Method and equipment for identifying particle size uniformity of coal sample particles

Abstract

The invention discloses a method and equipment for identifying particle size uniformity of coal sample particles, and belongs to the technical field of coal quality determination. The identification method comprises the following steps: acquiring baseline features and anchoring features of a coal sample image; calculating a topological parameter between the baseline feature and the anchoring feature based on a preset topological analysis unit; and calculating the particle size uniformity of the coal sample particles in the coal sample image according to the baseline characteristics and the topological parameters. According to the method, the baseline features and the anchoring features in the coal sample image are creatively extracted at the same time, and the topological parameters between the baseline features and the anchoring features are calculated, so that the algorithm is more suitable for the appearance characteristics of the coal particles, and the recognition precision of the uniformity of the coal particles is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal quality determination, and in particular relates to a method and device for identifying the uniformity of particle size of coal samples. Background Art

[0002] Currently, the methods commonly used in the industry to determine particle uniformity rely primarily on visual observation or simple screening. These methods are highly subjective and have poor reproducibility, making them difficult to achieve the sample uniformity required for high-precision testing. While some highly accurate analytical techniques exist for assessing particle size distribution, these methods typically rely on expensive specialized equipment and complex procedures, making them difficult to implement in routine, large-scale coal sampling and analysis processes.

[0003] Therefore, there is an urgent need to develop a coal particle uniformity identification technology that is easy to operate, cost-controlled, and has reliable accuracy, so as to address the significant shortcomings of existing methods in terms of accuracy, efficiency, and cost, and provide a solid sample foundation for high-precision detection of key coal quality and environmental protection indicators. Summary of the Invention

[0004] In view of this, the present invention provides a method and device for identifying the uniformity of particle size of coal samples, which improve the accuracy of identifying the uniformity of coal sample particles while keeping the cost low.

[0005] The technical solution is as follows: A method for identifying the uniformity of particle size of coal samples comprises the following steps:

[0006] Obtaining baseline features and anchor features of coal sample images;

[0007] Based on a preset topological analysis unit, a topological parameter between the baseline feature and the anchor feature is calculated; the topological parameter is used to characterize the spatial correlation relationship and structural mapping relationship between the baseline feature and the anchor feature;

[0008] The particle size uniformity of the coal sample particles in the coal sample image is calculated based on the baseline characteristics and the topological parameters.

[0009] In one embodiment of the present application, obtaining the baseline features of a coal sample image includes: obtaining a coal sample image, extracting the baseline features of the coal sample image based on multiple baseline feature extraction units connected end to end, and a convolution layer is provided inside the baseline feature extraction unit.

[0010] In one embodiment of the present application, the internal calculation process of the reference feature extraction unit includes:

[0011] Acquire data to be processed, perform pre-feature extraction on the data to be processed, and obtain a first basic feature;

[0012] Reducing the spatial dimension of the first basic feature to obtain a second basic feature;

[0013] Alignment processing is performed on the second basic features to obtain third basic features output by the reference feature extraction unit.

[0014] In one embodiment of the present application, performing pre-feature extraction on the data to be processed includes:

[0015] After the data to be processed is processed and activated by the first convolution layer, a first distribution feature is obtained, where the convolution kernel size of the first convolution layer is 1*1;

[0016] After the data to be processed is processed and activated by the second convolutional layer, a second distribution feature is obtained, and the convolution kernel size of the second convolutional layer is 3*3;

[0017] The first distribution feature and the second distribution feature are activated after performing a Hadamard product to obtain a third distribution feature;

[0018] The first distribution feature is subtracted from the data to be processed and then activated to obtain a fourth distribution feature;

[0019] The second distribution feature, the third distribution feature, and the fourth distribution feature are spliced ​​to obtain a fifth distribution feature;

[0020] After the fifth distribution feature is processed and activated by the third convolutional layer, the first basic feature is obtained.

[0021] In one embodiment of the present application, reducing the spatial dimension of the first basic feature includes: processing the first basic feature using a local maximum pooling layer;

[0022] In one embodiment of the present application, performing alignment processing on the second basic features includes:

[0023] After the second basic feature is processed and activated by the deformable convolution layer, the first alignment feature is obtained.

[0024] After the second basic feature is processed and activated by the fourth convolutional layer, the second alignment feature is obtained.

[0025] After the first alignment feature and the second alignment feature are spliced ​​together, a third alignment feature is obtained.

[0026] After the third alignment feature is processed and activated by the fifth convolutional layer, the third basic feature is obtained.

[0027] In one embodiment of the present application, obtaining anchor features of a coal sample image includes: obtaining a coal sample image, and obtaining a first anchor feature after the coal sample image is processed and activated by a sixth convolutional layer; obtaining a second anchor feature after the first anchor feature is processed and activated by a seventh convolutional layer; and obtaining the anchor feature by adding and activating the second anchor feature to the first anchor feature.

[0028] In one embodiment of the present application, the topological parameter between the baseline feature and the anchor feature is calculated based on a preset topological analysis unit, including:

[0029] Integrating the eigenvalues ​​of each channel position of the baseline feature to obtain a baseline vector;

[0030] Integrating the eigenvalues ​​of each spatial position of the anchor feature to obtain an anchor matrix;

[0031] Performing matching processing on the anchor matrix according to the baseline vector to obtain an anchor vector;

[0032] The baseline vector and the anchor vector are integrated to obtain the topological parameters.

[0033] In one embodiment of the present application, integrating the feature values ​​of each channel position of the baseline feature respectively includes: performing global average pooling and then activating the feature values ​​of each channel position of the baseline feature;

[0034] In one embodiment of the present application, integrating the feature values ​​of each spatial position of the anchor feature includes: performing global maximum pooling and then activating the feature values ​​of each spatial position of the anchor feature;

[0035] In one embodiment of the present application, matching processing is performed on the anchor matrix according to the baseline vector, including: expanding the anchor matrix and inputting it into a first fully connected layer, the first fully connected layer calculating and outputting a connection vector having the same length as the baseline vector, adding the connection vector to the baseline vector and activating the resultant to obtain the anchor vector;

[0036] In one embodiment of the present application, integrating the baseline vector with the anchor vector includes: performing a Hadamard product between the baseline vector and the anchor vector and then activating the resultant product.

[0037] In one embodiment of the present application, calculating the uniformity of the coal sample particles in the coal sample image according to the baseline characteristics and the topological parameters includes:

[0038] The feature values ​​of each channel position of the baseline feature are respectively activated by global maximum pooling to obtain the first target feature.

[0039] The first target feature is activated by performing a Hadamard product with the topological parameter to obtain a second target feature.

[0040] After the second target feature is calculated by the second fully connected layer, the third target feature is obtained.

[0041] After the third target feature is calculated by the classifier, the uniformity of the coal sample particles in the coal sample image is obtained.

[0042] The present invention also provides a device for identifying the uniformity of particle size of coal samples, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method when executing the computer program.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] Because coal particles are very dark in color and there is varying degrees of stacking between different particles, the edges of some coal particles are not obvious, resulting in poor performance of existing image-based particle uniformity recognition algorithms. The present invention creatively extracts baseline features and anchor features from coal sample images at the same time and calculates the topological parameters between them, making the algorithm more adaptable to the appearance characteristics of coal particles and improving the recognition accuracy of coal particle size uniformity. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The figure is a flow chart of a method for identifying the uniformity of particle size of coal samples according to the present invention. DETAILED DESCRIPTION

[0046] The present invention will be described in further detail below with reference to the accompanying drawings.

[0047] This embodiment provides a method for identifying the uniformity of particle size of coal samples. Figure 1 As shown, the following steps are included:

[0048] Step 1: Acquire a coal sample image and extract its baseline features. When capturing the coal sample image, the background should preferably be a pure color, preferably a light color (such as white), to increase the contrast between the coal and the background and reduce environmental interference.

[0049] Extracting baseline features from coal sample images typically requires multiple baseline feature extraction units connected end-to-end (i.e., arranged in series). In this embodiment, five baseline feature extraction units are used to extract baseline features from the coal sample images. For example, the first baseline feature extraction unit directly receives the coal sample image, extracts features from it, and outputs the results. The second unit no longer processes the coal sample image, but instead uses the features output by the first baseline feature extraction unit as input, continuing to extract features and outputting them. The third and fourth baseline feature extraction units follow the same logic, processing the features output by the previous unit. Finally, the fifth baseline feature extraction unit receives the features output by the fourth unit, and the resulting calculations are the baseline features extracted from the coal sample image. The baseline feature extraction units are internally configured with convolutional layers. These units can be implemented using existing feature extraction units based on convolutional neural networks. For example, in some embodiments, these units can be feature extraction modules used in GoogLeNet, MobileNet, or SENet.

[0050] In some other embodiments, the internal calculation process of the reference feature extraction unit includes:

[0051] Obtain the data to be processed (the data to be processed is the data input into the reference feature extraction unit from its head), perform pre-feature extraction on the data to be processed to obtain the first basic feature; reduce the spatial dimension of the first basic feature to obtain the second basic feature; align the second basic feature to obtain the third basic feature output by the end of the reference feature extraction unit. The third basic feature output by the end of the fifth reference feature extraction unit is the baseline feature.

[0052] In some embodiments, performing pre-feature extraction on the data to be processed includes:

[0053] After the data to be processed is processed and activated by the first convolutional layer, a first distribution feature is obtained, and the convolution kernel size of the first convolutional layer is 1*1; after the data to be processed is processed and activated by the second convolutional layer, a second distribution feature is obtained, and the convolution kernel size of the second convolutional layer is 3*3; the first distribution feature is activated by performing a Hadamard product with the second distribution feature to obtain a third distribution feature; the first distribution feature is activated by subtracting the data to be processed (for example, the first distribution feature minus the data to be processed) from the first distribution feature to obtain a fourth distribution feature; the second distribution feature, the third distribution feature, and the fourth distribution feature are concatenated to obtain a fifth distribution feature; the fifth distribution feature is processed and activated by the third convolutional layer (the convolution kernel size may be 3*3) to obtain a first basic feature. In some embodiments, the height and width of the first basic feature are respectively equal to the height and width of the data to be processed input to the reference feature extraction unit, and the number of channels of the first basic feature is twice the number of channels of the data to be processed input to the reference feature extraction unit.

[0054] In some embodiments, reducing the spatial dimension of the first basic feature includes processing the first basic feature using a local maximum pooling layer to obtain a second basic feature. For example, the pooling window size of the local maximum pooling layer may be 2*2. During the pooling operation, the pooling window slides along the spatial direction (i.e., the width and height directions) with a sliding step size of 2.

[0055] In some embodiments, the second basic feature is aligned, including: the second basic feature is processed and activated by the (existing) deformable convolution layer to obtain a first alignment feature, the second basic feature is processed and activated by the fourth convolution layer to obtain a second alignment feature, the first alignment feature and the second alignment feature are concatenated to obtain a third alignment feature, and the third alignment feature is processed and activated by the fifth convolution layer to obtain a third basic feature.

[0056] After adopting the above calculation process within the benchmark feature extraction unit, more refined feature extraction and robust alignment processing are achieved. While improving the feature expression ability and feature deformation adaptability, it realizes the efficient use of information in a hierarchical manner, which can effectively enhance the effect of subsequent recognition and classification tasks.

[0057] Step 2: Obtain a coal sample image. After the coal sample image is processed and activated by the sixth convolution layer (for example, the convolution kernel size can be 3*3), the first anchor feature is obtained. After the first anchor feature is processed and activated by the seventh convolution layer (for example, the convolution kernel size can be 3*3), the second anchor feature is obtained. The corresponding elements of the second anchor feature and the first anchor feature are added and activated to obtain the anchor feature.

[0058] Step 3: Both baseline and anchor features are three-dimensional (height, width, and channel) feature matrix data. Based on the pre-defined topological analysis unit, calculate the topological parameters between the baseline and anchor features. These topological parameters are used to characterize the spatial correlation and structural mapping relationship between the baseline and anchor features.

[0059] In some embodiments, based on a preset topological analysis unit, the topological parameters between the baseline feature and the anchor feature are calculated, including: integrating the eigenvalues ​​of each channel position of the baseline feature to obtain a baseline vector; integrating the eigenvalues ​​of each spatial position of the anchor feature to obtain an anchor matrix; matching the anchor matrix according to the baseline vector to obtain an anchor vector; integrating the baseline vector and the anchor vector to obtain the topological parameters.

[0060] In some embodiments, integrating the eigenvalues ​​of each channel position of the baseline feature includes performing global average pooling on the eigenvalues ​​of each channel position of the baseline feature and then activating them to obtain a baseline vector. For example, for a baseline feature with a height of A1, a width of A2, and a number of channels A3, each channel position of the baseline feature contains (A1*A2) eigenvalues. Global average pooling is performed on the eigenvalues ​​of each channel position of the baseline feature, that is, calculating the average of the (A1*A2) eigenvalues ​​in each channel, to obtain a baseline vector of length A3.

[0061] In some embodiments, integrating the eigenvalues ​​of each spatial position of the anchor feature includes: performing global maximum pooling on the eigenvalues ​​of each spatial position of the anchor feature and then activating them to obtain an anchor matrix. For example, for an anchor feature with a height of B1, a width of B2, and a number of channels of B3, the anchor feature contains a total of B1*B2 different spatial positions, each of which contains B3 eigenvalues. Global maximum pooling is performed on the eigenvalues ​​of each spatial position of the anchor feature, that is, the maximum value of the B3 eigenvalues ​​in each spatial position is calculated, and then an anchor matrix of size (B1*B2) is obtained.

[0062] In some embodiments, the anchor matrix is ​​matched according to the baseline vector, including: the anchor matrix is ​​expanded and input into the first fully connected layer, the first fully connected layer calculates and outputs a connection vector with the same length as the baseline vector, and the connection vector is activated after adding the corresponding elements of the baseline vector to obtain the anchor vector.

[0063] In some embodiments, integrating the baseline vector and the anchor vector includes: performing a Hadamard product on the baseline vector and the anchor vector and then activating the resultant to obtain a topological parameter.

[0064] Through the progressive design of "differentiated integration-precise matching-deep fusion", the topological analysis unit enables the topological parameters to not only reflect the overall characteristic benchmark of the coal sample, but also capture key local characteristics, while strengthening the correlation between the two, thereby greatly compensating for the shortcomings of the baseline characteristics.

[0065] Step 4: Calculate the uniformity of the coal sample particles in the coal sample image based on the baseline features and topological parameters. In some embodiments, calculating the uniformity of the coal sample particles in the coal sample image based on the baseline features and topological parameters includes: performing global maximum pooling on the feature values ​​of each channel position of the baseline features and then activating them to obtain a first target feature, performing a Hadamard product multiplication of the first target feature and the topological parameters and then activating them to obtain a second target feature, calculating the second target feature through a second fully connected layer to obtain a third target feature, and calculating the third target feature through a classifier (e.g., a softmax classifier) ​​to obtain the uniformity of the coal sample particle size in the coal sample image.

[0066] The above-mentioned third target feature is a one-dimensional vector data, and its length can be set to different values ​​according to actual needs, so as to adjust the number of classification categories of particle size uniformity. For example, the length of the third target feature can be set to 20, and the result output by the classifier will include 20 categories, which can be represented by numbers 1-20 respectively. The larger the number, the better the particle size uniformity. It is also possible to determine whether the particle size uniformity of the coal sample is qualified by setting an appropriate threshold (for example, the threshold can be 10). When it is identified that the particle uniformity is lower than the preset threshold, it means that the particle uniformity of the coal sample is unqualified and the sample needs to be processed again or replaced. Otherwise, it means that the particle size uniformity of the coal sample meets the requirements, which has little effect on the subsequent coal measurement results.

[0067] As is well known to those skilled in the art, the above activation refers to inputting relevant data into a nonlinear activation function for operation to obtain activated data. As an example but not limitation, the above activation can select one of ReLU, Sigmoid, Swish or ELU functions to perform the activation operation. In some embodiments, the above steps of extracting the baseline features and anchor features of the coal sample image, calculating the topological parameters between the baseline features and the anchor features based on a preset topological analysis unit, and calculating the uniformity of the particle size of the coal sample based on the baseline features and topological parameters can be achieved by constructing an artificial neural network through a computer program. These steps can be completed through calculation using electronic devices such as computers, tablet computers or mobile phone terminals.

[0068] Finally, it should be noted that the above description is only a preferred embodiment of the present invention. Under the guidance of the present invention, ordinary technicians in this field can make various similar expressions without violating the purpose and claims of the present invention. Such changes fall within the scope of protection of the present invention.

Claims

1. A method for identifying the uniformity of particle size of coal samples, characterized in that: The following steps are involved: Obtaining baseline features and anchor features of coal sample images; Based on a preset topological analysis unit, a topological parameter between the baseline feature and the anchor feature is calculated; the topological parameter is used to characterize the spatial correlation relationship and structural mapping relationship between the baseline feature and the anchor feature; The particle size uniformity of the coal sample particles in the coal sample image is calculated based on the baseline characteristics and the topological parameters.

2. The method for identifying the uniformity of particle size of coal samples according to claim 1, characterized in that: Obtaining the baseline features of the coal sample image includes: obtaining the coal sample image, and extracting the baseline features of the coal sample image based on a plurality of baseline feature extraction units connected end to end, wherein a convolution layer is provided inside the baseline feature extraction unit.

3. The method for identifying the uniformity of particle size of coal samples according to claim 2, characterized in that: The internal calculation process of the reference feature extraction unit includes: Acquire data to be processed, perform pre-feature extraction on the data to be processed, and obtain a first basic feature; Reducing the spatial dimension of the first basic feature to obtain a second basic feature; Alignment processing is performed on the second basic features to obtain third basic features output by the reference feature extraction unit.

4. The method for identifying the uniformity of particle size of coal samples according to claim 3, characterized in that: Performing pre-feature extraction on the data to be processed, including: After the data to be processed is processed and activated by the first convolution layer, a first distribution feature is obtained, where the convolution kernel size of the first convolution layer is 1*1; After the data to be processed is processed and activated by the second convolutional layer, a second distribution feature is obtained, and the convolution kernel size of the second convolutional layer is 3*3; The first distribution feature and the second distribution feature are activated after performing a Hadamard product to obtain a third distribution feature; The first distribution feature is subtracted from the data to be processed and then activated to obtain a fourth distribution feature; The second distribution feature, the third distribution feature, and the fourth distribution feature are spliced ​​to obtain a fifth distribution feature; After the fifth distribution feature is processed and activated by the third convolutional layer, the first basic feature is obtained.

5. The method for identifying the uniformity of particle size of coal samples according to claim 3, characterized in that: Reducing the spatial dimension of the first basic feature includes: processing the first basic feature using a local maximum pooling layer; or / and, performing alignment processing on the second basic features, including: After the second basic feature is processed and activated by the deformable convolution layer, the first alignment feature is obtained. After the second basic feature is processed and activated by the fourth convolutional layer, the second alignment feature is obtained. After the first alignment feature and the second alignment feature are spliced ​​together, a third alignment feature is obtained. After the third alignment feature is processed and activated by the fifth convolutional layer, the third basic feature is obtained.

6. The method for identifying the uniformity of particle size of coal samples according to claim 1, characterized in that: Obtaining anchor features of a coal sample image includes: obtaining a coal sample image, processing and activating the coal sample image through a sixth convolutional layer to obtain a first anchor feature; processing and activating the first anchor feature through a seventh convolutional layer to obtain a second anchor feature; and adding and activating the second anchor feature to the first anchor feature to obtain the anchor feature.

7. The method for identifying the uniformity of particle size of coal samples according to claim 1, characterized in that: Calculating a topological parameter between the baseline feature and the anchor feature based on a preset topological analysis unit includes: Integrating the eigenvalues ​​of each channel position of the baseline feature to obtain a baseline vector; Integrating the eigenvalues ​​of each spatial position of the anchor feature to obtain an anchor matrix; Performing matching processing on the anchor matrix according to the baseline vector to obtain an anchor vector; The baseline vector and the anchor vector are integrated to obtain the topological parameters.

8. The method for identifying the uniformity of particle size of coal samples according to claim 7, characterized in that: Integrating the feature values ​​of each channel position of the baseline feature respectively, including: performing global average pooling and activation on the feature values ​​of each channel position of the baseline feature respectively; or / and, integrating the feature values ​​of each spatial position of the anchor feature respectively, including: performing global maximum pooling and then activating the feature values ​​of each spatial position of the anchor feature respectively; or / and, performing matching processing on the anchor matrix according to the baseline vector, comprising: expanding the anchor matrix and inputting it into a first fully connected layer, the first fully connected layer calculating and outputting a connection vector having a length equal to that of the baseline vector, adding the connection vector to the baseline vector and activating the resultant to obtain the anchor vector; Or / and, integrating the baseline vector with the anchor vector, comprising: performing a Hadamard product between the baseline vector and the anchor vector and then activating the result.

9. The method for identifying the uniformity of particle size of coal samples according to claim 1, characterized in that: Calculating the uniformity of the coal sample particles in the coal sample image according to the baseline characteristics and the topological parameters includes: The feature values ​​of each channel position of the baseline feature are respectively activated by global maximum pooling to obtain the first target feature. The first target feature is activated by performing a Hadamard product with the topological parameter to obtain a second target feature. After the second target feature is calculated by the second fully connected layer, the third target feature is obtained. After the third target feature is calculated by the classifier, the uniformity of the coal sample particles in the coal sample image is obtained.

10. A device for identifying the uniformity of particle size of coal samples, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images