Tailing coal ash content dynamic detection method and system based on shallow layer sharing and deep layer expansion

Abstract

The invention discloses a tail coal ash content dynamic detection method and system based on shallow-layer sharing and deep-layer expansion, and the method comprises the steps: obtaining a coal slime mixed liquid data set, and carrying out the data preprocessing, and obtaining the two-dimensional spectrum image data of a coal slime mixed liquid; a large convolution kernel framework is introduced as a parameter sharing layer, a small convolution kernel framework is introduced as an independent expansion layer, and a tail coal ash content detection network model is constructed; and dynamically detecting the two-dimensional spectral image data of the coal slime mixed liquid based on the tail coal ash content detection network model to obtain a tail coal ash content detection result of the coal slime mixed liquid. By using the method, deep independent information mining can be performed on the image feature data, so that the accuracy of the detection result of the tail coal ash content of the coal slime mixed liquid is improved. The tail coal ash content dynamic detection method and system based on shallow layer sharing and deep layer extension can be widely applied to the technical field of tail coal ash content detection.

Description

Technical Field

[0001] The present invention relates to the technical field of tail coal ash detection, and particularly to a dynamic detection method and system for tail coal ash based on shallow sharing and deep expansion. Background Technique

[0002] The image method based on deep learning for the detection of tail coal ash in coal slime flotation mainly involves using machine vision technology to obtain flotation foam images and applying a deep learning model to extract and analyze the characteristic parameters in the images to predict the ash content of the flotation tail coal. This method collects the image data of the flotation foam and then uses deep learning algorithms, such as convolutional neural networks (CNNs), to automatically extract the features in the images and establish an association model between these features and the ash content. However, the related technologies generally only use small convolutional layers with similar kernel sizes to extract features, which can capture finer local features but do not consider the global features of larger receptive fields in the images, ignoring corresponding texture, shape and other features. Secondly, at the shallow layer of the model, the shallow features extracted from the images are similar and the distinguishability is not high, while the existing methods will retrain the entire model, wasting some computing resources. Summary of the Invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a dynamic detection method and system for tail coal ash based on shallow sharing and deep expansion, which can perform in-depth independent information mining on the image feature data, thereby improving the accuracy of the detection results of the tail coal ash in the coal slime mixture.

[0004] The first technical solution adopted by the present invention is: a dynamic detection method for tail coal ash based on shallow sharing and deep expansion, including the following steps: Obtain a coal slime mixture data set and perform data preprocessing to obtain two-dimensional spectral image data of the coal slime mixture; Introduce a large convolutional kernel framework as a parameter sharing layer and a small convolutional kernel framework as an independent expansion layer to construct a tail coal ash detection network model; Based on the tail coal ash detection network model, perform dynamic detection on the two-dimensional spectral image data of the coal slime mixture to obtain the detection result of the tail coal ash in the coal slime mixture.

[0005] Furthermore, the step of obtaining a coal slime mixture data set and performing data preprocessing to obtain two-dimensional spectral image data of the coal slime mixture specifically includes: Obtain several groups of coal slime mixtures with different concentrations and different ash contents; Drive the coal slime mixture to form a liquid flow through a peristaltic pump and set a flow-through colorimetric cell observation window for data collection to obtain a coal slime mixture image data set; Combined with an optical fiber spectrometer and a xenon light source, absorption spectral data of the coal slime mixture image dataset is collected to obtain the spectral time-series data of the coal slime mixture image; The spectral time-series data of the coal slime mixture image is converted into two-dimensional image data by the Gramian angular field method to obtain the two-dimensional spectral image data of the coal slime mixture.

[0006] Furthermore, the tail coal ash detection network model specifically includes a parameter sharing layer, an independent expansion layer, and an updated fully connected layer. The output end of the parameter sharing layer is connected to the input end of the independent expansion layer, and the output end of the independent expansion layer is connected to the input end of the updated fully connected layer.

[0007] Furthermore, the parameter sharing layer specifically includes a starting layer, a first stage layer, a first transition layer, a second stage layer, a second transition layer, and a third stage layer. The starting layer, the first stage layer, the first transition layer, the second stage layer, the second transition layer, and the third stage layer are connected in sequence, where: The starting layer includes a first two-dimensional convolutional layer, a first depthwise separable convolutional layer, a second two-dimensional convolutional layer, and a second depthwise separable convolutional layer. The first two-dimensional convolutional layer, the first depthwise separable convolutional layer, the second two-dimensional convolutional layer, and the second depthwise separable convolutional layer are connected in sequence; The first stage layer, the second stage layer, and the third stage layer all include a number of repeated large kernel modules and a number of convolutional feedforward network layers. The repeated large kernel module includes a first batch normalization layer, a third two-dimensional convolutional layer, a third depthwise separable convolutional layer, and a fourth two-dimensional convolutional layer. The convolutional feedforward network layer includes a second batch normalization layer, a fifth two-dimensional convolutional layer, a first activation function layer, and a sixth two-dimensional convolutional layer; Both the first transition layer and the second transition layer include a seventh two-dimensional convolutional layer and a fourth depthwise separable convolutional layer.

[0008] Furthermore, the independent expansion layer specifically includes a first residual layer, a second residual layer, and a third residual layer. The first residual layer, the second residual layer, and the third residual layer are connected in sequence, where: The first residual layer, the second residual layer, and the third residual layer all include an eighth two-dimensional convolutional layer, a third batch normalization layer, a second activation function layer, a ninth two-dimensional convolutional layer, a fourth batch normalization layer, and a third activation function layer.

[0009] Furthermore, the step of dynamically detecting the two-dimensional spectral image data of the coal slime mixture based on the tail coal ash detection network model to obtain the tail coal ash detection result of the coal slime mixture specifically includes: Input the two-dimensional spectral image data of the coal slime mixture into the tail coal ash detection network model; Based on the parameter sharing layer of the tail coal ash detection network model, perform shallow feature extraction on the two-dimensional spectral image data of the coal slurry mixture to obtain the shallow feature data of the coal slurry mixture; Based on the independent expansion layer of the tail coal ash detection network model, perform semantic feature extraction processing on the shallow feature data of the coal slurry mixture to obtain the deep feature data of the coal slurry mixture; Based on the updated fully connected layer of the tail coal ash detection network model, perform pooling processing on the deep feature data of the coal slurry mixture and convert it into a vector form to obtain the tail coal ash detection result of the coal slurry mixture.

[0010] Furthermore, the step of performing shallow feature extraction on the two-dimensional spectral image data of the coal slurry mixture based on the parameter sharing layer of the tail coal ash detection network model to obtain the shallow feature data of the coal slurry mixture specifically includes: Input the two-dimensional spectral image data of the coal slurry mixture into the parameter sharing layer of the tail coal ash detection network model; Based on the starting layer of the parameter sharing layer, perform local spatial feature extraction processing on the two-dimensional spectral image data of the coal slurry mixture to obtain the local spatial feature data of the coal slurry mixture; Based on the first stage layer of the parameter sharing layer, perform global feature extraction processing on the local spatial feature data of the coal slurry mixture to obtain the global spatial feature data of the coal slurry mixture; Based on the first transition layer of the parameter sharing layer, perform size reduction processing on the global spatial feature data of the coal slurry mixture to obtain the reduced global spatial feature data of the coal slurry mixture; Based on the second stage layer of the parameter sharing layer, perform intermediate feature extraction processing on the reduced global spatial feature data of the coal slurry mixture to obtain the intermediate feature data of the coal slurry mixture; Based on the second transition layer of the parameter sharing layer, perform resolution reduction processing on the intermediate feature data of the coal slurry mixture to obtain the reduced intermediate feature data of the coal slurry mixture; Based on the third stage layer of the parameter sharing layer, perform advanced feature extraction processing on the reduced intermediate feature data of the coal slurry mixture to obtain the shallow feature data of the coal slurry mixture.

[0011] The second technical solution adopted by the present invention is: a tail coal ash dynamic detection system based on shallow sharing and deep expansion includes: The first module is used to obtain the coal slurry mixture data set and perform data preprocessing to obtain the two-dimensional spectral image data of the coal slurry mixture; The second module is used to introduce a large convolution kernel framework as the parameter sharing layer and a small convolution kernel framework as the independent expansion layer to construct a tail coal ash detection network model; The third module is used to dynamically detect the two-dimensional spectral image data of the coal slime mixture based on the tail coal ash detection network model, and obtain the tail coal ash detection result of the coal slime mixture.

[0012] The beneficial effects of the method and system of the present invention are as follows: By obtaining the coal slime mixture data set and performing data preprocessing, the two-dimensional spectral image data of the coal slime mixture is obtained. Further, a large convolutional kernel framework is introduced as a parameter sharing layer, and a small convolutional kernel framework is used as an independent expansion layer to construct a tail coal ash detection network model. The large convolutional kernel framework is used for extracting the shallow layer features of the network mainly, while the deep layer of the network mainly uses the residual blocks of the small convolutional kernel layer as the module for extracting high-level semantic features. The shallow layer of the model with less distinct extracted features is used as the parameter sharing layer, and the deep layer uses a residual module group containing small convolutional kernels for in-depth independent information mining, which not only saves a part of computing resources but also ensures the performance of the model. Finally, based on the tail coal ash detection network model, the two-dimensional spectral image data of the coal slime mixture is dynamically detected, thereby improving the accuracy of the tail coal ash detection result of the coal slime mixture. Brief Description of the Drawings

[0013] Figure 1 is the flowchart of the steps of the tail coal ash dynamic detection method based on shallow layer sharing and deep layer expansion of the present invention; Figure 2 is the structural block diagram of the tail coal ash dynamic detection system based on shallow layer sharing and deep layer expansion of the present invention; Figure 3 is the structural schematic diagram of the tail coal ash detection network model provided by the specific embodiment of the present invention; Figure 4 is the structural schematic diagram of the detail level of the tail coal ash detection network model provided by the specific embodiment of the present invention. Detailed Embodiment

[0014] The following further describes the present invention in detail with reference to the drawings and specific embodiments. For the step numbers in the following embodiments, they are only set for the convenience of explanation and illustration, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0015] Refer to Figure 1 , the present invention provides a tail coal ash dynamic detection method based on shallow layer sharing and deep layer expansion, and the method includes the following steps: S100. Obtain the coal slime mixture data set and perform data preprocessing to obtain the two-dimensional spectral image data of the coal slime mixture; Specifically, a number of groups of slime mixed liquids with different concentrations and different ash contents are obtained; the slime mixed liquid is driven by a peristaltic pump to form a liquid flow, and a flow-through colorimetric cell observation window is set for data collection to obtain a slime mixed liquid image data set; the absorption spectrum data of the slime mixed liquid image data set is collected by combining an optical fiber spectrometer and a xenon light source to obtain the spectral time series data of the slime mixed liquid image; the spectral time series data of the slime mixed liquid image is converted into two-dimensional image data by the Gram angular field method to obtain the two-dimensional spectral image data of the slime mixed liquid.

[0016] In this embodiment, a total of 110 groups of slime mixed liquids with different concentrations and different ash contents are configured to simulate the flotation site, with a total of 11 kinds of ash contents and 10 concentrations for each kind of ash content. A peristaltic pump is used to drive the mixed liquid to form a liquid flow, and a flow-through colorimetric cell observation window is set on the liquid flow branch for data collection. A micro UV-Vis optical fiber spectrometer is used in combination with a xenon light source to continuously collect the absorption spectrum data of the suspension liquid flow. The collected absorption spectrum time series data is preprocessed, and the Gram angular field technology is applied to convert the time series data into two-dimensional image data, and the converted data is shuffled and divided into multiple task subsets for subsequent training.

[0017] In addition, it should be noted that when used in an industrial site, a branch can be erected beside the flotation machine, and an optical fiber spectrometer can be used to collect real-time spectral data of the liquid flow on the branch.

[0018] S200. Introduce a large convolutional kernel framework as a parameter sharing layer and a small convolutional kernel framework as an independent expansion layer to construct a tailings ash content detection network model; Specifically, as Figure 3 shown, the tailings ash content detection network model specifically includes a parameter sharing layer, an independent expansion layer, and an updated fully connected layer. The output end of the parameter sharing layer is connected to the input end of the independent expansion layer, and the output end of the independent expansion layer is connected to the input end of the updated fully connected layer.

[0019] In this embodiment, since the data sets used are all tailings samples, which have similar low-level information and the main distinguishing key points are concentrated on high-semantic features, we believe that when constructing a deep learning model, a hierarchical feature extraction strategy can be adopted, where parameters are shared in the shallow layer (i.e., the first few layers of the network) of the model, and high-semantic information is independently mined in the deep layer of the model. In this way, our model can maintain the generality of low-level features, enhance the sensitivity to high-semantic features, and significantly reduce the training memory occupancy.

[0020] Inspired by RepLKNet, a large convolutional kernel framework is used as the main shallow layer of the network. Using large convolutional kernels can extract these general features in a wider range, providing a sufficient receptive field and the ability to aggregate spatial information, ensuring the consistency of feature extraction, and at the same time helping to maintain the spatial hierarchy of the image, enabling the model to better understand the local features in the image and their relationships. The deep layer of the network mainly uses residual blocks with small convolutional kernels as modules for extracting high-semantic features.

[0021] Among them, it should be noted that as Figure 4 shown, the parameter sharing layer specifically includes a starting layer, a first stage layer, a first transition layer, a second stage layer, a second transition layer, and a third stage layer. The starting layer, the first stage layer, the first transition layer, the second stage layer, the second transition layer, and the third stage layer are connected in sequence. Among them, the starting layer includes a first two-dimensional convolutional layer, a first depthwise separable convolutional layer, a second two-dimensional convolutional layer, and a second depthwise separable convolutional layer. The first two-dimensional convolutional layer, the first depthwise separable convolutional layer, the second two-dimensional convolutional layer, and the second depthwise separable convolutional layer are connected in sequence; the first stage layer, the second stage layer, and the third stage layer each include a number of repeated large kernel modules and a number of convolutional feedforward network layers. The repeated large kernel module includes a first batch normalization layer, a third two-dimensional convolutional layer, a third depthwise separable convolutional layer, and a fourth two-dimensional convolutional layer. The convolutional feedforward network layer includes a second batch normalization layer, a fifth two-dimensional convolutional layer, a first activation function layer, and a sixth two-dimensional convolutional layer; the first transition layer and the second transition layer each include a seventh two-dimensional convolutional layer and a fourth depthwise separable convolutional layer.

[0022] In this embodiment, the first layer of the model is the starting layer Stem, which sequentially includes a two-dimensional convolutional kernel of size 3×3 , with the number of channels C1 = 128, a depthwise separable convolutional kernel of size 3×3 , a two-dimensional convolutional kernel of size 1×1 , and a depthwise separable convolutional kernel of size 3×3 , specifically as follows: ; The second, fourth, and sixth layers of the model are stage layers Stage 1, Stage 2, and Stage 3. Their structures include stacks of RepLKNet Blocks and ConvFFN blocks. One RepLKNet Block and one ConvFFN block are called stacked once. Among them, the stacking times of stage layers 1, 2, and 3 are all 2 times, that is, [2, 2, 2]. The corresponding number of channels is [256, 512, 1024].

[0023] The RepLKNet Block structure is a residual-like structure, and the features are batch-normalized and then passed through a 1×1 2D convolution , and then through a depthwise separable convolution of size , and then through a 1×1 convolution . The resulting output is added to the original input . Specifically, it is as follows: ; The depthwise separable convolution in the RepLKNet Block (Repeated Large Kernel module) uses a large convolution kernel structure. The convolution kernel sizes corresponding to stage layers 1, 2, and 3 are [31, 29, 27]. At the same time, the means of structural reparameterization are adopted: using parallel small convolution kernels to calculate and map to the parameters of the large convolution kernel, which can reduce the computational amount of the model and improve the inference speed while maintaining the strong feature extraction ability of the large convolution kernel.

[0024] The ConvFFN (Convolutional Feed-Forward Network layer) block structure is also a residual-like structure, and the features are batch-normalized and then passed through a 1×1 2D convolution , and then through an activation function, and then through a 1×1 convolution . The resulting output is added to the original input . Specifically, it is as follows: ; The third and fifth layers of the model are transition layers Transition 1 and Transition 2. The features are passed through a 1×1 2D convolution and then through a depthwise separable convolution kernel of size 3×3 . Specifically, it is as follows: ; In the above formula, represents a depthwise separable convolution kernel of size 3×3.

[0025] As Figure 4 shown, the independent expansion layer specifically includes a first residual layer, a second residual layer, and a third residual layer. The first residual layer, the second residual layer, and the third residual layer are connected in sequence. Among them, the first residual layer, the second residual layer, and the third residual layer all include an eighth 2D convolution layer, a third batch normalization layer, a second activation function layer, a ninth 2D convolution layer, a fourth batch normalization layer, and a third activation function layer.

[0026] In this embodiment, the seventh to ninth layers of the model are three classical residual structures, and the corresponding number of channels is [2048, 1024, 512]. The features pass through a two-dimensional convolutional layer with a size of 3×3 and then perform batch normalization , and then pass through an activation function , and then repeat the above steps once, and finally add it to the input as follows: ; In the above formula, represents the activation function.

[0027] S300. Dynamically detect the two-dimensional spectral image data of the coal slime mixture based on the tail coal ash detection network model to obtain the tail coal ash detection result of the coal slime mixture.

[0028] Specifically, input the two-dimensional spectral image data of the coal slime mixture into the tail coal ash detection network model; based on the parameter sharing layer of the tail coal ash detection network model, extract shallow features from the two-dimensional spectral image data of the coal slime mixture to obtain the shallow feature data of the coal slime mixture; based on the independent expansion layer of the tail coal ash detection network model, perform semantic feature extraction processing on the shallow feature data of the coal slime mixture to obtain the deep feature data of the coal slime mixture; based on the updated fully connected layer of the tail coal ash detection network model, perform pooling processing on the deep feature data of the coal slime mixture and convert it into a vector form to obtain the tail coal ash detection result of the coal slime mixture.

[0029] Further, it should be noted that the data processing process of the parameter sharing layer of the tail coal ash detection network model is as follows: input the two-dimensional spectral image data of the coal slime mixture into the parameter sharing layer of the tail coal ash detection network model; based on the starting layer of the parameter sharing layer, perform local spatial feature extraction processing on the two-dimensional spectral image data of the coal slime mixture to obtain the local spatial feature data of the coal slime mixture; based on the first-stage layer of the parameter sharing layer, perform global feature extraction processing on the local spatial feature data of the coal slime mixture to obtain the global spatial feature data of the coal slime mixture; based on the first transition layer of the parameter sharing layer, perform size reduction processing on the global spatial feature data of the coal slime mixture to obtain the reduced global spatial feature data of the coal slime mixture; based on the second-stage layer of the parameter sharing layer, perform intermediate feature extraction processing on the reduced global spatial feature data of the coal slime mixture to obtain the intermediate feature data of the coal slime mixture; based on the second transition layer of the parameter sharing layer, perform resolution reduction processing on the intermediate feature data of the coal slime mixture to obtain the reduced intermediate feature data of the coal slime mixture; based on the third-stage layer of the parameter sharing layer, perform high-level feature extraction processing on the reduced intermediate feature data of the coal slime mixture to obtain the shallow feature data of the coal slime mixture.

[0030] In summary, the data processing flow of the model of the present invention is as follows: Input layer: Receive input data, and the input data is a picture with a size of 224*224.

[0031] Starting layer: Use a 3×3 two-dimensional convolution operation for the extraction of preliminary features. Immediately followed by depthwise separable convolution and 1×1 convolution to reduce the computational complexity. The purpose is to capture the local spatial features in the image, and at the same time reduce the number of parameters through depthwise separable convolution and improve the computational efficiency.

[0032] First-stage layer: Includes multiple RepLK Blocks (Replicated Large Kernel Blocks), and each block contains: a large convolution kernel (K×K depthwise separable convolution) for capturing global context information. 1×1 convolution for adjusting the number of channels. Batch normalization and the non-linear activation function GELU to stabilize the training and enhance the features. While retaining local features, strengthen the global information modeling ability, which is suitable for complex feature representation.

[0033] First transition layer: Use 3×3 convolution combined with batch normalization and activation function to reduce the spatial resolution of the feature map. And use 1×1 convolution to adjust the number of channels. Reduce the size of the feature map, improve the compactness of feature representation, and provide input for subsequent stages.

[0034] The second-stage layer: Also composed of multiple RepLK Blocks, it increases the number of channels of the feature map to capture richer features. It extracts deeper intermediate features while retaining the input information.

[0035] The second transition layer: Similar to the first transition layer, it further reduces the resolution of the feature map through 3×3 convolution and batch normalization. It prepares for higher-level feature extraction and provides an optimized representation for the input of subsequent deeper networks.

[0036] The third-stage layer: Composed of RepLK Blocks with higher channel numbers, it includes large convolution kernels, batch normalization, and non-linear activation. It captures high-level semantic features and provides rich expressive power for the detection task.

[0037] The residual layer: Each residual layer contains two 3×3 convolutions, followed by batch normalization and the activation function. It uses residual connections to directly superimpose the input features onto the output. The purpose is to retain the input information, alleviate the vanishing gradient problem, and improve the optimization performance and feature expression ability of the model.

[0038] The updated fully connected layer: It converts the feature map into a vector form through global pooling or flattening operations, and then maps it to the number of categories of the final classification task through the fully connected layer. And it uses the Softmax activation function to output probabilities.

[0039] Therefore, in the embodiments of the present invention, by using a large convolution kernel framework as the main network for shallow feature extraction, while mainly using residual blocks with small convolution kernels in the deep network as modules for extracting high semantic features. The shallow layer of the model with less distinct feature extraction is used as a parameter sharing layer, and the deep layer uses a group of residual modules with small convolution kernels for in-depth independent information mining, which not only saves some computing resources but also ensures the performance of the model.

[0040] In summary, the difference between the embodiments of the present invention and the prior art lies in: 1) The model adopts a dynamic network form, that is, it follows the principle of "sharing large convolution kernels in the shallow layer and expanding small convolution kernels in the deep layer" to perform targeted feature extraction on the data.

[0041] 2) For the shallow layer of the model, both the front and back data use a specific large convolution kernel module group with parameter sharing for shallow feature extraction of information such as texture and contour. In the deep layer of the model, a group of residual modules with small convolution kernels is used for in-depth independent information mining.

[0042] Refer to Figure 2 , the dynamic detection system for tailings ash content based on shallow sharing and deep expansion includes: The first module 201 is used to obtain the coal slurry mixture dataset and perform data preprocessing to obtain the two-dimensional spectral image data of the coal slurry mixture; The second module 202 is used to introduce a large convolutional kernel framework as a parameter sharing layer and a small convolutional kernel framework as an independent expansion layer to construct a tail coal ash detection network model; The third module 203 is used to dynamically detect the two-dimensional spectral image data of the coal slurry mixture based on the tail coal ash detection network model to obtain the tail coal ash detection result of the coal slurry mixture.

[0043] The content in the above method embodiments is applicable to the system embodiments. The functions specifically implemented in the system embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0044] The above is a specific description of the preferred embodiments of the present invention. However, the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention. These equivalent deformations or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A dynamic detection method for tailings ash content based on shallow sharing and deep expansion, characterized in that It includes the following steps: Obtain the coal slime mixture dataset and perform data preprocessing to obtain the two-dimensional spectral image data of the coal slime mixture; Introduce a large convolution kernel framework as a parameter sharing layer and a small convolution kernel framework as an independent expansion layer to construct a tail coal ash detection network model; Based on the tail coal ash detection network model, dynamically detect the two-dimensional spectral image data of the coal slime mixture to obtain the tail coal ash detection result of the coal slime mixture.

2. The dynamic detection method for tail coal ash content based on shallow sharing and deep extension according to claim 1, wherein The step of obtaining the coal slime mixture dataset and performing data preprocessing to obtain the two-dimensional spectral image data of the coal slime mixture specifically includes: Obtain several groups of coal slime mixtures with different concentrations and different ash contents; Drive the coal slime mixture to form a liquid flow through a peristaltic pump and set up a flow-through colorimetric cell observation window for data collection to obtain a coal slime mixture image dataset; Combine a fiber optic spectrometer and a xenon light source to collect absorption spectral data of the coal slime mixture image dataset to obtain the spectral time series data of the coal slime mixture image; Convert the spectral time series data of the coal slime mixture image into two-dimensional image data through the Gram angular field method to obtain the two-dimensional spectral image data of the coal slime mixture.

3. The dynamic detection method for tail coal ash content based on shallow sharing and deep extension according to claim 2, wherein The tail coal ash detection network model specifically includes a parameter sharing layer, an independent expansion layer, and an updated fully connected layer. The output end of the parameter sharing layer is connected to the input end of the independent expansion layer, and the output end of the independent expansion layer is connected to the input end of the updated fully connected layer.

4. The dynamic detection method for tail coal ash content based on shallow sharing and deep extension according to claim 3, wherein The parameter sharing layer specifically includes a starting layer, a first stage layer, a first transition layer, a second stage layer, a second transition layer, and a third stage layer. The starting layer, the first stage layer, the first transition layer, the second stage layer, the second transition layer, and the third stage layer are connected in sequence, where: The starting layer includes a first two-dimensional convolutional layer, a first depthwise separable convolutional layer, a second two-dimensional convolutional layer, and a second depthwise separable convolutional layer. The first two-dimensional convolutional layer, the first depthwise separable convolutional layer, the second two-dimensional convolutional layer, and the second depthwise separable convolutional layer are connected in sequence; The first stage layer, the second stage layer, and the third stage layer each include several repeated large kernel modules and several convolutional feedforward network layers. The repeated large kernel module includes a first batch normalization layer, a third two-dimensional convolutional layer, a third depthwise separable convolutional layer, and a fourth two-dimensional convolutional layer. The convolutional feedforward network layer includes a second batch normalization layer, a fifth two-dimensional convolutional layer, a first activation function layer, and a sixth two-dimensional convolutional layer; The first transition layer and the second transition layer each include a seventh two-dimensional convolutional layer and a fourth depthwise separable convolutional layer.

5. The dynamic detection method for tail coal ash content based on shallow sharing and deep extension according to claim 4, wherein, The independent expansion layer specifically includes a first residual layer, a second residual layer, and a third residual layer. The first residual layer, the second residual layer, and the third residual layer are connected in sequence, where: The first residual layer, the second residual layer, and the third residual layer each include an eighth two-dimensional convolutional layer, a third batch normalization layer, a second activation function layer, a ninth two-dimensional convolutional layer, a fourth batch normalization layer, and a third activation function layer.

6. The dynamic detection method for tailings ash content based on shallow sharing and deep expansion according to claim 5, characterized in that The step of dynamically detecting the two-dimensional spectral image data of the coal slime mixture based on the tail coal ash detection network model to obtain the tail coal ash detection result of the coal slime mixture specifically includes: Input the two-dimensional spectral image data of the coal slime mixture into the tail coal ash detection network model; Based on the parameter sharing layer of the tail coal ash detection network model, perform shallow feature extraction on the two-dimensional spectral image data of the coal slime mixture to obtain the shallow feature data of the coal slime mixture; Based on the independent expansion layer of the tail coal ash detection network model, perform semantic feature extraction processing on the shallow feature data of the coal slime mixture to obtain the deep feature data of the coal slime mixture; Based on the updated fully connected layer of the tail coal ash detection network model, perform pooling processing on the deep feature data of the coal slime mixture and convert it into a vector form to obtain the tail coal ash detection result of the coal slime mixture.

7. The dynamic detection method for tailings ash content based on shallow sharing and deep expansion according to claim 6, characterized in that, The step of performing shallow feature extraction on the two-dimensional spectral image data of the coal slime mixture based on the parameter sharing layer of the tail coal ash detection network model to obtain the shallow feature data of the coal slime mixture specifically includes: Input the two-dimensional spectral image data of the coal slime mixture into the parameter sharing layer of the tail coal ash detection network model; Based on the starting layer of the parameter sharing layer, perform local spatial feature extraction processing on the two-dimensional spectral image data of the coal slime mixture to obtain the local spatial feature data of the coal slime mixture; Based on the first stage layer of the parameter sharing layer, perform global feature extraction processing on the local spatial feature data of the coal slime mixture to obtain the global spatial feature data of the coal slime mixture; Based on the first transition layer of the parameter sharing layer, perform size reduction processing on the global spatial feature data of the coal slime mixture to obtain the reduced global spatial feature data of the coal slime mixture; Based on the second stage layer of the parameter sharing layer, perform intermediate feature extraction processing on the reduced global spatial feature data of the coal slime mixture to obtain the intermediate feature data of the coal slime mixture; Based on the second transition layer of the parameter sharing layer, perform resolution reduction processing on the intermediate feature data of the coal slime mixture to obtain the reduced intermediate feature data of the coal slime mixture; Based on the third stage layer of the parameter sharing layer, perform advanced feature extraction processing on the reduced intermediate feature data of the coal slime mixture to obtain the shallow feature data of the coal slime mixture.

8. The dynamic detection system for tailings ash content based on shallow sharing and deep extension is characterized in that It includes the following modules: The first module is used to obtain the coal slime mixture data set and perform data preprocessing to obtain the two-dimensional spectral image data of the coal slime mixture; The second module is used to introduce a large convolution kernel framework as the parameter sharing layer and a small convolution kernel framework as the independent expansion layer to construct the tail coal ash detection network model; The third module is used to dynamically detect the two-dimensional spectral image data of the coal slime mixture based on the tail coal ash detection network model to obtain the tail coal ash detection result of the coal slime mixture.

Images