Fire coal calorific value detection method based on hyperspectral image technology

Abstract

The invention discloses a fire coal calorific value detection method based on a hyperspectral image technology, and relates to the field of coal calorific value detection. The problems that an existing coal quality evaluation method is complex in calculation process and low in detection efficiency are solved. The method comprises the following steps: taking coal samples with known coal calorific value parameters as correction set coal samples; a hyperspectral imaging system is used for collecting calibration set coal samples and corresponding black and white reference hyperspectral image data, and spectral data preprocessing is carried out; constructing a partial least square quantitative analysis model by taking hyperspectral data of the preprocessed calibration set as an input variable and coal calorific value parameters as dependent variables; the method comprises the following steps: acquiring a coal sample to be detected and corresponding black and white reference hyperspectral image data by using a hyperspectral imager, and preprocessing the spectral data; and detecting the calorific value of the to-be-detected coal sample by using the constructed partial least square quantitative analysis model by taking the hyperspectral data of the preprocessed to-be-detected coal sample as an input variable. The method is used for coal calorific value detection.

Description

Technical Field

[0001] The present invention relates to the technical field of coal calorific value detection, and in particular to a method for detecting the calorific value of coal based on hyperspectral imaging technology. Background Art

[0002] Coal, known as black "gold" and the "food" of industry, is a hidden treasure deep within the Earth, possessing immeasurable value for national and regional development. Currently, coal is the primary raw material for large-scale industrial production bases such as coal-fired power plants. Coal quality has a crucial impact on the production processes, efficiency, and quality of these companies. Coal quality is assessed across multiple indicators, with calorific value being a key key.

[0003] At present, there are three main methods for detecting the calorific value of coal: combustion method, spectral analysis method, and artificial intelligence method. All three methods have their own advantages and disadvantages. Combustion method: directly measuring the calorific value of coal by combustion method is the most accurate detection method at present. However, it also has obvious defects. The combustion method is time-consuming and complicated to operate. It is a destructive and non-real-time detection method that cannot meet the needs of rapid analysis on industrial sites. Spectral analysis methods mainly include: natural X-ray detection, multi-energy X-ray absorptiometry (MXRA), X-ray fluorescence (XRF), near-infrared spectroscopy, laser Raman spectroscopy, and laser-induced breakdown spectroscopy (LIBS) are among these methods. These methods determine the chemical composition and physical properties of coal by examining the spectral signatures of elements within it. Based on the relationship between coal quality and elements, these methods use algorithms to fit coal quality parameters. This method is rapid, non-destructive, and highly accurate. However, it also has significant drawbacks. Spectral analysis faces numerous challenges, including susceptibility to environmental influences, limited quantitative analysis capabilities, high equipment costs, and the need for highly specialized operators. Artificial intelligence methods, such as image analysis, leverage the relationship between coal composition and surface information to obtain coal quality parameters through machine vision and deep learning. This method offers real-time, non-destructive, efficient, accurate, and autonomous online testing. However, these methods also face significant drawbacks, including insufficient sample data, high equipment costs, and complex system integration.

[0004] Hyperspectral imaging, a new image acquisition technology, has been widely used in remote sensing analysis, information security, food safety, industrial testing, and other fields. Hyperspectral imaging can collect coal data across dozens to hundreds of wavelengths, outputting a three-dimensional data cube that fully exploits the coal's spectral and spatial information. Proper utilization of this information enables real-time, non-destructive, accurate, and efficient online measurement of coal calorific value.

[0005] For example, publication number CN119164899B, entitled "Method and System for Coal Quality Evaluation Based on Hyperspectral Imaging Technology," acquires visible-near infrared, shortwave infrared, and thermal infrared hyperspectral images of a region containing a target coal object; processes and fuses the multiple hyperspectral images to generate a raw hyperspectral image; utilizes a target detection algorithm to identify the coal region within the image; extracts spatial-spectral fusion features from the hyperspectral image of the coal region, and predicts the coal's ash content, moisture content, volatile matter content, and calorific value based on the relationship between the spatial-spectral fusion features and various basic coal quality indicators; calculates and corrects the predicted ash, moisture, volatile matter content, and calorific value to obtain a corrected calorific value; and evaluates and classifies the coal based on the moisture, ash, volatile matter content, and corrected calorific value. This invention enables more accurate coal quality detection and evaluation. However, its calculation process is complex and detection efficiency is low.

[0006] In summary, the existing coal quality evaluation methods have problems such as complex calculation process and low detection efficiency. Summary of the Invention

[0007] The purpose of the present invention is to solve the problems of complex calculation process and low detection efficiency in existing coal quality evaluation methods, and further provide a coal calorific value detection method based on hyperspectral imaging technology.

[0008] The technical solution of the present invention is:

[0009] A method for detecting the calorific value of coal based on hyperspectral imaging technology comprises the following steps:

[0010] Step 1: Use coal samples with known calorific value parameters as calibration set coal samples;

[0011] Step 2: Use the hyperspectral imaging system to collect the calibration set coal samples and the corresponding black and white reference hyperspectral image data, and perform spectral data preprocessing;

[0012] Step 3: Using the hyperspectral data of the calibration set after preprocessing in step 2 as input variables and the coal calorific value parameters as dependent variables, a partial least squares quantitative analysis model is constructed;

[0013] Step 4: Use a hyperspectral imager to collect the coal sample to be tested and the corresponding black and white reference hyperspectral image data, and perform spectral data preprocessing;

[0014] Step 5: Using the hyperspectral data of the pre-processed coal sample as input variables, the calorific value of the coal sample is detected using the partial least squares quantitative analysis model constructed in step 3.

[0015] Furthermore, in step 2, the hyperspectral imaging system uses a white board as a calibration reference when collecting the calibration set coal samples.

[0016] Furthermore, in step 2, when collecting the calibration set coal samples, the hyperspectral imaging system is placed in a black box to collect hyperspectral data.

[0017] Furthermore, the process of preprocessing the spectral data of the coal sample hyperspectral image data in step 2 is as follows:

[0018] Step 21: Use the collected black and white reference image to calibrate the original image data of the coal sample to obtain the real reflectivity data of the coal sample;

[0019]

[0020] Where, is the original image data of coal sample, is the white reference image data, is the black reference image data;

[0021] Step 22: Use bilateral filtering algorithm to perform spatial filtering on the reflectivity calibrated data;

[0022] Step 2 and 3: On the coal sample hyperspectral image, select a flat area as the region of interest, calculate all the pixels in the region of interest, and obtain the average spectral data;

[0023] Step 24: Use SG filtering algorithm to smooth and reduce noise of spectral data;

[0024] Step 25: Use the interquartile range algorithm to eliminate abnormal samples;

[0025] Step 26: Use the multivariate scatter correction algorithm to eliminate the influence of spectral scattering.

[0026] Furthermore, the formula of the bilateral filtering algorithm in step 22 is:

[0027] Formula (1)

[0028] In formula (1), is the pixel value after filtering, is the original pixel value, is the normalized weight, is the neighborhood window, is the spatial domain kernel function, is the pixel domain kernel function;

[0029] The specific forms of the spatial domain kernel function and the pixel value domain kernel function are Gaussian kernel functions, where:

[0030] The mathematical form of the spatial domain Gaussian kernel function is:

[0031] Formula (2)

[0032] In formula (2), is a point in the neighborhood, is the center point position, is the spatial domain standard deviation;

[0033] The mathematical form of the pixel value domain Gaussian kernel function is:

[0034] Formula (3)

[0035] Where, is the pixel value of a point in the neighborhood, is the center pixel value, is the pixel domain standard deviation.

[0036] Furthermore, the smoothing noise reduction process in step 24 includes the following steps:

[0037] Step 241: Divide all wavelength points of the coal sample hyperspectral data into p windows, and divide each window into z wavelength points at equal intervals;

[0038] Step 242: Perform m-order polynomial fitting on the reflectance values ​​of multiple wavelength points, and replace the original value with the fitted value of the central wave point of each window to achieve smoothing.

[0039] Preferably, the process of removing abnormal samples in step 25 is as follows:

[0040] Step 251: sort all sample spectral data corresponding to each wavelength point after smoothing filtering in ascending order;

[0041] Among them, the first quartile Q1 is the value corresponding to the 25% percentile in the ascending order data of the corresponding wavelength points, the third quartile Q3 is the value corresponding to the 75% percentile in the ascending order data of the corresponding wavelength points, and the second quartile Q2 is the value corresponding to the 50% percentile in the ascending order data of the corresponding wavelength points;

[0042] Step 252: Use the interquartile range (IQR) to set the upper and lower limits for eliminating abnormal samples:

[0043] According to: IQR = Q3 - Q1;

[0044] Upper=Q3+1.5IQR

[0045] Lower=Q1-1.5IQR

[0046] Here, Upper represents the upper limit of the interquartile range, and Lower represents the lower limit of the interquartile range.

[0047] Furthermore, the steps for eliminating spectral scattering in step 26 are as follows:

[0048] Step 261: Calculate the average spectrum of the coal sample as the ideal spectrum;

[0049] Step 262: Perform linear regression on each sample spectral curve;

[0050] Step 263: Perform multivariate scattering correction on each sample.

[0051] Furthermore, the hyperspectral imaging system in step 2 includes a hyperspectral imager, an electric-controlled displacement platform and computer hardware. The electric-controlled displacement platform is installed on the platform of the frame, and the hyperspectral imager is installed on the metal frame directly above the electric-controlled displacement platform. Two sets of halogen light sources are respectively installed on the metal frame, and the two sets of halogen light sources are respectively located on the left and right sides of the electric-controlled displacement platform. The two sets of halogen light sources are concentrated on the coal sample located on the electric-controlled displacement platform.

[0052] Preferably, the angle between the two groups of halogen light sources and the center of the electric-controlled displacement platform is 45 degrees.

[0053] Furthermore, the performance of the quantitative analysis model in step 5 is analyzed by the root mean square error RMSE, mean absolute error MAE and determination coefficient Three performance indicators are used to evaluate: the smaller the RMSE and MAE values, The closer the value is to 1, the better the performance of the analysis model. The calculation method of each indicator is as follows:

[0054]

[0055]

[0056]

[0057] Where n represents the sample size; and Respectively represent The true value and predicted value of the sample; Indicates the true mean value of all samples.

[0058] Compared with the prior art, the present invention has the following effects:

[0059] 1. Breaking through the limitations of traditional chemical analysis methods (such as oxygen bomb calorimetry) that require sample destruction and a testing cycle of up to 5-8 hours, the system uses a short-wave infrared hyperspectral imaging system (spectral range 1000-2500 nm) to achieve non-destructive testing of coal samples, completing spectrum acquisition and calorific value prediction within 3-5 minutes. The integrity rate of the coal sample after testing reaches 100%, and it can be directly used for industrial combustion or other physical and chemical analysis;

[0060] 2. Hyperspectral data integrates spatial and spectral information, covering key molecular vibration bands such as CH bonds, OH bonds, and CC bonds. Combining filtering algorithms with multivariate scattering correction preprocessing technology, it eliminates instrument noise and sample surface scattering interference, effectively improving the signal-to-noise ratio of key bands;

[0061] 3. The spectral data were reduced in dimension using the partial least squares regression model. The core characteristic wavelengths were screened out through variable importance projection. A quantitative analysis model for calorific value was established. The average percentage error of the prediction was less than 3%, and the prediction determination coefficient ( ) reaches 0.978, which can achieve high-precision prediction of coal calorific value. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of the structure of the hyperspectral imaging system of the present invention;

[0063] Figure 2 It is a schematic flow diagram of the present invention;

[0064] Figure 3 It is a work flow chart of Example 1;

[0065] Figure 4 It is the test result of Example 1;

[0066] In the figure: 1. Hyperspectral imager, 2. Electric-controlled displacement platform, 3. Computer hardware, 4. Halogen light source. DETAILED DESCRIPTION

[0067] Specific implementation method 1: Combination Figures 1 to 2 This embodiment describes a method for detecting the calorific value of coal based on hyperspectral imaging technology, which includes the following steps:

[0068] Step 1: Use coal samples with known calorific value parameters as calibration set coal samples;

[0069] Step 2: Use the hyperspectral imaging system to collect the calibration set coal samples and the corresponding black and white reference hyperspectral image data, and perform spectral data preprocessing;

[0070] Step 3: Using the hyperspectral data of the calibration set after preprocessing in step 2 as input variables and the coal calorific value parameters as dependent variables, a partial least squares quantitative analysis model is constructed;

[0071] Step 4: Use a hyperspectral imager to collect the coal sample to be tested and the corresponding black and white reference hyperspectral image data, and perform spectral data preprocessing;

[0072] Step 5: Using the hyperspectral data of the pre-processed coal sample as input variables, the calorific value of the coal sample is detected using the partial least squares quantitative analysis model constructed in step 3.

[0073] The hyperspectral imaging system in this embodiment is mainly used for collecting hyperspectral image data of a calibration set with known coal quality characteristic parameters and a coal sample to be tested with unknown coal quality characteristic parameters.

[0074] In step five of this embodiment, the calorific value of the coal sample to be tested is detected by using the pre-processed spectral data as the independent variable and the corresponding coal quality calorific value index parameter as the dependent variable to construct a partial least squares regression (PLSR) quantitative analysis model for prediction.

[0075] The partial least squares regression model is a multivariate regression analysis method that finds a linear regression model by projecting the independent and dependent variables into a new space. This method considers the information in both the independent and dependent variables, thereby maximizing the correlation between the independent and dependent variables while reducing the dimensionality.

[0076] Specific implementation method 2: Combination Figures 1 to 2 To illustrate this embodiment, in step 2 of this embodiment, the hyperspectral imaging system uses a white board as a calibration reference when collecting the calibration set coal samples.

[0077] With this setup, the raw image data obtained by the hyperspectral imager in this embodiment is actually the sensor's signal intensity. Calibration of the raw image to true reflectance data requires the acquisition of black and white reference image data. Therefore, during the hyperspectral image data acquisition process, a uniform, stable, and highly reflective white plate is used as a calibration reference. The white reference image is acquired under the same conditions as the raw coal image. Other components and connections are the same as those in Specific Embodiment 1.

[0078] Specific implementation method three: Combination Figures 1 to 2 To explain this embodiment, in step 2 of this embodiment, when collecting calibration set coal samples, the hyperspectral imaging system is placed in a black box for hyperspectral data acquisition. This setup eliminates the influence of dark current from the imaging sensor by requiring the acquisition of a black reference image with the light source turned off and the hyperspectral imager lens completely covered by a non-reflective, opaque mask. Other components and connections are identical to those in Specific Embodiments 1 or 2.

[0079] Specific implementation method four: Combination Figures 1 to 2 To illustrate this embodiment, the process of performing spectral data preprocessing on the black and white reference hyperspectral image data in step 2 of this embodiment is as follows:

[0080] Step 21: Use the collected black and white reference image to calibrate the original image data of the coal sample to obtain the real reflectivity data of the coal sample;

[0081]

[0082] Where, is the original image data of coal sample, is the white reference image data, is the black reference image data;

[0083] Step 22: Use bilateral filtering algorithm to perform spatial filtering on the reflectivity calibrated data;

[0084] Step 2 and 3: On the coal sample hyperspectral image, select a flat area as the region of interest, calculate all the pixels in the region of interest, and obtain the average spectral data;

[0085] Step 24: Use SG filtering algorithm to smooth and reduce noise of spectral data;

[0086] Step 25: Use the interquartile range algorithm to eliminate abnormal samples;

[0087] Step 26: Use the multivariate scatter correction algorithm to eliminate the influence of spectral scattering.

[0088] During the hyperspectral image acquisition process, this embodiment is inevitably affected by various external factors, such as random errors and instrument noise. When the spectral curve of the hyperspectral data appears to be non-smooth and contains many spikes or burrs, it usually indicates the presence of noise. These spikes or burrs may be caused by a variety of factors, including:

[0089] Measurement noise: Noise introduced during data acquisition and measurement can appear as spikes or glitches in the spectrum. This noise can come from sensor and instrument instability or environmental interference.

[0090] Instrument noise: The hyperspectral instrument used may itself introduce noise, such as electronic noise, thermal noise, detector noise, or light source noise.

[0091] Environmental factors: External factors such as illumination changes and atmospheric interference can also lead to non-smooth features in spectral data.

[0092] The accuracy and interpretability of hyperspectral data are closely related to noise management, making it a crucial component of hyperspectral analysis. This helps ensure the reliability of extracted information and features, resulting in pure data and an improved signal-to-noise ratio. Therefore, in practical applications, spatial filtering and spectral smoothing are often used to eliminate data noise.

[0093] Furthermore, during the hyperspectral image acquisition process in step 26 of this embodiment, the sample cannot be guaranteed to be absolutely uniform, so the light source irradiating the coal particles will cause scattering. The Multiple Scattering Correction (MSC) method primarily eliminates the effects of spectral scattering, while also mitigating some effects due to random variation.

[0094] The core concept of MSC is to perform correction based on an ideal spectrum, assuming that each spectral curve has a linear relationship with the ideal spectrum. However, in reality, it is impossible to obtain an absolutely ideal spectrum. Therefore, in practical applications, the average spectrum is generally approximated as the ideal spectrum, and then scatter correction is performed through mathematical transformation. The specific steps are as follows:

[0095]

[0096] Where, is the number of samples, is the average spectrum of all samples, For the Sample spectra, No. The samples are compared to the average spectrum The tilt and translation of is the corresponding linear offset, For the MSC calibration spectrum of each sample.

[0097] This arrangement can eliminate the effects of noise and spectral scattering in the hyperspectral data, and can also eliminate some effects caused by random variation. The other components and connection relationships are the same as any one of the specific embodiments 1 to 3.

[0098] Specific implementation method five: Combination Figures 1 to 2 To illustrate this embodiment, the formula of the bilateral filtering algorithm in step 22 of this embodiment is:

[0099] Formula (1)

[0100] In formula (1), is the pixel value after filtering, is the original pixel value, is the normalized weight, is the neighborhood window, is the spatial domain kernel function, is the pixel domain kernel function;

[0101] The specific forms of the spatial domain kernel function and the pixel value domain kernel function are Gaussian kernel functions, where:

[0102] The mathematical form of the spatial domain Gaussian kernel function is:

[0103] Formula (2)

[0104] In formula (2), is a point in the neighborhood, is the center point position, is the spatial domain standard deviation;

[0105] The mathematical form of the pixel value domain Gaussian kernel function is:

[0106] Formula (3)

[0107] Where, is the pixel value of a point in the neighborhood, is the center pixel value, is the pixel domain standard deviation.

[0108] With this setup, common spatial filtering methods include mean filtering, Gaussian filtering, median filtering, and bilateral filtering. Bilateral filtering is the preferred method for practical applications. Bilateral filtering is a filter that filters based on both spatial and pixel value information. By comprehensively considering both spatial and pixel value information, it smoothes the image while preserving edge information, effectively removing noise and maintaining image detail. The remaining structure and components are the same as any of the first to fourth embodiments.

[0109] Specific implementation method six: combination Figures 1 to 2 To illustrate this embodiment, the smoothing noise reduction process in step 24 of this embodiment includes the following steps:

[0110] Step 241: Divide all wavelength points of the coal sample hyperspectral data into p windows, and divide each window into z wavelength points at equal intervals;

[0111] Step 242: Perform m-order polynomial fitting on the reflectance values ​​of multiple wavelength points, and replace the original value with the fitted value of the central wave point of each window to achieve smoothing.

[0112] With this setup, common spectral smoothing methods include wavelet transform and Savitzky-Golay (SG) smoothing. In practical applications, SG smoothing is preferred. The SG smoothing method is based on local polynomial fitting, which smoothes the data by fitting a local polynomial to the data points, thereby filtering out noise in the spectral data. The remaining components and connections are the same as those in any of the first to third embodiments.

[0113] Specific implementation method seven: combination Figures 1 to 2 To illustrate this embodiment, the process of eliminating abnormal samples in step 25 of this embodiment is as follows:

[0114] Step 251: sort all sample spectral data corresponding to each wavelength point after smoothing filtering in ascending order;

[0115] Among them, the first quartile Q1 is the value corresponding to the 25% percentile in the ascending order data of the corresponding wavelength points, the third quartile Q3 is the value corresponding to the 75% percentile in the ascending order data of the corresponding wavelength points, and the second quartile Q2 is the value corresponding to the 50% percentile in the ascending order data of the corresponding wavelength points;

[0116] Step 252: Use the interquartile range (IQR) to set the upper and lower limits for eliminating abnormal samples:

[0117] According to: IQR = Q3 - Q1;

[0118] Upper=Q3+1.5IQR

[0119] Lower=Q1-1.5IQR

[0120] Here, Upper represents the upper limit of the interquartile range, and Lower represents the lower limit of the interquartile range.

[0121] With this setup, when collecting hyperspectral images of coal samples, measurements that deviate from the overall data distribution due to subjective and objective factors such as instrumentation, environment, and operating methods are considered outliers. If these outliers are not removed and directly used in subsequent modeling, not only will they bias model selection but also affect the accuracy and stability of prediction results, thereby reducing the robustness and generalization performance of the entire model. Therefore, in practical applications, the interquartile range (IQR) method is used to eliminate outliers. The remaining components and connections are the same as those in any of Specific Embodiments 1 to 6.

[0122] Specific implementation method eight: combination Figures 1 to 2 To illustrate this embodiment, the hyperspectral imaging system in step 2 of this embodiment includes a hyperspectral imager 1, an electrically controlled displacement platform 2 and computer hardware 3. The electrically controlled displacement platform 2 is installed on the platform of the frame, and the hyperspectral imager 1 is installed on a metal frame directly above the electrically controlled displacement platform 2. Two groups of halogen light sources 4 are respectively installed on the metal frame, and the two groups of halogen light sources 4 are respectively located on the left and right sides of the electrically controlled displacement platform 2. The two groups of halogen light sources 4 are concentratedly irradiated on the coal sample located on the electrically controlled displacement platform 2.

[0123] In this module, a hyperspectral imager and two halogen light sources are fixed to a horizontal metal frame. To achieve optimal lighting, the illumination direction of the two halogen light sources is adjusted to form a 45-degree angle with the center of the electrically controlled displacement platform. An experimental coal sample is placed in the center of the electrically controlled displacement platform, and the heights of the light sources and hyperspectral imager are further adjusted to ensure clear and complete sample acquisition. To prevent stray light from affecting experimental data acquisition, hyperspectral data acquisition is performed in a black box. The remaining components and connections are the same as those in any of the first to seventh embodiments.

[0124] Specific implementation method nine: combination Figures 1 to 2 To illustrate this embodiment, the angle between the two groups of halogen light sources 4 and the center of the electric-controlled displacement platform 2 is 45 degrees.

[0125] This arrangement facilitates the acquisition of clear data. Other components and connection relationships are the same as those in any one of the specific implementation methods 1 to 8.

[0126] Specific implementation method ten: Combination Figures 1 to 2 In this embodiment, the performance of the quantitative analysis model in step 5 of this embodiment is analyzed by the root mean square error RMSE, mean absolute error MAE and determination coefficient Three performance indicators are used to evaluate: the smaller the RMSE and MAE values, The closer the value is to 1, the better the performance of the analysis model. The calculation method of each indicator is as follows:

[0127]

[0128]

[0129]

[0130] Where n represents the sample size; and Respectively represent The true value and predicted value of the sample; Indicates the true mean value of all samples.

[0131] With this setting, the smaller the RMSE and MAE values ​​are, The closer the value of is to 1, the better the performance of the analysis model is, which can effectively evaluate the prediction effect of the model. The other components and connection relationships are the same as any one of the specific embodiments 1 to 9.

[0132] Combine Figures 1 to 4 Specific embodiment 1 of the present invention is described:

[0133] As attached Figure 3As shown, this embodiment uses 28 kinds of coal samples for experiments, and their calorific value reference values ​​are shown in Table 1.

[0134] Table 1 Reference values ​​of calorific values ​​of experimental coal samples

[0135]

[0136] Hyperspectral images of the experimental coal samples were collected using a hyperspectral imager. The hyperspectral imager used in this example is Model N25E-MCT-IS (ISUZU OPTICS), with a wavelength range of 1000-2500 nm, a spectral resolution of 10 nm, and a total of 256 spectral channels. The exposure time was set to 8.31 ms when collecting the hyperspectral images of the coal samples.

[0137] The light source is also an important component of the hyperspectral imaging system. The light source used in this embodiment is an IT 3900 150W halogen light source, which can provide continuous and smooth spectral information in the visible and near-infrared bands ranging from 400-2500nm.

[0138] When collecting hyperspectral images of coal samples, each sample is placed on an electrically controlled displacement platform, and each sample is collected three times. The average spectrum of the three collections is used as the final analytical spectrum of the coal sample.

[0139] The collected coal sample hyperspectral image is subjected to reflectivity correction, bilateral filtering, ROI extraction and SG smoothing to obtain the coal sample spectral data.

[0140] The IQR algorithm is used to eliminate abnormal samples. In this embodiment, the coal samples numbered 2 and 27 are eliminated as abnormal samples, and finally 26 coal sample spectral data are obtained.

[0141] The spectral data of 26 coal samples were corrected by multivariate scattering (MSC).

[0142] The spxy algorithm was used to divide the spectral data of the 26 coal samples into a calibration set and test samples, with the calibration set accounting for 80%. In this embodiment, five coal samples numbered 4, 7, 17, 18, and 22 were divided as test samples, and the remaining 21 coal samples were divided as calibration set samples.

[0143] The partial least squares regression (PLSR) model was constructed and trained using the hyperspectral data of the calibration set coal samples as input variables and the coal calorific value parameters as dependent variables.

[0144] The hyperspectral data of the coal sample to be tested is used as the input variable, and the calorific value of the coal sample is detected using the trained partial least squares regression (PLSR) quantitative analysis model.

[0145] In this embodiment, the root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination ( ) three performance indicators are used to evaluate the constructed partial least squares regression (PLSR) quantitative analysis model. The specific evaluation indicators are shown in Table 2.

[0146] Table 2 Model performance evaluation indicators

[0147]

[0148] In this embodiment, the test results of the calorific value of the coal sample to be tested are shown in the attached Figure 4 The comparison results of the predicted value and reference value of the coal sample to be tested are shown in Table 3.

[0149] Table 3 Predicted values ​​and reference values ​​of coal samples to be tested

[0150]

[0151] It can be seen that high-precision detection of coal calorific value can be achieved by using hyperspectral imaging technology.

[0152] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting the calorific value of coal based on hyperspectral imaging technology, characterized by: It includes the following steps: Step 1: Use coal samples with known calorific value parameters as calibration set coal samples; Step 2: Use the hyperspectral imaging system to collect the calibration set coal samples and the corresponding black and white reference hyperspectral image data, and perform spectral data preprocessing; Step 3: Using the hyperspectral data of the calibration set after preprocessing in step 2 as input variables and the coal calorific value parameters as dependent variables, a partial least squares quantitative analysis model is constructed; Step 4: Use a hyperspectral imager to collect the coal sample to be tested and the corresponding black and white reference hyperspectral image data, and perform spectral data preprocessing; Step 5: Using the hyperspectral data of the pre-processed coal sample as input variables, the calorific value of the coal sample is detected using the partial least squares quantitative analysis model constructed in step 3.

2. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 1, characterized in that: In step 2, the hyperspectral imaging system uses a white plate as a calibration reference when collecting the calibration set coal samples.

3. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 2, characterized in that: In step 2, when collecting the calibration set coal samples, the hyperspectral imaging system is placed in a black box to collect hyperspectral data.

4. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 1 or 3, characterized in that: The process of spectral data preprocessing of black and white reference hyperspectral image data in step 2 is as follows: Step 21: Use the collected black and white reference image to calibrate the original image data of the coal sample to obtain the real reflectivity data of the coal sample; Where, is the original image data of coal sample, is the white reference image data, is the black reference image data; Step 22: Use bilateral filtering algorithm to perform spatial filtering on the reflectivity calibrated data; Step 2 and 3: On the coal sample hyperspectral image, select a flat area as the region of interest, calculate all the pixels in the region of interest, and obtain the average spectral data; Step 24: Use SG filtering algorithm to smooth and reduce noise of spectral data; Step 25: Use the interquartile range algorithm to eliminate abnormal samples; Step 26: Use the multivariate scatter correction algorithm to eliminate the influence of spectral scattering.

5. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 4, characterized in that: The formula of the bilateral filtering algorithm in step 2 is: Formula (1) In formula (1), is the pixel value after filtering, is the original pixel value, is the normalized weight, is the neighborhood window, is the spatial domain kernel function, is the pixel domain kernel function; The specific forms of the spatial domain kernel function and the pixel value domain kernel function are Gaussian kernel functions, where: The mathematical form of the spatial domain Gaussian kernel function is: Formula (2) In formula (2), is a point in the neighborhood, is the center point position, is the spatial domain standard deviation; The mathematical form of the pixel value domain Gaussian kernel function is: Formula (3) Where, is the pixel value of a point in the neighborhood, is the center pixel value, is the pixel domain standard deviation.

6. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 4, characterized in that: The smoothing noise reduction process in step 24 includes the following steps: Step 241: Divide all wavelength points of the coal sample hyperspectral data into p windows, and divide each window into z wavelength points at equal intervals; Step 242: Perform m-order polynomial fitting on the reflectance values ​​of multiple wavelength points, and replace the original value with the fitted value of the central wave point of each window to achieve smoothing.

7. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 4, characterized in that: The process of eliminating abnormal samples in step 25 is as follows: Step 251: sort all sample spectral data corresponding to each wavelength point after smoothing filtering in ascending order; Among them, the first quartile Q1 is the value corresponding to the 25% percentile in the ascending order data of the corresponding wavelength points, the third quartile Q3 is the value corresponding to the 75% percentile in the ascending order data of the corresponding wavelength points, and the second quartile Q2 is the value corresponding to the 50% percentile in the ascending order data of the corresponding wavelength points; Step 252: Use the interquartile range (IQR) to set the upper and lower limits for eliminating abnormal samples: According to: IQR = Q3 - Q1; Upper=Q3+1.5IQR Lower=Q1-1.5IQR Here, Upper represents the upper limit of the interquartile range, and Lower represents the lower limit of the interquartile range.

8. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 4, characterized in that: The steps to eliminate spectral scattering in step 26 are as follows: Step 261: Calculate the average spectrum of the coal sample as the ideal spectrum; Step 262: Perform linear regression on each sample spectral curve; Step 263: Perform multivariate scattering correction on each sample.

9. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 1, characterized in that: The hyperspectral imaging system in step 2 includes a hyperspectral imager (1), an electrically controlled displacement platform (2) and computer hardware (3), wherein the electrically controlled displacement platform (2) is mounted on a platform of a frame, and the hyperspectral imager (1) is mounted on a metal frame directly above the electrically controlled displacement platform (2), and two sets of halogen light sources (4) are respectively mounted on the metal frames, and the two sets of halogen light sources (4) are respectively located on the left and right sides of the electrically controlled displacement platform (2), and the two sets of halogen light sources (4) are concentratedly irradiated on the coal sample located on the electrically controlled displacement platform (2), and the angle between the two sets of halogen light sources (4) and the center of the electrically controlled displacement platform (2) is 45 degrees.

10. The method for detecting the calorific value of coal based on hyperspectral imaging technology according to claim 1, characterized in that: In step 5, the performance of the quantitative analysis model is analyzed by the root mean square error RMSE, mean absolute error MAE and determination coefficient Three performance indicators are used to evaluate: the smaller the RMSE and MAE values, The closer the value is to 1, the better the performance of the analysis model. The calculation method of each indicator is as follows: Where n represents the sample size; and Respectively represent The true value and predicted value of the sample; Indicates the true mean value of all samples.

Images