Coal and rock microscopic component identification method and system based on channel-level image fusion

Abstract

The present invention provides a method and system for identifying coal rock microscopic components based on channel-level image fusion, which relates to the technical field of coal rock identification, including: obtaining oil-immersed reflected light coal rock microscopic images and fluorescent coal rock microscopic images; defining target pixels, obtaining pixel blocks similar to the target pixels in the oil-immersed reflected light coal rock microscopic images and the fluorescent coal rock microscopic images, and performing weighted averaging on similar pixel blocks by a non-local mean filtering method; then dividing the oil-immersed reflected light coal rock microscopic images and the fluorescent coal rock microscopic images into a plurality of non-overlapping sub-regions, performing histogram equalization on each sub-region, and enhancing the pixel contrast in the sub-region; splicing and fusing all the plurality of sub-regions at the channel level to form an overall fused image; inputting the fused image into a microscopic component identification model based on an improved Mask-RCNN network to achieve accurate segmentation of microscopic image components; the present invention effectively avoids the loss of information under a single modality and improves the accuracy of microscopic component identification.

Description

Technical Field

[0001] The present disclosure relates to the technical field of coal rock identification, and in particular to a method and system for identifying coal rock microscopic components based on channel-level image fusion. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Coal rock microscopic component identification is of great significance in the fields of coal mining, coalbed methane development, coal chemical industry, etc. It can determine the industrial use of coal, understand the coal-forming environment of coal, and guide the coal washing and processing process. In addition, microscopic component identification also helps the safety management of coal mines and prevents potential risks of spontaneous combustion or explosion in advance, thereby improving the efficiency and safety of coal resource development.

[0004] Traditional optical identification methods require researchers to use a microscope to identify the microscopic components of coal based on its transmittance, reflectivity, morphology, protrusions and other characteristics. Reflected light microscope analysis of coal slices is one of the most widely used methods for analyzing the microscopic components of coal rocks because it is applicable to a wide range of coal grades. In addition, the use of fluorescence microscopes to observe coal slices is particularly suitable for distinguishing the exinite and other components in lignite and low-rank coal. However, manual identification not only requires a solid foundation in coal petrology, but its results are also easily affected by the personal experience of the operator, which limits the widespread application of coal rock microscopic component identification in the field of modern coal chemical industry.

[0005] With the continuous development of artificial intelligence technology, the analysis and identification of coal and rock microscopic components are also continuously improving and optimizing. In recent years, computer vision technology has become a hot research direction in this field. The application of image clustering methods and machine learning models has achieved certain results in improving the degree of recognition automation, but it also faces some challenges. To this end, researchers have further introduced deep learning methods such as image segmentation, which has significantly improved the accuracy and application prospects of coal and rock microscopic component identification.

[0006] Existing image segmentation technology can segment coal and rock microscopic images into multiple pixel blocks, each of which corresponds to a different microscopic component. Through this method, microscopic components such as vitrinite, inertinite, and exinite can be accurately identified, achieving high-precision identification. However, there is still a lack of standardized coal and rock microscopic image datasets, as well as deep learning models that can simultaneously process different types of microscopic images. In addition, the intelligent identification method for microscopic components of a wider range of coal grades also needs to be improved, especially in terms of accuracy and processing speed. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes a method and system for identifying coal and rock microscopic components based on channel-level image fusion. The channel-level fusion method is adopted to perform channel-level splicing and fusion of different types of microscopic images (oil-immersed reflected light and fluorescence images) so that information under the two modes can interact. The improved Mask-RCNN network is used to realize fast and intelligent recognition, thereby realizing the rapid acquisition, analysis and archiving of coal and rock microscopic components.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] The method for identifying coal and rock microscopic components based on channel-level image fusion includes:

[0010] Obtain oil-immersion reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images;

[0011] Define the target pixel, obtain the pixel blocks similar to the target pixel in the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image, and perform weighted averaging on the similar pixel blocks by using the non-local mean filtering method; then divide the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image into multiple non-overlapping sub-regions, perform histogram equalization on each sub-region, and enhance the pixel contrast in the sub-region;

[0012] All the multiple sub-regions are spliced ​​and fused at the channel level to form an overall fused image;

[0013] The fused image is input into the microscopic component recognition model based on the improved Mask-RCNN network. By introducing feature pyramid and spatial attention mechanism in the backbone network, image features are extracted and ROI feature maps are generated. The fully connected layer and mask branch are used to generate ROI categories and segmentation masks to achieve accurate segmentation of microscopic image components.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions:

[0015] The coal and rock microscopic component recognition system based on channel-level image fusion includes:

[0016] A data acquisition module, used for acquiring oil-immersed reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images;

[0017] The image processing module is used to define the target pixel, obtain the pixel blocks of the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image that are similar to the target pixel, and perform weighted averaging on the similar pixel blocks by using the non-local mean filtering method; then divide the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image into multiple non-overlapping sub-regions, perform histogram equalization on each sub-region, and enhance the pixel contrast in the sub-region;

[0018] The microscopic component recognition module is used to stitch and fuse all the multiple sub-regions at the channel level to form an overall fused image; the fused image is input into the microscopic component recognition model based on the improved Mask-RCNN network, and the image features are extracted and the ROI feature map is generated by introducing the feature pyramid and spatial attention mechanism in the backbone network. The fully connected layer and mask branch are used to generate the ROI category and segmentation mask to achieve accurate segmentation of the microscopic image components.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for identifying coal and rock microscopic components based on channel-level image fusion is implemented.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the method for identifying coal and rock microscopic components based on channel-level image fusion.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The present invention discloses a method for identifying coal rock microscopic components based on channel-level image fusion. The method defines target pixels through an image fusion method, obtains pixel blocks similar to the target pixels in oil-immersed reflected light coal rock microscopic images and fluorescent coal rock microscopic images, and performs weighted averaging on similar pixel blocks through a non-local mean filtering method; then the oil-immersed reflected light coal rock microscopic images and fluorescent coal rock microscopic images are divided into multiple non-overlapping sub-regions, and each sub-region is subjected to histogram equalization to enhance the pixel contrast within the sub-region; all the multiple sub-regions are spliced ​​and fused according to the channel level to splice coal rock microscopic images of different modes, thereby achieving full information interaction. The method effectively avoids the loss of information under a single mode and significantly improves the accuracy of microscopic component identification.

[0025] The present invention discloses a method for identifying the microscopic components of coal and rock based on channel-level image fusion, which uses a deep learning model to analyze the fused image and can accurately identify the microscopic components of coal and rock, including vitrinite, inertinite, exinite, and background resin, etc. This intelligent identification method not only improves the efficiency of identification, but also reduces the errors caused by human intervention.

[0026] The present invention discloses a method for identifying coal and rock microscopic components based on channel-level image fusion, which integrates a data storage and management module and a user interaction and visualization module, can automatically save identification results, and provides convenient data management, retrieval and export functions, greatly facilitating subsequent analysis and research. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0028] Figure 1 This is a flow chart of a method for identifying coal and rock microscopic components based on channel-level image fusion according to an embodiment of the present disclosure;

[0029] Figure 2 A diagram of an intelligent identification system for coal and rock microscopic components according to an embodiment of the present disclosure;

[0030] Figure 3 Schematic diagram of the spatial attention structure of deep learning in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0033] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0034] Example 1

[0035] In one embodiment of the present disclosure, a method for identifying coal and rock microscopic components based on channel-level image fusion is provided, comprising:

[0036] Step 1: Obtain oil-immersion reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images;

[0037] Step 2: define the target pixel, obtain the pixel blocks similar to the target pixel in the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image, and perform weighted averaging on the similar pixel blocks by the non-local mean filtering method; then divide the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image into multiple non-overlapping sub-regions, perform histogram equalization on each sub-region, and enhance the pixel contrast in the sub-region;

[0038] Step 3: All the sub-regions are spliced ​​and fused at the channel level to form an overall fused image;

[0039] Step 4: Input the fused image into the microscopic component recognition model based on the improved Mask-RCNN network. By introducing the feature pyramid and spatial attention mechanism in the backbone network, the image features are extracted and the ROI feature map is generated. The fully connected layer and mask branch are used to generate the ROI category and segmentation mask to achieve accurate segmentation of the microscopic image components.

[0040] As an embodiment, the specific implementation process of a method for identifying coal and rock microscopic components based on channel-level image fusion disclosed in the present invention is as follows:

[0041] Step 1: Obtain oil-immersed reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images, and improve the preprocessing process of the oil-immersed reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images. The pixel blocks in the image similar to the target pixel are weighted averaged by the non-local mean filtering method to remove noise, and the image is enhanced by contrast-limited adaptive histogram equalization method.

[0042] Specifically, 1) the noise is removed by weighted averaging pixel blocks similar to the target pixel in the image using a non-local mean filtering method, including: defining the target pixel, obtaining pixel blocks similar to the target pixel in the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image, as follows:

[0043] (1) For any pixel in a given image p , defines a block containing the pixel , the center of which is the target pixel p .

[0044] (2) Find the corresponding All blocks similar to , these blocks may be located at different positions in the image. The similarity between blocks is usually measured by the Euclidean distance, which is as follows:

[0045]

[0046] in, Representation Block The starting coordinates in the image, Representation Block The starting coordinates in the image; i Represents a pixel offset within the block, relative to the starting coordinates of the block (i.e. p or q ) and Represents blocks and The pixel value within Indicates the size of the block, that is, the number of pixels in the block.

[0047] (3) Calculate the target pixel p With each similar block The weight between ( p , q ), the calculation of the weight value depends on the distance between blocks:

[0048] =

[0049] in, h It is a parameter that controls the filtering strength. The larger the weight, the more similar the two blocks are.

[0050] (4) Finally, the target pixel p The denoising result is obtained by weighted averaging the central pixel values ​​of all similar blocks:

[0051]

[0052] Further, 2) the image is enhanced contrast processed by contrast limited adaptive histogram equalization method, including: dividing the oil-immersed reflected light coal rock microscopic image and the fluorescent coal rock microscopic image into multiple non-overlapping sub-regions, performing histogram equalization on each sub-region, dividing each image into non-overlapping sub-regions, performing histogram equalization on each sub-region separately, calculating its grayscale histogram, and calculating the cumulative distribution function based on the histogram, and setting a contrast limit parameter clip_limit, when the pixel frequency of a certain gray level exceeds the contrast limit parameter, it is truncated, and the excess frequency value is evenly distributed to other gray levels, and then the truncated total frequency T is evenly distributed to all gray levels. The details are as follows:

[0053] (1) The image is divided into non-overlapping sub-regions (such as 8×8 or smaller blocks), and histogram equalization is performed on each sub-region separately. The pixel contrast in each sub-region is locally enhanced.

[0054] (2) For each sub-region, calculate its grayscale histogram , and calculate the cumulative distribution function based on the histogram .

[0055] (3) To prevent excessive contrast enhancement, a contrast limit parameter clip_limit is set. When the limit is exceeded, it is truncated and the excess frequency values ​​are evenly distributed to other gray levels. This avoids the phenomenon of excessive contrast in some areas during the equalization process:

[0056]

[0057] Then, the total frequency will be truncated T Distribute evenly to all gray levels:

[0058]

[0059] in L is the total number of gray levels.

[0060] (4) For each sub-region, according to the adjusted cumulative distribution function For grayscale value mapping, the formula is similar to the standard histogram equalization:

[0061]

[0062] (5) In order to avoid boundary effects between sub-regions, bilinear interpolation is performed on the grayscale value of each pixel. This step ensures smooth transitions between different sub-regions of the image and eliminates possible boundary artifacts.

[0063] Furthermore, all the multiple sub-regions are spliced ​​and fused at the RGB channel level to form an overall fused image, and the image is manually cropped to remove the image portion containing minerals and scratches.

[0064] Specifically, using discrete wavelet transform (DWT) to fuse different types of microscopic images (such as oil-immersion reflected light and fluorescence images) can effectively extract image edge, texture and other detail features, thereby processing coal and rock microscopic images with complex textures. The steps are as follows:

[0065] (1) Perform wavelet transform on the source image and decompose the image into low-frequency (smooth) and high-frequency (detail) sub-bands.

[0066]

[0067] in, represents the source image; LL Represents the low-frequency subband, that is, the smooth part of the image; LH Represents the horizontal high-frequency subband, that is, the detail information in the horizontal direction of the image; HL Represents the vertical high-frequency subband, that is, the detail information in the vertical direction of the image; HH Represents the diagonal high-frequency subband, that is, the detail information in the oblique direction (diagonal line) of the image.

[0068] (2) Each sub-band is fused, usually by weighted averaging the low-frequency components and selecting a large value for the high-frequency components.

[0069]

[0070] in, Represents the low-frequency component after fusion; and Respectively represent the wavelet low-frequency subband components of the two source images; and Represents the weight coefficient, which is used to control the proportion of the low-frequency components of the two images in the fusion result. .

[0071] For the high frequency part:

[0072]

[0073] in, represents the horizontal high-frequency component after fusion, and Respectively represent the two images in The absolute value of the horizontal high-frequency component of the position.

[0074] Similarly, for HL and HH The same is true.

[0075] (3) Perform inverse wavelet transform on the fused sub-bands to reconstruct the fused image.

[0076] .

[0077] in, Represents the fused image, which is obtained by combining the fused low-frequency and high-frequency components Perform inverse wavelet transform to obtain; represents the inverse discrete wavelet transform, which reassembles the fused low-frequency and high-frequency sub-bands back to the original image size to obtain the final fused image.

[0078] Step 2: Input the fused image into the microscopic component recognition model based on the improved Mask-RCNN network. By introducing the feature pyramid and spatial attention mechanism in the backbone network, the image features are extracted and the ROI feature map is generated. The fully connected layer and mask branch are used to generate the ROI category and segmentation mask to achieve accurate segmentation of the microscopic image components.

[0079] Specifically, the microscopic component recognition model adopts an improved Mask-RCNN algorithm, which is mainly composed of the following four parts: improved backbone network, region proposal network (RPN), ROI Align, and ROI segmentation and recognition module.

[0080] Furthermore, the improved backbone network consists of ResNet101 and Feature Pyramid Network (FPN). ResNet101 uses two residual network modules, Conv Block and Identity Block, to adjust the image dimension and network depth to extract image features and generate feature maps. FPN converts these feature layers into effective feature maps (such as , , , , ), complete the feature extraction.

[0081] In this improved backbone network, a spatial attention mechanism is introduced in the 2nd and 5th layers of the downsampling structure to control the complexity of the FPN structure and enhance its learning ability, thereby improving the accuracy of segmentation and recognition and reducing errors. The spatial attention mechanism converts the effective feature map ( H * W * C ) as input, and the spatial weight coefficients are generated through maximum pooling, average pooling, channel concatenation and convolution operations. Finally, these weight coefficients are multiplied by the input feature layer to obtain the enhanced feature map ( H * W *1).

[0082]

[0083]

[0084] in F Representation feature layer; represents average pooling; represents maximum pooling; Indicates the filter size; Indicates the size of the activation function; represents the attention parameter matrix.

[0085] Furthermore, the ROI region is obtained by sliding the RPN on the effective feature map. The RPN outputs the thin film component region by calculating the IOU value between the prior box and the real box. The RPN consists of a binary classification network and a regression network. The binary classification network is responsible for detecting whether the candidate region contains thin film components, while the regression network outputs the bounding boxes of these candidate regions. If the region detected by the binary classification network does not contain a component, the corresponding bounding box will be considered invalid.

[0086] In the ROI segmentation and identification module of microscopic image components, each ROI is first adjusted to the same size through ROI Align. Subsequently, the fully connected layer and mask branch are used to generate the ROI category and segmentation mask. The fully connected layer takes each adjusted ROI as input, identifies the component category of the microscopic image, and regresses and optimizes the bounding box of each component. The mask branch generates a pixel-level target mask for each ROI through a small fully connected network, thereby achieving accurate segmentation of microscopic image components.

[0087] The specific steps are as follows:

[0088] (1) The fused coal rock microscopic images under oil-immersion reflected light and fluorescence are sent to the microscopic component recognition module for feature extraction and generation of feature layers.

[0089] (2) FPN converts these feature layers into effective feature layers through downsampling, upsampling and maximum pooling operations to complete feature extraction.

[0090] (3) By introducing the spatial attention mechanism in the 2nd and 5th layers of the downsampling structure, the complexity of the FPN structure is controlled and its learning ability is enhanced, thereby obtaining an enhanced feature layer.

[0091] (4) Use RPN to slide on the effective feature layer to generate ROI.

[0092] (5) Use ROI Align to adjust each ROI to the same size.

[0093] (6) Use fully connected layers and mask branches to generate ROI category and pixel-level object masks to achieve accurate segmentation of microscopic image components.

[0094] Furthermore, the recognition results are automatically saved to the coal and rock microscopic image dataset, and data management, retrieval and export functions are provided to facilitate subsequent analysis and research;

[0095] It provides a user-friendly interface, displays recognition results and analysis data, and supports users' visual operation and analysis of images and data.

[0096] Example 2

[0097] In one embodiment of the present disclosure, a coal rock microscopic component identification system based on channel-level image fusion is provided, comprising:

[0098] A data acquisition module, used for acquiring oil-immersed reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images;

[0099] The image processing module is used to define the target pixel, obtain the pixel blocks of the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image that are similar to the target pixel, and perform weighted averaging on the similar pixel blocks by using the non-local mean filtering method; then divide the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image into multiple non-overlapping sub-regions, perform histogram equalization on each sub-region, and enhance the pixel contrast in the sub-region;

[0100] The microscopic component recognition module is used to stitch and fuse all the multiple sub-regions at the channel level to form an overall fused image; the fused image is input into the microscopic component recognition model based on the improved Mask-RCNN network, and the image features are extracted and the ROI feature map is generated by introducing the feature pyramid and spatial attention mechanism in the backbone network. The fully connected layer and mask branch are used to generate the ROI category and segmentation mask to achieve accurate segmentation of the microscopic image components.

[0101] It also includes a data storage and management module for automatically saving the identification results to the coal and rock microscopic image data set, and provides data management, retrieval and export functions to facilitate subsequent analysis and research;

[0102] And the user interaction and visualization module is used to provide a user-friendly interface, display recognition results and analysis data, and support users' visualization operations and analysis of images and data.

[0103] Example 3

[0104] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for identifying coal and rock microscopic components based on channel-level image fusion is implemented.

[0105] Example 4

[0106] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the method for identifying coal and rock microscopic components based on channel-level image fusion.

[0107] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0109] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A method for identifying coal and rock microscopic components based on channel-level image fusion, characterized in that: include: Obtain oil-immersion reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images; Define the target pixel, obtain the pixel blocks similar to the target pixel in the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image, and perform weighted averaging on the similar pixel blocks by using the non-local mean filtering method; then divide the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image into multiple non-overlapping sub-regions, perform histogram equalization on each sub-region, and enhance the pixel contrast in the sub-region; All the multiple sub-regions are spliced ​​and fused at the channel level to form an overall fused image; The fused image is input into the microscopic component recognition model based on the improved Mask-RCNN network. By introducing the feature pyramid and spatial attention mechanism into the backbone network, the image features are extracted and the ROI feature map is generated. The fully connected layer and mask branch are used to generate the ROI category and segmentation mask to achieve accurate segmentation of microscopic image components, including: All the multiple sub-regions are spliced ​​and fused at the RGB channel level to form an overall fused image, and the fused image is input into the microscopic component recognition model. The microscopic component recognition module is composed of an improved backbone network, a region proposal network, a ROI Align, and a ROI segmentation and recognition module. The improved backbone network is composed of ResNet101 and a feature pyramid FPN. ResNet101 adjusts the image dimension and network depth through two residual network modules, Conv Block and Identity Block, to extract image features and generate feature maps; In the improved backbone network, the spatial attention mechanism is introduced in the 2nd and 5th layers of the downsampling structure to control the complexity of the feature pyramid FPN structure and enhance its learning ability. The spatial attention mechanism takes the feature map H*W*C as input, generates spatial weight coefficients through maximum pooling, average pooling, channel splicing and convolution operations, and finally multiplies the weight coefficients with the input feature layer to obtain the enhanced feature map. The ROI region is obtained by sliding the RPN on the effective feature map. The RPN outputs the thin film component region by calculating the IOU value between the prior box and the real box. The RPN consists of a binary classification network and a regression network. The binary classification network is responsible for detecting whether the candidate region contains thin film components, while the regression network outputs the bounding boxes of these candidate regions. If the region detected by the binary classification network does not contain a component, the corresponding bounding box will be considered invalid. In the ROI segmentation and recognition module of the microscopic image component, each ROI is first adjusted to the same size through ROI Align. Subsequently, the fully connected layer and mask branch are used to generate the ROI category and segmentation mask. The fully connected layer uses each adjusted ROI as input to identify the component category of the microscopic image and regressively optimize the bounding box of each component. The mask branch generates a pixel-level target mask for each ROI through a small fully connected network, thereby achieving accurate segmentation of the microscopic image components.

2. The method for identifying coal and rock microscopic components based on channel-level image fusion according to claim 1, characterized in that: Define the target pixel and obtain the pixel blocks similar to the target pixel in the oil-immersion reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image, including: for any pixel in a given image p , define a pixel p Block , the center of which is the target pixel p , find the image with All similar blocks , the similarity between blocks is measured by the Euclidean distance: in, and Represents blocks and The pixel value within is the block size.

3. The method for identifying coal and rock microscopic components based on channel-level image fusion according to claim 1, characterized in that: Similar pixel blocks are weighted averaged by non-local mean filtering method, including: Calculate the target pixel p With each similar block The weight between ( p , q ), the calculation of the weight value depends on the distance between blocks: = in, h It is a parameter that controls the filtering strength. The larger the weight, the more similar the two blocks are. p The denoising result is obtained by weighted averaging the central pixel values ​​of all similar blocks: 。 4. The method for identifying coal and rock microscopic components based on channel-level image fusion according to claim 1, characterized in that: The oil-immersed reflected light coal rock microscopic image and the fluorescent coal rock microscopic image are divided into a plurality of non-overlapping sub-regions, and each sub-region is subjected to histogram equalization, including: dividing each image into non-overlapping sub-regions, performing histogram equalization on each sub-region separately, calculating its grayscale histogram, and calculating the cumulative distribution function based on the histogram, and setting a contrast limit parameter clip_limit, when the pixel frequency of a certain gray level exceeds the contrast limit parameter, it is truncated, and the excess frequency value is evenly distributed to other gray levels, and then, the truncated total frequency T is evenly distributed to all gray levels.

5. The method for identifying coal and rock microscopic components based on channel-level image fusion according to claim 4, characterized in that: For each sub-region, grayscale value mapping is performed according to the adjusted cumulative distribution function: Where L is the total number of gray levels, is the cumulative distribution function; And perform bilinear interpolation on the grayscale value of each pixel.

6. A coal-rock microscopic component identification system based on channel-level image fusion, which executes a coal-rock microscopic component identification method based on channel-level image fusion as claimed in any one of claims 1 to 5, characterized in that: include: A data acquisition module, used for acquiring oil-immersed reflected light coal-rock microscopic images and fluorescent coal-rock microscopic images; The image processing module is used to define the target pixel, obtain the pixel blocks of the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image that are similar to the target pixel, and perform weighted averaging on the similar pixel blocks by using the non-local mean filtering method; then divide the oil-immersed reflected light coal-rock microscopic image and the fluorescent coal-rock microscopic image into multiple non-overlapping sub-regions, perform histogram equalization on each sub-region, and enhance the pixel contrast in the sub-region; The microscopic component recognition module is used to stitch and fuse all the multiple sub-regions at the channel level to form an overall fused image; the fused image is input into the microscopic component recognition model based on the improved Mask-RCNN network, and the image features are extracted and the ROI feature map is generated by introducing the feature pyramid and spatial attention mechanism in the backbone network. The fully connected layer and mask branch are used to generate the ROI category and segmentation mask to achieve accurate segmentation of the microscopic image components.

7. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the method for identifying coal and rock microscopic components based on channel-level image fusion as described in any one of claims 1 to 5 is implemented.

8. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the method for identifying the microscopic components of coal and rock based on channel-level image fusion as described in any one of claims 1 to 5.

Images