Lead zinc ore separation method and electronic equipment

Abstract

According to the lead-zinc ore sorting method and the electronic equipment, in the preprocessing stage, an obtained raw ore high-energy diagram is registered with a low-energy diagram as the benchmark, physical response difference measurement is adopted, the low-energy diagram and the registered high-energy diagram are fused to obtain a fusion diagram, the fusion diagram is colored according to a color transition rule, a pseudo-color diagram is obtained, and a lead-zinc ore sorting result is obtained. And the pseudo-color image is used as input, and a target detection model based on deep learning is used for processing, so that accurate and efficient recognition of the lead-zinc ore in the raw ore is realized.

Description

Technical Field

[0001] This application relates to the field of mineral sorting technology, and in particular to a method and electronic equipment for sorting lead-zinc ore. Background Technology

[0002] Lead-zinc ore, as one of my country's important polymetallic mineral resources, is widely used in key fields such as batteries, cables, alloy smelting, anti-corrosion coatings, and photovoltaic materials, occupying an irreplaceable position in the national defense industry and high-tech industries. my country has abundant lead-zinc resources, diverse mineralization types, and widespread deposit distribution, but it also faces many challenges such as declining ore grades, complex associated and symbiotic deposits, and increasing beneficiation difficulties.

[0003] In the pre-selection and waste disposal process of lead-zinc mines, galena, with its high density, is relatively easy to identify. However, sphalerite and gangue minerals such as pyrite and dolomite have similar particle size distributions and overlapping densities. Traditional manual sorting and conventional image processing methods are difficult to achieve accurate and rapid identification of ore and waste rock, resulting in waste of mineral resources and a decline in corporate economic benefits. Summary of the Invention

[0004] This application proposes a lead-zinc ore sorting method and electronic equipment, which can solve one of the problems existing in the background art.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] Firstly, a method for separating lead-zinc ore is provided, including:

[0007] In the preprocessing stage, the original high-energy image and the original low-energy image of the raw ore are obtained, which are obtained by dual-energy X-ray projection imaging of the raw ore; using an enhanced correlation coefficient registration algorithm, the original high-energy image is registered with the original low-energy image as a reference to obtain a secondary high-energy image; the original low-energy image and the secondary high-energy image are fused using a physical response difference metric to obtain a fused image; and the fused image is colored according to color transition rules to obtain a pseudo-color image; and...

[0008] In the intelligent recognition stage, the pseudo-color image is used as the input to a deep learning-based target detection model. Through the processing of the target detection model, the sorting results of lead-zinc ore in the raw ore are obtained.

[0009] In one possible design approach of the first aspect, the physical response difference measure employs a normalized logarithmic difference. :

[0010] in, This is a secondary high-energy diagram. This is the original low-energy map. These are the maximum gray values ​​in the secondary high-energy image and the original low-energy image, respectively. It is a very small constant set to prevent the denominator from being zero.

[0011] In one possible design of the first aspect, the target detection model adopts the YOLO model, which includes a feature extraction network, a feature fusion network, and a detection head.

[0012] In one possible design approach of the first aspect, the YOLO model employs a weighted intersection-over-union (IoU) loss function L. WIoU :

[0013] in, Indicates the center of the prediction box Center of the real frame The square of the Euclidean distance This represents the diagonal length of the smallest bounding rectangle that can enclose both the predicted and ground truth bounding boxes. This is a dynamic weighting factor used to adjust the gradient contribution of different samples.

[0014] In one possible design of the first aspect, the feature fusion network includes: a C2f layer and a Concat layer, wherein an efficient multi-scale attention layer is set between the C2f layer and the Concat layer, the efficient multi-scale attention layer is used to perform grouped convolution and multi-scale modeling on the input features, extract channel attention and spatial attention information respectively, and fuse them through normalization and weight allocation mechanisms.

[0015] In one possible design of the first aspect, the detection head includes a lightweight adaptive spatial feature fusion module for adaptively assigning fusion weights to feature maps of different scales according to the importance of each location.

[0016] In one possible design of the first aspect, the feature extraction network employs ghost convolutional layers.

[0017] In one possible design approach of the first aspect, the YOLO model employs the Optuna hyperparameter search framework.

[0018] In one possible design of the first aspect, the preprocessing stage further includes: performing gamma correction on the secondary high-energy image and the original low-energy image. In the second aspect, a core quality index calculation device is provided, comprising:

[0019] In a second aspect, an electronic device is provided, comprising: a processor, and a memory coupled to the processor, the memory for storing a computer program; the processor for executing the computer program stored in the memory to cause the electronic device to perform the lead-zinc ore sorting method as described in any possible implementation of the first aspect.

[0020] Beneficial effects:

[0021] Based on the above technical solution, in the preprocessing stage, the obtained high-energy image of the raw ore is registered with the low-energy image as a reference. The physical response difference metric is used to fuse the low-energy image and the registered high-energy image to obtain a fused image. The fused image is then colored according to the color transition rules to obtain a pseudo-color image. The pseudo-color image is used as input and processed using a deep learning-based target detection model to achieve accurate and efficient identification of lead-zinc ore in the raw ore. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is the XRD quantitative analysis spectrum of lead-zinc ore provided in the embodiments of this application;

[0024] Figure 2 This is a dual-energy XRT grayscale image of lead-zinc ore provided in an embodiment of this application, wherein... Figure 2 (a) is the XRT grayscale image of high-energy lead-zinc ore. Figure 2 (b) is the XRT grayscale image of low-energy lead-zinc ore;

[0025] Figure 3 This is a two-dimensional joint histogram of grayscale in lead-zinc ore provided in the embodiments of this application;

[0026] Figure 4 This is the XRT image registration result of high-energy lead-zinc ore provided in the embodiments of this application;

[0027] Figure 5 This is the Gamma correction result of the high-energy image of lead-zinc ore provided in the embodiments of this application;

[0028] Figure 6 This is the pseudo-color mapping result of lead-zinc ore provided in the embodiments of this application;

[0029] Figure 7This is the result of denoising the pseudo-color image of lead-zinc ore provided in the embodiments of this application;

[0030] Figure 8 The L provided in the embodiments of this application WIoU Loss function schematic diagram;

[0031] Figure 9 This is a schematic diagram of the EMA module provided in an embodiment of this application;

[0032] Figure 10 This is a schematic diagram of the lightweight adaptive spatial feature fusion module provided in the embodiments of this application;

[0033] Figure 11 This is a schematic diagram of GhostConv provided in an embodiment of this application;

[0034] Figure 12 This is a diagram of the AI ​​recognition network structure for lead-zinc ore provided in an embodiment of this application;

[0035] Figure 13 This is a flowchart of the lead-zinc ore identification algorithm implementation scheme provided in this application embodiment;

[0036] Figure 14 This is the lead-zinc ore identification result provided in the embodiments of this application. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0038] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0040] Before introducing the embodiments of this application, a brief description of the current stage of technical research of this application will be given:

[0041] Process mineralogical studies were conducted on the lead-zinc ore samples collected in this embodiment. The results of multi-element chemical analysis of the raw ore are shown in Table 1, the chemical phase analysis results of lead, zinc, and silver are shown in Tables 2-4, and the X-ray diffraction (XRD) analysis results are shown in Table 1. Figure 1 .

[0042] Table 1 Multi-element chemical analysis of raw ore

[0043] Table 2 Chemical analysis of lead phases in the raw ore

[0044] Table 3. Chemical analysis of zinc phase in raw ore

[0045] Table 4. Chemical analysis of silver phases in the raw ore

[0046] Mineralogical studies show that the raw ore contains 215.88 g / t of valuable silver, 2.50% lead, and 3.73% zinc. Of this, 19.65% silver exists as elemental silver, 63.94% as silver sulfide, and 13.49% in sulfide ores; lead is mainly present as lead sulfide (85.25%), with cerussite accounting for 5.88%; zinc is mainly present as zinc sulfide (86.86%), with zinc oxide accounting for 12.65%. Figure 2 It can be seen that the main metallic minerals in the raw ore are sphalerite, galena, pyrite, and siderite; the main non-metallic minerals are quartz, chlorite, feldspar, and mica.

[0047] Based on process mineralogical analysis, X-ray imaging of lead-zinc ore exhibits the following characteristics: Pb, a high atomic number element, has an atomic number of 82. Galena, with a density of approximately 7.5–7.6 g / cm³, has a strong absorption capacity for X-rays, resulting in lower grayscale values ​​and a darker image in XRT images. This leads to a relatively strong contrast in XRT grayscale images at different energy levels. However, zinc (Zn) has a relatively low atomic number of only 30. Sphalerite, with a density of approximately 3.9–4.2 g / cm³, while lower than galena (PbS), still slightly higher than major non-metallic gangues such as quartz and chlorite. Its grayscale value may be close to that of interfering components such as trace amounts of pyrite in waste rock. Dual-energy XRT grayscale images of lead-zinc ore are shown below. Figure 2 As shown.

[0048] Depend on Figure 2It can be seen that in high-energy images, high-density minerals have higher imaging clarity and exhibit more obvious attenuation characteristics, showing darker imaging features in the image. In low-energy images, the weaker penetrating power of X-rays results in a more uniform overall image grayscale, but the grayscale of ores and waste rocks is similar, leading to significant identification interference.

[0049] The image preprocessing scheme for lead-zinc ore based on dual-energy XRT grayscale images constructed in this embodiment integrates physical modeling and visual mapping, greatly enhancing the visualization features of lead-zinc ore XRT images. This provides a high-quality dataset input foundation for the accurate identification of lead-zinc ore by deep learning AI algorithms, demonstrating significant engineering practical value and innovation. Figure 2 The gray-level two-dimensional joint histogram of dual-energy XRT images of lead-zinc ore is as follows: Figure 3 As shown.

[0050] As can be seen from Figure 4:

[0051] (1) The horizontal axis represents the gray value of the low-energy X-ray image, and the vertical axis represents the gray value of the high-energy X-ray image. The colors in the figure represent the pixel density of different gray-scale combinations. The brighter the color, the higher the frequency of the gray-scale combination. This figure can intuitively reflect the transmission response relationship and material absorption differences of the raw ore under different energy rays by statistically analyzing the one-to-one correspondence of dual-energy gray-scale pixels.

[0052] (2) From Figure 4 As can be seen, the overall pixel points are distributed in a banded pattern along the diagonal direction, indicating a strong positive correlation between high-energy and low-energy grayscale values, meaning that they maintain structural consistency. This shows that the same ore region exhibits similar transmission trends under different energy X-rays, verifying the imaging consistency of the dual-energy XRT image acquisition system. The densely distributed area of ​​the point cloud represents the main part of the ore, and its grayscale value changes synchronously in the high and low energy channels; while some discrete pixels exist below the banded area, indicating that some high-density minerals absorb more strongly and have lower transmittance under low-energy X-rays, resulting in a significant reduction in grayscale and exhibiting obvious energy spectrum differences.

[0053] (3) Figure 4 The high-density cluster in the upper right corner represents the background area. Its grayscale values ​​are close to saturation in both high- and low-energy images, indicating that this area absorbs virtually no X-rays. This invention, through analysis of this joint histogram, enables energy spectrum separation of the grayscale responses of different minerals, providing a basis for subsequent pseudo-color image mapping and algorithm optimization. This distribution pattern demonstrates that the dual-energy grayscale joint feature can effectively distinguish between high-density metallic minerals and low-density gangue, providing a high-quality input dataset for AI recognition algorithms.

[0054] In dual-energy X-ray transmission imaging of lead-zinc ore, slight spectral shifts or mechanical disturbances often occur between high-energy and low-energy images during the imaging process, resulting in sub-pixel-level spatial displacement errors between the images. To achieve accurate fusion and pseudo-color mapping, precise alignment of the image pairs is necessary.

[0055] The Enhanced Correlation Coefficient (ECC) image registration algorithm is a registration strategy based on maximizing the enhanced correlation coefficient in the image's grayscale space, particularly suitable for complex backgrounds and blurred boundaries in lead-zinc mine images. This algorithm effectively overcomes micro-displacement errors caused by material vibration and conveyor belt movement, thereby improving the accuracy and stability of subsequent image fusion and pseudo-color image generation. Given a reference image... and the image to be registered The main goal of the ECC algorithm is to achieve affine transformation. Will Transform into This maximizes the enhanced correlation coefficient between the two graphs. The ECC optimization objective function is as follows:

[0056] In the above formula, Represents image coordinates; Represents the pixel value of the reference image; Representative through affine parameters The transformed image to be registered; and These represent the means of the two graphs, respectively. This represents the enhanced correlation coefficient. ECC registration uses a low-energy image as a baseline and registers it with a high-energy image. It iteratively optimizes the affine transformation parameters to make the two images most similar in terms of grayscale statistics. This step is a crucial foundation for subsequent pseudo-color image generation and mineral identification. The high-energy image obtained after ECC registration is shown below. Figure 4 As shown.

[0057] Analysis of the above image shows that the ECC-based image registration algorithm achieves pixel-level precise alignment of dual-energy XRT images. Compared to traditional template matching and feature point methods, ECC achieves sub-pixel-level affine registration by maximizing the overall image grayscale covariance, effectively overcoming common material movement and image jitter problems in belt conveyors. The registered image possesses characteristics such as clear edges, consistent background, and no ghosting distortion, providing a high-quality image dataset for subsequent processes.

[0058] After the lead-zinc ore conveyed by the intelligent mineral processing equipment passes through the XRT detector and undergoes DA conversion, the grayscale values ​​ultimately appearing in the XRT image do not strictly reflect material differences, especially in the intermediate grayscale range where insufficient separability often occurs. Gamma correction, through a manually designed nonlinear mapping, stretches and compresses the mid-to-high grayscale range, such as where zinc minerals are located, to enhance the contrast with the background and / or Pb-rich areas. This can improve the performance of subsequent differential enhancement, pseudo-color mapping, and detection models. The mathematical formula for gamma correction is shown below.

[0059] In the above formula Indicates the original image The pixel value of the location, This represents the Gamma parameter, which typically ranges from [0.8, 2.0]. This represents the pixel value after Gamma correction. This represents the maximum grayscale value in the image. The high-energy image obtained with a value of 1.2 is corrected as follows: Figure 5 As shown.

[0060] After Gamma correction, Figure 5 The results show that the XRT-registered images of lead-zinc ore achieve nonlinear redistribution of brightness while maintaining structural continuity. This step significantly improves the robustness of subsequent differential enhancement maps, pseudo-color mapping, and AI recognition algorithms for identifying weak responses and small targets.

[0061] Physical response difference measurement is one of the core steps in pixel-level fusion analysis of dual-energy XRT images. It aims to amplify the response differences of different materials under high-energy and low-energy XRT radiation, facilitating subsequent pseudo-color image generation and AI recognition algorithm training and optimization. The high-energy image after ECC registration and Gamma correction is denoted as... The original low-energy image used as a reference is denoted as First, normalize both images to the [0,1] interval to unify the dynamic range. The mathematical formula is as follows:

[0062] In the above formula These are the maximum gray values ​​in the high-energy and low-energy images, respectively. This embodiment uses logarithmic normalized two-energy difference, and its mathematical calculation formula is as follows:

[0063] In the above formula, It is the normalized logarithmic difference. This is a very small constant set to prevent the denominator from being zero. Subsequently, the grayscale image, after difference calculation, will be mapped to a pseudo-color image based on a color lookup table to enhance the contrast between lead-zinc deposits and other waste rock.

[0064] This embodiment constructs a batch mapping process for pseudo-color images of lead-zinc ore. The mapping is performed in three channels based on pixel intensity, with color transitions following a visually appealing sequence: dark blue → light blue → cyan → yellow → orange → red. Low grayscale values ​​are mapped to dark blue; medium grayscale values ​​transition to green and yellow; and high grayscale values ​​are mapped to orange-red. This approach balances color contrast with feature continuity.

[0065] To enhance the algorithm's adaptability, this embodiment integrates piecewise nonlinear mapping and a custom color level lookup table to construct a pseudo-color image mapping function, significantly improving the visualization capabilities for different targets to be identified. Simultaneously, by combining batch image reading and automatic naming and saving mechanisms, high-quality pseudo-color images are uniformly output, such as... Figure 6 As shown, this effectively improves data visualization efficiency and model training usability.

[0066] Depend on Figure 6 As can be seen, the image contains a large number of irregularly distributed noise points, which not only affect the visual experience but may also interfere with the training and optimization of the algorithm. Therefore, this embodiment introduces a "red background denoising algorithm" based on channel rules during the pseudo-color image generation process. By setting a red channel threshold condition and combining it with a color difference judgment mechanism, it automatically identifies redundant and interfering dark red background areas in the image and performs pixel-level replacement processing. The denoised lead-zinc ore pseudo-color image is shown below. Figure 7 As shown.

[0067] Depend on Figure 7 It can be seen that this method effectively eliminates the false response of the image background, improves the overall image clarity and makes it easier to identify, and provides a purer and higher quality input pseudo-color image dataset as the basis for subsequent mineral identification algorithms.

[0068] By using the targeted preprocessing and pseudo-color mapping algorithms described above, all images are divided into training, validation, and test sets in an 8:1:1 ratio, which is beneficial for the training and optimization of the YOLO algorithm.

[0069] The bounding box loss function, a key component of AI recognition algorithms, measures the difference between the algorithm's predicted bounding box and the actual target bounding box. It also optimizes model parameters through backpropagation to improve recognition accuracy. While YOLOv8's CIoU (Complete Intersection over Union) loss function is sensitive to localization accuracy, it may negatively impact the overall performance of the lead-zinc ore recognition algorithm, affecting its generalization ability.

[0070] To address the instability of traditional CIoU loss functions in identifying lead-zinc ore, this embodiment introduces the Weighted Intersection over Union (WIoU) loss function. This function combines the overlap between the predicted and ground truth bounding boxes, boundary offset, and target density distribution, improving the model's regression accuracy for weakly boundaryed ore targets. Its principle diagram is shown below. Figure 8 As shown. The green rectangle represents the actual bounding box, with a width and height of [missing information]. and The center coordinates are The blue rectangle represents the prediction box, with a width and height of [missing information]. and The center coordinates are The red line segment represents the Euclidean distance between the center of the predicted box and the center of the ground truth box, used to characterize the target offset; the overlapping area represents the intersection area of ​​the predicted box and the ground truth box, reflecting the degree of overlap between the two; the union region is determined by the smallest bounding rectangle formed by the two, used to calculate the intersection-union ratio.

[0071] When calculating the overlap between the blue predicted bounding box and the green target bounding box, WIoU not only considers IoU but also introduces a distance weighting factor based on the center point. This avoids the algorithm overemphasizing "easily distinguishable samples" and improves its optimization ability for "difficult-to-distinguish samples." Its mathematical formula is shown below:

[0072] In the above formula Indicates the center of the prediction box Center of the real frame The square of the Euclidean distance This represents the diagonal length of the smallest bounding rectangle that can enclose both the predicted and ground truth bounding boxes. This is a dynamic weighting factor used to adjust the gradient contribution of different samples.

[0073] In this embodiment, to improve the AI ​​recognition algorithm's ability to model the color differences in different regions of the lead-zinc ore pseudocolor image, an efficient multi-scale attention (EMA) module is introduced into the Neck part of the YOLO network. The structural principle of the EMA module is as follows: Figure 9 As shown.

[0074] EMA aims to extract channel attention and spatial attention information by grouping and multi-scale modeling input features, and then fusing them through normalization and weight allocation mechanisms. This guides the neural network to focus more on high-response regions in lead-zinc ore images, suppressing background and noise interference. This structure effectively enhances the feature representation ability of small targets, improves overall recognition accuracy, and does not significantly increase computational cost. Specifically, the size of the input feature map is... ,in Indicates the number of feature map channels, Indicates feature map height, This represents the width of the feature map. First, the input feature map is divided along the channel dimension into... Groups, each group of features has a size of [size missing]. This achieves parallel computation in groups and separation of feature subspaces. Subsequently, each group of features undergoes feature recalibration processing through the following steps:

[0075] (1) Local feature aggregation stage: Perform two-dimensional average pooling (Avg Pool) on each group of features to obtain global statistical features compressed along the spatial dimension, forming a feature of size . The channel description vector is used to characterize the global response intensity of the group of channels.

[0076] (2) Multi-scale dependency modeling stage: The pooling features are input into two branches of different scales. The first branch generates local scale weight coefficients after 1×1 convolution and normalization. The second branch generates global scale weight representations by combining Softmax and average pooling. The features of the two scales are then fused through matrix multiplication to obtain a multi-scale attention map containing local and global information.

[0077] (3) Feature weight generation stage: The fused attention map is weighted channel by channel, and the weight distribution range is constrained by the Sigmoid function so that the output weight values ​​are within a certain range. Within the interval, the significance of each channel feature is controlled.

[0078] (4) Feature Reconstruction and Output Stage: Finally, the generated attention weights are multiplied element-wise with the original input features to enhance salient channels and suppress insignificant channels. The weighted feature map maintains the same spatial size as the input, thus providing... This is ultimately the output of the EMA module.

[0079] In the XRT pseudocolor image recognition task of lead-zinc ore, the targets to be identified at different scales have very significant size differences. Although the FPN-PAN structure used in YOLOv8 can fuse multi-scale features, it still has certain limitations in terms of feature alignment and target response consistency. To further improve the fusion representation capability of multi-scale targets, this embodiment introduces a lightweight adaptive spatial feature fusion module (ASFF) to optimize the feature fusion path of the Head detection head. Its principle diagram is shown below. Figure 10 As shown.

[0080] The ASFF module adaptively assigns fusion weights to feature maps of different scales based on the importance of each location, thereby achieving more accurate spatial feature reconstruction. In this implementation, the three-scale fusion output in the original YOLO Neck structure is replaced with the output path of the ASFF module, preserving the multi-scale features at the input end, and spatial alignment and fusion are achieved through lightweight weight learning. This module uses input feature maps from feature layers of different scales. , , The principle and process are as follows:

[0081] (1) Multi-scale feature sampling stage: sampling the input multi-layer feature map , , Perform scale alignment operations on each layer to adjust the features of each layer to the same spatial resolution through upsampling or downsampling, thus obtaining standardized features. , , Among them, the superscript " "" indicates the process of mapping the feature to the fusion scale, ensuring the alignment of features from different layers in spatial dimensions.

[0082] (2) Feature compression and unified mapping stage: The standardized feature maps are processed separately. Convolutional operations achieve channel compression and semantic alignment. This process eliminates dimensionality differences between different feature layers, enabling weighted fusion of multi-scale features at the same spatial scale.

[0083] (3) Feature concatenation and weight generation stage: The convolutional features of the three scales are concatenated along the channel dimension to obtain a comprehensive feature tensor; then, through a... The convolutional layer further integrates cross-layer information and uses the softmax activation function to generate adaptive fusion weight coefficients. , , Furthermore, the sum of the three is 1 to balance the contribution ratio of features at different scales. Adaptive weighting and fusion output stage: The generated weight coefficients are respectively weighted and fused with the corresponding scale features. , , Perform element-wise multiplication, then sum the weighted results element-wise to obtain the final fused feature output.

[0084] Ablation experiments show that the introduction of the ASFF module significantly improves the model's accuracy in recognizing small targets, especially under complex conditions such as blurred edges and dense targets, where it performs more stably. Furthermore, due to the ASFF module's strong structural versatility and lightweight nature, its parameter and computational costs increase only slightly, having virtually no impact on inference speed, making it suitable for deployment on resource-constrained industrial terminal equipment. The introduction of the ASFF module effectively improves the feature fusion quality of YOLO in multi-scale lead-zinc mine XRT images, enhances the model's adaptability to complex target scenes, and is one of the important optimization strategies for achieving high-precision intelligent photoelectric mineral processing.

[0085] In standard convolution operations, all output feature maps are directly calculated from the input feature maps using convolution kernels of the same size. While this approach has good feature extraction and representation capabilities, it inevitably leads to a large number of parameters. Its high computational cost creates a bottleneck for low-cost deployment of the model.

[0086] GhostConv is a lightweight convolutional module proposed to improve the inference efficiency of convolutional neural networks. Its core idea is to reduce redundant computation by separating the feature map generation process. A diagram of GhostConv is shown below. Figure 11 As shown, in Figure 12 In Chinese, the abbreviation "Ghost" is used instead of "GhostConv".

[0087] GhostConv differs from traditional convolutional generation strategies in its design. Specifically, this module first generates a subset of main feature maps using fewer standard convolutional operations. Then, it generates more pseudo-feature maps from these main feature maps through a series of linear transformations, thus forming a complete set of output feature maps. The different transformation operators marked in the diagram represent the generation functions of these pseudo-features. Essentially, this is a low-computation feature expansion strategy that improves feature diversity and expressive power without significantly increasing model complexity. Specifically, the input feature map is denoted as... The processing flow is as follows:

[0088] (1) Main convolution feature extraction stage: First, the input feature map is processed. Perform regular convolution operations to generate basic feature maps. ,in This step only includes some of the real feature channels. In this step, only a small number of convolutional kernels are used to extract the core features, thereby reducing the main computational overhead.

[0089] (2) Feature generation and mapping stage: Due to feature redundancy in the depthwise convolution process, this module uses a linear transformation function. For feature maps The different channels are expanded to generate several auxiliary feature maps.

[0090] (3) Feature aggregation and fusion stage: output the main convolution. With all generated auxiliary features The final output features are formed by concatenating or weighting the data along the channel dimension.

[0091] (4) Identity mapping connection: In some implementations, the module can introduce an identity mapping branch (Identity) to connect the input features. With output features Element-wise addition or concatenation is used to enhance gradient flow and feature fidelity, thereby improving the stability of neural network training and the ability to reuse features.

[0092] Before the lightweight convolution replacement operation, the hyperparameters were configured using the Optuna hyperparameter search framework. The recognition results showed that after introducing GhostConv to replace part of the standard convolution, the algorithm significantly reduced the pure inference time while maintaining stable recognition accuracy. This further verified that GhostConv can effectively improve the algorithm's inference efficiency without affecting recognition performance, and is suitable for the real-time accurate and fast identification task of lead-zinc ore in this study.

[0093] Compared with the prior art, the beneficial effects of this embodiment are as follows:

[0094] In the preprocessing stage of dual-energy XRT images, ECC image registration method and pseudo-color mapping technology were adopted to enhance the grayscale contrast and feature expression of the images.

[0095] In the algorithm optimization stage, the WIoU loss function, efficient multi-scale attention module (EMA), lightweight adaptive spatial feature fusion (ASSF) module, and GhostConv lightweight convolution and Optuna hyperparameter optimization framework were further introduced to improve the accuracy and speed of the YOLO-based lead-zinc ore identification algorithm. Through these optimizations, the model's identification accuracy was improved to over 95%.

[0096] This embodiment obtains a lightweight YOLO AI recognition algorithm for lead-zinc ore by performing multi-stage optimization processing on XRT images of lead-zinc ore. The network results are as follows: Figure 12 As shown in the figure. This algorithm has good versatility and scalability, and can be widely applied in industrial fields such as photoelectric intelligent sorting of lead-zinc mines.

[0097] like Figure 13 As shown, the lead-zinc ore identification algorithm implementation scheme of this embodiment includes three main steps: image data preprocessing, algorithm training and optimization, and identification result display. Specific implementation details are as follows:

[0098] Image data preprocessing: For the acquired dual-energy XRT images of lead-zinc ore, the high- and low-energy images were first spatially aligned using the ECC registration method to eliminate image offset errors caused by factors such as conveyor belt movement and acquisition jitter, ensuring the accuracy of subsequent analysis. Then, a Gamma correction method was introduced to perform nonlinear mapping enhancement on the mid- and high-grayscale regions to improve the contrast and feature representation of ore and gangue in the images. Finally, pseudo-color mapping technology was used to perform a visual transformation of the grayscale images, making the ore and waste rock regions exhibit obvious color differences in the images, facilitating subsequent deep learning feature extraction.

[0099] Algorithm Training and Optimization: In the deep learning model part, this embodiment uses the YOLO network as the basic framework, further introducing the WIoU loss function to improve the regression accuracy of the predicted boxes, and embedding the EMA efficient multi-scale attention module in the Neck part to guide the network to focus on high-response regions and suppress background noise interference. Simultaneously, the GhostConv lightweight convolutional structure is combined to reduce the model's computational cost, and the ASSF adaptive spatial feature fusion module is used to improve the fusion effect of multi-scale features. To further enhance the model's generalization ability, this embodiment uses the Optuna hyperparameter optimization framework to automatically search and adjust key training parameters, thereby effectively improving the model's accuracy and robustness while ensuring recognition speed.

[0100] Image recognition results display: The trained and optimized model is applied to the XRT image recognition task in lead-zinc ore mines, and the recognition results are displayed in real time through visualization software. The software automatically labels ore and waste rock areas on the pseudo-color image and displays category labels and confidence values, with recognition accuracy consistently improved to over 95%. This display method not only facilitates the intuitive presentation of automatic ore identification and sorting but also provides practical support for the engineering application of photoelectric intelligent mineral processing. The recognition results are as follows: Figure 14 As shown.

[0101] As shown in Figure 14, the optimized YOLOv8 recognition algorithm of this embodiment processes the dual-energy XRT pseudo-color image of lead-zinc ore, enabling automatic identification and classification of ore and waste rock targets. The rectangles in the figure mark the detected target areas, with the "ore" and "waste" labels inside corresponding to lead-zinc ore and waste rock categories, respectively. The numerical values ​​also represent the recognition confidence levels. The results show that the algorithm maintains high recognition accuracy across target areas of different sizes, shapes, and partial overlap, indicating good robustness and consistency in category differentiation.

[0102] Furthermore, the recognition results show that even in situations with a relatively simple background and blurred local target boundaries, the model can still accurately locate the target's central region and generate a reasonable bounding box. This demonstrates the effectiveness of the introduced WIoU loss function and EMA attention mechanism in enhancing target localization accuracy. Overall, the recognition results validate the potential of the proposed intelligent lead-zinc ore recognition algorithm under complex XRT imaging conditions and its engineering applications. It not only assists in manual image annotation but also ensures real-time performance meets the operational requirements of industrial XRT sorting machines through operator optimization and hardware acceleration strategies.

[0103] This application also provides an electronic device, including: a processor, and a memory coupled to the processor, the memory being used to store a computer program; the processor being used to execute the computer program stored in the memory, so that the electronic device performs the method as described in any of the above embodiments.

[0104] Electronic devices can be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These electronic devices may include, but are not limited to, processors and memory.

[0105] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the device via various interfaces and lines.

[0106] The memory can be used to store the computer program, and the processor implements various functions of the electronic device by running or executing the computer program stored in the memory and calling the data stored in the memory.

[0107] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0108] This application also provides a storage medium, which is a computer-readable storage medium. The computer program is stored in the computer-readable storage medium, and when executed by a processor, the computer program can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0109] This application also provides a computer program product, including: a computer program or instructions that, when the computer program or instructions are run on a computer, cause the computer to perform any of the above possible implementation methods.

[0110] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A method for separating lead-zinc ore, characterized in that, include: In the preprocessing stage, the original high-energy image and the original low-energy image of the raw ore are obtained. The original high-energy image and the original low-energy image are obtained by dual-energy X-ray projection imaging of the raw ore. Using the enhanced correlation coefficient registration algorithm, the original high-energy image is registered with the original low-energy image as a reference to obtain a secondary high-energy image. The original low-energy image and the secondary high-energy image are fused using a physical response difference metric to obtain a fused image. Finally, the fused image is colored according to color transition rules to obtain a pseudo-color image. as well as, In the intelligent recognition stage, the pseudo-color image is used as the input to a deep learning-based target detection model. Through the processing of the target detection model, the sorting results of lead-zinc ore in the raw ore are obtained.

2. The lead-zinc ore beneficiation method as described in claim 1, characterized in that, The physical response difference measure is the normalized logarithmic difference. : in, This is a secondary high-energy diagram. This is the original low-energy map. These are the maximum gray values ​​in the secondary high-energy image and the original low-energy image, respectively. It is a very small constant set to prevent the denominator from being zero.

3. The lead-zinc ore beneficiation method as described in claim 1, characterized in that, The target detection model adopts the YOLO model, which includes a feature extraction network, a feature fusion network, and a detection head.

4. The lead-zinc ore beneficiation method as described in claim 3, characterized in that, The YOLO model uses a weighted cross-union loss function L. WIoU : in, Indicates the center of the prediction box Center of the real frame The square of the Euclidean distance This represents the diagonal length of the smallest bounding rectangle that can enclose both the predicted and ground truth bounding boxes. This is a dynamic weighting factor used to adjust the gradient contribution of different samples.

5. The lead-zinc ore beneficiation method as described in claim 3, characterized in that, The feature fusion network includes a C2f layer and a Concat layer. An efficient multi-scale attention layer is set between the C2f layer and the Concat layer. The efficient multi-scale attention layer is used to perform grouped convolution and multi-scale modeling on the input features, extract channel attention and spatial attention information respectively, and fuse them through normalization and weight allocation mechanisms.

6. The lead-zinc ore beneficiation method as described in claim 3, characterized in that, The detection head includes a lightweight adaptive spatial feature fusion module for adaptively assigning fusion weights to feature maps of different scales according to the importance of each location.

7. The lead-zinc ore beneficiation method as described in claim 3, characterized in that, The feature extraction network employs ghost convolutional layers.

8. The lead-zinc ore beneficiation method as described in claim 3, characterized in that, The YOLO model uses the Optuna hyperparameter search framework.

9. The lead-zinc ore beneficiation method as described in claim 1, characterized in that, The preprocessing stage further includes performing gamma correction on the secondary high-energy image and the original low-energy image.

10. An electronic device, characterized in that, The electronic device includes: a processor, and a memory coupled to the processor. The memory is used to store computer programs; and The processor is configured to execute the computer program stored in the memory, so that the electronic device performs the lead-zinc ore sorting method as described in any one of claims 1-8.

Images