Coal quality evaluation method and system based on hyperspectral imaging technology

Abstract

The present invention discloses a method and system for evaluating coal quality based on hyperspectral imaging technology, and relates to the technical field of coal quality evaluation. The method comprises: obtaining visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of a region containing a coal target; performing image processing and fusion on a plurality of hyperspectral images to obtain an original hyperspectral image; using a target detection algorithm to identify the coal region in the image; extracting spatial-spectral fusion features based on the hyperspectral images of the coal region, and predicting the ash, moisture, volatile content, and calorific value of the coal according to the relationship between the spatial-spectral fusion features and a plurality of basic indicators of coal quality; performing calculation and correction according to the predicted ash, moisture, volatile content, and calorific value to obtain a corrected calorific value; comprehensively evaluating the moisture, ash, volatile content, and corrected calorific value of the coal to evaluate the coal quality and classify the coal. The present invention can achieve more accurate coal quality detection and evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of coal quality evaluation, and in particular to a coal quality evaluation method and system based on hyperspectral imaging technology. Background Art

[0002] As the main energy source, coal is one of the most important fuels in industrial production and life, and coal is a non-renewable energy source. The rational use of coal resources is a necessary condition for long-term development. At present, with the increasing requirements for environmental protection and energy conservation, it is very important to accurately and efficiently evaluate the quality of coal. The main purpose of coal quality evaluation is to determine the quality of coal, clarify the physical, chemical properties and process performance of coal, and then evaluate its utilization direction and economic value, so as to provide a scientific basis for coal mining, processing and utilization.

[0003] The basic indicators for coal quality detection and evaluation include moisture content, ash content, volatile matter content, fixed carbon content, etc. At the same time, the calorific value of coal is also an important indicator for evaluating the quality of coal. Its calorific value has a certain relationship with the ash content, moisture content, and volatile matter content of coal. Traditional coal quality detection methods can be divided into two types, namely direct detection method and indirect detection method. Among them, the direct detection method requires tedious and complicated processes such as preparing coal samples, burning, drying, and weighing in the coal quality laboratory. The measurement time is long and it is not real-time, and the test results cannot be obtained in time. Considering the disadvantages of the direct detection method, the indirect detection method can realize the real-time detection of coal quality, which includes radiation method, image method, microwave technology, radio and television, spectroscopy, etc. Among them, spectroscopy is one of the most commonly used detection methods. It mainly collects the spectral information of coal through a near-infrared spectrometer and completes the coal quality detection by processing the coal spectral curve.

[0004] However, the existing spectral method for detecting coal quality still has certain problems. On the one hand, the spectral data collected only by near-infrared spectrometer has certain range limitations and cannot fully and effectively reflect the coal quality. On the other hand, one-time processing of coal spectral data is difficult to ensure the accuracy of the final detection data, resulting in low coal quality detection accuracy. Summary of the invention

[0005] To address the deficiencies of the above-mentioned prior art, the present invention provides a coal quality evaluation method and system based on hyperspectral imaging technology, which obtains full-band hyperspectral data including visible-near infrared, short-wave infrared, and thermal infrared, detects coal areas based on more comprehensive and rich spectral data, and extracts spatial spectrum characteristics of coal areas, so as to achieve more accurate coal quality detection, detect and obtain the ash content, moisture content, volatile matter content and calorific value of the coal, and on this basis, combines the calculation of fixed carbon content to calibrate the detection results again, further correct the detection results of coal quality, improve the detection and calculation accuracy, and finally evaluate the coal quality according to various coal quality indicators to achieve accurate evaluation of coal quality; the above method not only avoids complex experimental processes and saves manpower and material resources, but also realizes real-time, accurate and large-scale intelligent monitoring of coal quality in mining areas, greatly shortens time consumption, and effectively improves work efficiency.

[0006] In a first aspect, the present invention provides a coal quality evaluation method based on hyperspectral imaging technology.

[0007] A coal quality evaluation method based on hyperspectral imaging technology, comprising:

[0008] Acquire visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of areas containing coal targets;

[0009] Perform image processing and fusion on multiple hyperspectral images to obtain the original hyperspectral image;

[0010] Based on the original hyperspectral image, the target detection algorithm is used to identify the coal area in the image;

[0011] Based on the hyperspectral image of the coal area, the spatial-spectral fusion features are extracted, and the ash, moisture, volatile matter content and calorific value of the coal are predicted according to the relationship between the spatial-spectral fusion features and various basic coal quality indicators;

[0012] Calculate and correct the predicted ash, moisture, volatile matter content and calorific value to obtain the corrected calorific value;

[0013] The moisture, ash, volatile matter content and corrected calorific value of the coal are comprehensively considered to evaluate the quality and classify the coal.

[0014] In a second aspect, the present invention provides a coal quality evaluation system based on hyperspectral imaging technology.

[0015] A coal quality evaluation system based on hyperspectral imaging technology, comprising:

[0016] Hyperspectral image acquisition module, used to acquire visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of the area containing coal targets;

[0017] An image processing module is used to process and fuse multiple hyperspectral images to obtain original hyperspectral images;

[0018] The coal area detection module is used to identify the coal area in the image based on the original hyperspectral image and using the target detection algorithm;

[0019] The coal quality evaluation module is used to extract spatial-spectral fusion features based on the hyperspectral images of the coal area, and predict the ash, moisture, volatile matter content and calorific value of the coal according to the relationship between the spatial-spectral fusion features and various basic coal quality indicators; calculate and correct the predicted ash, moisture, volatile matter content and calorific value to obtain the corrected calorific value; comprehensively evaluate the coal quality and classify the coal based on the moisture, ash, volatile matter content and corrected calorific value of the coal.

[0020] In a third aspect, the present invention further provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the method described in the first aspect are completed.

[0021] In a fourth aspect, the present invention further provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the method described in the first aspect.

[0022] One or more of the above technical solutions have the following beneficial effects:

[0023] 1. The present invention provides a coal quality evaluation method and system based on hyperspectral imaging technology, which obtains full-band hyperspectral data including visible-near infrared, short-wave infrared, and thermal infrared, so as to capture more comprehensive and rich spectral information of the target in different bands. It has the advantages of unified graph and spectrum, high spectral resolution, safety, speed, intelligence and efficiency. By fully mining the information of spatial and spectral dimensions of hyperspectral images, it can accurately detect coal areas and extract spatial-spectral characteristics of coal areas, so as to achieve more accurate coal quality detection and evaluation.

[0024] 2. The coal quality evaluation method applicable to mining areas provided by the present invention can accurately detect and comprehensively evaluate the ash content, moisture content, volatile matter content and calorific value of coal. On this basis, combined with the calculation of fixed carbon content, the detection results are corrected again, and the detection results of coal quality are further corrected to improve the detection and calculation accuracy; finally, the coal quality is evaluated according to various coal quality indicators to achieve accurate evaluation of coal quality, and then the real-time sorting of coal can be achieved. The above-mentioned method proposed by the present invention not only avoids the complicated experimental process and saves manpower and material resources, but also does not require auxiliary complex technology, reduces the complexity of coal quality evaluation, and realizes real-time, non-destructive, non-contact testing of coal quality in mining areas, and accurate large-scale intelligent monitoring, which greatly shortens the time consumption and effectively improves work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 This is an overall flow chart of the coal quality evaluation method based on hyperspectral imaging technology according to an embodiment of the present invention;

[0027] Figure 2 A schematic diagram of a coal quality evaluation system based on hyperspectral imaging technology according to an embodiment of the present invention;

[0028] Figure 3 Schematic diagram of the network structure for extracting spatial-spectral fusion features in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] It should be noted that the following detailed descriptions are exemplary only, are intended to describe specific embodiments, are intended to provide further explanation of the present invention, and are not intended to limit exemplary embodiments according to the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those of ordinary skill in the art to which the present invention belongs. In addition, it should also 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.

[0030] Embodiment 1

[0031] This embodiment provides a coal quality evaluation method based on hyperspectral imaging technology. Figure 1 As shown, the specific steps include:

[0032] Step S1, obtaining visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of the area containing the coal target;

[0033] Step S2, performing image processing and fusion on a plurality of hyperspectral images to obtain an original hyperspectral image;

[0034] Step S3, based on the original hyperspectral image, using a target detection algorithm to identify the coal area in the image;

[0035] Step S4, based on the hyperspectral image of the coal area, extracting the spatial-spectral fusion features, and predicting the ash content, moisture content, volatile matter content and calorific value of the coal according to the relationship between the spatial-spectral fusion features and various basic coal quality indicators;

[0036] Step S5, calculating and correcting according to the predicted ash content, moisture content, volatile matter content and calorific value to obtain a corrected calorific value;

[0037] Step S6: comprehensively evaluate the moisture, ash, volatile matter content and corrected calorific value of the coal, evaluate the coal quality and classify the coal.

[0038] The following content introduces the coal quality evaluation method based on hyperspectral imaging technology proposed in this embodiment in more detail.

[0039] In the above steps S1 and S2, visible-near infrared, short-wave infrared, and thermal infrared hyperspectral cameras are used to collect spectral data of the area containing the coal target, and visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of the area containing the coal target are obtained. Then, image processing and fusion of multiple (i.e., three) hyperspectral images are performed, including calibration, preprocessing, registration, fusion, and splicing of the multiple hyperspectral images to obtain the original hyperspectral image.

[0040] Specifically, (1) the calibration mentioned above refers to unifying the image size and scale, etc., so that the spatial coordinates of multiple hyperspectral images are consistent and the format is unified. (2) The preprocessing mentioned above includes image denoising and radiation correction, wherein one or more of mean filtering, median filtering, wavelet transform, principal component analysis and minimum noise separation transform are used for image denoising to reduce the noise in the image and improve the signal-to-noise ratio of the data; the correction method based on dark pixels and the correction method based on spectral standardization are used for radiation correction to eliminate the changes in radiation intensity caused by sensor characteristics, viewing angles and other factors, so that the data truly reflects the spectral characteristics of coal. (3) After completing the preprocessing, the three types of spectral data are aligned, fused and spliced ​​using spatial-spectral information fusion processing to obtain the original hyperspectral image, wherein the alignment, fusion and splicing are: for each hyperspectral image, the image is decomposed into sub-images of different scales using wavelet transform or Laplace pyramid, and fused at each scale, and then reconstructed to obtain the fused hyperspectral image, thereby retaining the edge and detail information of the image, and finally the multiple hyperspectral images are spliced ​​in the spectral dimension to obtain the original hyperspectral image. Preferably, during the splicing process, an interpolation method may be used to smooth the data at the splicing location to avoid discontinuity between bands.

[0041] Furthermore, based on the anomaly detection of spatial spectral consistency and local area consistency, the original hyperspectral image obtained by the fusion and stitching is subjected to bad pixel detection and compensation processing. The damaged pixels or bands in the image can be identified through bad pixel detection, and then the bad pixel compensation is used to improve the image and reduce the impact of bad pixels on subsequent analysis.

[0042] Specifically, the spatial consistency anomaly detection method includes spatial spectral consistency and local area consistency anomaly detection. First, based on spatial spectral consistency, the spectral consistency of each pixel and the pixels in the surrounding area is checked. For example, if the spectrum of pixel A is extremely inconsistent with the spectrum of multiple surrounding pixels, pixel A may be a bad pixel. Secondly, based on local area consistency, pixels with abnormal spectral characteristics are checked in each local area. For example, if in a local area, only pixel A shows significantly different spectral characteristics, while the spectral characteristics of other pixels are consistent, then A is further confirmed to be a bad pixel. By combining spatial and spectral information in the above way, bad pixel detection is achieved by considering the spectral consistency of spatially adjacent pixels and the spectral characteristics of pixels in the local range at the same time.

[0043] After the bad pixels are detected, they are compensated to reduce the impact of these bad pixels on subsequent analysis. The compensation includes: linear interpolation, polynomial interpolation, spatial interpolation, and joint interpolation based on spatial spectrum characteristics.

[0044] In the above step S3, based on the original hyperspectral image, the target detection algorithm is used to identify the coal area in the image. Specifically, based on the original hyperspectral image, the potential and hidden data features in the original hyperspectral image are excavated, and the vertical column division based on the spatial dimension decomposition and the horizontal row division based on the spectral dimension decomposition are performed, and the spatial features in the local area and the spectral features of each pixel are captured respectively, and the coal area in the image is identified by combining the target detection algorithm; wherein, the spatial features include texture features, shape features and structural features, etc., and the spectral features include absorption valleys, reflection peaks and spectral slopes, etc. Through the decomposition and division of the above spatial dimension and spectral dimension, the differences in quantity and density between the abnormal area and the background part, as well as the differences in spectral trends and morphologies can be highlighted, which is conducive to improving the accuracy of the target detection algorithm, and can effectively detect the hyperspectral targets in the coal area under the cluttered background, and identify and delineate the location of the coal area.

[0045] In this embodiment, the vertical column division based on spatial dimension decomposition and the horizontal row division based on spectral dimension decomposition are adopted, and the differences are highlighted after division, so as to improve the accuracy of the subsequent target detection algorithm. Among them, the vertical column division based on spatial dimension decomposition is used to analyze the spatial dimension of the image, divide the hyperspectral image into different regions, and capture the spatial features in the local region such as texture, shape and structural features. Specifically, the local binary pattern (LBP) can be used to capture the distribution characteristics or texture characteristics of the pixels in the hyperspectral image in its local area, or a filter group (such as Gabor filter, Laws texture energy filter, etc.) can be used to perform spatial segmentation and feature capture on the hyperspectral image.

[0046] Furthermore, the above local binary pattern means: for each pixel, a fixed-size neighborhood is selected (usually a 3×3 square area), and the difference between the grayscale values ​​of the central pixel and other pixels in its neighborhood is calculated. If the grayscale value of the neighborhood pixel is higher than that of the central pixel, it is marked as 1, otherwise it is marked as 0; then, the above binary results are arranged in order to obtain a binary number, and the decimal value of the binary number is the LBP value of the pixel. By calculating the LBP value of each pixel in the image, an LBP feature map can be obtained to capture the local texture features of the image. The above Gabor filter can capture edge information of specific directions and scales in the image, and help identify the directional texture features of the image. The above Laws texture energy filter extracts texture features of different scales and directions of the image through a set of filters to generate a texture energy map for capturing delicate texture changes.

[0047] In addition, through the horizontal division method of spectral dimension decomposition, the characteristic spectrum of coal is extracted, such as absorption peaks, absorption valleys, spectral slopes and reflection characteristics in specific bands. These characteristics are helpful in distinguishing different materials or substances. By comparing them with the background spectrum, the differences in the spectral trend and morphology of coal are highlighted. Specifically, the horizontal division based on spectral dimension decomposition is as follows: extracting spectral features such as absorption valley D, reflection peak P, slope Slope, etc., merging the extracted spectral features (absorption valley, reflection peak and slope, etc.) into a feature vector F=[D, P, Slope,…], performing horizontal division on the hyperspectral image, identifying targets with significant spectral characteristics, and facilitating the discovery of abnormal or target spectra. Among them, the calculation formula of the absorption valley is: , is the maximum reflectivity of the spectral curve within a specific wavelength range, is the minimum reflectivity; the calculation formula for the reflection peak is , is the maximum reflectivity of the spectral curve within a specific wavelength range, is the background value or minimum reflectance; the slope is calculated as , and are the wavelength and reflectance values ​​of the spectral curve, and are the means of these values, respectively.

[0048] Furthermore, through the above-mentioned vertical column division based on spatial dimension decomposition and horizontal row division based on spectral dimension decomposition, spatial-spectral information (referred to as spatial-spectral information) is extracted, that is, the position and texture relationship of the vertical column division of spatial dimension decomposition and the spectral information of the horizontal row division based on spectral dimension decomposition are extracted. Considering that the spatial-spectral information can be used to better separate the target and the background, and the matched filter is a filter used for target detection, which aims to enhance the matching degree with specific target features (such as coal) while suppressing other background features. Therefore, the spatial-spectral information is combined with the matched filter to enhance the distinction between the target and the background, so as to identify the abnormal area as a coal area.

[0049] Specifically, in order to highlight the differences in quantity, density, spectral trend and morphology between abnormal areas and background parts, statistics based on spectral dimensions (such as RX algorithm, mean-covariance) can be used to identify abnormal points, and target recognition can be achieved through abnormal point detection. In fact, in hyperspectral target detection, "abnormal" or "abnormal point" refers to points that are significantly different from the background in spectral and spatial characteristics. These points may represent specific target substances (such as coal areas) or other substances that are significantly different from the background. That is, the identified abnormal points are target points.

[0050] In this embodiment, the RX algorithm can be used to calculate the Mahalanobis distance of the mean-covariance matrix between the spectrum of each pixel and the background, calculate the abnormal probability of each pixel through the Mahalanobis distance, quantify the difference between each pixel and the background, map the calculated abnormal probability to each pixel of the image, generate an abnormal rate mapping diagram, and present potential abnormal targets in the hyperspectral data in the form of probability distribution, highlighting the density and distribution of the target, and identifying pixels with a large distance (distance exceeding the set value) as target points, thereby identifying the target area, i.e., the coal area.

[0051] As another implementation method, the mean-covariance method refers to the calculation of the mean and covariance matrix of the entire image, which can identify pixels that deviate from these statistical characteristics, i.e., target points; a density clustering algorithm can also be used to identify high-density areas and outliers in the data to highlight the spatial difference between the target and the background. In addition, the above target detection algorithm can also use deep learning models such as convolutional neural networks to perform coal target detection.

[0052] In the above step S4, based on the hyperspectral image of the coal area, the spatial-spectral fusion features are extracted, and the ash content, moisture content, volatile matter content and calorific value of the coal are predicted according to the relationship between the spatial-spectral fusion features and various basic coal quality indicators.

[0053] In this embodiment, a multi-index evaluation model of coal quality based on a neural network is constructed. The evaluation model includes a feature extraction module and multiple parallel evaluation modules. The multiple evaluation modules are an ash content evaluation module, a moisture content evaluation module, a volatile matter content evaluation module and a calorific value evaluation module.

[0054] like Figure 3 As shown, the hyperspectral image of the coal area is input into the multi-index evaluation model of coal quality. First, the spatial-spectral fusion feature is extracted through the feature extraction module. The feature extraction module adopts a dual-branch structure based on parallel spectral feature extraction branches and spatial feature extraction branches to extract the spatial features and spectral features of the input image respectively, and fully explore the contextual information at different levels in the image. The extracted spatial features and spectral features are then fused through the feature fusion block through a multi-level interaction mechanism to fuse the features with strong correlation and complementarity at different levels to obtain the fused spatial-spectral fusion features; wherein the multi-level interaction mechanism includes an attention mechanism and a category adaptive weighting mechanism. The extraction of the above-mentioned spatial-spectral fusion features specifically includes:

[0055] Step S4.1: Based on the dual-branch structure, preliminarily extract the multi-scale spectral characteristics and multi-scale spatial characteristics of coal.

[0056] Specifically, for the extraction of spectral features, a one-dimensional convolutional network is used to process the spectral information in the input full-spectral hyperspectral image, and the corresponding broadband and narrowband information of the spectral information are extracted through filters of different sizes. Among them, small-scale filters are used to capture microscopic spectral differences, such as identifying the absorption peak of coal or the reflectivity characteristics of key bands, and large-scale filters are used to identify the overall spectral trend. In this way, the spectral data can be analyzed from multiple scales (or frequency ranges), the spectral characteristics in different wavelength ranges can be captured, and the preliminary extraction of multi-scale spectral features can be achieved.

[0057] For the extraction of spatial features, a two-dimensional convolutional network is used to process the spatial information in the input full-spectral hyperspectral image, and convolution operations of different scales are used to capture spatial features from details to the global level. Specifically, a multi-scale feature extraction convolutional network such as pyramid convolution (or hole convolution) can be used, such as Figure 3 The embodiment shown in the figure adopts a. Pyramid convolution is used to capture the spatial features of different receptive fields and preliminarily extract multi-scale spatial features. Among them, small-scale convolution can capture local fine features, such as the edges of coal particles and local texture details, while large-scale convolution can capture global macro information, such as the overall shape, contour and large-scale spatial patterns of coal blocks.

[0058] Step S4.2: In the spatial feature extraction branch, for each scale of spatial features extracted, deformable convolution is used to process complex multi-scale spatial features, such as Figure 3 Shown b . Deformable convolution, whose convolution kernel can adaptively adjust the position and shape according to the input features, deeply extract the spatial features of each scale, and finally obtain the following Figure 3 Shown c The refined multi-scale spatial features can more comprehensively capture complex geometric structures and morphological information, such as the different shapes and form characteristics of coal in the image. Afterwards, the refined multi-scale spatial features are fused to output enhanced spatial features.

[0059] Step S4.3: In the spectral feature extraction branch, a spectral attention module (such as Figure 3 middle d. Spectral attention module), which includes attention mechanism, one-dimensional convolution layer, Softmax or Sigmoid function, so that the network can adaptively weight different bands according to the importance of spectral information in different bands when interacting with pixel space information. Among them, the attention mechanism assigns weights according to the importance of each band in identifying the target, and enhances the band with more information; the one-dimensional convolution layer processes the spectral data and generates a feature map to represent the relationship between different bands; the Softmax or Sigmoid function is applied to the output of the attention mechanism to generate Figure 3 In e .Weighted combination of spectral bands to output enhanced spectral features.

[0060] Step S4.4: Use Figure 3 shown f The interactive fusion module and category adaptive weighting module fuse the extracted deep spatial and spectral features and output the fused spatial-spectral fusion features.

[0061] The above-mentioned interactive fusion module is mainly used to promote the information interaction between spatial features and spectral features. During the fusion process, the module dynamically adjusts the importance of features through attention mechanisms or other weight allocation strategies to ensure that the most important information in spatial and spectral features is retained in the final fused features. The structure of the interactive fusion module is a bidirectional feature stream, an attention mechanism, and a fusion layer. The bidirectional feature stream consists of two main feature streams, one for spatial features and the other for spectral features. These two feature streams are feature fused through multiple levels of interactive mechanisms. The attention mechanism assigns weights to each channel (or each feature dimension) during the fusion process to highlight important features. Finally, the features from the spatial and spectral branches are combined through weighted averaging, element-wise multiplication, or other fusion strategies to generate a fused feature representation.

[0062] The above-mentioned category adaptive weighting module can dynamically adjust the weights of each feature dimension according to the characteristics of different categories in the input data, so that the most important features of targets of different categories can be highlighted. Its structure is a category-aware attention layer, dynamic weight allocation and output layer.

[0063] Step S4.5, establish the relationship between the fused spatial spectrum characteristics and the ash content, moisture content, volatile matter content and calorific value of the coal, and predict the ash content, moisture content, volatile matter content and calorific value of the coal based on the relationship between the spatial spectrum fusion characteristics and various basic coal quality indicators.

[0064] Among them, the ash content calculation method is: using coal hyperspectral image samples with known ash content, establishing training sets and validation sets, extracting the spatial-spectral fusion features of the training samples, selecting a suitable regression evaluation model as the ash content evaluation module, to fit the relationship between the spatial-spectral fusion features and the ash content, training the model and performing cross-validation to ensure the generalization ability of the model, applying the trained model to the input new samples, inputting the spatial-spectral fusion features of the samples, and outputting the predicted ash content. The above process is for each pixel of the coal hyperspectral image. For the entire coal sample, the predicted ash content of all pixels is weighted averaged to obtain the overall ash content.

[0065] Furthermore, in the calculation process of the above-mentioned overall ash content, the calculation of the weight of each pixel point is determined by the quality of the spectral information and the spatial position. The similarity between the spectral feature vector of each pixel (i.e., pixel point) and the spectral feature vector of the sample used in the training model is analyzed. Pixels with high similarity usually indicate that their spectral information is closer to the spectral characteristics of the samples in the training set, so they are given higher weights. At the same time, considering the influence of spatial distribution uniformity, the spectral information of certain areas in the coal sample is more uniform or representative, and the pixels in these areas can be given higher weights to ensure that the pixels in representative areas have a greater impact in the final ash content calculation. A weight distribution model is established by comprehensively considering the quality of the above-mentioned spectral information and the spatial position, and the weight of each pixel is automatically calculated.

[0066] Among them, the method for inverting the moisture content of coal is as follows: based on the above-mentioned spatial-spectral fusion features, the spectral information of each pixel and the spatial relationship between neighboring pixels are analyzed, and the spectral characteristics and spatial characteristics of each pixel are combined into spatial-spectral fusion features as the input of the inversion model. A sample data set with known moisture content is used to establish an inversion model of pixel-level moisture content (i.e., moisture content evaluation module) through a regression model, and the trained model is applied to each pixel in the hyperspectral image to predict its corresponding moisture content.

[0067] Furthermore, the moisture content of the entire coal sample area is inverted and mapped. Specifically, the moisture content of all pixels is spatially interpolated and mapped to generate a moisture content distribution map. The mapping process introduces spatial smoothing technology, spatial interpolation based on the Krigeing method, or spatial smoothing using a convolution filter.

[0068] Similarly, a quantitative relationship between the spatial-spectral fusion characteristics of the coal sample hyperspectral image and the coal volatile matter and calorific value is established, and the volatile matter and calorific value of the coal sample are inverted.

[0069] In the above step S5, calculation and correction are performed according to the predicted ash content, moisture content, volatile matter content and calorific value to obtain the corrected calorific value.

[0070] Step S5.1, calculate the fixed carbon content SC1 based on the predicted ash, moisture and volatile matter contents, and infer the solid carbon content SC2 based on the thermal infrared spectrum image of the coal area. After processing the mean of the fixed carbon content SC1 and the solid carbon content SC2, the corrected solid carbon content SC is obtained.

[0071] Specifically, in the industrial analysis process, fixed carbon = 100% - (coal moisture + ash + volatile matter). The fixed carbon content can be obtained according to the moisture, ash and volatile matter content of the coal. Based on the above ash content, moisture content and volatile matter content, the solid carbon content SC1 (SC1 1 , SC1 2 …SC1 n ).

[0072] Furthermore, the feasibility of the above calculation method is verified. Through experiments and data analysis, it is known that there is a close relationship between thermal infrared spectrum and solid carbon content. Based on a large amount of historical data and experimental data, regression analysis or machine learning model is used for fitting and verification to establish the relationship between spectral sensitive bands and solid carbon content, so that the solid carbon content SC2 (SC2 1 , SC2 2 …SC2 n ). Among them, the spectral sensitive bands that are highly correlated with the solid carbon content are extracted from the spectral data using competitive adaptive reweighted sampling, genetic algorithm, mutual information and other feature screening methods.

[0073] Afterwards, the solid carbon contents SC1 and SC2 are averaged to obtain the corrected final carbon content SC.

[0074] Step S5.2: Calculate the calorific value CV1 of the coal area based on the corrected solid carbon content SC, and combine it with the predicted output calorific value CV2 to obtain the corrected calorific value CV.

[0075] Specifically, since the content of the main components of coal directly affects the calorific value of coal, the relationship between the fixed carbon content and the calorific value of coal can be expressed by the empirical calculation formula of calorific value: CV = a ×SC+ b , where CV is the calorific value of coal, M is the fixed carbon content of coal, a , b is the empirical coefficient; the corrected solid carbon content is substituted into the above empirical formula to calculate the coal calorific value CV1, thereby verifying and correcting the calorific value CV2 inverted based on the hyperspectral data.

[0076] Furthermore, if the error of the coal calorific value CV1 and CV2 is within "±2%", the feasibility of this calorific value calculation method is proved, and CV2 can be selected as the final predicted calorific value; if there is a certain deviation between CV2 and CV1, the two can be used in combination, and CV1 can be used to correct CV2, that is, the correction model can be applied. ,in and is the regression coefficient, which is solved by minimizing the prediction error to improve the accuracy of the inversion model.

[0077] In the above step S6, the moisture content M, ash content A, volatile matter content VM and corrected calorific value CV of the coal are comprehensively considered to evaluate the coal quality, and appropriate ranges of the four indicators are set according to specific uses and needs to achieve coal classification.

[0078] Considering that high moisture content will reduce the calorific value of coal, coal with high ash content will produce more residue after combustion, affecting boiler efficiency and increasing the difficulty of flue gas treatment. Volatile matter determines the combustion characteristics of coal. Coal with high volatile matter is flammable but may not burn completely. Calorific value is the main quality indicator of coal, which directly determines the economic value and utilization efficiency of coal. In order to classify and evaluate the quality of coal, this embodiment constructs a comprehensive scoring model, which combines the weights of various indicators, calculates the comprehensive score, and classifies according to the score.

[0079] Specifically, according to the purpose and quality requirements of coal, a weight is assigned to each indicator to establish a weight model. For example, the calorific value has a larger weight and the volatile matter has a smaller weight. For example, the coal used in power plants can be set with the following weights: moisture 0.2, ash 0.3, volatile matter 0.2, calorific value 0.3; then the comprehensive score of the coal is calculated as Score=W m ×M+W a ×A+W vm ×VM+W cv ×CV,W m , W a , W vm , W cv They are the weights of moisture, ash, volatile matter and calorific value respectively.

[0080] Through the above method, coal areas can be identified and more accurate coal quality detection and evaluation can be achieved.

[0081] Embodiment 2

[0082] This embodiment provides a coal quality evaluation system based on hyperspectral imaging technology. Figure 2 As shown, including:

[0083] The data acquisition module is used to obtain visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of the area containing coal targets; specifically, the module includes a conveyor system, a camera system, and a light source system, which specifically includes a conveyor belt, a visible-near infrared hyperspectral camera, a short-wave infrared hyperspectral camera, a thermal infrared hyperspectral camera, and a halogen lamp light source, wherein the hyperspectral camera is set above the conveyor belt, and the halogen lamps are set on both sides of the camera, which can supplement the light source in bad weather conditions to ensure the quality of data acquisition;

[0084] A data processing module is used to process and fuse multiple hyperspectral images to obtain original hyperspectral images;

[0085] The coal area detection module is used to identify the coal area in the image based on the original hyperspectral image and using the target detection algorithm;

[0086] The coal quality evaluation module is used to extract the spatial-spectral fusion features based on the hyperspectral image of the coal area, and predict the ash, moisture, volatile content and calorific value of the coal according to the relationship between the spatial-spectral fusion features and various basic coal quality indicators; calculate and correct the predicted ash, moisture, volatile content and calorific value to obtain the corrected calorific value; comprehensively evaluate the coal quality and classify the coal based on the moisture, ash, volatile content and corrected calorific value of the coal. Among them, according to the various coal quality indicators, the coal quality evaluation module can be divided into coal moisture content evaluation submodule, ash content evaluation submodule, volatile content evaluation submodule, and calorific value evaluation submodule.

[0087] Preferably, in view of the classification of the coal, the above system can also be divided into a specific coal screening module, which is used to set appropriate ranges of multiple indicators in combination with the evaluated coal quality according to specific uses and needs to achieve the classification of the coal.

[0088] Embodiment 3

[0089] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the coal quality evaluation method based on hyperspectral imaging technology as described above are completed.

[0090] Embodiment 4

[0091] This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the coal quality evaluation method based on hyperspectral imaging technology as described above are completed.

[0092] The steps involved in the above embodiments 2 to 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0093] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0094] The above description is only a preferred embodiment of the present invention. Although the specific implementation mode of the present invention is described in conjunction with the accompanying drawings, it is not a limitation of the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the protection scope of the present invention.

Claims

1. A coal quality evaluation method based on hyperspectral imaging technology, characterized in that: include: Acquire visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of areas containing coal targets; Perform image processing and fusion on multiple hyperspectral images to obtain the original hyperspectral image; Based on the original hyperspectral image, the target detection algorithm is used to identify the coal area in the image; Based on the hyperspectral image of the coal area, the spatial-spectral fusion features are extracted, and the ash content, moisture content, volatile matter content and calorific value of the coal are predicted according to the relationship between the spatial-spectral fusion features and various basic indicators of coal quality; wherein a multi-indicator evaluation model of coal quality based on a neural network is constructed; the evaluation model includes a feature extraction module and multiple parallel evaluation modules, and the multiple evaluation modules are respectively an ash content evaluation module, a moisture content evaluation module, a volatile matter content evaluation module and a calorific value evaluation module; The hyperspectral image of the coal area is input into the multi-index evaluation model of coal quality, and the spatial-spectral fusion features are extracted through the feature extraction module; the feature extraction module adopts a dual-branch structure based on parallel spectral feature extraction branches and spatial feature extraction branches to extract the spatial features and spectral features of the input image respectively, and the extracted spatial features and spectral features are then fused through a multi-level interaction mechanism to obtain the fused spatial-spectral fusion features; the multi-level interaction mechanism includes an attention mechanism and a category adaptive weighting mechanism; Based on the spatial-spectral fusion features of each pixel in the extracted image, the ash content, moisture content, volatile matter content and calorific value of each pixel are predicted and output respectively; For the ash content of each pixel, the weight of the current pixel is determined according to the similarity between the spectral characteristics of the current pixel and the typical spectral characteristics, and the predicted ash content of all pixels is weighted averaged to obtain the overall ash content of the coal; According to the moisture content of each pixel, the moisture content of all pixel points in the entire coal area is spatially interpolated and mapped to generate a moisture content distribution map. The spatial interpolation method based on the Krigeing method is used, and the spatial smoothing technology is introduced in the mapping process, and the convolution filter is used for spatial smoothing. Establish the quantitative relationship between the spatial-spectral fusion characteristics of the hyperspectral image of the coal area and the volatile matter and calorific value of the coal, and invert the volatile matter content and calorific value of the coal; According to the predicted ash, moisture, volatile matter content and calorific value, calculation and correction are performed to obtain the corrected calorific value; wherein, according to the predicted ash, moisture and volatile matter content, the solid carbon content SC1 is calculated, and the solid carbon content SC2 is inferred based on the thermal infrared spectrum image of the coal area, and the corrected solid carbon content SC is obtained after processing based on the average value of the solid carbon content SC1 and the solid carbon content SC2; The relationship between solid carbon content and coal calorific value is expressed by the empirical calorific value calculation formula: CV= a ×SC+b; Among them, CV is the calorific value of coal, SC is the solid carbon content of coal, a, b is the empirical coefficient; The calorific value CV1 of the coal area is calculated based on the corrected solid carbon content SC, and the corrected calorific value CV is obtained by combining the predicted output calorific value CV2; The moisture content, ash, volatile matter content and corrected calorific value of the coal are comprehensively considered to evaluate the coal quality and classify the coal. A weight is assigned to each coal quality indicator according to the purpose and quality requirements of the coal. Based on the moisture content, ash, volatile matter content and corrected calorific value of the coal, combined with the weights, the comprehensive score of the coal quality is obtained through weighted sum calculation.

2. A method for evaluating coal quality based on hyperspectral imaging technology as claimed in claim 1, characterized in that: The image processing and fusion of multiple hyperspectral images includes: Calibrate and preprocess the multiple hyperspectral images respectively; the preprocessing includes image denoising and radiation correction, wherein one or more of mean filtering, median filtering, wavelet transform, principal component analysis and minimum noise separation transform are used for image denoising, and a correction method based on dark pixels and a correction method based on spectral standardization are used for radiation correction; The original hyperspectral image is obtained by registration, fusion and splicing; the registration, fusion and splicing are as follows: for each hyperspectral image, the image is decomposed into sub-images of different scales by using wavelet transform or Laplace pyramid, and fused at each scale to obtain a fused hyperspectral image after reconstruction, and then multiple hyperspectral images are spliced ​​in the spectral dimension to obtain the original hyperspectral image; Based on spatial consistency anomaly detection and local area consistency anomaly detection, bad pixel detection and compensation processing are performed on the original hyperspectral image.

3. The method for evaluating coal quality based on hyperspectral imaging technology according to claim 1, characterized in that: Based on the original hyperspectral image, the target detection algorithm is used to identify the coal area in the image, including: Based on the original hyperspectral image, vertical column division based on spatial dimension decomposition and horizontal row division based on spectral dimension decomposition are performed to capture the spatial features in the local area and the spectral features of each pixel point respectively, and combined with the target detection algorithm, the coal area in the image is identified; The spatial features include texture features, shape features and structural features, and the spectral features include absorption valleys, reflection peaks and spectral slopes.

4. A coal quality evaluation system based on hyperspectral imaging technology, characterized in that: include: Hyperspectral image acquisition module, used to acquire visible-near infrared, short-wave infrared, and thermal infrared hyperspectral images of the area containing coal targets; An image processing module is used to process and fuse multiple hyperspectral images to obtain original hyperspectral images; The coal area detection module is used to identify the coal area in the image based on the original hyperspectral image and using the target detection algorithm; The coal quality evaluation module is used to extract the spatial-spectral fusion features based on the hyperspectral images of the coal area, and predict the ash, moisture, volatile matter content and calorific value of the coal according to the relationship between the spatial-spectral fusion features and various basic coal quality indicators; calculate and correct the predicted ash, moisture, volatile matter content and calorific value to obtain the corrected calorific value; comprehensively evaluate the coal quality and classify the coal based on the moisture, ash, volatile matter content and corrected calorific value of the coal; Among them, a multi-index evaluation model of coal quality based on neural network is constructed; the evaluation model includes a feature extraction module and multiple parallel evaluation modules, and the multiple evaluation modules are respectively an ash content evaluation module, a moisture content evaluation module, a volatile content evaluation module and a calorific value evaluation module; The hyperspectral image of the coal area is input into the multi-index evaluation model of coal quality, and the spatial-spectral fusion features are extracted through the feature extraction module; the feature extraction module adopts a dual-branch structure based on parallel spectral feature extraction branches and spatial feature extraction branches to extract the spatial features and spectral features of the input image respectively, and the extracted spatial features and spectral features are then fused through a multi-level interaction mechanism to obtain the fused spatial-spectral fusion features; the multi-level interaction mechanism includes an attention mechanism and a category adaptive weighting mechanism; Based on the spatial-spectral fusion features of each pixel in the extracted image, the ash content, moisture content, volatile matter content and calorific value of each pixel are predicted and output respectively; For the ash content of each pixel, the weight of the current pixel is determined according to the similarity between the spectral characteristics of the current pixel and the typical spectral characteristics, and the predicted ash content of all pixels is weighted averaged to obtain the overall ash content of the coal; According to the moisture content of each pixel, the moisture content of all pixel points in the entire coal area is spatially interpolated and mapped to generate a moisture content distribution map. The spatial interpolation method based on the Krigeing method is used, and the spatial smoothing technology is introduced in the mapping process, and the convolution filter is used for spatial smoothing. Establish the quantitative relationship between the spatial-spectral fusion characteristics of the hyperspectral image of the coal area and the volatile matter and calorific value of the coal, and invert the volatile matter content and calorific value of the coal; The solid carbon content SC1 is calculated based on the predicted ash, moisture and volatile content, and the solid carbon content SC2 is inferred based on the thermal infrared spectrum image of the coal area. The corrected solid carbon content SC is obtained after processing based on the average of the solid carbon content SC1 and the solid carbon content SC2; The relationship between solid carbon content and coal calorific value is expressed by the empirical calorific value calculation formula: CV= a ×SC+b; Among them, CV is the calorific value of coal, SC is the solid carbon content of coal, a, b is the empirical coefficient; The calorific value CV1 of the coal area is calculated based on the corrected solid carbon content SC, and the corrected calorific value CV is obtained by combining the predicted output calorific value CV2; A weight is assigned to each coal quality indicator according to the purpose and quality requirements of the coal; based on the moisture, ash, volatile matter content and corrected calorific value of the coal, combined with the weights, a comprehensive score for the coal quality is obtained through weighted sum calculation.

5. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of a coal quality evaluation method based on hyperspectral imaging technology as described in any one of claims 1 to 3 are completed.

6. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a coal quality evaluation method based on hyperspectral imaging technology as described in any one of claims 1-3.

Images