Low-grade copper ore image sorting method based on deep learning

Abstract

The invention discloses a low-grade copper ore image sorting method based on deep learning. The low-grade copper ore image sorting method comprises the following steps that a dual-energy X-ray image of low-grade copper ore is obtained; constructing a time step embedded U-Net network, and inputting the dual-energy X-ray image of the low-grade copper ore into the time step embedded U-Net network to generate a diversified copper ore dual-energy X-ray image; constructing a self-adaptive copper ore sorting model; the diversified copper ore dual-energy X-ray images are input into a self-adaptive copper ore sorting model, and copper ore is sorted; according to the method, the separation process is optimized by fusing the dual-energy X-ray imaging technology and the data enhancement technology, compared with traditional single-energy imaging, the difference of copper minerals and waste rocks in an image feature space is enhanced, the recognition difficulty of a model on complex textures is reduced, the number of links of crushing and flotation of invalid ores can be reduced, and the separation efficiency is improved. The energy consumption and the use amount of chemical agents are obviously reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of copper ore separation, and specifically provides an image separation method for low-grade copper ore based on deep learning. Background Technique

[0002] As a key mineral resource, copper plays an irreplaceable core role in the modern industrial system. Its excellent electrical and thermal conductivity makes it the core raw material in key fields such as power transmission, new energy equipment manufacturing, and intelligent electronic devices. Especially in the key carbon-neutral fields such as the three electric systems of new energy vehicles, the cable network of photovoltaic power stations, and the conductive components of energy storage devices, the demand for copper materials shows a structural growth trend.

[0003] Traditional copper ore separation technologies have long relied on physical screening and chemical flotation methods, and their limitations run through the entire production chain: on the one hand, the energy consumption in the ore crushing and grinding links is too high, resulting in resource waste and a sharp increase in costs; on the other hand, the flotation process relies on a large number of chemical reagents (such as collectors and inhibitors), which not only increases the pressure of tailings treatment, but the residual heavy metal ions are more likely to pollute the soil and groundwater systems through percolation. At the technical level, traditional methods are difficult to effectively separate fine-grained disseminated structures (such as the symbiotic body of copper minerals and gangue) or minerals with similar physical and chemical properties (such as the similar density and surface activity of chalcopyrite and pyrite), resulting in large fluctuations in concentrate grade and low recovery rates. In addition, from equipment procurement to reagent supply, from waste disposal to process parameter regulation, the entire process requires high capital investment and professional manual intervention, and the flammable characteristics of some chemical reagents and the storage conditions of minerals are prone to get out of control. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides an image separation method for low-grade copper ore based on deep learning, aiming to solve the problems in the background technique.

[0005] To achieve the above object, the present invention provides the following technical solution: An image separation method for low-grade copper ore based on deep learning, including the following steps: Step S1: Obtain the dual-energy X-ray images of low-grade copper ore, where the dual-energy X-ray images of low-grade copper ore include copper-containing dual-energy X-ray images and waste rock dual-energy X-ray images; the dual-energy X-ray images of low-grade copper ore are the high-energy X-ray images and low-energy X-ray images of low-grade copper ore; Step S2: Embed a learnable time step embedding module in the U-Net model, and introduce cross-layer residual connections and self-attention mechanisms to construct a time step embedding U-Net network, and input the dual-energy X-ray images of low-grade copper ore into the time step embedding U-Net network to generate diverse dual-energy X-ray images of copper ore; Step S3: Construct an adaptive copper ore sorting model based on the VGG11 structure; Step S4: Input the diverse dual-energy X-ray images of copper ore into the adaptive copper ore sorting model to sort the copper ore.

[0006] Further, the time-step embedding U-Net network consists of a time-step embedding module, a convolutional module, a residual block, a self-attention module, a transposed convolution, and a convolutional layer; the processing flow of the time-step embedding U-Net network is as follows: First, encode the input, i.e., the dual-energy X-ray image of low-grade copper ore, into an initial noise image and a time-step scalar, input the time-step scalar into the time-step embedding module for dimensionality increase processing, input the initial noise image into the convolutional module, broadcast and add the output of the convolutional module and the output of the time-step embedding module to obtain a first concatenated feature, downsample the first concatenated feature and broadcast and add it to the output of the time-step embedding module to obtain a second concatenated feature, downsample the second concatenated feature and pass it through the residual block and the self-attention module together with the output of the time-step embedding module, concatenate the output of the self-attention module, the second concatenated feature, and the output of the time-step embedding module to obtain a third concatenated feature, input the third concatenated feature and the first concatenated feature into the transposed convolution together, concatenate the output of the transposed convolution and the output of the time-step embedding module to obtain a fourth concatenated feature, upsample the fourth concatenated feature and concatenate it with the output of the time-step embedding module to obtain a fifth concatenated feature, and map the fifth concatenated feature to the target dimension through the convolutional layer to obtain the final output of the time-step embedding U-Net network, i.e., the diverse dual-energy X-ray images of copper ore; The time-step embedding module uses sinusoidal positional encoding.

[0007] Further, the adaptive copper ore sorting model based on the VGG11 structure consists of a backbone network Stem, a first stage Stage1, a second stage Stage2, a main classifier, a feature projection module, and a domain discriminator; The processing flow of the adaptive copper ore sorting model based on the VGG11 structure is as follows: Input the diverse dual-energy X-ray images of copper ore into the backbone network, the first stage, and the second stage in sequence to obtain a second enhanced feature map, input the second enhanced feature map into the main classifier, the feature projection module, and the domain discriminator for processing respectively. The main classifier uses the output of the feature projection module to perform a classification task and predict the probability that the enhanced feature map belongs to each category. The domain discriminator uses the output of the feature projection module to perform a domain adaptation task and predict the probability that the enhanced feature map belongs to the source domain or the target domain. Finally, the output of the adaptive copper ore sorting model based on the VGG11 structure is the probability distribution that the enhanced feature map belongs to each category and the probability distribution that the enhanced feature map belongs to the source domain or the target domain.

[0008] Furthermore, the first stage and the second stage have the same structure, both consisting of a multi-scale context attention module, a dynamic feature compensation unit, and an enhanced feature convolution unit; the processing flow of the first stage is as follows: the input, which is the output of the backbone network, is successively passed through the multi-scale context attention module and the dynamic feature compensation unit, and the output of the dynamic feature compensation unit and the output of the multi-scale context attention module are jointly input into the enhanced feature convolution unit for processing to obtain the first enhanced feature map, which is the final output of the first stage. Among them, the enhanced feature convolution unit consists of a convolution layer, a batch normalization layer, and a ReLu activation function connected in sequence.

[0009] Furthermore, the dynamic feature compensation unit consists of a deformable convolution, a dense block, a channel attention module, a global average pooling module, a feature transformation unit, and a Sigmoid function; the processing flow of the dynamic feature compensation unit is as follows: the input, which is the output of the multi-scale context attention module, is successively passed through the deformable convolution and the dense block to obtain an adaptive feature, the input is successively passed through the channel attention module, the global average pooling module, the feature transformation unit, and the Sigmoid function to obtain a transformed feature, the transformed feature and the adaptive feature are concatenated and then channel-fused with the input of the dynamic feature compensation unit to obtain a fused feature, which is the final output of the dynamic feature compensation unit.

[0010] Furthermore, the main classifier consists of an adaptive average pooling layer, a flattening layer, and a fully connected layer connected in sequence. The domain discriminator consists of a gradient reversal layer, a global average pooling layer, a first dimensionality reduction feature transformation module, a first dropout layer, a second dimensionality reduction feature transformation module, a second dropout layer, and a fully connected layer connected in sequence. The first dimensionality reduction feature transformation module and the second dimensionality reduction feature transformation module have the same structure, both consisting of a fully connected layer, a batch normalization layer, and a ReLu activation function connected in sequence.

[0011] Furthermore, an intra-class - inter-class loss function is added to enhance the constraint, forcing the Euclidean distance between enhanced feature maps of the same class to be less than a threshold and forcing the Euclidean distance between enhanced feature maps of different classes to be greater than a threshold ; the intra-class - inter-class loss function is expressed as: ; In the formula, represents the intra-class - inter-class loss function; , are the i-th enhanced feature map and the j-th enhanced feature map respectively, , belong to enhanced feature maps of the same class; is the k-th enhanced feature map, belonging to an enhanced feature map of a different class; Represents the weight parameter.

[0012] An electronic device comprises a processor, a memory and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute a low-grade copper ore image sorting method based on deep learning.

[0013] A non-volatile computer storage medium stores computer executable instructions, which can execute a low-grade copper ore image sorting method based on deep learning.

[0014] Compared with the existing technology, the present invention has the following beneficial effects:

[0015] (1) The present invention optimizes the sorting process by integrating dual-energy X-ray imaging technology and data enhancement technology. Compared with traditional single-energy imaging, dual-energy X-ray sensors can simultaneously capture the density and composition characteristics of ore, enhance the difference between copper minerals and waste rock in the image feature space, and thus reduce the difficulty of the model to recognize complex textures; in view of the scarcity of low-grade ore samples and the imbalance of positive and negative sample distribution, the time-step embedded U-Net model is used to reconstruct the multi-scale features of the original ore image, generate synthetic samples consistent with the real mineral structure, and reduce the risk of overfitting of the model while expanding data diversity; the present invention reduces the dependence on manual experience debugging in traditional processes by constructing an automated closed loop of ore image recognition and sorting decision-making, and dynamically optimizes sorting parameters based on real-time feedback. In addition, the present invention can reduce the entry of invalid ore into the crushing and flotation links, significantly reduce energy consumption and the use of chemical agents, and reduce environmental pollution pressure from the source of the process.

[0016] (2) In order to solve the problem of decision bias in image sorting caused by the uneven distribution of copper ore and waste rock in low-grade copper ore samples, the present invention constructs a time-step embedded U-Net model based on the conditional diffusion process. The sample data generated by the time-step embedded U-Net model can effectively correct the decision bias of classification, reconstruct the discriminant feature space dominated by negative samples into a balanced representation space, and improve the model sorting accuracy and recall rate.

[0017] (3) In order to solve the problem that the features in low-grade copper ore images are sparsely distributed and difficult to be captured by conventional convolution models, the present invention constructs a multi-scale contextual attention module and a dynamic feature compensation unit, improves the fixed receptive field limitation of traditional convolution, and enhances the model's ability to model the association of global ore features. At the same time, the convolution sampling area is adjusted through deformable convolution to adapt to the tiny features in the ore image, and residual connection is used to strengthen the early feature expression, effectively improving the classification credibility of the model.

[0018] (4) In view of the problem that in the waste rock image, the presence of other metal impurities and the sudden change in thickness form confusing features, resulting in incorrect model recognition, the present invention uses a gradient reversal layer and an intra-class - inter-class contrast loss to achieve the stripping of confusing features, and by dynamically adjusting the adversarial coefficient, the interval between the two types of features is expanded, greatly improving the model accuracy and the overall sorting ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the method of the present invention.

[0020] Figure 2 It is a structural diagram of the time step embedding U-Net network of the present invention.

[0021] Figure 3 It is a structural diagram of the adaptive copper ore sorting model based on the VGG11 structure of the present invention.

[0022] Figure 4 It is a structural diagram of the dynamic feature compensation unit of the present invention.

[0023] Figure 5 It is a structural diagram of the domain discriminator of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0024] As Figure 1 shown, the present invention provides a technical solution: a method for sorting low-grade copper ore images based on deep learning, including the following steps:

[0025] Step S1: Obtain the dual-energy X-ray images of low-grade copper ore, where the dual-energy X-ray images of low-grade copper ore include the dual-energy X-ray images of copper-bearing and waste rock; the dual-energy X-ray images of low-grade copper ore are the high-energy X-ray images and low-energy X-ray images of low-grade copper ore.

[0026] Step S2: Embed a learnable time step embedding module in the U-Net model, and introduce cross-layer residual connections and self-attention mechanisms to construct a time step embedding U-Net network, and input the dual-energy X-ray images of low-grade copper ore into the time step embedding U-Net network to generate diverse dual-energy X-ray images of copper ore (enhanced images after denoising).

[0027] Step S3: Construct an adaptive copper ore sorting model based on the VGG11 structure. This model realizes multi-level structure evolution and dynamic parameter adaptation mechanism in the deep feature learning framework.

[0028] Step S4: Input the diverse dual-energy X-ray images of copper ore into the adaptive copper ore sorting model to sort the copper ore.

[0029] Among them, the time step embedding U-Net network is as Figure 2As shown, this model is based on the U-Net symmetric encoding-decoding architecture and is deeply modified for the time dependence of the diffusion model.

[0030] The time-step embedding U-Net network consists of a time-step embedding module, a convolutional module, a residual block, a self-attention module, a transposed convolution, and a convolutional layer. The processing flow of the time-step embedding U-Net network is as follows: First, the input (dual-energy X-ray image of low-grade copper ore) is encoded into an initial noise image and a time-step scalar t. The time-step scalar t is input into the time-step embedding module for dimensionality increase processing. The initial noise image is input into the convolutional module. The output of the convolutional module and the output of the time-step embedding module are broadcast-added to obtain the first concatenated feature. The first concatenated feature is downsampled and broadcast-added to the output of the time-step embedding module to obtain the second concatenated feature. The second concatenated feature is downsampled and passed through the residual block and the self-attention module together with the output of the time-step embedding module. The output of the self-attention module, the second concatenated feature, and the output of the time-step embedding module are concatenated to obtain the third concatenated feature. The third concatenated feature and the first concatenated feature are input into the transposed convolution together. The output of the transposed convolution and the output of the time-step embedding module are concatenated to obtain the fourth concatenated feature. The fourth concatenated feature is upsampled and concatenated with the output of the time-step embedding module to obtain the fifth concatenated feature. The fifth concatenated feature is mapped to the target dimension through a 1×1 convolutional layer to obtain the final output of the time-step embedding U-Net network, that is, the diverse dual-energy X-ray image of copper ore. The stability of mini-batch training is enhanced through group normalization (GroupNorm) and the SiLU activation function throughout the process. At the same time, the time-step embedding is used to guide the network to learn the temporal evolution law of the noise distribution during the diffusion process, forming the feature generation ability that can retain spatial details and adapt to time dynamics.

[0031] Among them, the time-step embedding module adopts sinusoidal positional encoding. Sinusoidal positional encoding embeds the time-step information into the model by using sine and cosine functions, enabling the model to understand the order relationship of elements in the sequence.

[0032] As Figure 3 shown, the adaptive copper ore sorting model based on the VGG11 structure consists of a backbone network (Stem), a first stage (Stage1), a second stage (Stage2), a main classifier, a feature projection module, and a domain discriminator.

[0033] The processing flow of the adaptive copper ore sorting model based on the VGG11 structure is as follows: The input (diverse dual-energy X-ray images of copper ore) passes through the backbone network (Stem), the first stage (Stage1), and the second stage (Stage2) in sequence to obtain the second enhanced feature map. The second enhanced feature map is respectively input into the main classifier, the feature projection module, and the domain discriminator for processing. The main classifier uses the output of the feature projection module to perform the classification task and predict the probability that the enhanced feature map belongs to each category. The domain discriminator uses the output of the feature projection module to perform the domain adaptation task and predict the probability that the enhanced feature map belongs to the source domain or the target domain. Finally, the output of the adaptive copper ore sorting model based on the VGG11 structure is the probability distribution that the enhanced feature map belongs to each category and the probability distribution that the enhanced feature map belongs to the source domain or the target domain.

[0034] Among them, the backbone network (Stem) consists of a two-dimensional convolutional layer (Conv2d), a batch normalization layer (BN), a ReLU activation function, and a two-dimensional max pooling layer (MaxPool2d) connected in sequence.

[0035] Among them, the first stage (Stage1) and the second stage (Stage2) have the same structure and are both composed of a multi-scale contextual attention module, a dynamic feature compensation unit, and an enhanced feature convolutional unit. The processing flow of the first stage (Stage1) is as follows: The input (the output of the backbone network) passes through the multi-scale contextual attention module and the dynamic feature compensation unit in sequence, and the output of the dynamic feature compensation unit and the output of the multi-scale contextual attention module are input into the enhanced feature convolutional unit for processing together to obtain the first enhanced feature map, which is the final output of the first stage (Stage1).

[0036] The second stage (Stage2) has the same structure as the first stage (Stage1), and its processing flow will not be elaborated here.

[0037] Among them, the enhanced feature convolutional unit consists of a convolutional layer, a batch normalization layer, and a ReLu activation function connected in sequence.

[0038] Such as Figure 4As shown, the dynamic feature compensation unit consists of a deformable convolution plus a residual structure, including a Deformable Convolution Module, a DenseBlock, a channel attention module, a global average pooling module, a feature transformation unit, and a Sigmoid function; the processing flow of the dynamic feature compensation unit is as follows: the input (the output of the multi-scale context attention module) passes through the deformable convolution and the DenseBlock in sequence to obtain adaptive features, the input passes through the channel attention module, the global average pooling module, the feature transformation unit, and the Sigmoid function in sequence to obtain transformed features, the transformed features and the adaptive features are concatenated and then subjected to channel fusion with the input of the dynamic feature compensation unit to obtain fused features, which are the final output of the dynamic feature compensation unit.

[0039] Among them, the DenseBlock consists of two fully connected layers (FC, Fully Connected Layer) and a ReLU activation function, and the ReLU activation function is between the two fully connected layers.

[0040] Among them, the main classifier consists of an AdaptiveAvgPool2d layer, a Flatten layer, and a fully connected layer connected in sequence; the loss function of the main classifier: (1); In the formula, represents the loss function of the main classifier, and represent the parameters of the second stage (Stage2) and the main classifier respectively; represents the input sampled from the data distribution and the expected value of the label , represents 's distribution, represents 's distribution; represents the cross-entropy loss, represents given and and , the probability that the main classifier predicts the label .

[0041] Such as Figure 5As shown in the figure, the domain discriminator consists of a Gradient Reversal Layer (GRL), a Global Average Pooling layer, a first Dimensionality Reduction Feature Transformation Module (DRFTModule), a first Dropout layer, a second Dimensionality Reduction Feature Transformation Module, a second Dropout layer, and a fully connected layer connected in sequence; the loss function of the domain discriminator is: (2); In the formula, represents the loss function of the domain discriminator, represents the parameters of the domain discriminator; represents the (true data distribution) expected value of; represents the expected value of the generated data distribution in ; represents the predicted generated label; represents the generator, which is used to receive and and generate data.

[0042] Adding an intra-class - inter-class loss function to enhance the constraint, for the intra-class enhanced feature maps, forcing their Euclidean distance to be less than the threshold and for the inter-class enhanced feature maps, forcing their Euclidean distance to be greater than the threshold ; the intra-class - inter-class loss function is expressed as: (3); In the formula, represents the intra-class - inter-class loss function; , are respectively the i-th enhanced feature map and the j-th enhanced feature map, , belong to the intra-class enhanced feature maps; is the k-th enhanced feature map, belonging to the inter-class enhanced feature map; represents the weight parameter.

[0043] This constraint enhances the discriminative power of the model for boundary samples by increasing the inter-class feature hypersphere interval and improving the intra-class feature compactness.

[0044] Among them, the first Dimensionality Reduction Feature Transformation Module (DRFT Module) and the second Dimensionality Reduction Feature Transformation Module have the same structure, and both consist of a fully connected layer, a batch normalization layer, and a ReLu activation function connected in sequence.

[0045] During forward propagation, the gradient reversal layer keeps the features unchanged, but during backpropagation, the gradient reversal layer negates the gradient and multiplies it by a dynamic coefficient , to achieve gradient reversal; this enables the second stage (Stage2) to learn to generate feature maps that the domain discriminator cannot correctly classify, thereby achieving domain invariance.

[0046] Among them, the dynamic coefficient varies dynamically with the number of iterations and is expressed as: (4); In the formula, represents the exponential function; represents the number of iterations.

[0047] An electronic device includes a processor, a memory, and a bus. The processor and the memory are connected through the bus. Among them, the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute a method for sorting low-grade copper ore images based on deep learning.

[0048] A non-volatile computer storage medium stores computer-executable instructions, and the computer-executable instructions execute a method for sorting low-grade copper ore images based on deep learning.

[0049] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for sorting low-grade copper ore images based on deep learning, characterized in that, It includes the following steps: Step S1: Obtain the dual-energy X-ray images of low-grade copper ore. The dual-energy X-ray images of low-grade copper ore include the copper-containing dual-energy X-ray images and the waste rock dual-energy X-ray images; the dual-energy X-ray images of low-grade copper ore are the high-energy X-ray image and the low-energy X-ray image of low-grade copper ore; Step S2: Embed a learnable time step embedding module in the U-Net model, and introduce cross-layer residual connections and self-attention mechanisms to construct a time step embedding U-Net network. Input the dual-energy X-ray images of low-grade copper ore into the time step embedding U-Net network to generate diverse dual-energy X-ray images of copper ore; Step S3: Construct an adaptive copper ore sorting model based on the VGG11 structure; Step S4: Input the diverse dual-energy X-ray images of copper ore into the adaptive copper ore sorting model to sort the copper ore.

2. The method for sorting low-grade copper ore images based on deep learning according to claim 1, wherein: The time step embedding U-Net network is composed of a time step embedding module, a convolution module, a residual block, a self-attention module, a transposed convolution, and a convolutional layer; the processing flow of the time step embedding U-Net network is as follows: First, encode the input, that is, the dual-energy X-ray images of low-grade copper ore, into an initial noise image and a time step scalar. Input the time step scalar into the time step embedding module for dimensionality increase processing. Input the initial noise image into the convolution module. Broadcast and add the output of the convolution module and the output of the time step embedding module to obtain a first concatenated feature. Downsample the first concatenated feature and broadcast and add it to the output of the time step embedding module to obtain a second concatenated feature. Downsample the second concatenated feature and pass it through the residual block and the self-attention module together with the output of the time step embedding module. Concatenate the output of the self-attention module, the second concatenated feature, and the output of the time step embedding module to obtain a third concatenated feature. Input the third concatenated feature and the first concatenated feature into the transposed convolution together. Concatenate the output of the transposed convolution and the output of the time step embedding module to obtain a fourth concatenated feature. Upsample the fourth concatenated feature and concatenate it with the output of the time step embedding module to obtain a fifth concatenated feature. Map the fifth concatenated feature to the target dimension through the convolutional layer to obtain the final output of the time step embedding U-Net network, that is, the diverse dual-energy X-ray images of copper ore; The time step embedding module uses sine position encoding.

3. A method for sorting low-grade copper ore images based on deep learning according to claim 2, characterized in that: The adaptive copper ore sorting model based on the VGG11 structure is composed of a backbone network Stem, a first stage Stage1, a second stage Stage2, a main classifier, a feature projection module, and a domain discriminator; The processing flow of the adaptive copper ore sorting model based on the VGG11 structure is as follows: The input, i.e., diverse dual-energy X-ray images of copper ore, is successively passed through the backbone network, the first stage, and the second stage to obtain the second enhanced feature map. The second enhanced feature map is respectively input into the main classifier, the feature projection module, and the domain discriminator for processing. The main classifier uses the output of the feature projection module to perform the classification task and predicts the probabilities that the enhanced feature map belongs to each category. The domain discriminator uses the output of the feature projection module to perform the domain adaptation task and predicts the probabilities that the enhanced feature map belongs to the source domain or the target domain. Finally, the output of the adaptive copper ore sorting model based on the VGG11 structure is the probability distribution that the enhanced feature map belongs to each category and the probability distribution that the enhanced feature map belongs to the source domain or the target domain.

4. The method for sorting low-grade copper ore images based on deep learning according to claim 3, characterized in that: The structures of the first stage and the second stage are the same, and both are composed of a multi-scale context attention module, a dynamic feature compensation unit, and an enhanced feature convolution unit. The processing flow of the first stage is as follows: The input, i.e., the output of the backbone network, is successively passed through the multi-scale context attention module and the dynamic feature compensation unit. The output of the dynamic feature compensation unit and the output of the multi-scale context attention module are jointly input into the enhanced feature convolution unit for processing to obtain the first enhanced feature map, which is the final output of the first stage. Among them, the enhanced feature convolution unit is composed of a convolution layer, a batch normalization layer, and a ReLu activation function connected in sequence.

5. A method for sorting low-grade copper ore images based on deep learning according to claim 4, characterized in that: The dynamic feature compensation unit is composed of a deformable convolution, a dense block, a channel attention module, a global average pooling module, a feature transformation unit, and a Sigmoid function. The processing flow of the dynamic feature compensation unit is as follows: The input, i.e., the output of the multi-scale context attention module, is successively passed through the deformable convolution and the dense block to obtain the adaptive feature. The input is successively passed through the channel attention module, the global average pooling module, the feature transformation unit, and the Sigmoid function to obtain the transformed feature. The transformed feature and the adaptive feature are concatenated and then subjected to channel fusion with the input of the dynamic feature compensation unit to obtain the fused feature, which is the final output of the dynamic feature compensation unit.

6. The method for sorting low-grade copper ore images based on deep learning according to claim 5, characterized in that: The main classifier is composed of an adaptive average pooling layer, a flattening layer, and a fully connected layer connected in sequence. The domain discriminator is composed of a gradient reversal layer, a global average pooling layer, a first dimensionality reduction feature transformation module, a first dropout layer, a second dimensionality reduction feature transformation module, a second dropout layer, and a fully connected layer connected in sequence. The first dimensionality reduction feature transformation module and the second dimensionality reduction feature transformation module have the same structure, and both are composed of a fully connected layer, a batch normalization layer, and a ReLu activation function connected in sequence.

7. A method for sorting low-grade copper ore images based on deep learning according to claim 6, characterized in that: Add intra-class and inter-class loss functions to enhance constraints, forcing the Euclidean distance of enhanced feature maps of the same class to be less than a threshold , and forcing the Euclidean distance of enhanced feature maps of different classes to be greater than a threshold ; The intra-class and inter-class loss function is expressed as: ； Wherein, represents the intra-class and inter-class loss function; and are the i-th enhanced feature map and the j-th enhanced feature map respectively, and belong to the same class of enhanced feature maps; is the k-th enhanced feature map, belonging to a different class of enhanced feature maps; represents the weight parameter.

8. An electronic device, characterized in that, It includes a processor, a memory, and a bus. The processor and the memory are connected through the bus. Among them, the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute a method for sorting low-grade copper ore images based on deep learning according to any one of claims 1-7.

9. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions execute a method for sorting low-grade copper ore images based on deep learning according to any one of claims 1-7.

Images