A method, device and system for detecting coal quality in coal mines

Through multi-base classifier training and fusion weight calculation, the problem of inaccurate determination of representative spectral curves in coal quality detection is solved, and the accuracy of coal quality detection is improved.

CN118883532BActive Publication Date: 2024-12-06SHAANXI ANBIAO INSPECTION & CERTIFICATION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411364897.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-29
Publication Date
2024-12-06
Estimated Expiration
2044-09-29

AI Technical Summary

Technical Problem

During the coal quality detection process, it is difficult for the prior art to accurately determine the representative spectral curves of each position, resulting in the lack of reference for the screened effective spectral curves and affecting the accuracy of coal quality detection.

Method used

By using the spectral curves of each sampling position on a coal mine product as a spectral category, using multiple base classifiers for training, analyzing the deviation rate and discrimination in the classification results, calculating the fusion weights of each spectral curve, and fusion through weighted fusion Representative spectral curves were obtained, and the effective spectral curves were screened for coal quality detection.

Benefits of technology

Effectively determine the representative spectral curve, improve the reference of the screened effective spectral curve, and ensure the detection accuracy of coal quality detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118883532B_ABST
    Figure CN118883532B_ABST
Patent Text Reader

Abstract

This invention relates to the field of materials testing technology, specifically to a method, apparatus, and system for coal quality testing in coal mines. The method treats the spectral curve at each sampling location as a spectral category. By analyzing the classification bias in the classification results of the classifier and the category distribution of misclassified spectral curves, the discriminative power of the base classifier for each spectral category is determined. Combining the results of multiple base classifiers and the classification correctness of each spectral curve yields the fusion weight of each spectral curve. Through weighted fusion with other spectral curves in the category, a representative spectral curve for each sampling location is obtained, and valid spectral curves are selected for coal quality testing. This invention determines the discriminative power of each spectral category by analyzing the distribution of misclassified spectral curves in the classification results, obtains the fusion weight of each spectral curve, effectively determines representative spectral curves, improves the reference value of the selected valid spectral curves, and thus ensures the accuracy of coal quality testing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of material detection, and in particular to a method, device and system for detecting coal quality in a coal mine. Background Art

[0002] In the process of coal quality detection, LIBS spectroscopy technology is often used for detection. When the spectrometer is performing laser irradiation, there will be problems such as laser irradiation angle and quality differences in coal powder sample preparation. The large number of spectra collected by the sensor will show three forms of breakdown, partial breakdown and non-breakdown. Among them, only the LIBS spectrum of breakdown is valid data. Therefore, when conducting coal quality detection, it is necessary to screen out valid spectral curves from multiple spectral curves. Commonly used screening methods usually select standard data and screen according to the difference between the standard data and the data to be screened. However, coal is a highly heterogeneous product, and the element distribution at different positions may vary greatly. If the accurate standard data at each position cannot be determined, the screened spectral curve cannot be directly used as data for coal quality detection, which will affect the accuracy of coal quality detection. Summary of the invention

[0003] In order to solve the technical problem that the representative spectral curve of each position is inaccurately determined in the prior art during the coal quality detection process, which leads to the lack of reference of the screened effective spectral curve and affects the accuracy of coal quality detection, the purpose of the present invention is to provide a coal mine coal quality detection method, device and system, and the technical scheme adopted is as follows:

[0004] The present invention proposes a method for detecting coal quality in a coal mine, the method comprising:

[0005] A plurality of spectral curves at each sampling position on the coal product are obtained; the spectral curve at each sampling position is used as a spectral category training classifier set to obtain a training result; the classifier set is composed of a plurality of base classifiers, and the training result includes a classification result of each spectral curve;

[0006] Taking any base classifier as the target base classifier, obtaining the deviation rate between the classification result of each spectral curve by the target base classifier and the true classification result, and screening out the misclassified spectral curves according to the deviation rate; taking any spectral category as the target category, obtaining the discrimination of the target base classifier for the target category according to the classification result distribution of the spectral curve of the target category in the misclassified spectral curve;

[0007] For each spectral curve, according to the distinguishability of the spectral category to which the spectral curve belongs in each base classifier and the classification correctness, a fusion weight of each spectral curve is obtained; in each spectral category, all spectral curves are weightedly fused according to the fusion weight to obtain a representative spectral curve of each sampling position;

[0008] An effective spectral curve at each sampling position is screened out in each spectral category according to the representative spectral curve, and coal quality detection is performed according to the effective spectral curve.

[0009] Furthermore, in the initial stage of training the classifier set, for each spectral curve, the cumulative sum of cosine similarities between the spectral curve and each other spectral curve is obtained; the spectral curve with the maximum value of the cumulative sum of cosine similarities is used as the reference curve, and the cosine similarity between each spectral curve and the reference curve is used as the initial weight of each spectral curve for training.

[0010] Furthermore, before training the classifier set, each spectrum category is labeled with a numerical value to obtain a label value for each spectrum category.

[0011] Furthermore, the method for obtaining the deviation rate includes:

[0012] For each spectral curve, a label prediction value when the target base classifier classifies the spectral curve is obtained, and the difference between the label prediction value and the label value is normalized to obtain the deviation rate.

[0013] Furthermore, the distinguishing acquisition method includes:

[0014] Each other spectral category except the target category is combined with the target category to obtain a plurality of category combinations, obtain the distinguishing ability in the category combinations, and use the mean of the distinguishing ability as the distinguishing ability of the target base classifier for the target category;

[0015] The method for acquiring the distinguishing ability includes:

[0016] For a category combination, one category in the category combination is selected as the first category and the other category is selected as the second category; the spectral curve of the first category in the misclassified spectral curve is used as the first misclassified spectral curve; the spectral curve of the second category in the misclassified spectral curve is used as the second misclassified spectral curve;

[0017] In the first misclassified spectral curve, the first misclassified spectral curve classified into the second category by the target base classifier is used as the first misclassified spectral curve to be analyzed; and a first distinguishing ability is obtained according to the proportion of the first misclassified spectral curve to be analyzed in the first misclassified spectral curve and the weight of the first misclassified spectral curve to be analyzed in the target base classifier;

[0018] In the second misclassified spectral curve, the second misclassified spectral curve classified into the first category by the target base classifier is used as the second misclassified spectral curve to be analyzed; and a second distinguishing ability is obtained according to the proportion of the second misclassified spectral curve to be analyzed in the second misclassified spectral curve and the weight of the second misclassified spectral curve to be analyzed in the target base classifier;

[0019] The minimum value of the first distinguishing ability and the second distinguishing ability is used as the distinguishing ability.

[0020] Furthermore, the method for acquiring the misclassified spectral curve includes: taking the spectral curve whose deviation rate is greater than a preset deviation rate threshold as the misclassified spectral curve.

[0021] Furthermore, the method for obtaining the fusion weight includes:

[0022] For any spectral curve, if it is correctly classified in the base classifier, the classification correctness is set to 1, and if it is incorrectly classified, the classification correctness is set to 0; for each base classifier, the discrimination of the spectral category to which the spectral curve belongs in the base classifier is multiplied by the classification correctness to obtain the feature prominence of the spectral curve in the base classifier; the cumulative sum of the feature prominences of all base classifiers is used as the numerator, and the cumulative sum of the discriminability of the spectral category to which the spectral curve belongs in each base classifier is used as the denominator to obtain the fusion weight of the spectral curve.

[0023] Furthermore, based on the representative spectral curve, the effective spectral curve is screened out using an absolute value intensity screening method.

[0024] The present invention also proposes a coal quality detection device for a coal mine, the device comprising:

[0025] A classifier training module is used to obtain multiple spectral curves at each sampling position on the coal product; each sampling position is used as a spectral category to train a classifier set to obtain a training result; the classifier set is composed of multiple base classifiers, and the training result includes a classification result of each spectral curve;

[0026] The distinguishability analysis module is used to take any base classifier as the target base classifier, obtain the deviation rate between the classification result of the target base classifier for each spectral curve and the true classification result, and screen out the misclassified spectral curve according to the deviation rate; take any spectral category as the target category, and obtain the distinguishability of the target base classifier for the target category according to the classification result distribution of the spectral curve of the target category in the misclassified spectral curve;

[0027] The representative spectrum curve acquisition module is used to obtain the fusion weight of each spectrum curve according to the distinguishability of the spectrum category to which the spectrum curve belongs in each base classifier and the classification correctness; in each spectrum category, all spectrum curves are weightedly fused according to the fusion weight to obtain the representative spectrum curve of each sampling position;

[0028] The coal quality detection module is used to screen out the effective spectrum curve of each sampling position in each spectrum category according to the representative spectrum curve, and perform coal quality detection according to the effective spectrum curve.

[0029] The present invention also proposes a coal mine coal quality detection system, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the coal mine coal quality detection method are implemented.

[0030] The present invention has the following beneficial effects:

[0031] The present invention regards the spectral curve of each sampling position as a spectral category. Considering that the representative spectral curve of each sampling position should be a spectral curve with obvious characteristics, the present invention analyzes the characteristic significance of the spectral curve in the category based on the classification result of the classifier, and determines the discrimination of the base classifier for each spectral category by analyzing the classification deviation in the classification result and the category distribution of the misclassified spectral curve. That is, the greater the discrimination, the more obvious the spectral curve characteristics corresponding to the category in the base classifier, and then the fusion weight of each spectral curve can be obtained by combining the results of multiple base classifiers and the classification correctness of each spectral curve. That is, the greater the fusion weight, the more obvious the spectral curve has in multiple base classifiers. Then, the representative spectral curve of each sampling position is obtained by weighted fusion with other spectral curves in the category, and the effective spectral curve is screened out for coal quality detection. The present invention uses the principle of extracting features by the classifier, determines the discrimination of each spectral category by the distribution of the misclassified spectral curve of the classification result, and further obtains the fusion weight of each spectral curve, which can effectively determine the representative spectral curve, improve the reference of the screened effective spectral curve, and thus ensure the detection accuracy of coal quality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1A flow chart of a method for detecting coal quality in a coal mine provided by an embodiment of the present invention;

[0034] Figure 2 A flow chart of a method for acquiring differentiation capability provided by an embodiment of the present invention;

[0035] Figure 3 A structural block diagram of a coal mine coal quality detection device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a coal mine coal quality detection method, device and system proposed according to the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0037] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0038] The embodiment of the present invention is aimed at the coal quality detection process of coal products in the production process of coal mines. Common coal products are products of fixed shapes pressed from coal powder extracted from coal mines. For example, a coal quality detection device proposed in the publication number CN117092034A, in which coal powder is pressed into coal cakes by a cake pressing mechanism, and then the formed coal cakes are subjected to spectral analysis by a spectral acquisition and analysis mechanism to detect coal quality characteristics. The embodiment of the present invention can randomly select 10 sampling areas when performing spectral acquisition and analysis on the basis of the existing device, and the size and shape of the sampling area can be set by itself. In the existing device, for each sampling area, the laser irradiates a high-energy laser beam onto the lens group, and the high-energy laser directly irradiates the coal sample on the placement table after being focused by the lens group. At this time, a high-temperature ionized plasma is generated on the surface of the irradiated coal sample, and then the light radiated outward by the plasma is introduced into the spectrometer along the optical fiber through the lighting device. In the embodiment of the present invention, each sampling area is punctured 100 times, that is, 100 spectral curves can be collected in each sampling area.

[0039] The specific scheme of a coal mine coal quality detection method, device and system provided by the present invention is described in detail below with reference to the accompanying drawings.

[0040] See also Figure 1 , which shows a flow chart of a coal mine coal quality detection method provided by an embodiment of the present invention, the method comprising:

[0041] Step S101: obtaining multiple spectral curves at each sampling position on the coal product; using the spectral curve at each sampling position as a spectral category training classifier set to obtain training results; the classifier set is composed of multiple base classifiers, and the training results include classification results of each spectral curve.

[0042] The embodiment of the present invention takes into account that the information at different positions is different due to the structural heterogeneity of coal products, so the spectral curve of each sampling position is used as a spectral category. In order to obtain the representative spectral curve in each spectral category, it is necessary to analyze the characteristic contingency of the spectral curve in the category. The embodiment of the present invention takes into account that when the classifier classifies the data, the contingency of the feature will cause the classifier to misclassify. If the feature of a certain spectral curve cannot represent the category to which it belongs, it means that the feature of the spectral curve has obvious contingency, which will cause misclassification when the classifier classifies it. Therefore, the embodiment of the present invention uses the collected spectral curve set for training the classifier set, wherein the classifier set is composed of multiple base classifiers. In the embodiment of the present invention, a classifier set is constructed based on the adaboost algorithm. The core idea is to train different weak classifiers for the same training set, and then the weak classifiers are assembled to form a stronger final classifier, that is, the base classifier proposed in the embodiment of the present invention is the weak classifier in the adaboost algorithm. The specific training method and the classifier construction method are technical means well known to those skilled in the art, and will not be repeated here. After the classifier set training is completed, each base classifier corresponds to a classification result for each spectral curve.

[0043] Preferably, in one embodiment of the present invention, in the initial stage of training of the classifier set, for each spectral curve, the cumulative sum of cosine similarities between the spectral curve and each other spectral curve is obtained. The spectral curve with the maximum value of the cumulative sum of cosine similarities is used as the reference curve. The reference curve can roughly represent the spectral characteristics of the entire coal mine product. Therefore, the cosine similarity between each spectral curve and the reference curve is used as the initial weight of each spectral curve for training. The specific training method is a technical means well known to those skilled in the art. During the training process, as the algorithm is iterated, the weight of each spectral curve will also change. After the training is completed, each spectral curve corresponds to a final weight.

[0044] Preferably, in order to facilitate the training of the post-classifier, the embodiment of the present invention marks each spectral category with a numerical value to obtain a label value for each spectral category. For example, in the embodiment of the present invention, there are 10 sampling positions, and they are marked with 1 to 10 in sequence. Similarly, other numerical values ​​can be used for marking in other embodiments, which are not limited here. When the classifier classifies the spectral curve, the output classification result is the label prediction value. For example, if the spectral curve with a spectral category label value of 1 has a label prediction value of 1 in the classifier, it means that the classifier is correctly classified, that is, the smaller the difference between the label prediction value and the actual label value, the more accurate the classification, and the larger the difference, the larger the classification error.

[0045] Step S102: Take any base classifier as the target base classifier, obtain the deviation rate between the classification result of the target base classifier for each spectral curve and the true classification result, and filter out the misclassified spectral curves according to the deviation rate; take any spectral category as the target category, and obtain the discrimination of the target base classifier for the target category according to the classification result distribution of the spectral curve of the target category in the misclassified spectral curve.

[0046] The embodiment of the present invention analyzes the classification results of each base classifier, takes any base classifier as the target base classifier, and then analyzes the deviation rate between the classification result of the target base classifier for each spectral curve and the true classification result. The larger the deviation rate, the worse the classification effect of the target base classifier for the spectral curve, and the more the spectral curve is a misclassified spectral curve.

[0047] Preferably, in one embodiment of the present invention, because each spectral category corresponds to a label value, for each spectral curve, the label prediction value when the target base classifier classifies the spectral curve is obtained, and the difference between the label prediction value and the label value is normalized to obtain the deviation rate. That is, the greater the difference between the label prediction value and the label value, the greater the deviation rate. In the embodiment of the present invention, the normalization method is to use the difference between the prediction value and the label value as the numerator, and the maximum value between the two as the denominator to obtain the normalized deviation rate. Other methods may also be used in other embodiments, which will not be described in detail here.

[0048] In the embodiment of the present invention, the spectrum curve whose deviation rate is greater than the preset deviation rate threshold is used as the misclassified spectrum curve. Because the deviation rate is data with a value range between 0 and 1, the deviation rate threshold is set to 0.2 in the embodiment of the present invention.

[0049] The misclassified spectral curve represents the data set of the spectral curve that is misclassified by the current target base classifier. The misclassification is because the spectral curve of a certain category is classified into another category. Therefore, each spectral category can be analyzed and any spectral category can be taken as the target category. By analyzing the classification result distribution of the spectral curve of the target category in the misclassified spectral curve, the discriminability of the target base classifier for the target category can be obtained.

[0050] Preferably, the embodiment of the present invention takes into account that when the base classifier classifies the spectral curve belonging to spectral category 1, if it is misclassified, it will not be classified into spectral category 2. Similarly, the spectral curve of spectral category 2 will not be misclassified into spectral category 1, which means that the base classifier has a strong ability to distinguish between spectral category 1 and spectral category 2. Therefore, the method for obtaining the distinguishability in the embodiment of the present invention includes:

[0051] In addition to the target category, each other spectral category is combined with the target category to obtain multiple category combinations and obtain the distinguishing ability in the category combination. That is, the distinguishing ability between the target category and each other category is analyzed separately, and the obtained distinguishing ability is further integrated, and the mean of the distinguishing ability is used as the distinguishing ability of the target base classifier for the target category. If the distinguishing ability in the category combination is strong, the final distinguishing ability will also be strong, indicating that the target base classifier can effectively classify the target category. The characteristics of the spectral curve in the target category in the target base classifier are more obvious, and the feature randomness is small.

[0052] Specifically, see Figure 2 , which shows a flow chart of a method for acquiring a distinguishing capability provided by an embodiment of the present invention, the method for acquiring a distinguishing capability includes:

[0053] Step S201: for a category combination, select one category in the category combination as the first category and the other as the second category; use the spectral curve of the first category in the misclassified spectral curve as the first misclassified spectral curve; use the spectral curve of the second category in the misclassified spectral curve as the second misclassified spectral curve.

[0054] Step S202: In the first misclassified spectral curve, the first misclassified spectral curve classified into the second category by the target base classifier is used as the first misclassified spectral curve to be analyzed. A first distinguishing ability is obtained according to the proportion of the first misclassified spectral curve to be analyzed in the first misclassified spectral curve and the weight of the first misclassified spectral curve to be analyzed in the target base classifier.

[0055] The larger the proportion is, the easier it is for the first category to be classified into the second category, and the worse the discrimination ability is. In the embodiment of the present invention, the final weight of the first misclassified spectral curve to be analyzed in the classifier set is further counted, and the accumulated value of the final weight is multiplied by the proportion as the weight. By weighting the proportion, the proportion can be weighted as the confidence, and the weighted result is further negatively correlated and normalized to obtain the first discrimination ability. In the embodiment of the present invention, the first discrimination ability is expressed by the formula:

[0056] ;in The first distinguishing ability, is a natural constant, is the number of the first misclassified spectral curves to be analyzed, N is the number of the first misclassified spectral curves, is the final weight of the i-th first misclassified spectrum curve to be analyzed. That is, in the embodiment of the present invention, an exponential function with a natural constant as the base is used to perform negative correlation mapping and normalization on the weighted result.

[0057] Step S203: In the second misclassified spectral curve, the second misclassified spectral curve classified into the first category by the target base classifier is used as the second misclassified spectral curve to be analyzed; and the second distinguishing ability is obtained according to the proportion of the second misclassified spectral curve to be analyzed in the second misclassified spectral curve and the weight of the second misclassified spectral curve to be analyzed in the target base classifier.

[0058] That is, the method for obtaining the second distinguishing ability is the same as the method for obtaining the first distinguishing ability, only the data analysis basis is different, which will not be elaborated here.

[0059] Step S204: taking the minimum value of the first distinguishing ability and the second distinguishing ability as the distinguishing ability.

[0060] Because the first distinguishing ability and the second distinguishing ability are results obtained using different data bases, the minimum value of the first distinguishing ability and the second distinguishing ability is taken as the distinguishing ability.

[0061] Step S103: For each spectral curve, the fusion weight of each spectral curve is obtained according to the distinguishability of the spectral category to which the spectral curve belongs in each base classifier and the classification correctness; in each spectral category, all spectral curves are weightedly fused according to the fusion weight to obtain a representative spectral curve at each sampling position.

[0062] The present invention aims to analyze the characteristic contingency of the spectral curve at each sampling position, and then determine the representative spectral curve. Therefore, it is necessary to further analyze the analysis of each spectral curve. For each spectral curve, the stronger the distinguishability of the spectral category to which it belongs in a base classifier, and the greater the classification correctness of the spectral curve, the stronger the characteristic obviousness of the spectral curve in the base classifier, and the fusion weight of each spectral curve can be obtained by further combining other base classifiers. The larger the fusion weight, the more obvious the characteristics of the spectral curve, then in each spectral category, all spectral curves are weightedly fused according to the fusion weight to obtain the representative spectral curve of each sampling position. Through the fusion weight, the representative spectral curve can express more obvious features that can represent the sampling position, reduce the information proportion of the characteristic contingency during fusion, and enable the representative spectral curve to better represent the features at the sampling position. That is, each sampling position corresponds to a representative spectral curve.

[0063] Preferably, in one embodiment of the present invention, the method for obtaining the fusion weight includes:

[0064] For any spectral curve, if it is correctly classified in the base classifier, the classification correctness is set to 1, and if it is wrongly classified, the classification correctness is set to 0; for each base classifier, the distinguishability of the spectral category to which the spectral curve belongs in the base classifier is multiplied by the classification correctness to obtain the feature prominence of the spectral curve in the base classifier. That is, the greater the feature prominence of the spectral curve in the base classifier, the more obvious the feature of the base classifier is when classifying it. If the feature prominence is 0, it means that the misclassification is caused by the randomness of the feature.

[0065] Further statistics are made by taking the sum of the feature prominences of all base classifiers as the numerator and the sum of the discriminativeness of the spectral category to which the spectral curve belongs in each base classifier as the denominator to obtain the fusion weight. That is, the sum of the feature prominences is normalized by taking the sum of the discriminativeness as the denominator, which can be expressed as: ;in is the fusion weight, f is the number of base classifiers, is the discriminability of the spectral category to which the spectral curve belongs in the d-th base classifier, is the classification correctness of the spectral curve in the dth base classifier.

[0066] In the embodiment of the present invention, in each spectrum category, all spectrum curves are weighted fused in a weighted average manner according to the fusion weights, so as to obtain a representative spectrum curve of each sampling position.

[0067] Step S104: Filter out the effective spectral curve of each sampling position in each spectral category according to the representative spectral curve, and perform coal quality detection according to the effective spectral curve.

[0068] In the embodiment of the present invention, based on the representative spectrum curve, the absolute value intensity screening method is used to screen out the valid spectrum curve. The absolute value intensity screening method is a technical means well known to those skilled in the art, and specifically, the spectral line intensity on the representative spectrum curve is used as a reference. If the spectral line intensity of the reference does not appear on the spectrum curve at the sampling position, it is identified as an invalid spectrum curve, otherwise it is identified as a valid spectrum curve.

[0069] The use of effective spectral curves for coal quality detection is a well-known technical means for those skilled in the art. Here, only a general method used in one embodiment of the present invention is briefly described, and the specific contents are not repeated. The method includes:

[0070] Characteristic parameters such as spectral slope and absorption valley depth are extracted from the effective spectral curve. These characteristic parameters can reflect the physical or chemical properties of coal, such as coalification degree and volatile matter content. The extracted characteristic parameters are correlated with the chemical composition analysis results of coal to determine which characteristic parameters have a significant correlation with coal quality indicators. Coal quality indicators include carbon content, hydrogen content, sulfur content, ash content, volatile matter, etc. A prediction model is established based on spectral data and chemical composition analysis results using statistical or machine learning methods. The model can be used to predict the coal quality indicators of unknown samples. The characteristic parameters represented by the spectral data are input into the model to output the corresponding coal quality indicators.

[0071] In summary, the embodiment of the present invention takes the spectral curve of each sampling position as a spectral category. Considering that the representative spectral curve of each sampling position should be a spectral curve with obvious characteristics, the present invention analyzes the characteristic significance of the spectral curve in the category based on the classification result of the classifier, and determines the discrimination of the base classifier for each spectral category by analyzing the classification deviation in the classification result and the category distribution of the misclassified spectral curve. That is, the greater the discrimination, the more obvious the spectral curve characteristics corresponding to the category in the base classifier, and then the fusion weight of each spectral curve can be obtained by combining the results of multiple base classifiers and the classification correctness of each spectral curve. That is, the greater the fusion weight, the more obvious the spectral curve has in multiple base classifiers. Then, the representative spectral curve of each sampling position is obtained by weighted fusion with other spectral curves in the category, and the effective spectral curve is screened out for coal quality detection. The present invention uses the principle of extracting features by the classifier, determines the discrimination of each spectral category by the distribution of the misclassified spectral curve of the classification result, and further obtains the fusion weight of each spectral curve, which can effectively determine the representative spectral curve, improve the reference of the screened effective spectral curve, and thus ensure the detection accuracy of coal quality detection.

[0072] Based on the same inventive concept, please refer to Figure 3, which shows a structural block diagram of a coal mine coal quality detection device provided by an embodiment of the present invention, the device comprises:

[0073] The classifier training module 301 is used to obtain multiple spectral curves at each sampling position on the coal product. Each sampling position is used as a spectral category to train a classifier set to obtain a training result. The classifier set is composed of multiple base classifiers, and the training result includes the classification result of each spectral curve.

[0074] The distinguishability analysis module 302 is used to take any base classifier as the target base classifier, obtain the deviation rate between the classification result of each spectral curve of the target base classifier and the true classification result, and screen the misclassified spectral curve according to the deviation rate. Take any spectral category as the target category, and obtain the distinguishability of the target base classifier for the target category according to the classification result distribution of the spectral curve of the target category in the misclassified spectral curve.

[0075] The representative spectrum curve acquisition module 303 is used to obtain the fusion weight of each spectrum curve according to the distinguishability of the spectrum category to which the spectrum curve belongs in each base classifier and the classification correctness. In each spectrum category, all spectrum curves are weightedly fused according to the fusion weight to obtain the representative spectrum curve of each sampling position.

[0076] The coal quality detection module 304 is used to select the effective spectrum curve of each sampling position in each spectrum category according to the representative spectrum curve, and perform coal quality detection according to the effective spectrum curve.

[0077] The present invention also proposes a coal mine coal quality detection system, the system includes a memory, a processor and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, any step of a coal mine coal quality detection method is implemented.

[0078] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0079] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for detecting coal quality in a coal mine, characterized in that: The method comprises: A plurality of spectral curves at each sampling position on the coal product are obtained; the spectral curve at each sampling position is used as a spectral category training classifier set to obtain a training result; the classifier set is composed of a plurality of base classifiers, and the training result includes a classification result of each spectral curve; Taking any base classifier as the target base classifier, obtaining the deviation rate between the classification result of each spectral curve by the target base classifier and the true classification result, and screening out the misclassified spectral curves according to the deviation rate; taking any spectral category as the target category, obtaining the discrimination of the target base classifier for the target category according to the classification result distribution of the spectral curve of the target category in the misclassified spectral curve; For each spectral curve, according to the distinguishability of the spectral category to which the spectral curve belongs in each base classifier and the classification correctness, a fusion weight of each spectral curve is obtained; in each spectral category, all spectral curves are weightedly fused according to the fusion weight to obtain a representative spectral curve of each sampling position; Screening out effective spectral curves at each sampling position in each spectral category according to the representative spectral curves, and performing coal quality detection according to the effective spectral curves; The distinguishing acquisition method includes: Each other spectral category except the target category is combined with the target category to obtain a plurality of category combinations, obtain the distinguishing ability in the category combinations, and use the mean of the distinguishing ability as the distinguishing ability of the target base classifier for the target category; The method for acquiring the distinguishing ability includes: For a category combination, one category in the category combination is selected as the first category and the other category is selected as the second category; the spectral curve of the first category in the misclassified spectral curve is used as the first misclassified spectral curve; the spectral curve of the second category in the misclassified spectral curve is used as the second misclassified spectral curve; In the first misclassified spectral curve, the first misclassified spectral curve classified into the second category by the target base classifier is used as the first misclassified spectral curve to be analyzed; and a first distinguishing ability is obtained according to the proportion of the first misclassified spectral curve to be analyzed in the first misclassified spectral curve and the weight of the first misclassified spectral curve to be analyzed in the target base classifier; In the second misclassified spectral curve, the second misclassified spectral curve classified into the first category by the target base classifier is used as the second misclassified spectral curve to be analyzed; and a second distinguishing ability is obtained according to the proportion of the second misclassified spectral curve to be analyzed in the second misclassified spectral curve and the weight of the second misclassified spectral curve to be analyzed in the target base classifier; The minimum value of the first distinguishing ability and the second distinguishing ability is used as the distinguishing ability.

2. A method for detecting coal quality in a coal mine according to claim 1, characterized in that: In the initial stage of training the classifier set, for each spectral curve, the cumulative sum of cosine similarities between the spectral curve and each other spectral curve is obtained; the spectral curve with the maximum value of the cumulative sum of cosine similarities is used as the reference curve, and the cosine similarity between each spectral curve and the reference curve is used as the initial weight of each spectral curve for training.

3. A coal mine coal quality detection method according to claim 1, characterized in that: Before training the classifier set, each spectral category is labeled with a numerical value to obtain a label value for each spectral category.

4. A coal mine coal quality detection method according to claim 3, characterized in that: The method for obtaining the deviation rate includes: For each spectral curve, a label prediction value when the target base classifier classifies the spectral curve is obtained, and the difference between the label prediction value and the label value is normalized to obtain the deviation rate.

5. A coal mine coal quality detection method according to claim 1, characterized in that: The method for acquiring the misclassified spectral curve includes: taking the spectral curve whose deviation rate is greater than a preset deviation rate threshold as the misclassified spectral curve.

6. A method for detecting coal quality in a coal mine according to claim 1, characterized in that: The method for obtaining the fusion weight includes: For any spectral curve, if it is correctly classified in the base classifier, the classification correctness is set to 1, and if it is incorrectly classified, the classification correctness is set to 0; for each base classifier, the discrimination of the spectral category to which the spectral curve belongs in the base classifier is multiplied by the classification correctness to obtain the feature prominence of the spectral curve in the base classifier; the cumulative sum of the feature prominences of all base classifiers is used as the numerator, and the cumulative sum of the discriminability of the spectral category to which the spectral curve belongs in each base classifier is used as the denominator to obtain the fusion weight of the spectral curve.

7. A method for detecting coal quality in a coal mine according to claim 1, characterized in that: Based on the representative spectral curve, the effective spectral curve is screened out using an absolute value intensity screening method.

8. A coal quality detection device for a coal mine, characterized in that: The device comprises: A classifier training module is used to obtain multiple spectral curves at each sampling position on the coal product; each sampling position is used as a spectral category to train a classifier set to obtain a training result; the classifier set is composed of multiple base classifiers, and the training result includes a classification result of each spectral curve; The distinguishability analysis module is used to take any base classifier as the target base classifier, obtain the deviation rate between the classification result of the target base classifier for each spectral curve and the true classification result, and screen out the misclassified spectral curve according to the deviation rate; take any spectral category as the target category, and obtain the distinguishability of the target base classifier for the target category according to the classification result distribution of the spectral curve of the target category in the misclassified spectral curve; The representative spectrum curve acquisition module is used to obtain the fusion weight of each spectrum curve according to the distinguishability of the spectrum category to which the spectrum curve belongs in each base classifier and the classification correctness; in each spectrum category, all spectrum curves are weightedly fused according to the fusion weight to obtain the representative spectrum curve of each sampling position; A coal quality detection module, used for screening out a valid spectrum curve of each sampling position in each spectrum category according to the representative spectrum curve, and performing coal quality detection according to the valid spectrum curve; The distinguishing acquisition method includes: Each other spectral category except the target category is combined with the target category to obtain a plurality of category combinations, obtain the distinguishing ability in the category combinations, and use the mean of the distinguishing ability as the distinguishing ability of the target base classifier for the target category; The method for acquiring the distinguishing ability includes: For a category combination, one category in the category combination is selected as the first category and the other category is selected as the second category; the spectral curve of the first category in the misclassified spectral curve is used as the first misclassified spectral curve; the spectral curve of the second category in the misclassified spectral curve is used as the second misclassified spectral curve; In the first misclassified spectral curve, the first misclassified spectral curve classified into the second category by the target base classifier is used as the first misclassified spectral curve to be analyzed; and a first distinguishing ability is obtained according to the proportion of the first misclassified spectral curve to be analyzed in the first misclassified spectral curve and the weight of the first misclassified spectral curve to be analyzed in the target base classifier; In the second misclassified spectral curve, the second misclassified spectral curve classified into the first category by the target base classifier is used as the second misclassified spectral curve to be analyzed; and a second distinguishing ability is obtained according to the proportion of the second misclassified spectral curve to be analyzed in the second misclassified spectral curve and the weight of the second misclassified spectral curve to be analyzed in the target base classifier; The minimum value of the first distinguishing ability and the second distinguishing ability is used as the distinguishing ability.

9. A coal mine coal quality detection system, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of a coal mine coal quality detection method as described in any one of claims 1 to 7 are implemented.