Method and system for estimating proportion of mineral components in coal

Abstract

The invention belongs to the technical field of coal quality detection, and particularly discloses a coal mineral component proportion estimation method and system, and the method comprises the steps: collecting an original coal sample hyperspectral image, and carrying out the preprocessing of the original coal sample hyperspectral image; the method comprises the following steps: determining types and proportions of minerals in a coal sample through a coal quality test method, collecting spectral data of each mineral in the coal sample by adopting a hyperspectral imager, and further establishing a spectral database of the minerals in the coal; constructing a 3D convolutional neural network model based on an encoder-decoder structure on the basis of the preprocessed coal sample hyperspectral image and the spectral database of the minerals in the coal; and inputting a to-be-detected original coal sample hyperspectral image into the 3D convolutional neural network model based on the encoder-decoder structure to carry out space-spectral feature fusion, introducing physical prior constraints, and estimating to obtain mineral proportion data in the coal sample. The problems that an existing coal quality detection method is low in precision and poor in applicability, and online real-time detection cannot be achieved are solved.

Description

Technical Field

[0001] This invention belongs to the field of coal quality testing technology, specifically relating to a method and system for estimating the proportion of mineral components in coal. Background Technology

[0002] As an important fossil fuel, coal quality evaluation is crucial for combustion efficiency and industrial applications. The types and contents of mineral components in coal directly affect key quality parameters such as ash content and calorific value. Accurately estimating the proportion of mineral components is a core technical issue in coal quality evaluation. Traditional coal mineral component analysis mainly relies on chemical analysis methods and X-ray analysis techniques. While chemical analysis methods offer high precision, they have long detection cycles and cannot be used for online detection. X-ray diffraction and fluorescence analysis techniques are expensive and have low detection efficiency, similarly failing to meet the needs of rapid online detection in industrial production.

[0003] Hyperspectral imaging technology can acquire characteristic spectral information of different mineral components in coal in the visible-near-infrared band, providing a theoretical basis for spectral-based mineral identification. However, multiple minerals in coal samples often exist in a mixed state, and the spectral signals of each component superimpose to form a mixed spectrum. Traditional spectral analysis methods struggle to accurately separate and quantify the contributions of each component, resulting in low accuracy in estimating mineral component proportions. Deep learning technology offers a new solution for hyperspectral image processing, with convolutional neural networks automatically extracting spectral-spatial features. However, existing deep learning methods often neglect physical constraints when dealing with the unmixing problem of mixed pixels and lack specific designs for the characteristics of mineral components in coal, leading to problems such as unstable accuracy and poor generalization ability in practical applications. Furthermore, the complex nonlinear relationship between coal quality parameters and mineral component proportions remains a technical challenge in establishing an effective coupling model for accurate prediction.

[0004] Therefore, there is an urgent need to develop an online estimation method for the proportion of mineral components in coal that integrates hyperspectral imaging and deep learning, in order to solve the problems of low detection efficiency, low accuracy and inability to detect online in traditional methods, and to provide an effective technical means for rapid evaluation of coal quality. Summary of the Invention

[0005] The purpose of this invention is to solve the problems of low accuracy, poor applicability, and inability to achieve online real-time detection in existing coal quality testing methods, and to propose a method and system for estimating the proportion of mineral components in coal.

[0006] The technical solution of the present invention is as follows: Firstly, a method for estimating the proportion of mineral components in coal, comprising the following steps: Acquired the original hyperspectral image of the coal sample and preprocessed the original hyperspectral image of the coal sample to obtain the preprocessed hyperspectral image of the coal sample; The types and proportions of minerals in coal samples are determined by coal quality testing methods, and the spectral data of each mineral in the coal sample are collected by a hyperspectral imager, thereby establishing a spectral database of minerals in coal. Based on the preprocessed hyperspectral images of coal samples and the spectral database of minerals in coal, a 3D convolutional neural network model based on an encoder-decoder structure was constructed. The hyperspectral image of the original coal sample to be tested is input into a 3D convolutional neural network model based on an encoder-decoder structure for spatial-spectral feature fusion, and physical prior constraints are introduced to estimate the mineral proportion data in the coal sample.

[0007] As a preferred method, when acquiring hyperspectral images of raw coal samples, standard white and black boards are used for radiometric and dark current corrections. The correction formula is as follows:

[0008] in, This represents the corrected reflectance. Represents the raw spectral data. Indicates dark current. This indicates the reflectivity of the whiteboard.

[0009] As a preferred method, the types and proportions of minerals in the coal sample are determined by coal quality analysis methods, and the spectral data of each mineral in the coal sample are acquired using a hyperspectral imager, thereby establishing a coal mineral spectral database. This specifically includes the following steps: The coal sample was ground to below 200 mesh using a ball mill and dried at 105℃ for 2 hours to remove moisture, thus obtaining the processed coal sample. The content of each element in the treated coal sample was determined by X-ray fluorescence analysis. X-ray diffraction analysis was used to identify the mineral types in the processed coal samples. Spectral data of various minerals in coal samples were acquired using a hyperspectral imager; Based on the analysis of the content of each element, mineral type and spectral data, the relationship between the variation of the spectral data of coal samples with different mineral proportions and the reflectance spectra of a single mineral is established to create a mineral spectral database in coal.

[0010] As a preferred embodiment, the 3D convolutional neural network model based on the encoder-decoder structure includes an encoder, a decoder, and a first output layer connected in sequence. The encoder is used to receive the hyperspectral image of the raw coal sample to be detected and extract its spatial-spectral features; Decoder, used to receive spatial-spectral features and reconstruct mineral abundance maps; The first output layer is used to receive the reconstructed mineral abundance map and obtain the abundance content of different minerals in each pixel of the original coal sample hyperspectral image through regression analysis, and output the spatial distribution map of mineral component proportions. The encoder comprises M stacked 3D convolutional units, a ReLU activation function layer, L stacked downsampling units, and a second output layer. Each 3D convolutional unit includes a first 3D convolutional layer and a 3D batch normalization layer connected in sequence. Each downsampling unit includes a second 3D convolutional layer and a downsampling layer connected in sequence. The input of the first 3D convolutional layer of the first 3D convolutional unit is the input of the encoder. The input of the first 3D convolutional layer of the Mth 3D convolutional unit is connected to the output of the (M-1)th 3D convolutional unit. The output of the 3D batch normalization layer of the Mth 3D convolutional unit is connected to the input of the ReLU activation function layer. The input of the second 3D convolutional layer of the first downsampling unit is connected to the output of the ReLU activation function layer. The input of the second 3D convolutional layer of the Lth downsampling unit is connected to the output of the (L-1)th downsampling unit. The output of the downsampling layer of the Lth downsampling unit is connected to the second output layer. The decoder consists of an average pooling layer and a fully connected layer connected in sequence. The input of the average pooling layer is connected to the output of the encoder, and the output of the fully connected layer is connected to the first output layer.

[0011] Preferably, the physical prior constraints include the abundance nonnegativity constraint and the abundance summation constraint of 1; the abundance nonnegativity constraint is implemented through the ReLU activation function of the ReLU activation function layer; the abundance summation constraint of 1 is implemented by setting the softmax normalization function in the output layer of the encoder.

[0012] Preferably, the method further includes: The estimated mineral proportion data in the coal sample is coupled with the spectral data to reconstruct a reconstructed hyperspectral image. The similarity between the reconstructed hyperspectral image and the original coal sample hyperspectral image is measured by a self-supervised learning algorithm, and a 3D convolutional neural network model based on an encoder-decoder structure is trained to form a closed-loop optimization system.

[0013] As a preferred option, the specific formula for reconstructing the hyperspectral image is:

[0014] in, This indicates the reconstruction of the hyperspectral image. Indicates the first coal sample Proportional data of various minerals Indicates the first coal sample Pure material spectral data of the mineral, This indicates the quantity of minerals in a coal sample.

[0015] Preferably, the total loss function used during training of the 3D convolutional neural network model based on the encoder-decoder structure is:

[0016]

[0017] in, Represents the total loss function. Indicates the predicted loss based on mineral proportions. Indicates the reconstruction loss. The weighting coefficients representing the loss in mineral proportion prediction. The weighting coefficients representing the reconstruction loss. This represents a hyperspectral image of a coal sample. This represents the reconstructed hyperspectral image of the coal sample. This represents the structural similarity index obtained through similarity measurement. This represents the weighting coefficient corresponding to the structural similarity index.

[0018] The beneficial effects of this invention are: This invention achieves accurate estimation of the mineral proportion in coal by combining hyperspectral images of coal with a mineral spectral database and a 3D convolutional neural network model based on an encoder-decoder structure, effectively improving detection accuracy. Through coal quality parameter-spectral coupling, the interpretability and reliability of coal quality prediction are enhanced. Furthermore, by integrating a similarity metric between the reconstructed hyperspectral image and the original coal sample hyperspectral image using self-supervised learning to train the 3D convolutional neural network model based on the encoder-decoder structure, the robustness and generalization ability of the model under multiple coal types and scenarios are significantly improved, making it suitable for online real-time monitoring.

[0019] Secondly, a system for estimating the proportion of mineral components in coal includes: The first module is used to acquire the hyperspectral image of the original coal sample and preprocess the hyperspectral image of the original coal sample to obtain the preprocessed hyperspectral image of the coal sample. The second module is used to determine the types and proportions of minerals in coal samples through coal quality testing methods, and to collect spectral data of each mineral in the coal sample using a hyperspectral imager, thereby establishing a coal mineral spectral database. The third module is used to construct a 3D convolutional neural network model based on an encoder-decoder structure, based on the preprocessed hyperspectral images of coal samples and the spectral database of minerals in coal. The fourth module is used to input the hyperspectral image of the original coal sample to be detected into a 3D convolutional neural network model based on an encoder-decoder structure for spatial-spectral feature fusion, and to introduce physical prior constraints to estimate the mineral proportion data in the coal sample.

[0020] Thirdly, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to perform the method as described in the first aspect. Attached Figure Description

[0021] Figure 1 The diagram shows a flowchart of a method for estimating the proportion of mineral components in coal.

[0022] Figure 2 The diagram shown is a network model diagram in spectral unmixing. Detailed Implementation

[0023] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, and are not intended to limit the scope of the invention.

[0024] Example 1: like Figure 1 As shown, a method for estimating the proportion of mineral components in coal includes the following steps: S1. Acquire the original hyperspectral image of the coal sample and preprocess the original hyperspectral image of the coal sample to obtain the preprocessed hyperspectral image of the coal sample; Representative samples of various coal types, including lignite, bituminous coal, and anthracite, were selected. The coal samples were crushed and sieved to a particle size of less than 0.5 mm to ensure uniform particle size. A hyperspectral imager was used to scan and acquire the coal samples under a stable light source. The spectral range covered the near-infrared band of 900-1700 nm, with a spectral resolution of 2-5 nm and a spatial resolution of 0.1 mm.

[0025] During the data acquisition process, standard white and black boards were used for radiometric and dark current corrections to eliminate the effects of ambient light variations and system noise. The correction formula is as follows:

[0026] in, This represents the corrected reflectance. Represents the raw spectral data. Indicates dark current. This indicates the reflectivity of the whiteboard.

[0027] Further preprocessing of the acquired hyperspectral images includes, but is not limited to, steps such as: Savitzky-Golay filter denoising, normalizing spectral data to the 0-1 range, using polynomial fitting to correct baseline drift, and using Gaussian filtering to smooth spectral curves, thereby improving signal-to-noise ratio and spectral consistency.

[0028] S2. Determine the types and proportions of minerals in coal samples through coal quality testing methods, and use a hyperspectral imager to collect spectral data of each mineral in the coal sample, thereby establishing a coal mineral spectral database; In the mineral spectral data acquisition stage, coal samples underwent grinding and drying pretreatment. The samples were ground to below 200 mesh using a ball mill and dried at 105℃ for 2 hours to remove moisture. X-ray fluorescence analysis (XRF) was used to determine the contents of major elements such as Si, Al, Fe, Ca, Mg, K, and Na in the coal samples. X-ray diffraction analysis (XRD) was used to identify mineral types, mainly including clay minerals (kaolinite, illite, montmorillonite), carbonate minerals (calcite, dolomite, siderite), sulfide minerals (pyrite), and other minerals such as quartz and feldspar. Based on the elemental contents and mineral types, the relationship between the spectral variations of coal samples with different mineral proportions and the reflectance spectra of a single mineral was analyzed, establishing a mineral spectral database for coal.

[0029] S3. Based on the preprocessed hyperspectral images of coal samples and the spectral database of minerals in coal, a 3D convolutional neural network model based on an encoder-decoder structure is constructed. S4. The hyperspectral image of the original coal sample to be tested is input into a 3D convolutional neural network model based on an encoder-decoder structure for spatial-spectral feature fusion. Physical prior constraints are introduced to estimate the mineral proportion data in the coal sample. The encoder of the 3D convolutional neural network model based on the encoder-decoder structure is used to extract spatial-spectral features, and the decoder is used to reconstruct the mineral abundance map. Physical prior constraints are introduced, including the non-negativity of abundance constraints and the constraint that the sum of abundances is 1. The network is trained using the test results as supervised labels to establish the mapping relationship between spectral features and mineral proportions, thereby realizing the estimation of the proportion of different minerals in the coal sample.

[0030] like Figure 2 As shown, the 3D convolutional neural network model based on the encoder-decoder structure includes an encoder, a decoder, and a first output layer connected in sequence. The encoder is used to receive the hyperspectral image of the raw coal sample to be detected and extract its spatial-spectral features; Decoder, used to receive spatial-spectral features and reconstruct mineral abundance maps; The first output layer is used to receive the reconstructed mineral abundance map and obtain the abundance content of different minerals in each pixel of the original coal sample hyperspectral image through regression analysis, and output the spatial distribution map of mineral component proportions. The encoder comprises M stacked 3D convolutional units, a ReLU activation function layer, L stacked downsampling units, and a second output layer. Each 3D convolutional unit includes a first 3D convolutional layer and a 3D batch normalization layer connected in sequence. Each downsampling unit includes a second 3D convolutional layer and a downsampling layer connected in sequence. The input of the first 3D convolutional layer of the first 3D convolutional unit is the input of the encoder. The input of the first 3D convolutional layer of the Mth 3D convolutional unit is connected to the output of the (M-1)th 3D convolutional unit, and the output of the 3D batch normalization layer of the Mth 3D convolutional unit is connected to the input of the ReLU activation function layer. The input of the second 3D convolutional layer of the first downsampling unit is connected to the output of the ReLU activation function layer. The input of the second 3D convolutional layer of the Lth downsampling unit is connected to the output of the (L-1)th downsampling unit, and the output of the downsampling layer of the Lth downsampling unit is connected to the second output layer. The decoder comprises an average pooling layer and a fully connected layer connected in sequence. The input of the average pooling layer is connected to the output of the encoder, and the output of the fully connected layer is connected to the first output layer. The input layer of the 3D convolutional neural network model based on an encoder-decoder structure receives a preprocessed hyperspectral image of a coal sample. The data dimension is H×W×B (height×width×number of bands), where H and W are the spatial dimensions of the image, and B is the number of spectral bands. The encoder part includes multiple 3D convolutional layers. The first 3D convolutional layer uses a 3×3×3 convolutional kernel, extracting features in both the spatial and spectral dimensions to capture joint spatial-spectral information. Each 3D convolutional layer is followed by a 3D batch normalization layer to normalize the convolutional output, accelerating network convergence and improving training stability. A ReLU activation function layer is then connected to introduce a nonlinear transformation, enhancing the network's expressive power while ensuring the output value is non-negative. Physical prior constraints are introduced, including using the ReLU activation function to ensure non-negativity of abundance and adding softmax normalization to the encoder's output layer to ensure the abundance sum is 1.

[0031] The encoder's downsampling layer reduces the spatial resolution of the feature map through pooling operations or stride convolutions, thereby reducing computational cost and expanding the receptive field. The network employs a combination of multi-layer 3D convolutions and downsampling to progressively extract and abstract spatial-spectral features, forming a compact high-dimensional feature representation. The decoder first performs global average pooling on the features output by the encoder through an average pooling layer to generate a fixed-length feature vector. Then, a fully connected layer maps the pooled features to the dimension of mineral abundance, establishing a non-linear mapping relationship between features and mineral abundance.

[0032] The network output layer outputs the abundance content of different minerals in each pixel through regression analysis, forming a spatial distribution map of mineral component proportions. The entire network structure makes full use of the three-dimensional characteristics of hyperspectral data, and can simultaneously handle spatial correlation and spectral correlation. Through an end-to-end learning approach, it achieves a direct mapping from hyperspectral images to mineral component proportions, providing an effective technical means for accurately estimating the mineral component proportions in coal.

[0033] In this embodiment, the method for estimating the proportion of mineral components in coal further includes the following steps: Based on the nonnegativity constraint of abundance and the condition that the sum of abundances is 1, the estimated mineral proportion data in the coal sample is coupled with the spectral data for reconstruction to obtain the reconstructed hyperspectral image. The specific formula is as follows:

[0034] in, This indicates the reconstruction of the hyperspectral image. Indicates the first coal sample Proportional data of various minerals Indicates the first coal sample Pure material spectral data of the mineral, This indicates the quantity of minerals in a coal sample.

[0035] Specifically, the predicted proportions of the five substances by the proportional encoder are multiplied pixel by pixel with their corresponding standard spectral data (dimension [1,215]), and the results are added together to generate a reconstructed hyperspectral image with the same dimension as the original coal sample hyperspectral image. The low-dimensional features generated at this time, namely the reconstructed hyperspectral image, are used as input for self-supervised learning optimization.

[0036] A self-supervised learning algorithm is used to measure the similarity between the reconstructed hyperspectral image and the original coal sample hyperspectral image, and a 3D convolutional neural network model based on an encoder-decoder structure is trained. Alternating between supervised and self-supervised learning forms a closed-loop optimization system. The total loss function used during training of the encoder-decoder-based 3D convolutional neural network model is:

[0037]

[0038] in, Represents the total loss function. Indicates the predicted loss based on mineral proportions. Represents the reconstruction loss function. The weighting coefficients representing the loss in mineral proportion prediction. The weighting coefficients representing the reconstruction loss. This represents a hyperspectral image of a coal sample. This represents the reconstructed hyperspectral image of the coal sample. This represents the structural similarity index obtained through similarity measurement. This represents the weight coefficient corresponding to the structural similarity index. The reconstruction loss function combines the mean squared error and the structural similarity index. This process is added as a constraint to the model training, forming a closed-loop optimization system that improves the model's generalization ability and adaptability.

[0039] Experimental verification used 200 coal samples, 150 for training and 50 for testing. Results showed that the mean absolute error (MAE) of mineral component estimation was <2%, and the coefficient of determination (R²) was >0.95. The method of this invention was applied to an online laboratory testing platform, which acquires hyperspectral images of coal samples in real time and outputs predicted results for mineral component proportions and coal quality parameters. Practical application shows that the method of this invention can operate stably, and the predicted results are in good agreement with laboratory testing results, effectively improving the efficiency and accuracy of coal quality monitoring and providing technical support for the intelligent development of the coal industry.

[0040] Example 2: Based on Example 1, this embodiment of the invention provides a coal mineral component ratio estimation system, which can be used to implement the coal mineral component ratio estimation method as described in the foregoing embodiments. The system includes: The first module is used to acquire the hyperspectral image of the original coal sample and preprocess the hyperspectral image of the original coal sample to obtain the preprocessed hyperspectral image of the coal sample. The second module is used to determine the types and proportions of minerals in coal samples through coal quality testing methods, and to collect spectral data of each mineral in the coal sample using a hyperspectral imager, thereby establishing a coal mineral spectral database. The third module is used to construct a 3D convolutional neural network model based on an encoder-decoder structure, based on the preprocessed hyperspectral images of coal samples and the spectral database of minerals in coal. The fourth module is used to input the hyperspectral image of the original coal sample to be detected into a 3D convolutional neural network model based on an encoder-decoder structure for spatial-spectral feature fusion, and to introduce physical prior constraints to estimate the mineral proportion data in the coal sample.

[0041] According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0042] In an exemplary embodiment, the electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method for estimating the proportion of mineral components in coal as described in Embodiment 1 above.

[0043] In an exemplary embodiment, the readable storage medium may be a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method for estimating the proportion of mineral components in coal according to Embodiment 1 above.

[0044] In an exemplary embodiment, the computer program product includes a computer program that, when executed by a processor, implements the method for estimating the proportion of mineral components in coal according to Embodiment 1 above.

[0045] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0046] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0047] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0048] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0049] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0050] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method of estimating proportions of mineral components in coal, characterized by, The method comprises the following steps: Collecting a hyperspectral image of a raw coal sample, and pre-processing the hyperspectral image of the raw coal sample to obtain a pre-processed hyperspectral image of the coal sample; Determining the types and proportions of minerals in the coal sample by a coal quality testing method, and collecting spectral data of each mineral in the coal sample by using a hyperspectral imager, and then establishing a mineral spectral database in the coal; Based on the pre-processed hyperspectral image of the coal sample and the mineral spectral database in the coal, a 3D convolutional neural network model based on an encoder-decoder structure is constructed; The original hyperspectral image of the coal sample to be detected is input into the 3D convolutional neural network model based on the encoder-decoder structure for spatial-spectral feature fusion, and a physical prior constraint is introduced to estimate the mineral proportion data in the coal sample.

2. The method of estimating the proportion of mineral matter in coal according to claim 1, characterized by, When collecting the hyperspectral image of the raw coal sample, a standard white board and a black board are used for radiation correction and dark current correction, and the correction formula is: wherein, represents the corrected reflectance, represents the original spectral data, represents the dark current, represents the whiteboard reflectance.

3. The method of estimating the proportion of mineral matter in coal according to claim 1, characterized by, The types and proportions of minerals in the coal sample are determined by a coal quality testing method, and the spectral data of each mineral in the coal sample are collected by using a hyperspectral imager, and then a mineral spectral database in the coal is established, which specifically comprises the following steps: The coal sample is ground to 200 mesh or less by using a ball mill, and dried at 105 DEG C for 2 hours to remove water, to obtain a treated coal sample; The elemental content of the treated coal sample is determined by X-ray fluorescence analysis; The mineral types in the treated coal sample are identified by X-ray diffraction analysis; The spectral data of each mineral in the coal sample are collected by using a hyperspectral imager; Based on the elemental content, mineral types and spectral data, the change relationship between the spectral of different mineral proportion coal samples and the reflection spectrum of single mineral is analyzed, and the mineral spectral database in the coal is established.

4. The method of estimating the proportion of mineral matter in coal according to claim 1, characterized by, The 3D convolutional neural network model based on the encoder-decoder structure comprises an encoder, a decoder and a first output layer connected in sequence; The encoder is used for receiving the original hyperspectral image of the coal sample to be detected and extracting spatial-spectral features; The decoder is used for receiving the spatial-spectral features and reconstructing a mineral abundance map; The first output layer is used for receiving the reconstructed mineral abundance map, and the abundance content of different minerals in each pixel of the original hyperspectral image of the coal sample is obtained by regression analysis, and a spatial distribution map of the mineral component proportion is output; The encoder comprises M stacked 3D convolution units, a ReLU activation function layer, L stacked down-sampling units and a second output layer; the 3D convolution unit comprises a first 3D convolution layer and a 3D batch normalization layer connected in sequence; the down-sampling unit comprises a second 3D convolution layer and a down-sampling layer connected in sequence; the input end of the first 3D convolution layer of the first 3D convolution unit is the input end of the encoder, the input end of the first 3D convolution layer of the Mth 3D convolution unit is connected to the output end of the M-1th 3D convolution unit, and the output end of the 3D batch normalization layer of the Mth 3D convolution unit is connected to the input end of the ReLU activation function layer; the input end of the second 3D convolution layer of the first down-sampling unit is connected to the output end of the ReLU activation function layer, the input end of the second 3D convolution layer of the Lth down-sampling unit is connected to the output end of the L-1th down-sampling unit, and the output end of the down-sampling layer of the Lth down-sampling unit is connected to the second output layer; The decoder comprises a mean pooling layer and a full connection layer connected in sequence, an input end of the mean pooling layer is connected with an output end of the encoder, and an output end of the full connection layer is connected with the first output layer.

5. The method of estimating the proportion of mineral matter in coal according to claim 4, characterized by, The physical prior constraint comprises an abundance non-negativity constraint and an abundance sum-to-one constraint; the abundance non-negativity constraint is realized by a ReLU activation function of a ReLU activation function layer; and the abundance sum-to-one constraint is realized by setting a softmax normalization function in an output layer of the encoder.

6. The method of estimating the proportion of mineral matter in coal according to claim 1, characterized by, The method further comprises: coupling and reconstructing the estimated mineral proportion data in the coal sample with the spectral data to obtain a reconstructed hyperspectral image; performing similarity measurement on the reconstructed hyperspectral image and the original coal sample hyperspectral image by a self-supervised learning algorithm, and training the 3D convolutional neural network model based on the encoder-decoder structure to form a closed-loop optimization system.

7. The method according to claim 6, characterized by, The specific formula of the reconstructed hyperspectral image is: wherein, represents a reconstructed hyperspectral image, represents the proportion data of the th mineral in the coal sample, represents the pure substance spectral data of the th mineral in the coal sample, represents the number of minerals in the coal sample.

8. The method according to claim 6, characterized by, When training the 3D convolutional neural network model based on the encoder-decoder structure, a total loss function used is: wherein, denotes the total loss function, denotes the mineral proportion prediction loss, denotes the reconstruction loss, denotes the weight coefficient of the mineral proportion prediction loss, denotes the weight coefficient of the reconstruction loss, denotes the hyperspectral image of the coal sample, denotes the reconstructed hyperspectral image of the coal sample, denotes the structural similarity index obtained by the similarity measure, denotes the weight coefficient corresponding to the structural similarity index.

9. A system for estimating proportions of mineral components in coal, characterized by comprising: comprises: a first module configured to collect an original coal sample hyperspectral image, and pre-process the original coal sample hyperspectral image to obtain a pre-processed coal sample hyperspectral image; a second module configured to determine the types and proportions of minerals in the coal sample by a coal quality testing method, and collect spectral data of each mineral in the coal sample by using a hyperspectral imager, and further establish a coal mineral spectral database; a third module configured to construct a 3D convolutional neural network model based on the encoder-decoder structure based on the pre-processed coal sample hyperspectral image and the coal mineral spectral database; a fourth module configured to input an original coal sample hyperspectral image to be detected into the 3D convolutional neural network model based on the encoder-decoder structure for spatial-spectral feature fusion, and introduce a physical prior constraint to estimate mineral proportion data in the coal sample.

10. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to make the computer execute the method according to any one of claims 1-8.

Images