Deep residual network model for coal gangue recognition, coal gangue recognition method and system

Abstract

The application discloses a deep residual network model for coal gangue identification, a coal gangue identification method and a system. The model comprises a 3*3 convolution layer, a first mixed attention layer, a first Layer layer, a second Layer layer, a third Layer layer, a fourth Layer layer, a second mixed attention layer, a processing layer, a full connection layer and a Softmax layer. The 3*3 convolution layer is used for extracting shallow feature information of a target data set. Through processing of the first mixed attention layer, a first mixed weight feature map is obtained. Through processing of a plurality of convolution layers, a second mixed weight feature map is obtained. Through processing of the second mixed attention layer, a third mixed weight feature map is obtained. Through Dropout processing of the processing layer, a fourth mixed weight feature map is obtained. Through processing of the full connection layer, coal gangue information is obtained. The Softmax layer is used for outputting a coal gangue identification result. The application can effectively improve the coal gangue identification accuracy.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of coal gangue recognition, and particularly relates to a deep residual network model for coal gangue recognition, a coal gangue recognition method and system. BACKGROUND

[0002] Coal is one of the traditional energy sources in China, is closely related to the development of national economy and people's life, and occupies an important strategic position in China's energy industry. With the improvement of social modernization, the demand for energy is also increasing.

[0003] In recent years, the national coal mining and processing system has been continuously improved, which promotes economic development while paying attention to efficient and clean production. Coal gangue is a low-carbon solid waste produced in the process of coal mining and processing.

[0004] Due to the complex characteristics of coal gangue, it is easily affected by factors such as light and angle. The existing technology discloses that the convolutional neural network is used to recognize and classify coal and gangue images, which cannot well handle long-distance dependence, resulting in poor processing effect of sequence data, and relatively poor effect of processing high-dimensional data or unstructured data, resulting in poor reliability of processing results and low accuracy of coal gangue recognition. SUMMARY

[0005] Therefore, the present disclosure provides a deep residual network model for coal gangue recognition, a coal gangue recognition method and system, which can solve the problem of low accuracy of coal gangue recognition in the prior art.

[0006] In a first aspect, the present disclosure provides a deep residual network model for coal gangue recognition, comprising:

[0007] a 3x3 convolutional layer for extracting shallow feature information of a target data set;

[0008] a first mixed attention layer for processing the shallow feature information to obtain a first mixed weight feature map;

[0009] a first Layer layer for performing convolutional processing on the first mixed weight feature map to obtain a first spectral convolutional feature;

[0010] a second Layer layer for performing convolutional processing on the first spectral convolutional feature to obtain a second spectral convolutional feature;

[0011] a third Layer layer for performing convolutional processing on the second spectral convolutional feature to obtain a third spectral convolutional feature;

[0012] a fourth Layer layer for performing convolutional processing on the fourth spectral convolutional feature to obtain a second mixed weight feature map;

[0013] a second mixed attention layer, configured to process the second mixed weight feature map to obtain a third mixed weight feature map;

[0014] a processing layer, configured to perform Dropout processing on the third mixed weight feature map to obtain a fourth mixed weight feature map;

[0015] a full connection layer, configured to process the fourth mixed weight feature map to obtain coal gangue information;

[0016] a Softmax layer, configured to process the coal gangue information to output a coal gangue recognition result.

[0017] Optionally, the first mixed attention layer comprises a first channel attention module, a first fusion module, a first spatial attention module and a second fusion module.

[0018] The first channel attention module is configured to extract first channel feature information of the shallow feature information.

[0019] The first fusion module is configured to fuse the shallow feature information and the first channel feature information to obtain first feature information.

[0020] The first spatial attention module is configured to extract first spatial feature information of the first feature information.

[0021] The second fusion module is configured to fuse the first spatial feature information and the first feature information to obtain the first mixed weight feature map.

[0022] Optionally, the first Layer layer comprises three first residual modules connected in series; the first residual module comprises a first convolutional layer, a second convolutional layer and a third convolutional layer connected in series; the first convolutional layer is provided with 64 convolutional kernels, and each convolutional kernel has a size of 1x1; the second convolutional layer is provided with 64 convolutional kernels, and each convolutional kernel has a size of 3x3; and the third convolutional layer is provided with 128 convolutional kernels, and each convolutional kernel has a size of 1x1.

[0023] The second Layer layer comprises four second residual modules connected in series; the second residual module comprises a fourth convolutional layer, a fifth convolutional layer and a sixth convolutional layer connected in series; the fourth convolutional layer is provided with 128 convolutional kernels, and each convolutional kernel has a size of 1x1; the fifth convolutional layer is provided with 128 convolutional kernels, and each convolutional kernel has a size of 3x3; and the sixth convolutional layer is provided with 256 convolutional kernels, and each convolutional kernel has a size of 1x1.

[0024] The third Layer layer includes six third residual modules connected in series; the third residual module includes a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer connected in series; the seventh convolutional layer is provided with 256 convolutional kernels, and each convolutional kernel has a size of 1*1; the eighth convolutional layer is provided with 256 convolutional kernels, and each convolutional kernel has a size of 3*3; and the ninth convolutional layer is provided with 512 convolutional kernels, and each convolutional kernel has a size of 1*1.

[0025] The fourth Layer layer includes three fourth residual modules connected in series; the fourth residual module includes a tenth convolutional layer, an eleventh convolutional layer, and a twelfth convolutional layer connected in series; the tenth convolutional layer is provided with 512 convolutional kernels, and each convolutional kernel has a size of 1*1; the eleventh convolutional layer is provided with 512 convolutional kernels, and each convolutional kernel has a size of 3*3; and the twelfth convolutional layer is provided with 1024 convolutional kernels, and each convolutional kernel has a size of 1*1.

[0026] Optionally, the second hybrid attention layer includes a second channel attention module, a third fusion module, a second spatial attention module, and a fourth fusion module.

[0027] The second channel attention module is configured to extract second channel feature information of the second hybrid weight feature map.

[0028] The third fusion module is configured to fuse the second hybrid weight feature map and the second channel feature information to obtain third feature information.

[0029] The second spatial attention module is configured to extract second spatial feature information of the third feature information.

[0030] The fourth fusion module is configured to fuse the second spatial feature information and the third feature information to obtain the third hybrid weight feature map.

[0031] The second aspect of the application discloses a coal gangue recognition method, comprising:

[0032] Performing spectral dimension transformation processing on the collected coal gangue hyperspectral data to obtain a coal gangue RGB image;

[0033] Based on the coal gangue RGB image, a coal gangue HSV image dataset is obtained;

[0034] Based on the coal gangue HSV image dataset, a training set and a test set are constructed;

[0035] Training a new residual network model on the training set, and verifying the trained new residual network model by using a loss function;

[0036] Test the trained fusion residual network model on the test set to obtain a coal gangue recognition result;

[0037] The new residual network model is the deep residual network model for coal gangue recognition.

[0038] Optionally, the method for obtaining the coal gangue RGB image comprises:

[0039] The near-infrared spectrometer is used to collect coal gangue spectral data multiple times, and the average value of the obtained data is taken as first data;

[0040] The PCA algorithm is used to perform dimension reduction processing on the first data to obtain the coal gangue RGB image.

[0041] Optionally, the coal gangue HSV image data set comprises an H pixel matrix, an S pixel matrix and a V pixel matrix.

[0042]

[0043]

[0044] V = max(R, G, B);

[0045] Wherein, R, G and B are R pixel matrix, G pixel matrix and B pixel matrix in the coal gangue RGB image respectively.

[0046] max(R, G, B) is the maximum matrix among R, G and B;

[0047] min(R, G, B) is the minimum matrix among R, G and B;

[0048] Delta is the difference between max(R, G, B) and min(R, G, B).

[0049] Optionally, the construction of the training set and the test set comprises:

[0050] The coal gangue HSV image data set is randomly allocated according to a ratio of 6:4 to obtain the training set and the test set.

[0051] Optionally, the training of the fusion residual network model on the training set comprises:

[0052] The number of batch training data is set to 64;

[0053] The initial learning rate is set to 0.001, and the learning rate is reduced by 0.5 times every 10 iterations;

[0054] When the actual training iteration number reaches the training iteration number threshold, the training is stopped.

[0055] Preferably, the training iteration threshold is 100.

[0056] The loss function is a cross-entropy loss function.

[0057] A third aspect of the present application discloses a coal gangue recognition system, which comprises:

[0058] A first acquisition module configured to perform spectral dimension transformation processing on the collected coal gangue hyperspectral data to obtain a coal gangue RGB image.

[0059] A second acquisition module configured to obtain a coal gangue HSV image dataset based on the coal gangue RGB image.

[0060] A construction module configured to construct a training set and a test set based on the coal gangue HSV image dataset.

[0061] A training module configured to train a new residual network model on the training set and verify the trained new residual network model using a loss function.

[0062] A recognition module configured to test the trained fusion residual network model on the test set to obtain a coal gangue recognition result.

[0063] The new residual network model is the deep residual network model for coal gangue recognition.

[0064] In a fourth aspect, the embodiments of the present disclosure further provide an electronic device, which adopts the following technical scheme:

[0065] The electronic device comprises:

[0066] at least one processor; and

[0067] a memory in communication connection with the at least one processor; wherein

[0068] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the coal gangue recognition method described in any one of the above.

[0069] In a fifth aspect, the embodiments of the present disclosure further provide a computer readable storage medium storing computer instructions for causing a computer to execute the coal gangue recognition method described in any one of the above.

[0070] The deep residual network model for coal gangue recognition disclosed in the application has the following advantages: 1) the 3*3 convolution layer can effectively extract the shallow feature information of the target data set, which helps to capture the key features of the coal gangue and improve the accuracy of coal gangue recognition; 2) the first mixed attention layer can process the shallow feature information and generate the first mixed weight feature map, and this attention mechanism can make the network pay more attention to important features and improve the recognition ability of the coal gangue; 3) through the processing of multiple convolution layers, the model can gradually extract higher-level features, making the extracted features more accurate; 4) through the processing of the second mixed attention layer and the processing layer, the model can generate the third mixed weight feature map and the fourth mixed weight feature map, which can further extract the key features of the coal gangue and enhance the ability of coal gangue recognition; 5) through the processing of the full connection layer and the Softmax layer, the model can convert the fourth mixed weight feature map into coal gangue information and output the coal gangue recognition result, and this structure can convert the feature map into an interpretable result, which is convenient for users to understand and apply. In summary, the deep residual network model for coal gangue recognition provided in the application can improve the accuracy and reliability of coal gangue recognition to the greatest extent by extracting feature information, attention mechanism, multi-layer convolution processing and the generation of mixed weight feature maps, while reducing the amount of parameters and calculations, and can play an important role in the field of coal gangue recognition.

[0071] The coal gangue recognition method disclosed in the application has the following advantages: 1) the spectral dimension transformation processing of the collected coal gangue hyperspectral data can convert multi-dimensional hyperspectral data into three-dimensional RGB images, which helps to simplify data representation, reduce data dimension and improve the efficiency of subsequent processing; 2) converting the coal gangue RGB image into an HSV image can more comprehensively describe the characteristics of the coal gangue from different color space angles, which helps to extract more rich feature information and enhance the ability of coal gangue recognition; 3) based on the coal gangue HSV image data set, the data is divided into a training set and a test set, which helps to train and evaluate the model, and this division can ensure that the model is trained on the training set to learn the feature pattern of the coal gangue, and at the same time verify the generalization ability of the model on the test set; 4) by constructing a new residual network model, the advantages of residual blocks can be fully utilized to improve the accuracy and stability of coal gangue recognition, and the new residual network model can effectively solve the problems of gradient disappearance and gradient explosion, and improve the efficiency of information transmission; 5) the new residual network model is trained on the training set, and the model is verified by using a loss function, which can optimize the parameters of the model by minimizing the loss function and improve the performance of coal gangue recognition, which helps the network model to adapt to the feature distribution of the coal gangue data set and improve the recognition ability of the coal gangue, greatly improving the accuracy of coal gangue recognition, i.e. improving the reliability of the processing result.

[0072] The above description is only a summary of the technical solutions of the present disclosure. In order to make the technical means of the present disclosure more clearly understood, the contents of the specification can be implemented, and in order to make the above and other purposes, characteristics and advantages of the present disclosure more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0073] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0074] Figure 1 The logical flow chart of the coal gangue recognition method provided by the embodiments of the present disclosure.

[0075] Figure 2 The processing schematic diagram of the novel residual network model in the present application.

[0076] Figure 3 The constituent schematic diagram of the first mixed attention layer in Figure 2

[0077] The constituent schematic diagram of the first Layer layer in Figure 4 Figure 2 The constituent schematic diagram of the second Layer layer in

[0078] Figure 5 Figure 2 The constituent schematic diagram of the third Layer layer in

[0079] Figure 6 The constituent schematic diagram of the fourth Layer layer in Figure 2

[0080] The constituent schematic diagram of the second mixed attention layer in Figure 7 Figure 2 The constituent schematic diagram of the coal gangue recognition system in the present application.

[0081] Figure 8 Figure 2 The structure schematic diagram of an electronic device provided by the embodiments of the present disclosure.

[0082] Figure 9 The structure schematic diagram of an electronic device provided by the embodiments of the present disclosure.

[0083] Figure 10 The structure schematic diagram of an electronic device provided by the embodiments of the present disclosure.

[0084] BRIEF DESCRIPTION OF DRAWINGS

[0085] ​​​​1, 3x3 convolutional layer; 2, first mixed attention layer; 3, first Layer layer; 4, second Layer layer; 5, third Layer layer; 6, fourth Layer layer; 7, second mixed attention layer; 8, processing layer; 9, fully connected layer; 10, Softmax layer. DETAILED DESCRIPTION

[0086] The embodiments of the present disclosure will be described in detail below with reference to the drawings.

[0087] It should be apparent that the following describes embodiments of the present disclosure by way of specific examples, and that one of ordinary skill in the art can readily derive other advantages and benefits from this disclosure without departing from the spirit of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, and are not all the embodiments. The present disclosure can also be implemented or applied by other different specific embodiments, and various modifications or changes can be made to the details in this specification without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by one of ordinary skill in the art without creative labor are within the scope of protection of the present disclosure.

[0088] It should be noted that the various aspects of the embodiments described below are within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms and that any specific structure and / or function described herein is merely illustrative. Based on the teachings provided herein one skilled in the art should be able to contemplate these and similar aspects of the present disclosure. For example, the various aspects described herein can be implemented across many differing platforms. Depending on the embodiment, certain of the aspects can be implemented via software, hardware or a combination thereof. Moreover, the various aspects can be implemented using any number of techniques, and using any number of programming languages, tools, libraries and / or other development utilities and / or platforms.

[0089] It should also be noted that the drawings provided in the following embodiments are only schematically illustrating the basic concepts of the present disclosure, and only the components related to the present disclosure are shown in the drawings, not drawn according to the number, shape and size of the components in actual implementation, and the shape, number and proportion of each component in actual implementation can be arbitrarily changed, and the layout pattern of the components can also be more complex.

[0090] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, one skilled in the art will understand that the described aspects can be practiced without these specific details.

[0091] Reference Figure 1The first aspect of the application discloses a coal gangue recognition method, which comprises the following steps:

[0092] S100, performing spectral dimension transformation processing on the collected coal gangue hyperspectral data to obtain a coal gangue RGB image.

[0093] Specifically, the method for obtaining the coal gangue RGB image comprises the following steps: collecting coal gangue spectral data multiple times by using a near-infrared spectrometer, and taking the average value of the obtained data as first data;

[0094] Performing dimension reduction processing on the first data by using a PCA (Principal components analysis) algorithm to obtain a coal gangue RGB image.

[0095] Through this step, the spectral dimension transformation is performed on the coal gangue hyperspectral data, the low-dimensional signal with original information is generated based on principal component analysis, the dimension of the features can be effectively reduced, and the total number of the features can be reduced.

[0096] For the acquisition of the coal gangue RGB image, continuous wavelet transformation (CWT) and short-time Fourier transform (STFT) can also be used.

[0097] S200, obtaining a coal gangue HSV image data set based on the coal gangue RGB image.

[0098] Specifically, the coal gangue HSV image data set comprises an H pixel matrix, an S pixel matrix and a V pixel matrix.

[0099]

[0100]

[0101] V = max (R, G, B).

[0102] Wherein, R, G and B are R pixel matrix, G pixel matrix and B pixel matrix in the coal gangue RGB image, which have been normalized.

[0103] max (R, G, B) is the maximum matrix in R, G and B;

[0104] min (R, G, B) is the minimum matrix in R, G and B;

[0105] Delta is the difference between max (R, G, B) and min (R, G, B).

[0106] By converting the coal gangue RGB image into the coal gangue HSV image data set, that is, by converting the RGB color space into the HSV color space, the random noise in the spectral signal is effectively eliminated, and the recognition rate is improved.

[0107] S300, constructing a training set and a test set based on the coal gangue HSV image dataset.

[0108] Specifically, the coal gangue HSV image dataset is randomly allocated in a ratio of 6:4 to obtain the training set and the test set; that is, in this embodiment, 60% of the coal gangue HSV image dataset is used as a sample of the training set, and the remaining 40% of the coal gangue HSV image dataset is used as a sample of the test set.

[0109] S400, constructing a new residual network model.

[0110] S500, training the new residual network model on the training set, and verifying the trained new residual network model by using a loss function.

[0111] Specifically, at the beginning of training, the number of batch training data is set to 64; the initial learning rate is set to 0.001, and the learning rate is reduced by 0.5 times every 10 iterations.

[0112] When the actual training iteration number reaches the training iteration number threshold, the training is stopped.

[0113] In this embodiment, the training iteration number threshold is preferably 100; and the loss function is preferably a cross-entropy loss function.

[0114] S600, testing the trained fusion residual network model on the test set to obtain a coal gangue recognition result.

[0115] The coal gangue recognition method disclosed in the application has the following advantages: 1) The multi-dimensional hyperspectral data can be converted into a three-dimensional RGB image through spectral dimension transformation processing of the collected coal gangue hyperspectral data, which helps to simplify data representation, reduce data dimension, and improve the efficiency of subsequent processing; 2) The characteristics of the coal gangue can be more comprehensively described from different color space angles by converting the coal gangue RGB image into an HSV image, which helps to extract more rich feature information and enhance the ability of coal gangue recognition; 3) Based on the coal gangue HSV image data set, the data is divided into a training set and a test set, which helps to train and evaluate the model. This division can ensure that the model is trained on the training set, so as to learn the feature mode of the coal gangue, and verify the generalization ability of the model on the test set; 4) By constructing a new residual network model, the advantages of the residual block can be fully utilized to improve the accuracy and stability of coal gangue recognition. The new residual network model can effectively solve the problems of gradient disappearance and gradient explosion, and improve the efficiency of information transmission; 5) The new residual network model is trained on the training set, and the model is verified by using a loss function. The parameters of the model can be optimized by minimizing the loss function to improve the performance of coal gangue recognition. This helps the network model to adapt to the feature distribution of the coal gangue data set and improve the recognition ability of the coal gangue, greatly improving the accuracy of coal gangue recognition, that is, improving the reliability of the processing result.

[0116] Specifically, referring to Figure 2 , the new residual network model is a deep residual network model for coal gangue recognition. The model includes a 3x3 convolution layer 1, a first mixed attention layer 2, a first Layer layer 3, a second Layer layer 4, a third Layer layer 5, a fourth Layer layer 6, a second mixed attention layer 7, a processing layer 8, a fully connected layer 9, and a Softmax layer 10.

[0117] The 3x3 convolution layer is used to extract shallow feature information of the target data set. In this embodiment, the target data set refers to the coal gangue HSV image data set.

[0118] The first mixed attention layer is used to process the shallow feature information to obtain a first mixed weight feature map.

[0119] The first Layer layer is used to perform convolution processing on the first mixed weight feature map to obtain a first spectral convolution feature.

[0120] The second Layer layer is used to perform convolution processing on the first spectral convolution feature to obtain a second spectral convolution feature.

[0121] The third Layer layer is used to perform convolution processing on the second spectral convolution feature to obtain a third spectral convolution feature.

[0122] The fourth Layer is configured to perform convolution processing on the fourth spectral convolution feature to obtain a second mixed weight feature map.

[0123] The second mixed attention layer is configured to process the second mixed weight feature map to obtain a third mixed weight feature map.

[0124] The processing layer is configured to perform Dropout processing on the third mixed weight feature map to obtain a fourth mixed weight feature map, so as to reduce the too close interaction between the feature information and avoid overfitting.

[0125] The full connection layer is configured to process the fourth mixed weight feature map to obtain coal gangue information; in this embodiment, the coal gangue information refers to a 2*1 first feature vector and a 2*1 second feature vector.

[0126] The Softmax layer is configured to process the coal gangue information to output a coal gangue recognition result, that is, the data with the highest probability is taken as the output result, that is, an accurate coal gangue recognition result is obtained.

[0127] Referring to Figure 3 The first mixed attention layer includes a first channel attention module, a first fusion module, a first spatial attention module and a second fusion module.

[0128] The first channel attention module is configured to extract first channel feature information of shallow feature information.

[0129] The first fusion module is configured to fuse the shallow feature information and the first channel feature information to obtain first feature information.

[0130] The first spatial attention module is configured to extract first spatial feature information of the first feature information.

[0131] The second fusion module is configured to fuse the first spatial feature information and the first feature information to obtain a first mixed weight feature map.

[0132] Further, the method for extracting the first channel feature information specifically includes: compressing global spatial information through global maximum pooling (MaxPool) and global average pooling (Avg Pool) to generate two feature maps with a size of C*1*1.

[0133] In order to fully utilize the feature information extracted by the compression operation, the obtained feature maps are then input into a multi-layer perceptron (MLP) to obtain two one-dimensional feature maps, the multi-layer perceptron being composed of two fully connected layers and a ReLU activation function. After summing the two one-dimensional feature maps by channel, a Sigmoid function is used for normalization to obtain the output result of Channel Attention, and then the output result is multiplied by the original image to return to the size of C*H*W.

[0134] The calculation formula is as follows:

[0135]

[0136] The feature map obtained after recalibration of the channel features is taken as input, and global maximum pooling and average pooling operations are respectively performed on the channel dimension to obtain two 1*H*W feature maps.

[0137] Then, the two feature maps are spliced through channel concatenation. The information at different positions is encoded and fused through a convolution layer to obtain spatial weighting information M S , which can be used to distinguish the importance of different spatial positions of the image.

[0138] The calculation formula is as follows:

[0139]

[0140] After the maximum pooling downsampling processing removes the redundant information, the mixed features are sent to the first Layer layer of the model.

[0141] Referring to Figure 4 , the first Layer layer includes three first residual modules connected in series.

[0142] The first residual module includes a first convolution layer, a second convolution layer, and a third convolution layer connected in series; the first convolution layer is provided with 64 convolution kernels, and each convolution kernel has a size of 1*1; the second convolution layer is provided with 64 convolution kernels, and each convolution kernel has a size of 3*3; the third convolution layer is provided with 128 convolution kernels, and each convolution kernel has a size of 1*1.

[0143] Specifically, the first convolution layer is configured to perform a convolution operation on the first mixed weight feature map using 64 1*1 convolution kernels to obtain first convolution information.

[0144] The second convolution layer is configured to perform a convolution operation on the first convolution information using 64 3*3 convolution kernels to obtain second convolution information.

[0145] The third convolution layer is configured to perform a convolution operation on the second convolution information using 128 1*1 convolution kernels to obtain third convolution information.

[0146] Then, the third convolution information is processed through another two first residual modules connected in series to obtain the first spectral convolution feature.

[0147] Referring to Figure 5 , the second Layer layer includes four second residual modules connected in series.

[0148] The second residual module comprises a fourth convolutional layer, a fifth convolutional layer, and a sixth convolutional layer connected in sequence; the fourth convolutional layer is provided with 128 convolutional kernels, each of which has a size of 1*1; the fifth convolutional layer is provided with 128 convolutional kernels, each of which has a size of 3*3; and the sixth convolutional layer is provided with 256 convolutional kernels, each of which has a size of 1*1.

[0149] Specifically, the fourth convolutional layer is configured to perform a convolution operation on the first spectral convolutional feature by using 128 1*1 convolutional kernels to obtain fourth convolutional information.

[0150] The fifth convolutional layer is configured to perform a convolution operation on the fourth convolutional information by using 128 3*3 convolutional kernels to obtain fifth convolutional information.

[0151] The sixth convolutional layer is configured to perform a convolution operation on the fifth convolutional information by using 256 1*1 convolutional kernels to obtain sixth convolutional information.

[0152] The sixth convolutional information is then processed by another two second residual modules connected in sequence to obtain the second spectral convolutional feature.

[0153] Referring to Figure 6 , the third Layer layer comprises six third residual modules connected in sequence.

[0154] The third residual module comprises a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer connected in sequence; the seventh convolutional layer is provided with 256 convolutional kernels, each of which has a size of 1*1; the eighth convolutional layer is provided with 256 convolutional kernels, each of which has a size of 3*3; and the ninth convolutional layer is provided with 512 convolutional kernels, each of which has a size of 1*1.

[0155] Specifically, the seventh convolutional layer is configured to perform a convolution operation on the second spectral convolutional feature by using 256 1*1 convolutional kernels to obtain seventh convolutional information.

[0156] The eighth convolutional layer is configured to perform a convolution operation on the seventh convolutional information by using 256 3*3 convolutional kernels to obtain eighth convolutional information.

[0157] The ninth convolutional layer is configured to perform a convolution operation on the eighth convolutional information by using 512 1*1 convolutional kernels to obtain ninth convolutional information.

[0158] The ninth convolutional information is then processed by another two third residual modules connected in sequence to obtain the third spectral convolutional feature.

[0159] Referring to Figure 7 , the fourth Layer layer comprises three fourth residual modules connected in sequence.

[0160] The fourth residual module comprises a tenth convolutional layer, an eleventh convolutional layer and a twelfth convolutional layer connected in sequence; the tenth convolutional layer is provided with 512 convolutional kernels, each of which has a size of 1*1; the eleventh convolutional layer is provided with 512 convolutional kernels, each of which has a size of 3*3; and the twelfth convolutional layer is provided with 1024 convolutional kernels, each of which has a size of 1*1.

[0161] Specifically, the tenth convolutional layer is configured to perform a convolution operation on the third spectral convolutional feature by using 256 1*1 convolutional kernels to obtain tenth convolutional information.

[0162] The eleventh convolutional layer is configured to perform a convolution operation on the tenth convolutional information by using 256 3*3 convolutional kernels to obtain eleventh convolutional information.

[0163] The twelfth convolutional layer is configured to perform a convolution operation on the tenth convolutional information by using 512 1*1 convolutional kernels to obtain twelfth convolutional information.

[0164] The twelfth convolutional information is then processed by another two fourth residual modules connected in sequence to obtain the second mixed weight feature map.

[0165] With reference to Figure 8 The second mixed attention layer comprises a second channel attention module, a third fusion module, a second spatial attention module and a fourth fusion module.

[0166] The second channel attention module is configured to extract second channel feature information of the second mixed weight feature map.

[0167] The third fusion module is configured to fuse the second mixed weight feature map and the second channel feature information to obtain third feature information.

[0168] The second spatial attention module is configured to extract second spatial feature information of the third feature information.

[0169] The fourth fusion module is configured to fuse the second spatial feature information and the third feature information to obtain a third mixed weight feature map.

[0170] The deep residual network model for coal gangue recognition disclosed in the present application combines the improved mixed attention layer with the residual structure, and through the setting of the first mixed attention layer and the second mixed attention layer, important local feature information with discriminability can be extracted, and the residual modules in the first Layer layer, the second Layer layer, the third Layer layer and the fourth Layer layer can well combine the correlation between these information, so that the network pays more attention to important feature information, suppresses the interference of irrelevant information, allocates limited computing resources to key areas with larger weights, speeds up the convergence speed of the model, improves the network recognition performance, and thus improves the coal gangue recognition accuracy.

[0171] The deep residual network model for coal gangue recognition disclosed in the application has the following advantages: 1) the 3*3 convolution layer can effectively extract the shallow feature information of the target data set, which helps to capture the key features of the coal gangue and thus improves the accuracy of the coal gangue recognition; 2) the first mixed attention layer can process the shallow feature information and generate the first mixed weight feature map, and this attention mechanism can make the network pay more attention to important features and improve the recognition ability of the coal gangue; 3) through the processing of multiple convolution layers, the model can gradually extract higher-level features, making the extracted features more accurate; 4) through the processing of the second mixed attention layer and the processing layer, the model can generate the third mixed weight feature map and the fourth mixed weight feature map, which can further extract the key features of the coal gangue and enhance the recognition ability of the coal gangue; 5) through the processing of the full connection layer and the Softmax layer, the model can convert the fourth mixed weight feature map into coal gangue information and output the coal gangue recognition result, and this structure can convert the feature map into an interpretable result, which is convenient for users to understand and apply.

[0172] In summary, the deep residual network model for coal gangue recognition provided by the application can reduce the parameter quantity and the calculation amount while maximizing the accuracy and reliability of the coal gangue recognition through the extraction of feature information, the attention mechanism, the multi-layer convolution processing and the generation of mixed weight feature maps, and can play an important role in the field of coal gangue recognition.

[0173] Reference Figure 9 Another aspect of the application discloses a coal gangue recognition system, which comprises:

[0174] The first acquisition module is configured to perform spectral dimension transformation processing on the collected coal gangue hyperspectral data to obtain a coal gangue RGB image.

[0175] The second acquisition module is configured to obtain a coal gangue HSV image data set based on the coal gangue RGB image.

[0176] The construction module is configured to construct a training set and a test set based on the coal gangue HSV image data set.

[0177] The training module is configured to train the new residual network model on the training set and verify the trained new residual network model using a loss function.

[0178] The recognition module is configured to test the trained fusion residual network model on the test set to obtain a coal gangue recognition result.

[0179] An electronic device according to an embodiment of the disclosure includes a memory and a processor. The memory is configured to store non-transitory computer-readable instructions. Specifically, the memory can include one or more computer program products that can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory, for example, can include random access memory (RAM), cache memory, and / or the like. The non-volatile memory, for example, can include read-only memory (ROM), hard disk, flash memory, and / or the like.

[0180] The processor can be a central processing unit (CPU) or other form of processing unit that has data processing and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions. In an embodiment of the disclosure, the processor is configured to execute the computer-readable instructions stored in the memory, so that the electronic device performs all or part of the steps of the coal gangue identification method according to the embodiments of the disclosure described above.

[0181] Those skilled in the art will understand that, in order to solve the technical problem of how to obtain a good user experience effect, the embodiment can also include well-known structures such as a communication bus, an interface, and the like, which should also be included in the protection scope of the disclosure.

[0182] As Figure 10 A structural schematic diagram of an electronic device according to an embodiment of the disclosure is provided. It shows a structural schematic diagram suitable for implementing the electronic device in the embodiments of the disclosure. Figure 10 The electronic device shown is only an example and should not impose any limitation on the functions and use range of the embodiments of the disclosure.

[0183] As Figure 10 As shown, the electronic device can include a processor (such as a central processing unit, a graphics processing unit, and the like) that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) or loaded from a storage device into a random access memory (RAM). In the RAM, various programs and data required for the operation of the electronic device are also stored. The processor, the ROM, and the RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0184] Generally, the following devices can be connected to the I / O interface: input devices including, for example, sensors or visual information acquisition devices; output devices including, for example, display screens; storage devices including, for example, magnetic tapes, hard disks, and the like; and communication devices. The communication devices can allow the electronic device to communicate with other devices (such as edge computing devices) to exchange data wirelessly or through wires. Although Figure 10Electronic devices with various arrangements are shown, but it should be understood that not all of the arrangements shown are required, or necessary, to practice or have the various embodiments. More or less arrangements can alternatively be practiced or had.

[0185] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the coal gangue identification method of the embodiments of the present disclosure are performed.

[0186] Detailed descriptions of the present embodiments can refer to the corresponding descriptions of the previous embodiments, which will not be repeated here.

[0187] According to the computer-readable storage medium of the embodiments of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are run by a processor, all or part of the steps of the coal gangue identification method of the embodiments of the present disclosure are performed.

[0188] The computer-readable storage medium described above includes, but is not limited to, optical storage media (such as CD-ROM and DVD), magneto-optical storage media (such as MO), magnetic storage media (such as magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (such as memory card), and media with built-in ROM (such as ROM cartridge).

[0189] Detailed descriptions of the present embodiments can refer to the corresponding descriptions of the previous embodiments, which will not be repeated here.

[0190] The basic principles of the present disclosure are described above in conjunction with specific embodiments, but it should be noted that the advantages, advantages, effects, etc. mentioned in the present disclosure are only examples and are not limiting, and these advantages, advantages, effects, etc. cannot be considered as the various embodiments of the present disclosure must have. In addition, the above specific details are only for the purpose of example and for the purpose of understanding, and the above details do not limit the present disclosure to be necessarily implemented with the above specific details.

[0191] In this disclosure, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. The block diagram of the devices, apparatus, equipment, systems referred to in this disclosure is merely illustrative and not intended to imply the necessity or arrangement of the connections, arrangement, configuration as shown in the block diagram. As will be appreciated by those skilled in the art, the devices, apparatus, equipment, systems can be connected, arranged, configured in any manner. The words comprising, including, having and the like are to be open ended. As used in this document, the conjunction "or" is to be interpreted in the inclusive sense, i.e. as meaning one or the other, or both. As used in this document, the words "and" and "or" are to be interpreted as having the meaning indicated in the phrase "and / or". As used in this document, the word "such as" is to be interpreted as meaning "such as, but not limited to".

[0192] Also, as used in this document, the word "or" in the cases used to introduce an enumeration of several items, for example, a list of items, is to be interpreted in the inclusive sense, i.e. as meaning one or more, or any combination thereof, of the listed items. Additionally, the phrase "example of" as used in this document is not meant to be limiting in any way. It is not meant to imply that the described example is the only example of the described feature.

[0193] It is also important to note that the systems and methods of the present disclosure can be embodied in a variety of forms including, but not limited to, a data processor, a computer program product, a computer, one or more tangible computer readable storage devices, one or more computer memories, one or more programmable logic devices, one or more application specific devices, one or more computers, one or more processors, one or more microprocessors, one or more microcomputers, one or more microcontrollers, one or more microcontrollers, one or more microprocessors, one or more state machines, one or more integrated circuits, one or more other components, or any combination thereof, and that the systems and methods can comprise, consist of, or consist essentially of such forms.

[0194] Various changes, modifications and alterations in the teachings and techniques described herein can be made without departing from the teachings and techniques defined by the appended claims. Moreover, the scope of the claims of this disclosure is not limited to the specific aspects described above. Rather, the aspects of the disclosure are meant to cover all alternatives, modifications, and equivalents falling within the scope of the claims. Accordingly, the appended claims are intended to encompass all such changes, modifications and alterations.

[0195] The above description of the disclosed aspects is meant to be illustrative only and not limiting as to the scope of the disclosure. Many variations of these aspects are possible, as those skilled in the art will readily understand upon reading this document. Changes in form and detail can be made without departing from the scope of the disclosure. Accordingly, the disclosed aspects are not limited to the aspects set forth herein for purposes of describing the aspects of the disclosure, but are meant to cover all alternatives, modifications, and equivalents falling within the scope of the claims.

[0196] The foregoing description has been presented for the purposes of illustration and description. Furthermore, the description is not intended to limit the embodiments of the disclosure to the forms disclosed herein. Although the various example aspects and embodiments have been described herein with regard to particular aspects and embodiments, those skilled in the art will recognize that certain modifications, changes, substitutions, additions and sub-combinations can be made without departing from the spirit of the disclosure.

Claims

1. A deep residual network model for coal gangue identification, characterized in that, include: A 3×3 convolutional layer is used to extract shallow feature information from the target dataset, which refers to the coal gangue HSV image dataset. The first hybrid attention layer is used to process the shallow feature information to obtain the first hybrid weight feature map; The first layer is used to perform convolution processing on the first hybrid weight feature map to obtain the first spectral convolution feature; The second layer is used to perform convolution processing on the first spectral convolution features to obtain the second spectral convolution features. The third layer is used to perform convolution processing on the second spectral convolution features to obtain the third spectral convolution features; The fourth layer is used to perform convolution processing on the third spectral convolution features to obtain the second mixed weight feature map; The second hybrid attention layer is used to process the second hybrid weight feature map to obtain the third hybrid weight feature map; The processing layer is used to perform Dropout processing on the third mixed weight feature map to obtain the fourth mixed weight feature map. A fully connected layer is used to process the fourth hybrid weight feature map to obtain coal gangue information; The Softmax layer is used to process the coal and gangue information and output the coal and gangue identification results. The first hybrid attention layer includes a first channel attention module, a first fusion module, a first spatial attention module, and a second fusion module; the first channel attention module is used to extract the first channel feature information of the shallow feature information; the first fusion module is used to fuse the shallow feature information with the first channel feature information to obtain the first feature information; The first spatial attention module is used to extract the first spatial feature information of the first feature information; the second fusion module is used to fuse the first spatial feature information with the first feature information to obtain the first hybrid weight feature map. The first layer includes three first residual modules connected in series, each of which includes a first convolutional layer, a second convolutional layer, and a third convolutional layer connected in series. The second layer includes four second residual modules connected in series. Each second residual module includes a fourth convolutional layer, a fifth convolutional layer, and a sixth convolutional layer connected in series. The third layer includes six third residual modules connected in series. Each third residual module includes a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer connected in series. The fourth layer includes three fourth residual modules connected in series. Each fourth residual module includes a tenth convolutional layer, an eleventh convolutional layer, and a twelfth convolutional layer connected in series.

2. The deep residual network model for coal gangue identification according to claim 1, characterized in that, The first convolutional layer has 64 convolutional kernels, each with a size of 1×1; the second convolutional layer has 64 convolutional kernels, each with a size of 3×3; and the third convolutional layer has 128 convolutional kernels, each with a size of 1×1. The fourth convolutional layer has 128 convolutional kernels, each with a size of 1×1; the fifth convolutional layer has 128 convolutional kernels, each with a size of 3×3; and the sixth convolutional layer has 256 convolutional kernels, each with a size of 1×1. The seventh convolutional layer has 256 convolutional kernels, each with a size of 1×1; the eighth convolutional layer has 256 convolutional kernels, each with a size of 3×3; and the ninth convolutional layer has 512 convolutional kernels, each with a size of 1×1. The tenth convolutional layer has 512 convolutional kernels, each with a size of 1×1; the eleventh convolutional layer has 512 convolutional kernels, each with a size of 3×3; and the twelfth convolutional layer has 1024 convolutional kernels, each with a size of 1×1.

3. The deep residual network model for coal gangue identification according to claim 2, characterized in that, The second hybrid attention layer includes a second channel attention module, a third fusion module, a second spatial attention module, and a fourth fusion module; The second channel attention module is used to extract the second channel feature information of the second hybrid weight feature map; The third fusion module is used to fuse the second mixed weight feature map with the second channel feature information to obtain the third feature information; The second spatial attention module is used to extract the second spatial feature information of the third feature information; The fourth fusion module is used to fuse the second spatial feature information with the third feature information to obtain the third hybrid weighted feature map.

4. A method for identifying coal gangue, characterized in that, include: The collected hyperspectral data of coal gangue were processed by spectral dimension transformation to obtain RGB images of coal gangue. Based on the RGB image of the coal gangue, a coal gangue HSV image dataset is obtained; A training set and a test set were constructed based on the aforementioned coal gangue HSV image dataset; The novel residual network model is trained on the training set, and the trained novel residual network model is validated using a loss function. The trained fusion residual network model was tested on the test set to obtain coal gangue identification results; The novel residual network model is the deep residual network model for coal gangue identification as described in any one of claims 1-3.

5. The coal gangue identification method according to claim 4, characterized in that, The method for obtaining the RGB image of the coal gangue includes: The spectral data of coal gangue were collected multiple times using a near-infrared spectrometer, and the average value of the data was used as the first data. The first data is subjected to dimensionality reduction processing using the PCA algorithm to obtain the RGB image of the coal gangue.

6. The coal gangue identification method according to claim 5, characterized in that, The coal gangue HSV image dataset includes an H pixel matrix, an S pixel matrix, and a V pixel matrix; ； ； ； in, 、 、 These are the R pixel matrix, G pixel matrix, and B pixel matrix in the RGB image of the coal gangue, respectively. for , 、 The largest matrix in; for , 、 The smallest matrix in; for and The difference.

7. The coal gangue identification method according to claim 6, characterized in that, The construction of the training set and the test set includes: The coal gangue HSV image dataset is randomly allocated in a 6:4 ratio to obtain the training set and the test set.

8. The coal gangue identification method according to claim 4, characterized in that, Training the fusion residual network model on the training set includes: Set the batch training data size to 64; The initial learning rate is set to 0.001, and the learning rate decreases by 0.5 times every 10 iterations. Training stops when the actual number of training iterations reaches the training iteration threshold.

9. A coal gangue identification system, characterized in that, The system includes: The first acquisition module is configured to perform spectral dimension transformation processing on the collected coal gangue hyperspectral data to obtain coal gangue RGB images; The second acquisition module is configured to obtain a coal gangue HSV image dataset based on the coal gangue RGB image; The module is configured to build training and testing sets based on the coal gangue HSV image dataset; The training module is configured to train a novel residual network model on the training set and to validate the trained novel residual network model using a loss function. The identification module is configured to test the trained fusion residual network model on the test set to obtain coal gangue identification results; The novel residual network model is the deep residual network model for coal gangue identification as described in any one of claims 1-3.

Images