Method and device for extracting spectral characteristics of water content

Abstract

The invention relates to the technical field of multi-modal learning and coal analysis, in particular to a water content spectral feature extraction method and device. Coal spectral feature matrixes under different measurement distances and initial air-dry basis moisture labels corresponding to the coal spectral feature matrixes are extracted from target coal spectral data; determining a final characteristic wave band based on the importance rank of the initial characteristic wave band of the coal spectral characteristic matrix; and determining a final characteristic wave band interval based on the final characteristic wave band. Therefore, the problems that in the related technology, most of visible light image fusion schemes depend on multi-modal data synchronous calibration, the calculation complexity is high, and noise is sensitive are solved; the feature engineering in the near infrared spectrum depends on prior knowledge, the anti-interference capability is limited, and the classification model adaptability is insufficient.

Description

Technical Field

[0001] This application relates to the fields of multimodal learning and coal analysis technology, and in particular to a method and apparatus for extracting the spectral characteristics of moisture content. Background Technology

[0002] In related technologies, the moisture content of coal can be located by fusing visible light image semantic segmentation, extracting moisture-related spectral features by combining hyperspectral data sliding window and attention mechanism, and generating a feature matrix by fusing spatial-spectral features using a transformer model. After pooling, the matrix is ​​input into a least squares support vector machine model to achieve pixel-level moisture content prediction. Finally, the moisture content and its spatial distribution of coal are accurately presented through graded color rendering. Alternatively, near-infrared spectra of control and test samples can be collected, absorption vectors can be constructed, and absorption peak significance can be quantified to screen target bands. Further extraction of absorption level vectors, absorption peak recognition, and potential information feature values ​​can be performed. Finally, a classification model is trained based on the coal gangue recognition index of multi-dimensional feature fusion to achieve high-precision intelligent identification of coal gangue with resistance to noise interference.

[0003] However, among related technologies, most schemes that fuse visible light images rely on synchronous calibration of multimodal data, which has high computational complexity and is sensitive to noise; while feature engineering in near-infrared spectroscopy relies on prior knowledge, has limited anti-interference ability, and insufficient adaptability of classification models, which urgently need improvement. Summary of the Invention

[0004] This application provides a method and apparatus for extracting water content spectral features, in order to solve the problems in related technologies, such as the fact that most schemes that fuse visible light images rely on synchronous calibration of multimodal data, which has high computational complexity and is sensitive to noise; while feature engineering in near-infrared spectroscopy relies on prior knowledge, has limited anti-interference ability, and insufficient adaptability of classification models.

[0005] The first aspect of this application provides a method for extracting moisture content spectral features, comprising the following steps: extracting coal spectral feature matrices at different measurement distances and initial air-dried basis moisture labels corresponding to the coal spectral feature matrices from target coal spectral data under air-dried baseline moisture content conditions; determining final feature bands based on the importance ranking of the initial feature bands of the coal spectral feature matrix; and determining the final feature band interval based on the final feature bands.

[0006] The above technical solution enables the extraction of coal spectral feature matrices and corresponding initial air-dry basis moisture labels under air-drying baseline moisture content conditions. This is achieved by introducing different measurement distance dimensions and then selecting final feature band intervals based on the importance ranking of the initial feature bands. This allows for the establishment of a high-precision air-dry basis moisture prediction model. By introducing feature training at different measurement distances, the model learns the influence of distance changes on the spectrum, thereby reducing errors caused by probe distance fluctuations in actual detection and improving the model's robustness and adaptability. Selecting feature bands and intervals based on importance rankings eliminates redundant and noisy bands, retaining the core spectral information most sensitive to moisture, making the model lighter and with clearer physical meaning. Establishing a model based on an air-drying baseline eliminates interference from external environmental humidity fluctuations, establishes a unified moisture measurement standard, and facilitates horizontal comparison and standardized management of coal moisture data from different batches and under different environments.

[0007] Optionally, in one embodiment of this application, before determining the final characteristic band, the method further includes: calculating a first initial importance ranking of the initial characteristic band based on the initial characteristic band and the initial air-dried basis moisture label; calculating a second initial importance ranking of the initial characteristic band based on the interaction between different initial characteristic bands; obtaining a comparison result between the first initial importance ranking and the second initial importance ranking; and determining the importance ranking of the initial characteristic band in response to the comparison result satisfying a preset condition.

[0008] The above technical solution allows for the calculation of a first initial importance ranking and a second initial importance ranking before determining the final characteristic bands. Based on the comparison between the first and second initial importance rankings, and provided the comparison results meet preset conditions, the importance ranking is determined. By fusing dual importance assessments based on label association and band interaction, and dynamically determining the priority of characteristic bands based on the comparison results, the scientific nature and anti-interference capability of feature selection are effectively improved.

[0009] Optionally, in one embodiment of this application, the step of determining the importance ranking of the initial feature band in response to the comparison result satisfying a preset condition includes: obtaining the predicted air-dry basis moisture label corresponding to the coal spectral feature matrix; calculating the mean square error value between the initial air-dry basis moisture label and the predicted air-dry basis moisture label; determining a first weight value corresponding to the first initial importance ranking and a second weight value corresponding to the second initial importance ranking based on the mean square error value; calculating the total score of the corresponding initial feature band based on the first weight value, the second weight value, and the importance ranking; and determining the final feature band in response to the total score being greater than a preset score.

[0010] The above technical solution can determine the first weight value corresponding to the first initial importance ranking and the second weight value corresponding to the second initial importance ranking based on the mean square error between the predicted air-dried basis moisture label and the initial air-dried basis moisture label corresponding to the coal spectral feature matrix. Then, the total score of the corresponding initial feature bands can be calculated. When the total score is greater than the preset score, the final feature band is determined. By dynamically adjusting the weight of the dual importance rankings in combination with the prediction error and screening feature bands with high total scores, the accuracy of feature selection and the robustness of model prediction are effectively improved.

[0011] Optionally, in one embodiment of this application, determining the final characteristic band based on the importance ranking of the initial characteristic bands of the coal spectral feature matrix includes: calculating a third initial importance ranking of the initial characteristic bands based on the coal spectral feature matrix; and determining the final characteristic band based on the third initial importance ranking.

[0012] The above technical solution can be used to calculate the third initial importance ranking of the initial feature bands based on the coal spectral feature matrix, thereby determining the final feature bands. By introducing the third initial importance ranking to conduct a secondary evaluation of the initial feature bands, the feature bands that have a significant impact on the target task (such as coal moisture content prediction) can be screened more comprehensively and accurately, thus improving the model performance.

[0013] Optionally, in one embodiment of this application, determining the final characteristic band interval based on the final characteristic band includes: calculating the mean square error value of the final characteristic band; and determining the final characteristic band interval based on the mean square error value of the final characteristic band.

[0014] The above technical solution can determine the final feature band interval by calculating the mean square error value of the final feature band. By dynamically determining the final feature band interval based on the mean square error value, noise interference can be effectively eliminated and the key band range can be focused, thereby improving the stability of feature expression and the accuracy of model prediction.

[0015] A second aspect of this application provides an apparatus for extracting spectral features of moisture content, comprising: an extraction module for extracting coal spectral feature matrices at different measurement distances and initial air-dried basis moisture labels corresponding to the coal spectral feature matrices from target coal spectral data under air-dried baseline moisture content conditions; a first determination module for determining final feature bands based on the importance ranking of the initial feature bands of the coal spectral feature matrix; and a second determination module for determining a final feature band interval based on the final feature bands.

[0016] The above technical solution enables the extraction of coal spectral feature matrices and corresponding initial air-dry basis moisture labels under air-drying baseline moisture content conditions. This is achieved by introducing different measurement distance dimensions and then selecting final feature band intervals based on the importance ranking of the initial feature bands. This allows for the establishment of a high-precision air-dry basis moisture prediction model. By introducing feature training at different measurement distances, the model learns the influence of distance changes on the spectrum, thereby reducing errors caused by probe distance fluctuations in actual detection and improving the model's robustness and adaptability. Selecting feature bands and intervals based on importance rankings eliminates redundant and noisy bands, retaining the core spectral information most sensitive to moisture, making the model lighter and with clearer physical meaning. Establishing a model based on an air-drying baseline eliminates interference from external environmental humidity fluctuations, establishes a unified moisture measurement standard, and facilitates horizontal comparison and standardized management of coal moisture data from different batches and under different environments.

[0017] Optionally, in one embodiment of this application, it further includes: a first calculation module, configured to calculate a first initial importance ranking of the initial characteristic band based on the initial characteristic band and the initial air-dried basis moisture label before determining the final characteristic band; a second calculation module, configured to calculate a second initial importance ranking of the initial characteristic band based on the interaction between different initial characteristic bands; an acquisition module, configured to acquire a comparison result between the first initial importance ranking and the second initial importance ranking; and a third determination module, configured to determine the importance ranking of the initial characteristic band in response to the comparison result satisfying a preset condition.

[0018] The above technical solution allows for the calculation of a first initial importance ranking and a second initial importance ranking before determining the final characteristic bands. Based on the comparison between the first and second initial importance rankings, and provided the comparison results meet preset conditions, the importance ranking is determined. By fusing dual importance assessments based on label association and band interaction, and dynamically determining the priority of characteristic bands based on the comparison results, the scientific nature and anti-interference capability of feature selection are effectively improved.

[0019] Optionally, in one embodiment of this application, the third determining module includes: an acquisition unit, configured to acquire the predicted air-dry basis moisture label corresponding to the coal spectral feature matrix; a first calculation unit, configured to calculate the mean square error value between the initial air-dry basis moisture label and the predicted air-dry basis moisture label; a first determining unit, configured to determine a first weight value corresponding to the first initial importance ranking and a second weight value corresponding to the second initial importance ranking based on the mean square error value; a second calculation unit, configured to calculate the total score of the corresponding initial feature band based on the first weight value, the second weight value, and the importance ranking; and a second determining unit, configured to determine the final feature band in response to the total score being greater than a preset score.

[0020] The above technical solution can determine the first weight value corresponding to the first initial importance ranking and the second weight value corresponding to the second initial importance ranking based on the mean square error between the predicted air-dried basis moisture label and the initial air-dried basis moisture label corresponding to the coal spectral feature matrix. Then, the total score of the corresponding initial feature bands can be calculated. When the total score is greater than the preset score, the final feature band is determined. By dynamically adjusting the weight of the dual importance rankings in combination with the prediction error and screening feature bands with high total scores, the accuracy of feature selection and the robustness of model prediction are effectively improved.

[0021] Optionally, in one embodiment of this application, the first determining module includes: a third calculation unit, configured to calculate the third initial importance ranking of the initial feature band based on the coal spectral feature matrix; and a third determining unit, configured to determine the final feature band based on the third initial importance ranking.

[0022] The above technical solution can be used to calculate the third initial importance ranking of the initial feature bands based on the coal spectral feature matrix, thereby determining the final feature bands. By introducing the third initial importance ranking to conduct a secondary evaluation of the initial feature bands, the feature bands that have a significant impact on the target task (such as coal moisture content prediction) can be screened more comprehensively and accurately, thus improving the model performance.

[0023] Optionally, in one embodiment of this application, the second determining module includes: a fourth calculation unit, configured to calculate the mean square error value of the final characteristic band; and a fourth determining unit, configured to determine the final characteristic band interval based on the mean square error value of the final characteristic band.

[0024] The above technical solution can determine the final feature band interval by calculating the mean square error value of the final feature band. By dynamically determining the final feature band interval based on the mean square error value, noise interference can be effectively eliminated and the key band range can be focused, thereby improving the stability of feature expression and the accuracy of model prediction.

[0025] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for extracting water content spectral features as described in the above embodiments.

[0026] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for extracting water content spectral features.

[0027] A fifth aspect of this application provides a computer program product, including a computer program that, when executed, implements the above-described method for extracting water content spectral features.

[0028] This application's embodiments can establish a high-precision air-dried basis moisture prediction model by extracting the spectral feature matrix of coal and corresponding initial air-dried basis moisture labels under air-dried baseline moisture content conditions, and selecting the final feature band intervals based on the importance ranking of the initial feature bands. By introducing feature training at different measurement distances, the model can learn the influence of distance changes on the spectrum, thereby reducing errors caused by probe distance fluctuations in actual detection and improving the robustness and adaptability of the model. Using importance rankings to select feature bands and intervals can eliminate redundant and noisy bands, retaining the core spectral information most sensitive to moisture, making the model lighter and with clearer physical meaning. Establishing the model based on an air-dried baseline eliminates the interference of external environmental humidity fluctuations, establishes a unified moisture measurement standard, and facilitates horizontal comparison and standardized management of coal moisture data from different batches and under different environments. This solves the problems in related technologies, where schemes fusing visible light images mostly rely on multimodal data synchronous calibration, resulting in high computational complexity and sensitivity to noise; while feature engineering in near-infrared spectroscopy relies on prior knowledge, has limited anti-interference capabilities, and insufficient adaptability of classification models.

[0029] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0030] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for extracting water content spectral features according to an embodiment of this application; Figure 2 This is a schematic diagram showing the comparison of mean squared error values ​​of nine methods provided according to an embodiment of this application on a dataset; Figure 3 The image shows the importance curves of each band in four modal datasets based on PLS (Partial Least Squares) + VIP (Variable Importance in the Projection) + correlation stability analysis provided according to an embodiment of this application. Figure 4 A flowchart illustrating the working principle of a method for extracting water content spectral features according to an embodiment of this application; Figure 5 This is a block diagram of an extraction device for water content spectral characteristics according to an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0031] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0032] The following description, with reference to the accompanying drawings, illustrates a method and apparatus for extracting moisture content spectral features according to embodiments of this application. Addressing the issues raised in the background art, where most visible light image fusion schemes rely on synchronous calibration of multimodal data, resulting in high computational complexity and sensitivity to noise, and near-infrared spectral feature engineering relies on prior knowledge, exhibiting limited anti-interference capabilities and insufficient adaptability of classification models, this application provides a method for extracting moisture content spectral features. In this method, under air-dried baseline moisture content conditions, different measurement distance dimensions are introduced to extract the coal spectral feature matrix and corresponding initial air-dried basis moisture labels. The final feature band intervals are then selected based on the importance ranking of the initial feature bands, thereby establishing a high-precision air-dried basis moisture prediction model. This method, by introducing feature training at different measurement distances, can learn the influence of distance changes on the spectrum, thereby reducing errors caused by probe distance fluctuations in actual detection and improving the robustness and adaptability of the model. By using importance ranking to select feature bands and intervals, redundant and noisy bands can be eliminated, retaining the core spectral information most sensitive to moisture, making the model lighter and with clearer physical meaning. Based on an air dryness benchmark, the model eliminates interference from external environmental humidity fluctuations, establishing a unified moisture measurement standard, which is conducive to horizontal comparison and standardized management of coal moisture data from different batches and under different environments. Therefore, it solves the problems in related technologies, such as the fact that most schemes integrating visible light images rely on multimodal data synchronous calibration, resulting in high computational complexity and sensitivity to noise; and that feature engineering in near-infrared spectroscopy relies on prior knowledge, has limited anti-interference capabilities, and insufficient adaptability of classification models.

[0033] Specifically, Figure 1 This is a flowchart of a method for extracting water content spectral features according to an embodiment of this application.

[0034] like Figure 1 As shown, the method for extracting the spectral characteristics of water content includes the following steps: In step S101, under the air-dried baseline moisture content condition, the coal spectral feature matrix at different measurement distances and the initial air-dried baseline moisture label corresponding to the coal spectral feature matrix are extracted from the target coal spectral data.

[0035] It is understood that, in the embodiments of this application, the air-drying baseline moisture content condition can be understood as the moisture content state of coal when it reaches equilibrium with the surrounding air humidity under standard environmental conditions (temperature 20-30℃, relative humidity 50-70%) through natural or forced ventilation. At this time, the moisture content of the coal is calculated based on the air-dried baseline mass, that is, the percentage of moisture mass to the mass of the coal sample after air drying.

[0036] In some embodiments, this application can extract the coal spectral feature matrix at different measurement distances from the target coal spectral data under air-dried baseline moisture content conditions, and simultaneously determine the corresponding initial air-dried baseline moisture label.

[0037] It should be noted that, in the embodiments of this application, the coal spectral feature matrix can be understood as a two-dimensional matrix formed by arranging the key features (such as absorption peak intensity, wavelength point reflectance, derivative spectrum, etc.) extracted from the spectral data of coal samples in a specific band range after preprocessing (such as baseline correction, noise filtering, normalization, etc.) according to the sample dimensions. The rows represent samples (such as coal from different origins and with different moisture contents), and the columns represent characteristic bands (such as wavelengths near the moisture absorption peaks of 1450nm and 1940nm in the near-infrared region).

[0038] For example, embodiments of this application can be experimentally verified on a coal spectral dataset with air-dried basis moisture content. This dataset contains four modal datasets, each representing a coal spectral feature matrix measured by a spectrometer at different distances: 5mm, 10mm, 15mm, and 20mm. Each modal dataset matrix is ​​781*125 in size, where 785 represents the number of samples and 125 represents the spectral feature dimension of each sample, i.e., the total number of feature bands. It can be, but is not limited to, represented as: ,in, These are the spectral characteristic matrices of coal with moisture content on an air-drying basis. This represents the spectral feature matrix obtained at a measurement distance of j millimeters. ,in, Indicates the number of samples. This indicates the dimensionality of the spectral features for each sample.

[0039] The initial air-dried basis moisture label corresponding to the coal spectral dataset of air-dried basis moisture can be represented as: ,in, This represents the label corresponding to the spectral feature matrix obtained at a measurement distance of j millimeters. In step S102, the final characteristic bands are determined based on the importance ranking of the initial characteristic bands in the coal spectral characteristic matrix.

[0040] It is understood that, in the embodiments of this application, the importance ranking of characteristic bands can be calculated using five correlation analysis methods and four model-driven methods.

[0041] Among them, the five correlation analysis methods may include, but are not limited to, Kendall's rank correlation coefficient, Spearman's correlation coefficient, distance correlation, maximum information coefficient, and mutual information, etc., and the embodiments of this application do not impose specific limitations.

[0042] Among them, Kendall's rank correlation coefficient calculates the correlation by comparing the order consistency of data pairs, and is suitable for small samples or situations where there are ties.

[0043] The Spearman rank correlation coefficient calculates the strength of monotonic relationships using rank and is suitable for non-normally distributed or ordered data.

[0044] Distance correlation is based on distance covariance to detect any type of dependency (including nonlinearity) and is suitable for multidimensional variable analysis.

[0045] The maximum information coefficient is calculated by dividing the data into grids to obtain the maximum mutual information between variables. It can detect linear and nonlinear relationships (such as periodic and functional relationships) in a balanced way and has fairness (the coefficients are similar for different types of relationships).

[0046] Mutual information directly measures the amount of information shared between features and the target variable, and is suitable for classification problems (such as feature selection).

[0047] The four model-driven methods may include, but are not limited to, RF (Random Forest)-MDI (Mean Decrease Impurity), RF-MDA (Mean Decrease Accuracy), extreme gradient boosting, and linear regression, etc., and this application does not impose specific limitations.

[0048] RF-MDI measures the improvement in data purity when a feature splits at a decision tree node, based on impurity metrics such as Gini impurity or information gain. In a random forest, each tree selects a split node based on the reduction in feature impurity; RF-MDI, however, evaluates feature importance by calculating the average reduction in impurity across all trees for that feature.

[0049] RF-MDA measures feature importance by assessing the decrease in model accuracy after randomly permuting feature values. Specifically, for each feature, its value in the dataset is randomly permuted, the model's accuracy is recalculated, and the decrease in accuracy is calculated. The average decrease in accuracy for all features is then calculated to obtain the average accuracy decrease for that feature.

[0050] Extreme gradient boosting models (such as XGBoost) also provide methods for evaluating feature importance, which typically reflect the contribution of features to model prediction based on metrics such as feature gain, coverage, or frequency in the model.

[0051] Linear regression predicts a target variable by fitting a linear equation. The importance of features can be measured by the absolute value of the coefficients or the standardized coefficients of the linear equation. The larger the absolute value of the coefficient or the larger the standardized coefficient, the greater the influence of the feature on the target variable.

[0052] The final feature bands can be selected from the optimal subset of the initial feature bands based on their importance ranking, and used to build a high-precision, low-complexity water content prediction model.

[0053] In some embodiments, the importance ranking of the initial characteristic bands in the coal spectral feature matrix can be calculated first, and then the final characteristic bands can be determined based on the importance ranking.

[0054] Optionally, in one embodiment of this application, before determining the final characteristic band, the method further includes: calculating a first initial importance ranking of the initial characteristic band based on the initial characteristic band and the initial air-dried basis moisture label; calculating a second initial importance ranking of the initial characteristic band based on the interaction between different initial characteristic bands; obtaining a comparison result between the first initial importance ranking and the second initial importance ranking; and determining the importance ranking of the initial characteristic band in response to the comparison result meeting a preset condition.

[0055] Understandably, correlation analysis methods can be used to capture the relationship between individual spectral feature bands and labels.

[0056] Therefore, in this application embodiment, the first initial importance ranking of the initial characteristic band can be calculated using five correlation analysis methods based on the initial characteristic band and the initial air-dried basis moisture label.

[0057] Model-driven methods can be used to capture the interactions between multiple spectral feature bands and explore the relationship between multispectral feature bands and labels.

[0058] Therefore, embodiments of this application can calculate the second initial importance ranking of the initial characteristic band based on the interaction between the initial characteristic bands using four model-driven methods.

[0059] For example, in this application embodiment, network training and performance verification are performed on an air-dried basis moisture coal spectral dataset using these nine methods (five correlation analysis methods and four model-driven methods) to obtain the importance ranking of initial feature bands, extract the top 10 feature bands most relevant to the labels, and obtain the initial feature bands. ,in, Indicates the measurement using the i-th method. The top 10 most critical feature bands under the matrix are shown in Table 1. Table 1 is a schematic table of the importance ranking of the top 10 features and nine methods according to an embodiment of this application.

[0060] Table 1

[0061] Where K represents the Kendall rank correlation coefficient, Sp represents the Spearman correlation coefficient, D represents the distance correlation, Mic represents the maximum information coefficient, Mi represents mutual information, Mdi represents RF-MDI, Mda represents RF-MDA, Xgb represents extreme gradient boosting, Lr represents linear regression, _r represents the spectral band order corresponding to the m-th method, and band number i is related to band length. relation .

[0062] Furthermore, embodiments of this application can perform regional marking on the band, marking... ,in, The ranking of the characteristic bands measured by the i-th method is shown in Table 1, thus obtaining a new label table as shown in Table 2. The average label value is calculated. Table 2 is a label illustration of the importance ranking of the top 10 features among the nine methods according to an embodiment of this application.

[0063] Table 2

[0064] In some embodiments, this application can obtain a comparison result between a first initial importance ranking and a second initial importance ranking, and determine the importance ranking of the initial characteristic band in response to the comparison result meeting a preset condition. The preset condition can be set by those skilled in the art according to actual circumstances, and this application does not impose specific limitations.

[0065] For example, embodiments of this application compare the initial feature bands found by nine different methods. Coal Spectrum Dataset Labels As shown in Table 2, the key feature bands predicted by mutual information and model-driven linear regression in correlation analysis differ significantly from those predicted by other methods. Therefore, mutual information and linear regression were excluded from the subsequent analysis, and the other seven methods were used for comprehensive analysis.

[0066] Furthermore, embodiments of this application can utilize these seven methods to determine the importance ranking of the initial characteristic bands.

[0067] Optionally, in one embodiment of this application, in response to the comparison result satisfying a preset condition, determining the importance ranking of the initial feature band includes: obtaining the predicted air-dried basis moisture label corresponding to the coal spectral feature matrix; calculating the mean square error between the initial air-dried basis moisture label and the predicted air-dried basis moisture label; based on the mean square error, determining a first weight value corresponding to the first initial importance ranking and a second weight value corresponding to the second initial importance ranking; calculating the total score of the corresponding initial feature band based on the first weight value, the second weight value, and the importance ranking; and determining the final feature band in response to the total score being greater than a preset score.

[0068] In some embodiments, this application can dynamically allocate weights to the initial importance rankings of feature bands and calculate a total score by calculating the mean squared error (MSE) between the initial air-dried basis moisture label and the predicted air-dried basis moisture label of the coal spectral feature matrix. Feature bands with a total score greater than a preset score are then selected as the final feature bands. The preset score can be set by those skilled in the art according to actual conditions, and this application does not impose specific limitations.

[0069] For example, embodiments of this application perform air-dry basis moisture label prediction analysis using nine methods and plot MSE images, such as... Figure 2 As shown, by Figure 2 It is evident that model-driven methods outperform correlation analysis methods on MSE. Therefore, in comprehensive analysis, model-driven methods carry a higher weight (e.g., 0.8, this application does not impose specific limitations), while correlation analysis carries a lower weight (e.g., 0.2, this application does not impose specific limitations). In this case, the formula for calculating the initial total score of characteristic bands can be, but is not limited to, as follows: , in, , , This represents the total score of the relevant characteristic bands. This represents the sum of the rankings of all model-driven methods. This represents the sum of the rankings of all correlation analysis methods.

[0070] Furthermore, embodiments of this application can calculate the total ranking score of the top 10 features at measurement distances of 5mm, 10mm, 15mm, and 20mm, as shown in Table 3. Table 3 is a schematic table illustrating the initial feature band comprehensive scores at different measurement distances according to an embodiment of this application.

[0071] Table 3

[0072] As shown in Table 3, the initial characteristic bands (i.e. key characteristic bands) found in the embodiments of this application under different modes (different measurement distances) are similar, and are concentrated between the 0-20 characteristic bands and the 115-125 characteristic bands, which is consistent with the spectral characteristic mechanism of coal.

[0073] In summary, the embodiments of this application can predict air-dried basis moisture labels based on different methods. In the comprehensive analysis process, the weight values ​​corresponding to different methods are determined by the mean square error value, and then the total score of the corresponding initial characteristic band is calculated. If the total score is greater than a preset score, the initial empty-base moisture label is located. Relevant characteristic bands This allows for the determination of the final characteristic bands.

[0074] Optionally, in one embodiment of this application, determining the final characteristic band based on the importance ranking of the initial characteristic bands of the coal spectral feature matrix includes: calculating the third initial importance ranking of the initial characteristic bands based on the coal spectral feature matrix; and determining the final characteristic band based on the third initial importance ranking.

[0075] In some embodiments of this application, PLS+VIP+correlation stability analysis can be used to calculate the third initial importance ranking of the initial characteristic bands and to find PLS-related bands. ,in, This indicates that PLS+VIP+correlation stability analysis was used in... The relevant bands found by the matrix, and further, the relevant characteristic bands obtained by comprehensive analysis in this embodiment of the application. The correlation mechanism with coal spectra was investigated, thereby determining the final characteristic bands. .

[0076] It is understood that in this embodiment of the application, since the key bands of the modes in the coal spectral dataset with air-dried basis moisture content continuously increase in range with increasing distance, as shown in Table 4, this is related to the correlation mechanism of coal spectra. Therefore, this embodiment of the application introduces the PLS+VIP+correlation stability analysis method for guidance to determine the final characteristic bands. Table 4 is a schematic table of the top 10 key characteristic bands under different mode datasets provided according to an embodiment of this application.

[0077] Table 4

[0078] Furthermore, embodiments of this application plotted the importance curves of the PLS+VIP+correlation stability analysis method for each band on four modal datasets, such as... Figure 3 As shown. By Figure 3It can be seen that the characteristic bands in the embodiments of this application show a consistent trend in the spectral ranges of 900–960 nm and 1400–1500 nm, that is, the features in these characteristic bands have good stability and are highly correlated with the air-dried basis moisture label, thus becoming the key spectral bands related to the air-dried basis moisture label in coal spectroscopy. Meanwhile, the embodiments of this application found that the spectral ranges of 1025–1150 nm, 1211–1335 nm, 1211–1460 nm, and 1397–1520 nm are the same as the 900–960 nm range, all being the most important characteristic regions—consistent with the trend analyzed in Table 4, thus verifying... Figure 3 The persuasiveness of the selected important characteristic bands.

[0079] In step S103, the final characteristic band interval is determined based on the final characteristic band.

[0080] In some embodiments of this application, the final characteristic band interval can be determined based on the final characteristic band.

[0081] Optionally, in one embodiment of this application, determining the final characteristic band interval based on the final characteristic band includes: calculating the mean square error value of the final characteristic band; and determining the final characteristic band interval based on the mean square error value of the final characteristic band.

[0082] In some embodiments of this application, when determining the final characteristic band interval based on the final characteristic band, the mean square error value of the final characteristic band can be calculated first, and then the final characteristic band interval can be determined based on the mean square error value.

[0083] For example, embodiments of this application may use relevant characteristic bands obtained from comprehensive analysis. The most important top 30 characteristic bands and the final characteristic bands obtained The predictions were compared using RF-MDA and LR methods to obtain... ,Will Comparative analysis was conducted to verify... The effectiveness.

[0084] This application embodiment can perform effectiveness verification experiments on the selected characteristic band intervals of 900–960 nm and 1400–1500 nm. Since the selected characteristic band intervals have a total of npre_m = ( +1)+( +1)=28, therefore, the embodiments of this application can use the data obtained in the comprehensive analysis. The top 30 most important feature bands in the analysis were compared with the prediction results, where nfold=30 represents the number of the most important feature bands in the comprehensive analysis.

[0085] To comprehensively assess the importance of the selected bands and reduce the bias that may be introduced by a single prediction method, two comparative methods, RF-MDA and linear regression, were used for analysis. These two methods showed significant differences, thus obtaining more reliable evaluation results. Here, `msefold_m` and `msepre_m` represent the MSE values ​​predicted using the 30 most important feature bands and the selected feature band interval, respectively. The experimental results are shown in Table 5. Table 5 is a comparative table of MSE values ​​obtained by RF-MDA and linear regression under different distance datasets according to an embodiment of this application.

[0086] Table 5

[0087] In the comparative prediction using the RF-MDA method, the maximum reduction of msepre_m compared to msefold_m was only 0.84. When using the linear regression method for comparative prediction, the maximum reduction of msepre_m compared to msefold_m was only 0.42. Notably, in the linear regression prediction, msepre_m even outperformed msefold_m by 2.5 times. These results strongly demonstrate the critical importance and effectiveness of the selected feature bands.

[0088] In the MSEfold assessment, this embodiment selects the top 30 key bands from 125 discrete small feature bands for predictive assessment. This method performs particularly well because the bands selected by RF-MDA are highly consistent with those selected by the comprehensive analysis method. Therefore, this embodiment can form continuous intervals for the selected feature bands within the set of 125 discrete small bands, thereby mitigating the secondary influence of other discrete bands to some extent. However, since these feature band intervals are applicable to all measurement distances and have strong transferability, as long as the difference between MSEpre and MSEfold remains within a small range, the critical importance and effectiveness of the selected feature band intervals can be effectively demonstrated.

[0089] The working principle of the water content spectral feature extraction method proposed in this application will be introduced below with reference to a specific embodiment.

[0090] in, Figure 4 This is a flowchart illustrating the working principle of a method for extracting water content spectral features according to an embodiment of this application.

[0091] Step S401: Extract the coal spectral feature matrix and the initial air-dried basis moisture label.

[0092] Step S402: Calculate the importance ranking of the initial characteristic bands.

[0093] In this application embodiment, five correlation analysis methods can be selected, such as Kendall rank correlation coefficient, Spearman correlation coefficient, distance correlation, maximum information coefficient and mutual information, to calculate the first initial importance rank of the initial feature bands, and four model-driven methods can be selected, such as RF-MDI, RF-MDA, extreme gradient boosting and linear regression, to calculate the second initial importance rank of the initial feature bands.

[0094] Step S403: Comprehensive analysis.

[0095] In this embodiment of the application, the relationship between the initial feature bands found by nine methods and the labels of the coal spectral dataset is compared. As shown in Table 2, the key feature bands predicted by mutual information and model-driven linear regression in the correlation analysis differ significantly from those predicted by other methods. Therefore, mutual information and linear regression are excluded in the subsequent analysis, and the other seven methods are used for comprehensive analysis.

[0096] Furthermore, embodiments of this application perform air-dry basis moisture label prediction analysis using nine methods, and plot MSE images, such as... Figure 2 As shown, by Figure 2 It can be seen that the model-driven method performs better than the correlation analysis method on MSE. Therefore, in the comprehensive analysis, the model-driven method has a higher weight (e.g., 0.8, which is not specifically limited in this application), while the correlation analysis has a lower weight (e.g., 0.2, which is not specifically limited in this application).

[0097] The embodiments of this application can predict the moisture content of air-dried basis labels based on different methods. In the comprehensive analysis process, the weight values ​​corresponding to different methods are determined by the mean square error value, and then the total score of the corresponding initial characteristic band is calculated. If the total score is greater than a preset score, the initial empty-base moisture label is located. Relevant characteristic bands .

[0098] Step S404: PLS+VIP+correlation stability analysis.

[0099] In this embodiment, PLS+VIP+correlation stability analysis can be used to calculate the third initial importance ranking of the initial characteristic bands and find PLS related bands. .

[0100] Step S405: Relevant mechanisms of coal spectroscopy.

[0101] In this application, the embodiments can be based on relevant characteristic bands obtained from comprehensive analysis. and related bands By combining the relevant mechanisms of coal spectroscopy, the final characteristic bands were determined.

[0102] Step S406: Determine the final characteristic band interval.

[0103] In this application embodiment, relevant characteristic bands obtained from comprehensive analysis can be used. The most important top 30 characteristic bands and the final characteristic bands obtained The predictions were compared using RF-MDA and LR methods to obtain... ,Will Comparative analysis was conducted to verify... The effectiveness.

[0104] The method for extracting spectral features of moisture content proposed in this application can extract the spectral feature matrix of coal and the corresponding initial air-dried moisture label by introducing different measurement distance dimensions under air-dried baseline moisture content conditions. It then selects the final feature band intervals based on the importance ranking of the initial feature bands, thereby establishing a high-precision air-dried moisture prediction model. By introducing feature training at different measurement distances, the method can learn the influence of distance changes on the spectrum, thereby reducing errors caused by probe distance fluctuations in actual detection and improving the robustness and adaptability of the model. Using importance ranking to select feature bands and intervals can eliminate redundant and noisy bands, retaining the core spectral information most sensitive to moisture, making the model lighter and with clearer physical meaning. Establishing a model based on an air-dried baseline eliminates interference from external environmental humidity fluctuations, establishes a unified moisture measurement standard, and facilitates horizontal comparison and standardized management of coal moisture data from different batches and under different environments. This solves the problems in related technologies, such as the fact that most schemes for fusing visible light images rely on synchronous calibration of multimodal data, which has high computational complexity and is sensitive to noise; while feature engineering in near-infrared spectroscopy relies on prior knowledge, has limited anti-interference ability, and insufficient adaptability of classification models.

[0105] Next, referring to the accompanying drawings, an extraction apparatus for water content spectral characteristics according to an embodiment of this application is described.

[0106] Figure 5 This is a block diagram of an extraction device for water content spectral characteristics according to an embodiment of this application.

[0107] like Figure 5 As shown, the extraction device 10 for the spectral characteristics of water content includes: an extraction module 100, a first determination module 200, and a second determination module 300.

[0108] The extraction module 100 is used to extract the coal spectral feature matrix and the initial air-dried basis moisture label corresponding to the coal spectral feature matrix at different measurement distances from the target coal spectral data under air-dried baseline moisture content conditions.

[0109] The first determining module 200 is used to determine the final characteristic bands based on the importance ranking of the initial characteristic bands in the coal spectral characteristic matrix.

[0110] The second determining module 300 is used to determine the final characteristic band interval based on the final characteristic band.

[0111] Optionally, in one embodiment of this application, it further includes: a first calculation module, a second calculation module, an acquisition module, and a third determination module.

[0112] The first calculation module is used to calculate the first initial importance ranking of the initial characteristic band based on the initial characteristic band and the initial air-dried basis moisture label before determining the final characteristic band.

[0113] The second calculation module is used to calculate the second initial importance ranking of the initial characteristic bands based on the interaction between different initial characteristic bands.

[0114] The acquisition module is used to obtain the comparison results between the first initial importance ranking and the second initial importance ranking.

[0115] The third determination module is used to determine the importance ranking of the initial characteristic bands in response to the comparison results meeting preset conditions.

[0116] Optionally, in one embodiment of this application, the third determining module includes: an acquisition unit, a first calculation unit, a first determining unit, a second calculation unit, and a second determining unit.

[0117] The acquisition unit is used to acquire the predicted air-dried basis moisture label corresponding to the coal spectral feature matrix.

[0118] The first calculation unit is used to calculate the mean square error between the initial air-dried basis moisture label and the predicted air-dried basis moisture label.

[0119] The first determining unit is used to determine, based on the mean square error value, a first weight value corresponding to the first initial importance ranking and a second weight value corresponding to the second initial importance ranking.

[0120] The second calculation unit is used to calculate the total score of the corresponding initial feature band based on the first weight value, the second weight value, and the importance ranking.

[0121] The second determining unit is used to determine the final characteristic band in response to the total score being greater than the preset score.

[0122] Optionally, in one embodiment of this application, the first determining module 200 includes: a third calculation unit and a third determining unit.

[0123] The third calculation unit is used to calculate the third initial importance ranking of the initial characteristic bands based on the coal spectral feature matrix.

[0124] The third determining unit is used to determine the final characteristic bands based on the third initial importance ranking.

[0125] Optionally, in one embodiment of this application, the second determining module 300 includes: a fourth calculation unit and a fourth determining unit.

[0126] The fourth calculation unit is used to calculate the mean square error value of the final characteristic band.

[0127] The fourth determining unit is used to determine the final characteristic band interval based on the mean square error value of the final characteristic band.

[0128] It should be noted that the explanation of the above-described method for extracting water content spectral features also applies to the device for extracting water content spectral features in this embodiment, and will not be repeated here.

[0129] The moisture content spectral feature extraction device proposed in this application can extract the coal spectral feature matrix and corresponding initial air-dried basis moisture label by introducing different measurement distance dimensions under air-dried baseline moisture content conditions. It then selects the final feature band intervals based on the importance ranking of the initial feature bands, thereby establishing a high-precision air-dried basis moisture prediction model. By introducing feature training at different measurement distances, it can learn the influence of distance changes on the spectrum, thereby reducing errors caused by probe distance fluctuations in actual detection and improving the robustness and adaptability of the model. Using importance ranking to select feature bands and intervals can eliminate redundant and noisy bands, retaining the core spectral information most sensitive to moisture, making the model lighter and with clearer physical meaning. Establishing a model based on an air-dried baseline eliminates interference from external environmental humidity fluctuations, establishes a unified moisture measurement standard, and facilitates horizontal comparison and standardized management of coal moisture data from different batches and under different environments. This solves the problems in related technologies, such as the fact that most schemes for fusing visible light images rely on synchronous calibration of multimodal data, which has high computational complexity and is sensitive to noise; while feature engineering in near-infrared spectroscopy relies on prior knowledge, has limited anti-interference ability, and insufficient adaptability of classification models.

[0130] Figure 6 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. The electronic device may include: The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0131] When the processor 602 executes the program, it implements the method for extracting water content spectral features provided in the above embodiments.

[0132] Furthermore, electronic devices also include: Communication interface 603 is used for communication between memory 601 and processor 602.

[0133] The memory 601 is used to store computer programs that can run on the processor 602.

[0134] Memory 601 may include high-speed RAM memory, and may also include non-volatile memory, such as disk storage.

[0135] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0136] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0137] The processor 602 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0138] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for extracting water content spectral features.

[0139] This application also provides a computer program product, including a computer program that, when executed, implements the above-described method for extracting water content spectral features.

[0140] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in the embodiments or examples of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0141] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0142] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0143] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0144] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0145] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0146] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0147] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method of extracting a water content spectral feature, characterized by, The method comprises the following steps: extracting, under the condition of air-dried basis moisture content, a coal spectrum feature matrix at different measurement distances and an initial air-dried basis moisture label corresponding to the coal spectrum feature matrix from target coal spectrum data; determining a final feature band based on the importance ranking of the initial feature bands of the coal spectrum feature matrix; determining a final feature band interval based on the final feature band.

2. The method of claim 1, wherein, Before determining the final feature band, the method further comprises: calculating a first initial importance ranking of the initial feature bands based on the initial feature bands and the initial air-dried basis moisture label; calculating a second initial importance ranking of the initial feature bands based on the interaction between different initial feature bands; obtaining a comparison result between the first initial importance ranking and the second initial importance ranking; determining the importance ranking of the initial feature bands in response to the comparison result meeting a preset condition.

3. The method of claim 2, wherein, The determination of the importance ranking of the initial feature bands in response to the comparison result meeting the preset condition comprises: obtaining a predicted air-dried basis moisture label corresponding to the coal spectrum feature matrix; calculating a mean square error value between the initial air-dried basis moisture label and the predicted air-dried basis moisture label; determining a first weight value corresponding to the first initial importance ranking and a second weight value corresponding to the second initial importance ranking based on the mean square error value; calculating a total score of the corresponding initial feature band based on the first weight value, the second weight value and the importance ranking; determining the final feature band in response to the total score being greater than a preset score.

4. The method of claim 2, wherein, The determination of the final feature band based on the importance ranking of the initial feature bands of the coal spectrum feature matrix comprises: calculating a third initial importance ranking of the initial feature bands based on the coal spectrum feature matrix; determining the final feature band based on the third initial importance ranking.

5. The method of claim 4, wherein, The determination of the final feature band interval based on the final feature band comprises: calculating a mean square error value of the final feature band; determining the final feature band interval based on the mean square error value of the final feature band.

6. An apparatus for extracting a water content spectral feature, characterized by, The method comprises: an extraction module configured to extract, under the condition of air-dried basis moisture content, a coal spectrum feature matrix at different measurement distances and an initial air-dried basis moisture label corresponding to the coal spectrum feature matrix from target coal spectrum data; a first determination module configured to determine a final feature band based on the importance ranking of the initial feature bands of the coal spectrum feature matrix; a second determination module configured to determine a final feature band interval based on the final feature band.

7. The apparatus of claim 6, wherein, The method further comprises: a first calculation module configured to, before determining the final feature band, calculate a first initial importance ranking of the initial feature bands based on the initial feature bands and the initial air-dried basis moisture label; a second calculation module configured to calculate a second initial importance ranking of the initial feature bands based on the interaction between different initial feature bands; and an obtaining module configured to obtain a comparison result between the first initial importance ranking and the second initial importance ranking. A third determining module is configured to determine the importance ranking of the initial characteristic wave band in response to the comparison result satisfying a preset condition.

8. An electronic device, comprising: The application relates to a method for extracting a water content spectrum feature, comprising the following steps: A computer program product, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for extracting a water content spectrum feature according to any one of claims 1-5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method for extracting a water content spectrum feature according to any one of claims 1-5.

10. A computer program product, characterised in that, The application relates to a computer program, which is executed to implement the method for extracting a water content spectrum feature according to any one of claims 1-5.

Images