A method of ore sorting, a sorting system and a sorting machine

Abstract

This disclosure relates to the field of sorting equipment technology, specifically to an ore sorting method, classification system, and sorting machine. The ore sorting method includes: acquiring a surface image of the ore to be sorted; determining a primary classification result of the ore to be sorted based on the surface image using a primary classifier; responding to the primary classification result being a textured ore, determining a secondary classification result of the ore to be sorted based on the texture features of the surface image using a secondary classifier; determining a tertiary classifier from multiple tertiary classifiers as the final classifier based on the primary and secondary classification results; and determining the final classification result of the ore to be sorted based on the surface image using the final classifier. According to this disclosure, the ore to be sorted can be finely grouped by texture based on the surface image, and the corresponding tertiary classifier can be used to accurately determine the final classification result based on the fine texture grouping, thereby achieving accurate classification of the ore to be sorted.

Description

Technical Field

[0001] This application mainly relates to the field of sorting equipment technology, and in particular to an ore sorting method, classification system and sorting machine. Background Technology

[0002] In the field of ore sorting, some ore resources exist in the raw ore in multiple forms, and these forms are significantly complex and diverse. In response, many related technologies use a single overall classifier to sort all types of raw ore uniformly. However, for example, in the field of metal ore, especially gold ore sorting, gold resources are mostly found in the raw ore as associated minerals, and their associated combination forms are significantly complex and diverse, resulting in scattered and overlapping discriminative feature dimensions. This leads to the following technical defects in such an overall classifier: (1) The model is extremely difficult to build, requiring simultaneous adaptation to multi-dimensional heterogeneous features, which leads to complex feature fusion logic and a sharp increase in parameter debugging costs; (2) Model optimization is stuck in a bottleneck. Optimizing the sorting accuracy for one type of ore can easily lead to an increase in the misjudgment rate of another type of ore, making it difficult to achieve the optimal across all scenarios; (3) The classification accuracy is low. The mutual interference of heterogeneous features makes the classifier's ability to discriminate boundary cases insufficient, failing to meet the dual requirements of classification efficiency and accuracy for industrial production. Summary of the Invention

[0003] To overcome the problems existing in related technologies, an exemplary embodiment of this disclosure provides an ore sorting method in a first aspect. The method includes: acquiring a surface image of the ore to be sorted; determining a primary classification result of the ore to be sorted based on the surface image using a primary classifier, wherein the primary classification result includes solid-color ore or textured ore; in response to the primary classification result being textured ore, determining a secondary classification result of the ore to be sorted based on the texture features of the surface image using a secondary classifier, wherein the secondary classification result includes multiple different types of sub-textured ore; determining a tertiary classifier as a final classifier from multiple tertiary classifiers according to the primary and secondary classification results; and determining a final classification result of the ore to be sorted based on the surface image using the final classifier, wherein the final classification result includes concentrate and waste ore.

[0004] In some embodiments, based on the surface image, the primary classification result of the ore to be sorted is determined by a primary classifier, including: determining the texture uniformity of the surface image by the primary classifier; if the texture uniformity is greater than or equal to the texture uniformity threshold, the primary classification result is a solid color ore; if the texture uniformity is less than the texture uniformity threshold, the primary classification result is a textured ore.

[0005] In some embodiments, multiple different types of sub-texture ores include: striped ores, mottled ores, and dotted ores; multiple three-level classifiers include: a solid color-specific classifier, a striped ores-specific classifier, a mottled ores-specific classifier, and a dotted ores-specific classifier; based on the surface image, the final classification result of the ores to be sorted is determined by the final classifier, including: determining the final classification result of the ores to be sorted based on the surface image of solid color ores using the solid color-specific classifier; determining the final classification result of the ores to be sorted based on the surface image of striped ores using the striped ores-specific classifier; determining the final classification result of the ores to be sorted based on the surface image of mottled ores using the mottled ores-specific classifier; and determining the final classification result of the ores to be sorted based on the surface image of dotted ores using the dotted ores-specific classifier.

[0006] In some embodiments, the final classification result of the ore to be sorted is determined based on the surface image of the ore belonging to the solid color by using a solid color-specific classifier, including: determining the background color of the surface image by using the solid color-specific classifier; if the background color belongs to a preset concentrate color range, the final classification result is concentrate; if the background color does not belong to the preset concentrate color range, the final classification result is waste ore.

[0007] In some embodiments, the final classification result of the ore to be sorted is determined based on the surface image of the ore belonging to the spotted ore by using a spotted-specific classifier, including: determining the spotted color and background color of the surface image data by using the spotted-specific classifier; if the combination of background color and spotted color belongs to a preset color combination, the final classification result is concentrate; if the combination of background color and spotted color does not belong to the preset color combination, the final classification result is waste ore.

[0008] In some embodiments, the final classification result of the ore to be sorted is determined based on the surface image of the point-specific ore using a point-specific classifier, including: determining the color of the surface associated mineral region of the surface image data using the point-specific classifier; if the matching degree between the color of the surface associated mineral region and the characteristic color of pyrite is not less than a preset matching degree, the final classification result is concentrate; if the matching degree between the color of the surface associated mineral region and the characteristic color of pyrite is less than the preset matching degree, the final classification result is waste ore.

[0009] In some embodiments, the secondary classifier includes a neural network model trained on a training dataset. The training dataset includes multiple training images, each labeled with one of the following: striped ore, mottled ore, or dotted ore. The training images are labeled using the following methods: if one or more of the following parameters in the training image—the percentage of striped pixels, the consistency of the direction of each stripe, the contrast between the stripes and the background color, and the distribution density of pyrite—are greater than the corresponding thresholds, then striped ore is used as the label for the training image; if the texture type in the training image is irregular patchy, and one or more of the following parameters—the number of patches, the percentage of the area of ​​a single patch, the contrast between the patch and the background color, and the overlap rate between the pyrite distribution area and the patch area—are greater than the corresponding thresholds, then mottled ore is used as the label for the training image; if the texture type in the training image is discrete dotted, the dotted distribution type is non-directional, and one or more of the following parameters—the contrast between the dotted area and the background color, and the spectral matching degree between the dotted area and pyrite—are greater than the corresponding thresholds—then dotted ore is used as the label for the training image.

[0010] In some embodiments, the method further includes: determining the error based on the primary classification result, the secondary classification result, and the final classification result; determining the error type based on the error; and adjusting the parameters of the corresponding primary classifier, secondary classifier, or tertiary classifier based on the error type.

[0011] Secondly, this disclosure also provides a classification system, comprising: a data acquisition device for acquiring a surface image of the ore to be sorted; and a classification device for determining the final classification result of the ore to be sorted according to the ore sorting method of the first aspect.

[0012] Thirdly, this disclosure also provides a sorting machine, including: a conveying device for conveying ore to be sorted; a classification system as described in the second aspect for acquiring a surface image of the ore to be sorted and determining the corresponding final classification result; and a sorting device for sorting the ore to be sorted according to the final classification result.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0014] According to the ore sorting method provided in this disclosure, a hierarchical combination of primary and secondary classifiers can be used to handle surface images of ores with complex surface features, accurately classifying the ores into solid-color ores and multiple sub-texture ores. Furthermore, by setting multiple tertiary classifiers corresponding to solid-color ores and each sub-texture ore, each tertiary classifier can focus on the relevant surface features of the corresponding ores, resulting in a more accurate and reliable final classification result. This not only simplifies the overall complexity of individual classifiers, enabling convenient model building and reducing debugging costs, but also allows for individual optimization of the classification accuracy of each classifier without mutual interference, thereby improving the accuracy of the final classification result by enhancing the local classification accuracy of each classifier. Attached Figure Description

[0015] The accompanying drawings are included to provide a further understanding of this application; they are incorporated into and constitute a part of this application. The drawings illustrate embodiments of this application and, together with this specification, serve to explain the principles of this application. In the drawings:

[0016] Figure 1 This is a flowchart illustrating an ore sorting method according to a disclosed exemplary embodiment;

[0017] Figure 2 This is a schematic diagram of a solid-color mineral according to an exemplary embodiment of a published document;

[0018] Figure 3 This is a schematic diagram of mottled mineral according to an exemplary embodiment of a published document;

[0019] Figure 4 This is a schematic diagram of striped ore shown according to an exemplary embodiment of a disclosed document;

[0020] Figure 5 This is a schematic diagram of dot-shaped ore according to an exemplary embodiment disclosed in a book;

[0021] Figure 6 This is a flowchart illustrating an ore sorting method according to a disclosed exemplary embodiment;

[0022] Figure 7 This is a flowchart illustrating a training image annotation method according to a publicly available exemplary embodiment;

[0023] Figure 8 This is a flowchart illustrating an ore sorting method according to a disclosed exemplary embodiment;

[0024] Figure 9 This is a flowchart illustrating an ore sorting method according to a disclosed exemplary embodiment;

[0025] Figure 10This is a flowchart illustrating an ore sorting method according to a disclosed exemplary embodiment;

[0026] Figure 11 This is a flowchart illustrating an ore sorting method according to a disclosed exemplary embodiment;

[0027] Figure 12 This is a flowchart illustrating an ore sorting method according to an exemplary embodiment disclosed in a book. Detailed Implementation

[0028] The following describes specific embodiments of this disclosure. It should be noted that, in order to maintain brevity, this specification cannot provide a detailed description of all features of the actual embodiments. It should be understood that, in the actual implementation of any embodiment, just as in any engineering or design project, various specific decisions are often made to achieve the developer's specific goals and to meet system-related or business-related constraints, and this can change from one embodiment to another. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content of this disclosure, changes in design, manufacturing, or production based on the technical content disclosed herein are merely conventional technical means and should not be construed as insufficient content of this disclosure.

[0029] Unless otherwise defined, the technical or scientific terms used in the claims and description shall have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” and similar terms used in the specification and claims of this patent application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the element or object preceding “comprising” or “including” encompasses the element or object listed following “comprising” or “including” and its equivalents, and do not exclude other elements or objects. The terms “connected” or “linked” and similar terms are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.

[0030] In ore sorting, ores are typically classified based on their surface images. For example, in binary classification, the surface images of the ore are used to separate it into concentrate or waste ore. However, some ores exhibit significant complexity and diversity in their surface images. For instance, gold resources, often found as associated minerals in the ore, frequently coexist with pyrite, and the ore often displays various inclusions, including typical types such as gray single ore, gray quartz-intercalated ore, and brown ore with black lines. In many related technologies, a single, holistic classifier is used to uniformly analyze the ore of the same metal resource. Clearly, for gold resources, such a single, holistic classifier requires collecting and processing multiple completely different features from the surface images of the gold ore, leading to complex internal processing logic that is difficult to debug effectively. This results in extremely high construction costs and difficulties for a single, holistic classifier. Furthermore, because a single holistic classifier needs to be applicable to various types of gold ore, the optimizable parameters within it are interconnected, making it difficult to effectively determine parameters that can be optimized for a specific type of gold ore. This results in difficulty in accurately and effectively improving the identification accuracy for a particular type of gold ore. Consequently, single holistic classifiers also suffer from low classification accuracy, difficulty in adjusting parameters, and low identification and sorting efficiency.

[0031] To solve the above technical problems, such as Figure 1 As shown, this disclosure provides an ore sorting method that can be applied to the sorting of gold ore. The ore sorting method may include steps S110 to S150.

[0032] Step S110: Acquire a surface image of the ore to be sorted. During the transport of the ore, a surface image of the ore can be acquired using a data acquisition device, such as a camera. Multiple data acquisition devices can be set up at different locations during transport to acquire surface images of multiple different sides of the ore, resulting in more comprehensive and complete surface image data to improve the accuracy of subsequent classification results. The image acquisition device can output high-resolution and high-pixel color surface images, allowing the surface images to contain more color and texture information of the ore to be sorted, thus enabling subsequent processing steps to extract more detailed features and improve classification accuracy. The surface image can be a standardized image obtained by preprocessing the initial image acquired by the image acquisition device. Preprocessing can include one or more operations such as image noise reduction, feature normalization, outlier removal, and background color region extraction. Background color region extraction can extract the region containing the background color of the ore to be sorted.

[0033] Step S120: Based on the surface image, a primary classifier is used to determine the primary classification result of the ore to be sorted, wherein the primary classification result includes solid-color ore or textured ore. For example, Figure 2 Images (a) through (f) are surface images corresponding to solid-colored minerals. Figure 3 (a) to (c) Figure 4 Middle (a) to (e) and Figure 5 All images are surface images corresponding to textured ores. In this embodiment, the surface image of a single ore to be sorted can be input into a primary classifier to obtain the primary classification result of that ore. This allows the primary classifier to have high classification accuracy while maintaining a small size, facilitating overall migration and subsequent optimization. In other embodiments, surface images of multiple ores to be sorted can be input into the primary classifier to obtain the primary classification result of each ore, thereby improving the classification efficiency of the primary classifier to meet the needs of efficient industrial sorting. Since gold ore contains not only a type of ore with a single-color surface feature but also another type with complex textured surface features, the primary classifier can effectively perform preliminary classification of gold ore types, avoiding interference between the surface features of different types of ores, thereby improving the accuracy of subsequent classification operations.

[0034] Step S130: In response to the primary classification result being textured ore, a secondary classifier is used to determine the secondary classification result of the ore to be sorted based on the texture features of the surface image. The secondary classification result includes multiple sub-textured ores of different types. In this embodiment, the surface image of the ore to be sorted, determined as textured ore by the primary classifier, can be input into the secondary classifier to obtain the secondary classification result of the ore to be sorted. This allows the secondary classifier to have high classification accuracy while maintaining a small size, facilitating overall migration and subsequent optimization. In other embodiments, a fused image containing surface images of multiple textured ores can be input into the secondary classifier to obtain the secondary classification result of each ore to be sorted, thereby improving the classification efficiency of the secondary classifier to meet the needs of efficient industrial sorting. For gold ore, there are various different texture morphologies, such as stripes and spots. Therefore, a secondary classifier can be used to further classify the screened textured ores, thereby further subdividing the gold ore with different texture morphologies into multiple sub-textured ores of different types. This allows for detailed classification of a large number of textured ores, reducing the complexity and difficulty of subsequent classification. Because the secondary classifier is completely independent of the primary classifier, it can perform secondary classification solely based on the relevant detailed features of the textured mineral, unaffected by the primary classifier. This reduces the structural complexity of the secondary classifier while increasing its adjustability, thus contributing to improved accuracy. Similarly, the primary classifier is not limited by the secondary classifier, allowing it to simplify its structure while improving the accuracy of its classification results.

[0035] Step S150: Based on the primary and secondary classification results, determine one tertiary classifier from multiple tertiary classifiers as the final classifier. In this embodiment, multiple tertiary classifiers can be set to correspond to different ores to be sorted, thus enabling the selection of the most suitable tertiary classifier based on the primary and secondary classification results of the ores. The secondary classification result for solid-color ores can be empty, ensuring that each ore to be sorted has primary and secondary classification results that can be recognized by the program. Furthermore, by setting multiple tertiary classifiers, the structural complexity of a single tertiary classifier can be reduced, thereby improving the classification effect of each tertiary classifier on specific ores to be sorted, and ultimately improving the classification accuracy of the ore sorting method for all types of ores to be sorted.

[0036] Step S150: Based on the surface image, the final classification result of the ore to be sorted is determined by the final classifier. The final classification result includes concentrate and waste ore. In this embodiment, the surface image corresponding to a single ore to be sorted can be input into the final classifier to obtain the final classification result for that ore. This allows the final classifier to have high classification accuracy while maintaining a small size, facilitating overall migration and subsequent optimization. In other embodiments, a fused image containing surface images of multiple ores with the same sub-texture can be input into the final classifier to obtain the final classification result for each ore to be sorted, thereby improving the classification efficiency of the final classifier to meet the needs of efficient industrial sorting.

[0037] The ore sorting method provided in this embodiment can handle surface images of gold ore with complex surface features by using a hierarchical combination of primary and secondary classifiers. This allows for accurate classification of the gold ore to be sorted into solid-color ores and ores with multiple sub-textures. Furthermore, by setting multiple tertiary classifiers corresponding to solid-color ores and each sub-texture, each tertiary classifier can focus on the relevant surface features of the corresponding ore, resulting in a more accurate and reliable final classification result. Therefore, replacing a single overall classifier with primary, secondary, and tertiary classifiers not only simplifies the overall complexity of a single classifier, facilitating model building and reducing debugging costs, but also allows for targeted adjustments to the classification results of each classifier to ensure high classification accuracy for each classifier, thereby resulting in a high overall accuracy of the final classification result. Furthermore, since the first-level classifier, the second-level classifier, and multiple third-level classifiers are independent of each other, each classifier has sufficient judgment ability for boundary cases and is not affected or interfered with by other classifiers. This improves the accuracy of the classification results of each classifier and makes it easier to effectively optimize and adjust the classification after errors are found.

[0038] In some embodiments, such as Figure 6 As shown, step S120, based on the surface image, determines the primary classification result of the ore to be sorted through a primary classifier, which may include steps S121 to S123.

[0039] Step S121: Determine the texture uniformity of the surface image using a first-level classifier. After preprocessing with grayscale conversion, denoising, and normalization, the surface image is divided into multiple non-overlapping sub-images, and texture features are extracted from each sub-image using a Gray Level Co-occurrence Matrix (GLCM). The dispersion of all texture features is then calculated to determine the texture uniformity. Texture uniformity can be a value between 0 and 1. A uniformity closer to 1 indicates a more even texture distribution with no significant local differences, suggesting a preference for solid-color minerals. Conversely, a uniformity closer to 0 indicates a more uneven texture distribution with significant local differences, suggesting a preference for textured minerals.

[0040] Step S122: If the texture uniformity is greater than or equal to the texture uniformity threshold, the primary classification result is a solid-color ore. Step S122 uses a primary classifier to judge the relationship between the texture uniformity of the surface image and the texture uniformity threshold, thereby determining whether the texture uniformity is greater than or equal to the texture uniformity threshold. A texture uniformity greater than or equal to the texture uniformity threshold indicates that the color distribution in the surface image is uniform, and there are not too many textures of other colors that stand out from the background color, thus determining that the ore to be sorted is a solid-color ore.

[0041] Step S123: If the texture uniformity is less than the texture uniformity threshold, the primary classification result is textured ore. In step S123, the primary classifier can judge the relationship between the texture uniformity of the surface image and the texture uniformity threshold to determine whether the texture uniformity is less than the threshold. A texture uniformity less than the threshold indicates uneven color distribution in the surface image, with textures of other colors significantly protruding from the background color, thus identifying the ore to be sorted as textured ore.

[0042] According to the ore sorting method provided in this embodiment, by determining the texture uniformity of the surface image and combining it with a texture uniformity threshold, it is possible to easily and effectively distinguish between solid-color ores and textured ores. Thus, the primary classifier can further simplify the overall architecture while ensuring recognition accuracy. In some embodiments, a lightweight CNN (Convolutional Neural Network) model can be used as the primary classifier to reduce the model size while maintaining high classification accuracy.

[0043] In some embodiments, the secondary classifier may include a neural network model trained on a training dataset, such as a trained convolutional neural network model. The training dataset includes multiple training images, each labeled with one of the following: striped ore, mottled ore, or dotted ore. To address the complexity of surface features when gold ore exhibits surface texture, the textured ore can be further categorized into striped ore, mottled ore, and dotted ore. This classifies the surface texture of the gold ore based on its essential characteristics, facilitating accurate classification results when applying the corresponding final classifier to each sub-textured ore. For example, Figure 3 In the middle (a), (b) and (c), respectively, are surface images of mottled ores with different mottled features, where the position of the dashed circle is a patch on the surface of the corresponding mottled ores; Figure 4 (a), (b), (c), (d) and (e) are surface images of striped ores with different stripe characteristics, where the dashed circle represents a stripe on the surface of the corresponding striped ores. Figure 5 This is a surface image corresponding to a dotted ore, where the black dashed circle represents a dotted area on the surface of the striped ore.

[0044] The neural network model of a two-level classifier can include an input layer, a feature extraction layer, a fully connected layer, and an output layer. The input layer receives a tensor of the surface image; the feature extraction layer extracts associated features from the tensor; the fully connected layer performs dimensionality compression and fusion of the associated features to obtain fused features; and the output layer generates predicted probabilities for striped ore, mottled ore, and dotted ore corresponding to the surface image based on the fused features. The feature extraction layer can include three convolutional blocks, each consisting of two convolutional layers and one pooling layer. The convolutional layers can use 3×3 kernels, and the pooling layer can use 2×2 max pooling. By gradually increasing the number of convolutional kernels (e.g., from 32 to 64 kernels, and then from 64 to 128 kernels), the feature extraction capability is improved, thereby automatically learning associated features such as stripe width, stripe direction, texture morphology, and pyrite distribution. A fully connected layer can include two fully connected sub-layers, for example, a first fully connected sub-layer with 1024 dimensions and a second fully connected sub-layer with 256 dimensions. The fully connected layer can also introduce a Dropout layer (random deactivation layer), such as a Dropout layer with a dropout rate of 0.5, to avoid overfitting while compressing and fusing the extracted associated features. The output layer can use the Softmax (normalized exponential) activation function to output predicted probabilities for multiple different categories, and select the category with the highest probability as the secondary classification result. In other embodiments, the neural network model can also output the predicted probability of abandoned mines to filter out some abandoned mines with significant features, reducing the data processing complexity of the subsequent tertiary classifier, thereby improving the accuracy of the final classification result.

[0045] The training dataset can be divided into training, validation, and test sets in a 7:1.5:1.5 ratio to iteratively train the neural network model until the training termination condition is met. The training dataset contains at least 5000 training data sets to ensure the neural network model has good generalization ability. The loss function for the neural network model can include the cross-entropy loss function, and the optimizer can include the Adam optimizer (Adaptive Moment Estimation optimizer). The Adam optimizer has an initial learning rate of 0.001 and adaptively adjusts the learning rate. The training termination condition can include stopping training when the accuracy of the neural network model on the validation set does not improve for 10 consecutive epochs (traversal periods), or when the accuracy of the neural network model on the training set is ≥98%, and saving the optimal model parameters.

[0046] like Figure 7 As shown, the training images are labeled using the following method, which may include steps S210 to S230.

[0047] Step S210: If one or more of the following parameters in the training image—the percentage of striped pixels, the consistency of stripe direction, the contrast between stripes and background color, and the pyrite distribution density—are greater than the corresponding thresholds, then the striped ore is used as the label for the training image. In some embodiments, for gold ore, the following parameters are determined for the training image: the percentage of striped pixels ≥ 5%, the consistency of stripe direction ≥ 80%, the contrast between stripes and background color ≥ 30%, and the pyrite distribution density ≥ 0.5 pixels / cm². 2 If so, striped ore can be used as the label for the training image. Specifically, the percentage of striped pixels, the consistency of stripe direction, and the contrast between the stripes and the background color can be used to verify whether the gold ore corresponding to the training image meets the core indicators of striped ore, thus determining whether the gold ore corresponding to the training image belongs to striped ore. Furthermore, the distribution density of pyrite can exclude pure striped ore without associated pyrite, i.e., waste ore. This allows the trained neural network model, i.e., the second-level classifier, to further exclude some waste ore with significant characteristics, reducing the classification complexity of the subsequent third-level classifier and improving its classification accuracy.

[0048] Step S220: If the texture type in the training image is irregular patchy, and one or more of the following—the number of patches, the area ratio of a single patch, the contrast between the patch and the background color, and the overlap rate between the pyrite distribution area and the patch area—are greater than the corresponding thresholds, then the mottled ore is used as the label for the training image. In some embodiments, for gold ore, the irregular patchy texture type in the training image can serve as the basis for identifying mottled ore. Based on this, if the number of patches in the training image is ≥3, the area ratio of a single patch (i.e., the ratio of the area of ​​each patch to the entire gold ore surface) is ≥3%, the contrast between the patch and the background color is ≥25%, and the overlap rate between the pyrite distribution area and the patch area is ≥20%, then the mottled ore can be used as the label for that training image. The number of patches, the area ratio of a single patch, and the contrast between the patch and the background color can be used to verify whether the gold ore corresponding to the training image meets the core indicators of mottled ore, thereby determining whether the gold ore corresponding to the training image belongs to the mottled ore category. Based on this, by using the overlap rate between the distribution area of ​​pyrite and the patch area, the relevant characteristics of associated pyrite can be used to determine whether the gold ore is waste ore. This allows the trained neural network model, i.e., the second-level classifier, to exclude some waste ore with significant characteristics, thereby reducing the classification complexity of the subsequent third-level classifier and improving the classification accuracy of the third-level classifier.

[0049] Step S230: If the texture type in the training image is discrete dots, the dot distribution type is non-directional, and one or more of the following—the contrast between the dot region and the background color, and the spectral matching degree between the dot region and pyrite—are greater than the corresponding thresholds, then the dotted ore is used as the label for the training image. For gold ore, the texture type in the training image is discrete dots, which can serve as the basis for identifying dotted ore. Based on this, in some embodiments, the dot distribution type in the training image is determined to be non-directional, and the dot density is ≥10 dots / cm². 2 If the contrast between the dotted region and the background color is ≥20% and the spectral matching degree between the dotted region and pyrite is ≥50%, then the dotted ore can be used as the label for the training image. Specifically, the dotted distribution type, dot density, and contrast between the dotted region and the background color can verify whether the gold ore corresponding to the training image meets the core indicators of dotted ore, thus determining whether the gold ore corresponding to the training image belongs to the dotted ore category. Furthermore, by using the spectral matching degree between the dotted region and pyrite, the relevant characteristics of associated pyrite can be used to determine whether the gold ore is waste ore. This allows the trained neural network model (secondary classifier) ​​to exclude some waste ore with significant characteristics, reducing the classification complexity of the subsequent tertiary classifier and improving its classification accuracy.

[0050] According to the training image annotation method provided in this embodiment, it is possible to distinguish different types of ores with obvious characteristics in ore sorting (especially gold ore sorting), and to classify ores in detail according to their texture so that ores with the same texture features are classified into the same category. This simplifies the feature information contained in ores in the same category, allowing the three-level classifiers for different categories to focus on the simplified feature information, avoiding interference from too much feature information, improving the classification accuracy of the subsequent three-level classifiers, and ensuring the accuracy and reliability of the final classification results.

[0051] In some embodiments, multiple different types of sub-texture ores may include: striped ores, mottled ores, and dotted ores. Correspondingly, multiple tertiary classifiers include: a solid color-specific classifier, a striped-specific classifier, a mottled-specific classifier, and a dotted-specific classifier. For example... Figure 8 As shown, step S150, based on the surface image, determines the final classification result of the ore to be sorted through the final classifier, which may include steps S151 to S154.

[0052] Step S151: Using a solid color-specific classifier, the final classification result of the ore to be sorted is determined based on the surface image of the solid color ore. In step S151, the surface image of the ore to be sorted corresponding to the solid color ore can be input into the solid color-specific classifier, thereby outputting the final classification result of the ore to be sorted.

[0053] Step S152 involves using a stripe-specific classifier to determine the final classification result of the ore to be sorted based on the surface image of the striped ore. In step S152, the surface image of the ore to be sorted corresponding to the striped ore can be input into the stripe-specific classifier, which then outputs the final classification result. The stripe-specific classifier may include a CNN model, which is trained using a training dataset. This allows the CNN model to determine whether the corresponding striped ore is concentrate or waste ore based on different stripe patterns. Therefore, the stripe-specific classifier containing the trained CNN model can accurately identify concentrate or waste ore based on the stripe pattern in the surface image.

[0054] Step S153: Using a mottled-specific classifier, the final classification result of the ore to be sorted is determined based on the surface image of the mottled ore. In step S153, the surface image of the ore to be sorted corresponding to the mottled ore can be input into the mottled-specific classifier to output the final classification result of the ore to be sorted.

[0055] Step S154: Using a point-specific classifier, the final classification result of the ore to be sorted is determined based on the surface image of the point-type ore. In step S154, the surface image of the ore to be sorted corresponding to the point-type ore can be input into the point-specific classifier to output the final classification result of the ore to be sorted.

[0056] According to the ore sorting method provided in this embodiment, three-level classifiers are set for solid-color ore, striped ore, mottled ore, and dotted ore respectively. These are dedicated classifiers for solid color, striped, mottled, and dotted ore, corresponding to the four main types of ore textures in gold ore. This allows for effective classification of the gold ore to be sorted. Furthermore, the total number of features required for each three-level classifier is reduced, simplifying the model size of each classifier. As a result, each three-level classifier has a smaller model size, reducing the number of parameters that need to be adjusted. This allows for more efficient and accurate optimization of all three-level classifiers, thereby improving their accuracy.

[0057] In some embodiments, such as Figure 9 As shown, step S151, which uses a solid color-specific classifier to determine the final classification result of the ore to be sorted based on the surface image of the solid color ore, may include steps S1511 to S1513.

[0058] Step S1511: Determine the background color of the surface image using a solid color-specific classifier. In step S1511, the surface image of the ore to be sorted corresponding to a solid color can be input into the solid color-specific classifier, thereby determining the background color of the surface image. The background color can be the color with the largest proportion of the ore to be sorted in the surface image. The surface image can be in RGB format, allowing it to reflect more color information, thus helping to improve the accuracy of color-based recognition results.

[0059] Step S1512: If the background color belongs to the preset concentrate color range, the final classification result is concentrate.

[0060] Step S1513: If the background color does not belong to the preset concentrate color range, the final classification result is waste ore. For ore sorting, especially for gold ore, the color range belonging to the concentrate can be determined as the preset concentrate color range based on the actual recoverable gold ore results of existing pure-colored ore. This allows for quick and accurate determination of whether the pure-colored ore contains recoverable gold or is waste ore without recoverable gold based on its surface color. The pure-color dedicated classifier can include a logistic regression model, which can accurately determine whether the background color belongs to the preset concentrate color range while reducing the number of model parameters, thereby reducing the model debugging and maintenance costs, and has a faster deployment speed to meet the needs of industrial production.

[0061] According to the ore sorting method provided in this embodiment, by setting a corresponding solid-color-specific classifier for solid-color ores, the method can judge based on the unique characteristics of solid-color ores, unaffected by the related characteristics of other types of ores, thus simplifying the classification logic of the solid-color-specific classifier. Based on this, by comparing the background color of the concentrate with the background color of the ore to be sorted, the final classification result for the solid-color ores can be obtained simply and effectively, thereby ensuring the accuracy and reliability of the final classification result. In some embodiments, such as... Figure 10 As shown, step S153, which uses a mottled-specific classifier to determine the final classification result of the ore to be sorted based on the surface image of the mottled ore, may include steps S1531 to S1533.

[0062] Step S1531 involves determining the mottled color and background color of the surface image data using a mottled-specific classifier. In step S1531, the surface image of the ore to be sorted, corresponding to the mottled ore, can be input into the mottled-specific classifier to determine the background color and mottled color of the surface image. The background color can be the color with the highest proportion in the ore to be sorted in the surface image. The mottled color can be the color corresponding to each mottled shape in the surface image. The surface image can be in RGB format, allowing it to reflect more color information and thus improving the accuracy of color-based recognition results.

[0063] Step S1532: If the combination of the base color and the mottled color belongs to the preset color combination, the final classification result is concentrate.

[0064] Step S1533: If the combination of background color and speckled color does not belong to the preset color combination, the final classification result is waste ore. For gold ore, the combinations of speckled color and background color belonging to concentrate can be determined based on the actual recoverable gold ore results of existing speckled ore, thus forming a preset color combination. That is, the preset color combination can include multiple sets of speckled color and background color corresponding to concentrate. For example, yellow speckled color and brown background correspond to concentrate, while red speckled color and brown background correspond to waste ore. Therefore, by comparing the combination of speckled color and background color in the surface image of speckled ore with the preset color combination, it is possible to quickly and accurately determine whether the speckled ore contains recoverable gold concentrate or not. The speckled-specific classifier can include a semantic segmentation model, which can accurately extract speckled color and background color from the surface image to ensure the accuracy of the final classification result and meet the needs of industrial production.

[0065] According to the ore sorting method provided in this embodiment, by setting a corresponding pattern-specific classifier for patterned ore, it is possible to judge based on the unique characteristics of patterned ore, unaffected by the related characteristics of other types of ore, thus simplifying the classification logic of the solid color-specific classifier. Based on this, by comparing the combination of the base color and pattern color of the concentrate with the combination of the base color and pattern color of the ore to be sorted, the final classification result corresponding to the patterned ore can be obtained simply and effectively, thereby ensuring the accuracy and reliability of the final classification result. In some embodiments, such as... Figure 11 As shown, step S154, which determines the final classification result of the ore to be sorted based on the surface image of the ore belonging to the dot-type ore by using a dot-type dedicated classifier, may include steps S1541 to S1543.

[0066] Step S1541 involves determining the color of the surface associated mineral region in the surface image data using a point-specific classifier. In step S1541, the surface image of the ore to be sorted, corresponding to the point-specific ore, can be input into the point-specific classifier to determine the color of the surface associated mineral region. The color of the surface associated mineral region can be the color with the largest proportion in the surface image that is identified as the surface associated mineral region. The surface image can be in RGB format, allowing it to reflect more color information and thus improving the accuracy of color-based recognition results. The point-specific classifier may include a segmentation model, which can accurately segment the surface associated mineral region from the background region using its segmentation capabilities, thereby accurately obtaining the color of the surface associated mineral region corresponding to the surface associated mineral region.

[0067] Step S1542: If the color of the surface associated mineral area matches the characteristic color of pyrite with a degree of matching that is not less than the preset matching degree, then the final classification result is concentrate.

[0068] Step S1543: If the color matching degree between the surface associated mineral region and the characteristic color of pyrite is less than the preset matching degree, the final classification result is waste ore. For gold ore, its associated ore is usually pyrite, and when pyrite is distributed in a dotted form on the surface of the gold ore, the gold ore can be identified as concentrate. Therefore, by determining the matching degree between the color of the surface associated mineral region and the characteristic color of pyrite, it is possible to determine whether the surface associated mineral is pyrite, and thus determine whether the dotted ore is concentrate. For example, if the matching degree between the color of the surface associated mineral region and the characteristic color of pyrite is greater than 75%, then the dotted ore has a high probability of including pyrite, and is thus identified as an associated ore containing gold, and thus determined as concentrate. Thus, the dot-specific classifier can effectively identify whether dotted ore is a gold associated mineral by matching the color of the surface associated mineral region and the characteristic color of pyrite, thereby accurately determining whether the dotted ore is concentrate or waste ore.

[0069] According to the ore sorting method provided in this embodiment, by setting a corresponding point-specific classifier for point-like ores, it is possible to judge based on the unique characteristics of point-like ores, unaffected by the related characteristics of other types of ores, thus simplifying the classification logic of the point-specific classifier. Based on this, by comparing the color matching degree between the surface associated minerals and pyrite of the ore to be sorted, it is possible to easily and effectively determine whether the surface associated minerals are pyrite, thereby determining whether the ore to be sorted is a gold-containing pyrite-associated ore, thus ensuring the accuracy and reliability of the final classification result. In some embodiments, such as... Figure 12 As shown, the ore sorting method may further include steps S160 to S180.

[0070] Step S160: Based on the primary classification result, secondary classification result, and final classification result, determine the error. The error may include one or more of the following: the difference between the primary classification result and the result actually manually determined as solid-color ore or textured ore; the difference between the secondary classification result and the result actually manually determined as striped ore, mottled ore, or dotted ore; and the difference between the final classification result and the result actually manually determined as concentrate or waste ore. In other words, the difference between the primary classification result, secondary classification result, and final classification result and the actual result determined manually.

[0071] Step S170: Based on the error, determine the error type. The error type includes one or more of the following: first-level classifier error, second-level classifier error, and third-level classifier error. This allows for the effective identification of the specific classifier that needs to be optimized subsequently, thereby efficiently and accurately improving the overall classification accuracy.

[0072] Step S180: Based on the error type, adjust the parameters of the corresponding first-level classifier, second-level classifier, or third-level classifier. Since each classifier is independent, a separate interface can be set for each classifier. This allows for targeted adjustment of only the relevant classifier based on the error type, avoiding the drastic increase in complexity caused by joint adjustment and thus improving adjustment efficiency. Furthermore, the independent interface facilitates the subsequent application of all classifiers to different mining areas. It allows for the construction of corresponding training data based on the actual ore conditions of each mining area, enabling optimization and fine-tuning of some or all classifiers according to actual needs, ensuring that all classifiers achieve good classification accuracy in different mining areas.

[0073] According to the ore sorting method provided in this embodiment, the primary, secondary, and tertiary classifiers of the mining area using the ore sorting method can be specifically optimized based on the primary classification results, secondary classification results, and final classification results. This allows for adaptation to the characteristic differences of ores in different mining areas, thereby improving the accuracy and reliability of identifying the ores to be sorted in the current mining area. Since the primary, secondary, and tertiary classifiers can be configured with independent interfaces, they can be easily upgraded, optimized, or retrained independently, reducing the difficulty of model training in industrial production and improving transferability.

[0074] Based on the same inventive concept, this disclosure also provides a classification system, which may include: a collection device and a classification device.

[0075] An acquisition device is used to acquire surface images of the ore to be sorted. The acquisition device may include a high-resolution camera to acquire high-definition surface images, enabling subsequent sorting devices to extract more effective surface feature data. The acquisition device can be positioned above the transport path of the ore to be sorted, thus ensuring stable acquisition of surface images. The acquisition device may include multiple acquisition sub-devices distributed at different angles to acquire surface images of the ore from multiple different perspectives, forming a three-dimensional surface image of the ore to be sorted, thereby providing richer surface feature data for subsequent sorting devices.

[0076] A classification device is used to determine the final classification result of the ore to be classified according to the ore sorting method provided in any of the foregoing embodiments. The classification device can be communicatively connected to the acquisition device to receive surface images and determine the final classification result according to the ore sorting method. Specifically, the classification device can extract texture features based on the surface image to determine the sub-type of the ore to be classified, and obtain the final classification result of the ore to be classified according to the three-level classifier corresponding to the sub-type.

[0077] According to this embodiment, through the integrated acquisition and classification devices, the classification system can acquire and classify surface images of the ore to be sorted, thereby efficiently and accurately determining the final sorting result for each ore. The classification device, through a primary classifier, a secondary classifier, and a tertiary classifier, first divides all the ore to be sorted into multiple ore groups according to texture features, and then uses the tertiary classifier corresponding to each ore group to determine the accurate final sorting result. Therefore, setting primary and secondary classifiers significantly reduces the classification logic complexity of each tertiary classifier, thereby simplifying the number of parameters in the tertiary classifier while effectively improving its classification accuracy. Furthermore, by using the primary classifier to coarsely distinguish between solid-color and textured ores, and the secondary classifier to finely distinguish textured ores, the classification logic complexity of the primary and secondary classifiers can be significantly reduced, thereby improving processing efficiency and classification accuracy. Therefore, the classification system can efficiently and accurately classify the ore to be sorted in industrial ore mining.

[0078] Based on the same inventive concept, this disclosure also provides a sorting machine, which may include: a conveying device, a classification system as described in any of the foregoing embodiments, and a sorting device.

[0079] A conveying device is used to transport ore to be sorted. The conveying device can be a belt conveyor, the surface of which is used to carry and transport the material, so that the material passes sequentially along a predetermined direction through the identification area corresponding to the classification system and the sorting area corresponding to the sorting device.

[0080] A classification system is used to acquire surface images of the ore to be sorted and determine the corresponding final classification result. The classification system can be positioned above the conveying device to effectively acquire surface images of the ore. The system processes the surface images of the ore through primary and secondary classifiers to accurately identify the type of ore. The sorting system can then select the appropriate tertiary classifier to process the surface image based on the type of ore to obtain the corresponding final classification result.

[0081] A sorting device is used to separate ores according to the final classification result. The sorting device can communicate with the classification system and is used to sort the ores according to the final classification result determined by the classification system. The sorting device may include an air jet assembly, a feeding assembly, or other actuators that apply forces to the ores at different positions or along different trajectories according to the characteristics of different ores, thereby causing ores corresponding to different final classification results to fall into the corresponding collection areas.

[0082] According to this embodiment, by integrating the conveying device, classification system, and sorting device, the sorting machine can achieve complete closed-loop control from surface image acquisition and final classification result determination to physical sorting during the continuous conveying of ore to be sorted. Specifically, the classification system simultaneously acquires surface images during material conveying and decouples texture recognition and classification recognition based on primary, secondary, and tertiary classifiers. This simplifies the individual structural complexity of each classifier, thereby improving the classification accuracy and processing efficiency of the corresponding classifier. As a result, the classification system can more accurately and quickly determine the final classification result of the ore to be sorted and promptly transmit the final classification result to the sorting device. The sorting device then applies corresponding sorting actions to different categories of ore to be sorted, effectively distinguishing the ore on its spatial trajectory. This comprehensively improves the sorting machine's accuracy in classifying ores with complex characteristics and its operational stability, meeting the practical application requirements of high-precision ore sorting.

[0083] This application uses specific terms to describe embodiments of the application. Terms such as "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of the application. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Furthermore, certain features, structures, or characteristics in one or more embodiments of the application can be appropriately combined.

[0084] In the context of this application, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0085] Similarly, it should be noted that, in order to simplify the description of the present application and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of the present application sometimes combines multiple features into a single embodiment, drawing, or description thereof. However, this disclosure method does not imply that the subject matter of the present application requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of the single embodiments disclosed above.

[0086] The basic concepts have been described above. Obviously, for those skilled in the art, the above disclosure is merely illustrative and does not constitute a limitation of this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, and therefore remain within the spirit and scope of the embodiments of this application.

Claims

1. A method for separating ores, characterized in that, The method includes: Acquire a surface image of the ore to be sorted; Based on the surface image, a primary classifier is used to determine the primary classification result of the ore to be sorted, wherein the primary classification result includes solid-color ore or textured ore; In response to the primary classification result being the textured ore, a secondary classification result for the ore to be sorted is determined based on the texture features of the surface image using a secondary classifier. The secondary classifier is independent of the primary classifier, and the secondary classification result includes multiple different types of sub-textured ore, including: striped ore, mottled ore, and dotted ore. Based on the primary classification results and the secondary classification results, one of the tertiary classifiers corresponding to the solid color ore and each of the sub-textured ores is determined as the final classifier. The primary classifier, the secondary classifier, and the multiple tertiary classifiers are independent of each other. The multiple tertiary classifiers include: a solid color exclusive classifier, a stripe exclusive classifier, a mottled exclusive classifier, and a dotted exclusive classifier. Based on the surface image, the final classifier determines the final classification result of the ore to be sorted, wherein the final classification result includes concentrate and waste ore. The step of determining the final classification result of the ore to be sorted based on the surface image and through the final classifier includes: The final classification result of the ore to be sorted is determined based on the surface image belonging to the solid color ore using a solid color-specific classifier. The final classification result of the ore to be sorted is determined based on the surface image belonging to the striped ore using a stripe-specific classifier. Using a mottled-specific classifier, the final classification result of the ore to be sorted is determined based on the surface image belonging to the mottled ore; The final classification result of the ore to be sorted is determined based on the surface image belonging to the dot-shaped ore using a dot-specific classifier.

2. The ore sorting method as described in claim 1, characterized in that, The step of determining the primary classification result of the ore to be sorted based on the surface image using a primary classifier includes: The texture uniformity of the surface image is determined using the first-level classifier. If the texture uniformity is greater than or equal to the texture uniformity threshold, then the primary classification result is the solid-color ore; If the texture uniformity is less than the texture uniformity threshold, then the primary classification result is the textured mineral.

3. The ore sorting method as described in claim 1, characterized in that, The step of determining the final classification result of the ore to be sorted based on the surface image belonging to the solid-color ore using a solid-color dedicated classifier includes: The base color of the surface image is determined using the solid color-specific classifier. If the background color belongs to the preset concentrate color range, then the final classification result is the concentrate; If the base color does not belong to the preset concentrate color range, the final classification result is the waste ore.

4. The ore sorting method as described in claim 1, characterized in that, The step of determining the final classification result of the ore to be sorted based on the surface image belonging to the spotted ore using a spotted-specific classifier includes: The spot color and background color of the surface image data are determined using the spot-specific classifier. If the combination of the base color and the speckled color belongs to a preset color combination, then the final classification result is the concentrate; If the combination of the base color and the mottled color does not belong to the preset color combination, then the final classification result is the waste mine.

5. The ore sorting method as described in claim 1, characterized in that, The step of determining the final classification result of the ore to be sorted based on the surface image belonging to the dot-shaped ore using a dot-specific classifier includes: The color of the surface associated mineral region in the surface image data is determined by the point-specific classifier. If the color of the surface associated mineral area matches the characteristic color of pyrite with a degree of not less than a preset degree of matching, then the final classification result is the concentrate. If the color of the surface associated mineral area matches the characteristic color of pyrite less than a preset matching degree, then the final classification result is the waste ore.

6. The ore sorting method as described in claim 1, characterized in that, The secondary classifier includes a neural network model trained on a training dataset. The training dataset includes multiple training images, each labeled with one of the following: striped ore, mottled ore, or dotted ore. The training images are labeled using the following method: If one or more of the following in the training image—the percentage of striped pixels, the consistency of the direction of each stripe, the contrast between the stripe and the background color, and the distribution density of pyrite—are greater than the corresponding threshold, then the striped ore is used as the label of the training image. If the texture type in the training image is irregular patchy, and one or more of the following parameters are greater than the corresponding threshold: number of patches, percentage of single patch area, contrast between patch and background color, and overlap rate between pyrite distribution area and patch area, then the mottled ore is used as the label of the training image. If the texture type of the training image is discrete dots, the dot distribution type is non-directional, and one or more of the dot density, the contrast between the dot region and the background color, and the spectral matching degree between the dot region and pyrite are greater than the corresponding threshold, then the dot ore is used as the label of the training image.

7. The ore sorting method as described in claim 1, characterized in that, The method further includes: Based on the primary classification result, the secondary classification result, and the final classification result, the error is determined; Based on the error, determine the error type; Based on the error type, adjust the parameters of the corresponding first-level classifier, second-level classifier, or third-level classifier.

8. A classification system, characterized in that, include: Acquisition device, used to acquire surface images of the ore to be sorted; A classification device for determining the final classification result of the ore to be classified according to the ore sorting method as described in any one of claims 1 to 7.

9. A sorting machine, characterized in that, include: Conveying device, used to transport ore to be sorted; The classification system as described in claim 8 is used to acquire a surface image of the ore to be sorted and determine the corresponding final classification result; The sorting device is used to sort the ore to be sorted according to the final classification result.

Images