Copper ore / waste ore classification method and system based on multi-energy-spectrum image

Abstract

The invention discloses a copper ore / waste ore classification method and system based on a multi-energy-spectrum image. The method comprises the steps of obtaining multi-energy-spectrum image data information of existing copper ore and waste ore and preprocessing the multi-energy-spectrum image data information to obtain a training data set; based on the convolution, the residual error and the attention scheme, constructing a copper ore / waste ore classification primary model based on the multi-energy-spectrum image, and training to obtain a copper ore / waste ore classification model based on the multi-energy-spectrum image; and adopting the obtained copper ore / waste ore classification model based on the multi-energy spectrum image to carry out actual copper ore / waste ore classification identification. According to the method, a training data set is constructed for the multi-energy-spectrum image of the copper ore / waste ore, a classification model including a convolution scheme, a residual scheme and an attention scheme is designed, and the trained classification model is adopted to perform classification of the copper ore / waste ore based on the multi-energy-spectrum image; therefore, the copper ore / waste ore can be classified, the reliability is high, and the accuracy is good.

Description

Technical Field

[0001] The invention belongs to the field of ore sorting, and in particular relates to a copper ore / waste ore classification method and system based on multi-energy spectrum images. Background Art

[0002] Copper ore is a key material for global industrial and electrical infrastructure and is vital to the development of modern economy and technology. my country's copper production is relatively low, and there is a large amount of residual copper in the huge amount of waste ore produced during the mining process of copper mines that has not been further utilized. Therefore, further distinguishing between copper ore and waste ore can save more copper resources.

[0003] At present, in the field of ore classification, the research on non-ferrous metals such as lead-zinc ore and tungsten ore has achieved remarkable results, and there are relatively mature sorting processes. However, there is no precedent in the field of photoelectric sorting of copper ore in mines. In the existing photoelectric sorting technology, general visible light, short-wave infrared, near-infrared and other methods can only obtain information on the surface of copper ore; however, in the task of classifying copper ore and waste ore, it is impossible to distinguish them based on the appearance of the ore alone. Although X-ray imaging can penetrate the surface of the ore and obtain its internal structure information, due to the characteristics of copper ore such as low grade, fine embedded particle size, complex mineral composition, and complex symbiotic relationship, the information obtained by a single energy spectrum image is not enough to effectively classify copper ore and waste ore. Summary of the invention

[0004] One of the purposes of the present invention is to provide a copper ore / waste ore classification method based on multi-energy spectrum images with high reliability and good accuracy.

[0005] A second object of the present invention is to provide a system for implementing the copper ore / waste ore classification method based on multi-energy spectrum images.

[0006] The copper ore / waste ore classification method based on multi-energy spectrum image provided by the present invention comprises the following steps: S1. Obtaining multi-spectral image data information of existing copper ore and waste ore; S2. Preprocessing the multi-spectral image data information obtained in step S1 to construct a training data set; S3. Based on the convolution scheme, residual scheme and attention scheme, a primary model for copper ore / waste ore classification based on multi-spectral images was constructed; The constructed primary model for copper ore / waste ore classification based on multi-energy spectrum images includes an image processing layer, a residual connection layer, an attention layer and a classification layer connected in series in sequence; the image processing layer is used to extract the preliminary features of the input ore image; the residual connection layer is used to extract the deep features of the input ore image and alleviate the problem of gradient disappearance; the attention layer enhances the influence of key information in the input ore image in a feature weighted manner; the classification layer is used to classify and identify the feature data output by the attention layer to achieve the final classification of copper ore / waste ore; S4. Using the training data set constructed in step S2, the primary model for copper ore / waste ore classification based on multi-spectral images constructed in step S3 is trained to obtain a copper ore / waste ore classification model based on multi-spectral images; S5. Using the copper ore / waste ore classification model based on the multi-energy spectrum image obtained in step S4, actual copper ore / waste ore classification and identification is performed.

[0007] The step S2 specifically includes the following steps: Performing data preprocessing on the multi-energy spectrum image data information obtained in step S1; The preprocessing includes image denoising, image enhancement, image segmentation and data enhancement; wherein, the image denoising includes image denoising using a Gaussian filtering method; the image enhancement includes image enhancement using a contrast enhancement technique; the image segmentation includes segmenting and acquiring all ore areas in the image, and uniformly adjusting the acquired images to a set size; the data enhancement is used to improve the generalization performance of the model.

[0008] The step S3 comprises the following steps: An image processing layer is constructed based on a convolution scheme to perform preliminary convolution and downsampling processing on the input multi-energy spectrum image data; A residual connection layer is constructed based on the residual scheme to perform residual processing on the feature data output by the image processing layer; An attention layer is constructed based on the channel attention scheme and the spatial attention scheme to perform channel attention and spatial attention processing on the feature data output by the residual connection layer; A classification layer is constructed based on the fully connected scheme to classify and identify the feature data output by the attention layer.

[0009] The image processing layer specifically includes the following contents: The image processing layer includes a concatenated convolution kernel, a batch normalization layer, and an activation function layer; the convolution kernel is The convolution kernel with an output channel number of 32; the activation function layer is the ReLU activation function layer.

[0010] The residual connection layer specifically includes the following contents: The residual connection layer includes a first residual layer and a second residual layer connected in series; the number of output channels of the first residual layer is 64, and the number of output channels of the second residual layer is 128; In the first residual layer, the input of the first residual layer is After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the first residual sub-feature of the first layer; the first residual sub-feature of the first layer is obtained by After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the second residual sub-feature of the first layer; the input of the first residual layer is processed through the fully connected layer to obtain the first layer input sub-feature, and the fully connected layer is used to ensure that the shape of the first input sub-feature is the same as the second residual sub-feature of the first layer; the second residual sub-feature of the first layer and the input sub-feature of the first layer are superimposed as the output of the first residual layer; In the second residual layer, the input of the second residual layer is passed through After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the first residual sub-feature of the second layer; the first residual sub-feature of the second layer is obtained by After the convolution kernel extracts the features, it is processed through a batch normalization layer and a ReLU activation function layer to obtain the second residual sub-features of the second layer; the input of the second residual layer is processed through a fully connected layer to obtain the second layer input sub-features. The fully connected layer is used to ensure that the shape of the second input sub-features is the same as that of the second residual sub-features of the second layer; the second residual sub-features of the second layer and the second layer input sub-features are superimposed as the output of the second residual layer.

[0011] The attention layer specifically includes the following contents: The attention layer includes a channel attention layer, a spatial attention layer and a preliminary classification layer connected in series in sequence; the input of the channel attention layer is processed by the channel attention layer to obtain the channel attention feature, and the channel attention feature is multiplied by the input of the channel attention layer as the input of the spatial attention layer; the input of the spatial attention layer is processed by the spatial attention layer to obtain the spatial attention feature, and the spatial attention feature is multiplied by the input of the spatial attention layer as the input of the preliminary classification; the preliminary classification layer is used to perform preliminary classification on the feature data input by the attention layer; The channel attention layer includes a global layer and a maximum layer; the input of the channel attention layer is divided into two paths, one path is processed by the global layer to obtain global features, and the other path is processed by the maximum layer to obtain maximum features, and finally the global features and the maximum features are added to obtain the channel attention features; wherein the global layer includes a global average pooling layer, a global first fully connected layer, a global ReLU activation function layer, a global second fully connected layer and a global Sigmiod activation function layer; the input of the global layer is processed by the global average pooling layer, and then processed by the global first fully connected layer to reduce the number of channels, and then processed by the global ReLU activation function layer for channel feature weighting, and then processed by the global second fully connected layer. The second fully connected layer is processed to restore the number of channels, and finally the channel features are weighted through the global Sigmiod activation function layer to obtain the global features; the maximum layer includes the maximum pooling layer, the maximum first fully connected layer, the maximum ReLU activation function layer, the maximum second fully connected layer and the maximum Sigmiod activation function layer; the input of the maximum layer is processed by the maximum pooling layer, and then processed by the maximum first fully connected layer to reduce the number of channels, and then processed by the maximum ReLU activation function layer for channel feature weighting, and then processed by the maximum second fully connected layer to restore the number of channels, and finally the channel features are weighted through the maximum Sigmiod activation function layer to obtain the maximum feature; In the spatial attention layer, the input of the spatial attention layer calculates the mean and maximum value along the channel dimension, and then concatenates the mean and maximum value and passes Convolution generates a spatial weight map, which is then processed by the Sigmoid activation function to weight the spatial features to obtain the spatial attention features. The preliminary classification layer includes an average pooling layer and a fully connected layer connected in series; the input of the preliminary classification layer is processed by the average pooling layer and the fully connected layer to achieve preliminary classification of the feature data input by the attention layer.

[0012] The classification layer specifically includes the following contents: In the classification layer, the preliminary classification results of different energy spectra output by the attention layer are concatenated and then processed through the fully connected layer for classification and recognition to obtain the final classification result.

[0013] The present invention also provides a system for implementing the copper ore / waste ore classification method based on multi-energy spectral images, comprising a data acquisition module, a data processing module, a model construction module, a model training module and an ore classification module; the data acquisition module, the data processing module, the model construction module, the model training module and the ore classification module are connected in series in sequence; the data acquisition module is used to acquire the existing multi-energy spectral image data information of copper ore and waste ore, and upload the data information to the data processing module; the data processing module is used to perform data preprocessing on the acquired multi-energy spectral image data information according to the received data information to construct a training data set, and upload the data information to the model construction module; the model construction module is used to construct a primary model for copper ore / waste ore classification based on multi-energy spectral images based on the received data information, based on a convolution scheme, a residual scheme and an attention scheme, and upload the data information to the model training module; wherein the constructed copper ore / waste ore classification based on multi-energy spectral images is The primary model includes an image processing layer, a residual connection layer, an attention layer and a classification layer connected in series in sequence; the image processing layer is used to extract the preliminary features of the input ore image; the residual connection layer is used to extract the deep-level features of the input ore image and alleviate the problem of gradient disappearance; the attention layer enhances the influence of key information in the input ore image in a feature-weighted manner; the classification layer is used to classify and identify the feature data output by the attention layer to achieve the final classification of copper ore / waste ore; the model training module is used to train the constructed copper ore / waste ore classification primary model based on multi-energy spectral images according to the received data information using the constructed training data set to obtain the copper ore / waste ore classification model based on the multi-energy spectral image, and upload the data information to the ore classification module; the ore classification module is used to perform actual classification and identification of copper ore / waste ore according to the received data information using the obtained copper ore / waste ore classification model based on the multi-energy spectral image.

[0014] The copper ore / waste ore classification method and system based on multi-energy spectral images provided by the present invention construct a training data set for the multi-energy spectral images of copper ore / waste ore, and design a classification model including a convolution scheme, a residual scheme and an attention scheme, and use the trained classification model to classify the copper ore / waste ore based on the multi-energy spectral images; therefore, the present invention can not only realize the classification of copper ore / waste ore based on ore multi-energy spectral images, but also has high reliability and good accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The figure is a schematic diagram of the method flow of the present invention.

[0016] Figure 2 Schematic diagram of the functional modules of the system of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 The method flow diagram of the method of the present invention is shown as follows: The copper ore / waste ore classification method based on multi-energy spectrum images disclosed in the present invention comprises the following steps: S1. Obtain the multi-energy spectrum image data information of the existing copper ore and waste ore; in the specific implementation, it is necessary to simultaneously obtain high-energy spectrum X-ray images and low-energy spectrum X-ray images; through the penetration difference of the multi-energy spectrum X-ray images on the tiny copper fibers in the copper ore, more comprehensive internal information of the copper ore can be obtained, thereby providing technical support for improving the classification accuracy.

[0018] S2. Preprocessing the multi-spectral image data information obtained in step S1 to construct a training data set; specifically comprising the following steps: Performing data preprocessing on the multi-energy spectrum image data information obtained in step S1; The preprocessing includes image denoising, image enhancement, image segmentation and data enhancement; wherein the image denoising includes image denoising by using Gaussian filtering method; the image enhancement includes image enhancement by using contrast enhancement technology; the image segmentation includes segmenting and acquiring all ore regions in the image, and uniformly adjusting the acquired images to a set size (preferably ); the data enhancement is used to improve the generalization performance of the model.

[0019] S3. Based on the convolution scheme, residual scheme and attention scheme, a primary model for copper ore / waste ore classification based on multi-spectral images was constructed; The constructed primary model for copper ore / waste ore classification based on multi-energy spectrum images includes an image processing layer, a residual connection layer, an attention layer and a classification layer connected in series in sequence; the image processing layer is used to extract the preliminary features of the input ore image; the residual connection layer is used to extract the deep features of the input ore image and alleviate the problem of gradient disappearance; the attention layer enhances the influence of key information in the input ore image in a feature weighted manner; the classification layer is used to classify and identify the feature data output by the attention layer to achieve the final classification of copper ore / waste ore.

[0020] In specific implementation, an image processing layer is constructed based on a convolution scheme to perform preliminary convolution and down-sampling processing on the input multi-energy spectrum image data; A residual connection layer is constructed based on the residual scheme to perform residual processing on the feature data output by the image processing layer; An attention layer is constructed based on the channel attention scheme and the spatial attention scheme to perform channel attention and spatial attention processing on the feature data output by the residual connection layer; A classification layer is constructed based on the fully connected scheme to classify and identify the feature data output by the attention layer.

[0021] Among them, the image processing layer specifically includes the following contents: The image processing layer includes a concatenated convolution kernel, a batch normalization layer, and an activation function layer; the convolution kernel is The convolution kernel has a 32 output channels; the activation function layer is the ReLU activation function layer; the batch normalization layer and the activation function layer are used to downsample the input data.

[0022] The residual connection layer specifically includes the following: The residual connection layer includes a first residual layer and a second residual layer connected in series; the number of output channels of the first residual layer is 64, and the number of output channels of the second residual layer is 128; In the first residual layer, the input of the first residual layer is After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the first residual sub-feature of the first layer; the first residual sub-feature of the first layer is obtained by After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the second residual sub-feature of the first layer; the input of the first residual layer is processed through the fully connected layer to obtain the first layer input sub-feature, and the fully connected layer is used to ensure that the shape of the first input sub-feature is the same as the second residual sub-feature of the first layer; the second residual sub-feature of the first layer and the input sub-feature of the first layer are superimposed as the output of the first residual layer; In the second residual layer, the input of the second residual layer is passed through After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the first residual sub-feature of the second layer; the first residual sub-feature of the second layer is obtained by After the convolution kernel extracts the features, it is processed through a batch normalization layer and a ReLU activation function layer to obtain the second residual sub-features of the second layer; the input of the second residual layer is processed through a fully connected layer to obtain the second layer input sub-features. The fully connected layer is used to ensure that the shape of the second input sub-features is the same as that of the second residual sub-features of the second layer; the second residual sub-features of the second layer and the second layer input sub-features are superimposed as the output of the second residual layer.

[0023] The attention layer specifically includes the following: The attention layer includes a channel attention layer, a spatial attention layer and a preliminary classification layer connected in series in sequence; the input of the channel attention layer is processed by the channel attention layer to obtain the channel attention feature, and the channel attention feature is multiplied by the input of the channel attention layer as the input of the spatial attention layer; the input of the spatial attention layer is processed by the spatial attention layer to obtain the spatial attention feature, and the spatial attention feature is multiplied by the input of the spatial attention layer as the input of the preliminary classification; the preliminary classification layer is used to perform preliminary classification on the feature data input by the attention layer; The channel attention layer includes a global layer and a maximum layer; the input of the channel attention layer is divided into two paths, one path is processed by the global layer to obtain global features, and the other path is processed by the maximum layer to obtain maximum features, and finally the global features and the maximum features are added to obtain the channel attention features; wherein the global layer includes a global average pooling layer, a global first fully connected layer, a global ReLU activation function layer, a global second fully connected layer and a global Sigmiod activation function layer; the input of the global layer is processed by the global average pooling layer, and then processed by the global first fully connected layer to reduce the number of channels, and then processed by the global ReLU activation function layer for channel feature weighting, and then processed by the global second fully connected layer. The second fully connected layer is processed to restore the number of channels, and finally the channel features are weighted through the global Sigmiod activation function layer to obtain the global features; the maximum layer includes the maximum pooling layer, the maximum first fully connected layer, the maximum ReLU activation function layer, the maximum second fully connected layer and the maximum Sigmiod activation function layer; the input of the maximum layer is processed by the maximum pooling layer, and then processed by the maximum first fully connected layer to reduce the number of channels, and then processed by the maximum ReLU activation function layer for channel feature weighting, and then processed by the maximum second fully connected layer to restore the number of channels, and finally the channel features are weighted through the maximum Sigmiod activation function layer to obtain the maximum feature; In the spatial attention layer, the input of the spatial attention layer calculates the mean and maximum value along the channel dimension, and then concatenates the mean and maximum value and passes Convolution generates a spatial weight map, which is then processed by the Sigmoid activation function to weight the spatial features to obtain the spatial attention features. The preliminary classification layer includes an average pooling layer and a fully connected layer connected in series; the input of the preliminary classification layer is processed by the average pooling layer and the fully connected layer to achieve preliminary classification of the feature data input by the attention layer.

[0024] The classification layer specifically includes the following: In the classification layer, the preliminary classification results of different energy spectra output by the attention layer are concatenated and then processed through the fully connected layer for classification and recognition to obtain the final classification result.

[0025] The classification model provided by the present invention not only has good reliability and accuracy, but also has a small number of network parameters and a fast running speed.

[0026] S4. Using the training data set constructed in step S2, the primary model for copper ore / waste ore classification based on multi-spectral images constructed in step S3 is trained to obtain a copper ore / waste ore classification model based on multi-spectral images; S5. Using the copper ore / waste ore classification model based on the multi-energy spectrum image obtained in step S4, actual copper ore / waste ore classification and identification is performed.

[0027] The method of the present invention is further described below in conjunction with an embodiment: First, a training data set of copper ore / waste ore was constructed by ourselves, and then the method of the present invention was compared with several existing solutions.

[0028] Among them, the existing solutions include Resnet-34, SwinT-Tiny, ConvNeXt and Efficientnet-V2; among them, the Resnet-34 solution is the solution proposed by Microsoft Research in the paper "Deep Residual Learning for Image Recognition" in 2016; the SwinT-Tiny solution is the solution proposed by Microsoft Research in the paper "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" in 2021; the ConvNeXt solution is the solution proposed by Facebook AIResearch and UC Berkeley in the paper "A ConvNet for the 2020s" in 2022; the Efficientnet-V2 solution is the solution proposed by Google Research in the paper "EfficientNet-V2: SmallerModels and Faster Training" in 2021.

[0029] When conducting comparative experiments, since the existing scheme is only applicable to single-energy spectrum images, only single-energy spectrum images are input for the comparative scheme to conduct experiments; at the same time, in order to illustrate the comprehensive performance of the scheme of the present invention, the scheme of the present invention is divided into two categories during the experiment, one category only inputs single-energy spectrum images, and the other category inputs multi-energy spectrum images.

[0030] In the comparative experiment, the experimental indicators include accuracy, precision, recall, F1 score and parameter quantity; the specific experimental results of the comparative experiment are shown in Table 1 below: Table 1 Schematic diagram of comparative experimental indicators

[0031] From the comparative data in Table 1, it can be seen that when using multi-spectral images for classification experiments, the scheme of the present invention performs well in the four indicators of accuracy, precision, recall rate and F1 score; in addition, in terms of the amount of training parameters, the scheme of the present invention shows obvious advantages. Therefore, the scheme of the present invention has the advantages of high reliability, good accuracy and good efficiency.

[0032] like Figure 2 The functional module schematic diagram of the system of the present invention is shown as follows: the system for realizing the copper ore / waste ore classification method based on multi-energy spectral images disclosed in the present invention comprises a data acquisition module, a data processing module, a model construction module, a model training module and an ore classification module; the data acquisition module, the data processing module, the model construction module, the model training module and the ore classification module are connected in series in sequence; the data acquisition module is used to acquire the multi-energy spectral image data information of the existing copper ore and waste ore, and upload the data information to the data processing module; the data processing module is used to perform data preprocessing on the acquired multi-energy spectral image data information according to the received data information to construct a training data set, and upload the data information to the model construction module; the model construction module is used to construct a primary model for copper ore / waste ore classification based on multi-energy spectral images based on the received data information, based on a convolution scheme, a residual scheme and an attention scheme, and upload the data information to the model training module; wherein the constructed copper ore / waste ore classification model based on multi-energy spectral images The primary model for ore / waste ore classification includes an image processing layer, a residual connection layer, an attention layer and a classification layer connected in series in sequence; the image processing layer is used to extract preliminary features of the input ore image; the residual connection layer is used to extract deep features of the input ore image and alleviate the problem of gradient disappearance; the attention layer enhances the influence of key information in the input ore image in a feature-weighted manner; the classification layer is used to classify and identify the feature data output by the attention layer to achieve the final classification of copper ore / waste ore; the model training module is used to train the primary model for copper ore / waste ore classification based on multi-energy spectral images using the constructed training data set according to the received data information, obtain the copper ore / waste ore classification model based on multi-energy spectral images, and upload the data information to the ore classification module; the ore classification module is used to perform actual classification and identification of copper ore / waste ore using the obtained copper ore / waste ore classification model based on multi-energy spectral images according to the received data information.

Claims

1. A copper ore / waste ore classification method based on multi-spectral images, characterized in that The steps include: S1. Obtaining multi-spectral image data information of existing copper ore and waste ore; S2. Preprocessing the multi-spectral image data information obtained in step S1 to construct a training data set; S3. Based on the convolution scheme, residual scheme and attention scheme, a primary model for copper ore / waste ore classification based on multi-spectral images was constructed; The constructed primary model for copper ore / waste ore classification based on multi-energy spectrum images includes an image processing layer, a residual connection layer, an attention layer and a classification layer connected in series; the image processing layer is used to extract the preliminary features of the input ore image; the residual connection layer is used to extract the deep features of the input ore image and alleviate the problem of gradient disappearance; the attention layer enhances the influence of key information in the input ore image in a feature-weighted manner; The classification layer is used to classify and identify the feature data output by the attention layer to achieve the final classification of copper ore / waste ore; S4. Using the training data set constructed in step S2, the primary model for copper ore / waste ore classification based on multi-spectral images constructed in step S3 is trained to obtain a copper ore / waste ore classification model based on multi-spectral images; S5. Using the copper ore / waste ore classification model based on the multi-spectral image obtained in step S4, actual copper ore / waste ore classification and identification is performed.

2. The copper ore / waste ore classification method based on multi-spectral images according to claim 1 is characterized in that The step S2 specifically includes the following steps: Performing data preprocessing on the multi-energy spectrum image data information obtained in step S1; The preprocessing includes image denoising, image enhancement, image segmentation and data enhancement; wherein, the image denoising includes image denoising using a Gaussian filtering method; the image enhancement includes image enhancement using a contrast enhancement technique; the image segmentation includes segmenting and acquiring all ore areas in the image, and uniformly adjusting the acquired images to a set size; the data enhancement is used to improve the generalization performance of the model.

3. The copper ore / waste ore classification method based on multi-spectral images according to claim 1 is characterized in that The step S3 comprises the following steps: An image processing layer is constructed based on a convolution scheme to perform preliminary convolution and downsampling processing on the input multi-energy spectrum image data; A residual connection layer is constructed based on the residual scheme to perform residual processing on the feature data output by the image processing layer; An attention layer is constructed based on the channel attention scheme and the spatial attention scheme to perform channel attention and spatial attention processing on the feature data output by the residual connection layer; A classification layer is constructed based on the fully connected scheme to classify and identify the feature data output by the attention layer.

4. The copper ore / waste ore classification method based on multi-spectral images according to claim 3 is characterized in that The image processing layer specifically includes the following contents: The image processing layer includes a concatenated convolution kernel, a batch normalization layer, and an activation function layer; the convolution kernel is The convolution kernel with an output channel number of 32 is used; the activation function layer is the ReLU activation function layer.

5. The copper ore / waste ore classification method based on multi-spectral images according to claim 4 is characterized in that The residual connection layer specifically includes the following contents: The residual connection layer includes a first residual layer and a second residual layer connected in series; the number of output channels of the first residual layer is 64, and the number of output channels of the second residual layer is 128; In the first residual layer, the input of the first residual layer is After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the first residual sub-feature of the first layer; The first residual sub-feature of the first layer is obtained by After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the first layer second residual sub-features; the input of the first residual layer is processed through the fully connected layer to obtain the first layer input sub-features. The fully connected layer is used to ensure that the shape of the first input sub-features is the same as the first layer second residual sub-features; The second residual sub-feature of the first layer and the input sub-feature of the first layer are superimposed as the output of the first residual layer; In the second residual layer, the input of the second residual layer is passed through After the convolution kernel extracts the features, it is processed through the batch normalization layer and the ReLU activation function layer to obtain the first residual sub-feature of the second layer; The first residual sub-feature of the second layer is obtained by After the convolution kernel extracts the features, it is processed through a batch normalization layer and a ReLU activation function layer to obtain the second residual sub-features of the second layer; the input of the second residual layer is processed through a fully connected layer to obtain the second layer input sub-features. The fully connected layer is used to ensure that the shape of the second input sub-features is the same as that of the second residual sub-features of the second layer; the second residual sub-features of the second layer and the second layer input sub-features are superimposed as the output of the second residual layer.

6. The copper ore / waste ore classification method based on multi-spectral images according to claim 5 is characterized in that The attention layer specifically includes the following contents: The attention layer includes a channel attention layer, a spatial attention layer and a preliminary classification layer connected in series in sequence; the input of the channel attention layer is processed by the channel attention layer to obtain the channel attention feature, and the channel attention feature is multiplied by the input of the channel attention layer as the input of the spatial attention layer; the input of the spatial attention layer is processed by the spatial attention layer to obtain the spatial attention feature, and the spatial attention feature is multiplied by the input of the spatial attention layer as the input of the preliminary classification; the preliminary classification layer is used to perform preliminary classification on the feature data input by the attention layer; The channel attention layer includes a global layer and a maximum layer; the input of the channel attention layer is divided into two paths, one path is processed by the global layer to obtain global features, and the other path is processed by the maximum layer to obtain maximum features, and finally the global features and the maximum features are added to obtain the channel attention features; wherein the global layer includes a global average pooling layer, a global first fully connected layer, a global ReLU activation function layer, a global second fully connected layer and a global Sigmiod activation function layer; the input of the global layer is processed by the global average pooling layer, and then processed by the global first fully connected layer to reduce the number of channels, and then processed by the global ReLU activation function layer for channel feature weighting, and then processed by the global second fully connected layer. The second fully connected layer is processed to restore the number of channels, and finally the channel features are weighted through the global Sigmiod activation function layer to obtain the global features; the maximum layer includes the maximum pooling layer, the maximum first fully connected layer, the maximum ReLU activation function layer, the maximum second fully connected layer and the maximum Sigmiod activation function layer; the input of the maximum layer is processed by the maximum pooling layer, and then processed by the maximum first fully connected layer to reduce the number of channels, and then processed by the maximum ReLU activation function layer for channel feature weighting, and then processed by the maximum second fully connected layer to restore the number of channels, and finally the channel features are weighted through the maximum Sigmiod activation function layer to obtain the maximum feature; In the spatial attention layer, the input of the spatial attention layer calculates the mean and maximum value along the channel dimension, and then concatenates the mean and maximum value and passes Convolution generates a spatial weight map, which is then processed by the Sigmoid activation function to weight the spatial features to obtain the spatial attention features. The preliminary classification layer includes an average pooling layer and a fully connected layer connected in series; the input of the preliminary classification layer is processed by the average pooling layer and the fully connected layer to achieve preliminary classification of the feature data input by the attention layer.

7. The copper ore / waste ore classification method based on multi-spectral images according to claim 6 is characterized in that The classification layer specifically includes the following contents: In the classification layer, the preliminary classification results of different energy spectra output by the attention layer are concatenated and then processed through the fully connected layer for classification and recognition to obtain the final classification result.

8. A system for implementing the copper ore / waste ore classification method based on multi-spectral images as claimed in any one of claims 1 to 7, characterized in that It includes a data acquisition module, a data processing module, a model building module, a model training module and an ore classification module; the data acquisition module, the data processing module, the model building module, the model training module and the ore classification module are connected in series in sequence; the data acquisition module is used to obtain the multi-energy spectrum image data information of the existing copper ore and waste ore, and upload the data information to the data processing module; The data processing module is used to perform data preprocessing on the acquired multi-energy spectrum image data information according to the received data information to construct a training data set, and upload the data information to the model construction module; The model building module is used to construct a primary model for copper ore / waste ore classification based on multi-energy spectrum images based on the convolution scheme, residual scheme and attention scheme according to the received data information, and upload the data information to the model training module; wherein the primary model for copper ore / waste ore classification based on multi-energy spectrum images comprises an image processing layer, a residual connection layer, an attention layer and a classification layer connected in series in sequence; the image processing layer is used to extract the preliminary features of the input ore image; the residual connection layer is used to extract the deep-level features of the input ore image and alleviate the problem of gradient disappearance; the attention layer improves the input ore image in a feature-weighted manner. The influence of key information in the attention layer; the classification layer is used to classify and identify the feature data output by the attention layer to achieve the final classification of copper ore / waste ore; the model training module is used to train the constructed copper ore / waste ore classification primary model based on multi-energy spectral images according to the received data information, using the constructed training data set to obtain the copper ore / waste ore classification model based on multi-energy spectral images, and upload the data information to the ore classification module; the ore classification module is used to perform actual copper ore / waste ore classification and identification according to the received data information using the obtained copper ore / waste ore classification model based on multi-energy spectral images.

Images