ​​the triangular mesh unit A in the depth map i Calculated by the following formula:

[0115]

[0116] Among them, (x1, y1), (x2, y2), and (x3, y3) are the vertex coordinates of the triangle mesh.

[0117] The volume of particles in the depth map is calculated by a point-by-point integration method based on a triangular mesh, and the local surface characteristics of the particles are accurately described using the formula for the area of ​​the triangular mesh unit, achieving high-precision measurement of the volume of irregular particles. This method fully considers the three-dimensional geometric characteristics of the particle surface, avoids the error of the traditional volume calculation method in approximating the particle shape, and is particularly suitable for the measurement of ore particles with complex shapes. This calculation method significantly improves the comprehensiveness and accuracy of particle size measurement, and provides more real and reliable data support for particle size distribution statistics.

[0118] Multimodal fusion: The multimodal data collected in step 1 are fused according to the following formula to generate an enhanced image I fused

[0119] I fused (x, y) = α1I v (x,y)+α2I ir (x,y)+α3I d (x, y),

[0120] Among them, α1, α2, α3 are modal weights, I fused (x, y) represents the intensity value of the fused image, I fused (x, y) represents the intensity value of the visible light image, I ir (x, y) represents the intensity value of the infrared light image, I d (x, y) represents other types of intensity values;

[0121] Enhanced Image I fused The calculation of (x, y) takes into account the grayscale contrast of the infrared image and compensates it through the following formula:

[0122]

[0123] Among them, G ir and G v is the grayscale average of the infrared image and the visible light image,

[0124] α1, α2, α3 are modal weights,

[0125] I fused(x, y) represents the intensity value of the fused image,

[0126] I fused (x, y) represents the intensity value of the visible light image,

[0127] I ir (x, y) represents the intensity value of the infrared light image,

[0128] I d (x, y) represents other types of intensity values.

[0129] By generating enhanced images through multimodal fusion, combined with dynamic adjustment of modal weights and compensation calculation of infrared image grayscale contrast, the clarity and boundary saliency of ore particle images are significantly improved. This method makes full use of the detail information of visible light images, the contour capture ability of infrared images in low-light environments, and the spatial characteristics of other modalities (such as laser depth). Grayscale contrast compensation further balances the information intensity differences between the modalities, optimizes the quality and stability of the fused image, and provides high-quality input images in complex environments such as dust and insufficient light, laying a reliable foundation for subsequent particle segmentation and particle size calculation.

[0130] Gaussian filter formula: By dynamically adjusting the standard deviation, it achieves a balance between denoising and retaining edge details, adapting to complex mining environments;

[0131] Morphological post-processing formula: Combine corrosion and expansion operations to optimize particle segmentation results and improve segmentation quality and stability;

[0132] Equivalent diameter formula: standardizes particle size for quick statistical analysis;

[0133] Volume calculation formula: Based on the point-by-point integration method of the depth map, the three-dimensional volume of particles is accurately measured;

[0134] Modal weight formula: dynamically adjust the multi-modal contribution ratio to improve the clarity of the enhanced image;

[0135] Distributed parameters and feedback formula: used for real-time dynamic control to ensure the intelligent and efficient operation of the crusher;

[0136] Strong environmental adaptability: Multimodal acquisition and fusion technology significantly enhances image quality and adapts to complex environments such as dust and weak light;

[0137] High segmentation accuracy: Combining deep learning models with morphological optimization, it can effectively handle the segmentation of irregularly shaped and adherent particles;

[0138] Comprehensive particle size measurement: The combination of two-dimensional particle size and three-dimensional volume calculation can reflect the true characteristics of particles;

[0139] Intelligent control: Real-time feedback mechanism dynamically regulates the operation of processing equipment to improve ore processing efficiency;

[0140] Cloud storage and analysis: supports long-term trend analysis and process optimization, laying a data foundation for intelligent mine management;

[0141] Summary: This ore particle size detection method constructs a complete technical chain from data collection to dynamic control, which significantly improves the detection accuracy, real-time and environmental adaptability. It has broad application potential in the field of mining processing, especially in complex environments and irregular particle characteristics. This method provides advanced technical support for intelligent mining production and promotes the development of ore processing towards efficiency, precision and intelligence.

[0142] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting ore particle size, characterized in that: include: Step 1: Data collection: collect visible light images of ore particles using industrial cameras. v , collect infrared images through infrared cameras I ir , collect depth map I through laser structured light module d , multimodal data; Step 2: Multimodal fusion: fuse the multimodal data collected in step 1 to generate an enhanced image I fused ; Step 3: Image preprocessing: the enhanced image I generated in step 2 fused Gaussian filtering is used to remove noise. The filtering formula is: Where σ is the filter standard deviation, I smooth (x, y) represents the smoothing result of the fused image. represents the Gaussian kernel function, and * represents the convolution operation; Step 4: Particle segmentation: Use the deep learning segmentation model to perform particle segmentation on the image preprocessed in step 3, extract the particle region mask M(x, y), and combine morphological post-processing to optimize the particle segmentation result. The morphological post-processing formula is: in, represents the corrosion operation, represents the dilation operation, and B is the morphological structure element; Step 5: Particle size calculation: for the particle area segmented and optimized in step 4, calculate the particle equivalent diameter D according to the particle area F. eq : Among them, D eq represents the equivalent diameter, F represents the area of ​​the target area, and π represents pi; And the particle volume V is calculated by the laser depth map collected in step 1: Among them, A i is the area of ​​the triangular mesh unit, h i is the corresponding depth value, N is the number of layers; Step 6: Transmit the particle size distribution parameters D10, D50, and D90 calculated in step 5 to the control system for dynamic adjustment of the crusher, and upload to the cloud for storage.

2. A method for detecting ore particle size according to claim 1, characterized in that: The calculation formulas of the modal weights α1, α2, and α3 are: Among them, σ i represents the noise standard deviation of the i-th modality image, σ j represents the noise standard deviation of the j-th modality image.

3. The method for detecting ore particle size according to claim 1, characterized in that: The particle segmentation adopts the MaskR-CNN deep learning model, and the loss function after segmentation result optimization is: G=G cls +G box +G mask , Among them, G cls is the classification loss, G box is the bounding box loss, G mask is the mask loss.

4. The method for detecting ore particle size according to claim 1, characterized in that: The particle size distribution parameter is calculated by the cumulative distribution function F(D), and the formula is: Where D is the particle size, N i is the number of particles in the i-th particle size range, M is the total particle size range, F(D) is the cumulative distribution function, and D i is a threshold or a specific value.

5. A method for detecting ore particle size according to claim 4, characterized in that: The key characteristic values ​​D10, D50 and D90 of the particle size distribution represent the percentage points corresponding to the particle size in the cumulative distribution, which are the particle size values ​​when the cumulative distribution of particle size reaches 10%, 50% and 90% respectively.

6. A method for detecting ore particle size according to claim 1, characterized in that: The real-time feedback process adjusts the crusher's rotation speed ω according to the particle size distribution result, and the adjustment formula is: ω′=ω+k·(D target -D measured ), Among them, k is the adjustment coefficient, D target is the target particle size, D measured is the current detected particle size.

7. A method for detecting ore particle size according to claim 1, characterized in that: The standard deviation σ of the Gaussian filter is dynamically adjusted according to the global noise level σ of the image. noise Calculate and satisfy the following relationship: σ=c·σ noise , Where c is the filter adjustment coefficient.

8. The method for detecting ore particle size according to claim 1, characterized in that: The particle volume V calculated from the depth map uses a point-by-point integration method based on a triangular mesh. The area of ​​the triangular mesh unit A in the depth map is i Calculated by the following formula: Among them, (x1, y1), (x2, y2), and (x3, y3) are the vertex coordinates of the triangle mesh.

9. The method for detecting ore particle size according to claim 1, characterized in that: The multimodal fusion generates an enhanced image I by fusing the multimodal data collected in step 1 according to the following formula: fused I fused (x,y)=α1I v (x,y)+α2I ir ((x,y)+α3I d (x,y), Among them, α1, α2, α3 are modal weights, I fused (x, y) represents the intensity value of the fused image, I fused (x, y) represents the intensity value of the visible light image, I ir (x, y) represents the intensity value of the infrared light image, I d (x, y) represents other types of intensity values.

10. A method for detecting ore particle size according to claim 9, characterized in that: The enhanced image I fused The calculation of (x, y) takes into account the grayscale contrast of the infrared image and compensates it through the following formula: Among them, G ir and G v is the grayscale average of the infrared image and the visible light image, α1, α2, α3 are modal weights, I fused (x, y) represents the intensity value of the fused image, I fused (x, y) represents the intensity value of the visible light image, I ir (x, y) represents the intensity value of the infrared light image, I d (x, y) represents other types of intensity values.

Images