Audible fully-mechanized caving coal gangue recognition algorithm

Abstract

The invention relates to an auditory fully-mechanized coal caving gangue recognition algorithm, which comprises the following steps of: acquiring a vibration signal and a sound signal in fully-mechanized coal caving operation through a sensor, and preprocessing the acquired original signal; variational mode decomposition is carried out on the preprocessed signal, the preprocessed signal is decomposed into k intrinsic mode components, k is a preset mode number, and k is larger than 1; performing Gramer angle difference field conversion on each intrinsic mode component, and mapping a one-dimensional time domain signal of the intrinsic mode component into a two-dimensional feature matrix; and on the basis of the two-dimensional feature matrix, coal and gangue in the fully mechanized coal caving operation are identified and classified through a classification network. According to the method, the coal gangue can be efficiently and accurately identified in the fully mechanized coal caving process, and the method has relatively high robustness and practicability.

Description

Technical Field

[0001] The present invention relates to the field of automated monitoring technology, and in particular to an auditory fully-mechanized caving coal gangue recognition algorithm. Background Art

[0002] With the continuous deepening of coal mining, the proportion of gangue in fully-mechanized caving coal is gradually increasing. The presence of gangue not only affects coal quality but also may cause environmental pollution. Therefore, accurately identifying the distribution of gangue in fully-mechanized caving coal is a key task in coal mine production and resource management. Currently, traditional gangue identification methods mostly rely on manual vision or simple sensor-based analysis, but these methods have certain limitations in accuracy and efficiency. In recent years, data-driven intelligent algorithms, such as variational mode decomposition and machine learning methods, have been gradually applied to the field of coal quality analysis. However, the complex characteristics of gangue and external interference make accurate identification more difficult. Therefore, how to improve the accuracy of gangue identification and reduce the false positive rate have become urgent technical challenges. Summary of the Invention

[0003] In order to solve the problems existing in the prior art, the embodiments of the present application provide an auditory fully-mechanized coal gangue identification algorithm, device, computing equipment, computer storage medium and product containing a computer program, which can efficiently and accurately identify coal gangue from the fully-mechanized coal caving process and has strong robustness and practicality.

[0004] In a first aspect, an embodiment of the present application provides an auditory fully-mechanized coal-caving gangue identification algorithm, comprising: collecting vibration signals and sound signals in fully-mechanized coal caving operations through sensors, and preprocessing the collected original signals; performing variational modal decomposition on the preprocessed signals, decomposing them into k eigenmodal components, where k is a preset number of modes and k is greater than 1; performing Gram angle difference field conversion on each eigenmodal component, mapping the one-dimensional time domain signal of the eigenmodal component into a two-dimensional feature matrix; based on the two-dimensional feature matrix, identifying and classifying coal and gangue in fully-mechanized coal caving operations through a classification network.

[0005] In some possible implementations, the preprocessed signal is subjected to variational modal decomposition to decompose it into k eigenmodal components, including: initializing k modal components and the center frequencies corresponding to the modal components; iteratively updating the modal components, center frequencies and Lagrange multipliers through the alternating direction multiplier method until the convergence conditions are met.

[0006] In some possible implementations, the Gram angle difference field conversion of each eigenmode component includes: normalizing each eigenmode component to the interval [0,1]; mapping the normalized data to polar angles through an inverse cosine function, and generating a polar coordinate sequence in combination with a timestamp; calculating the sine value of the polar angle difference between any two time points to generate a symmetric two-dimensional feature matrix.

[0007] In some possible implementations, the classification network adopts a convolutional neural network, which includes: 3 to 5 convolutional layers, each layer is followed by batch normalization and ReLU activation function; a fully connected layer and a Softmax classifier, which outputs the classification probability of coal and gangue.

[0008] In some possible implementations, the sensor includes a sound sensor and a vibration sensor installed at the rear of the hydraulic support tail beam, and the vibration signal and the sound signal are fused after being time-synchronized in a preprocessing stage.

[0009] In some possible implementations, the variational mode decomposition is solved by optimizing a constrained variational model, and the constrained variational model is expressed as follows:

[0010]

[0011] Where u k Characterizes the kth modal component, W k Characterize the center frequency, represents the partial derivative with respect to time t, δ(t) represents the Dirac function, j represents the complex number, u k (t) represents the kth modal component at time t, and f(t) represents the original signal.

[0012] In some possible implementations, to solve the constrained variational model, the original constrained optimization problem is transformed into an unconstrained form by introducing a quadratic penalty term and Lagrange multipliers, and the following objective function is constructed for iterative optimization:

[0013]

[0014] Where λ represents the Lagrange multiplier, α represents the quadratic penalty term, and f(t) represents the original signal.

[0015] In a second aspect, an embodiment of the present application provides an auditory fully-mechanized coal-caving gangue identification device, comprising: an acquisition module for collecting vibration signals and sound signals in fully-mechanized coal-caving operations, and preprocessing the collected original signals; a processing module for performing variational modal decomposition on the preprocessed signals, decomposing them into k eigenmodal components, where k is a preset number of modes, and k is greater than 1; the processing module is also used to perform Gram angle difference field conversion on each eigenmodal component, and map the one-dimensional time domain signal of the eigenmodal component into a two-dimensional feature matrix; the processing module is also used to identify and classify coal and gangue in fully-mechanized coal-caving operations through a classification network based on the two-dimensional feature matrix.

[0016] In a third aspect, an embodiment of the present application provides a computer-readable storage medium comprising computer-readable instructions. When a computer reads and executes the computer-readable instructions, the computer executes the method as described in any one of the first aspects.

[0017] In a fourth aspect, an embodiment of the present application provides a computing device comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method as described in any one of the first aspects is executed.

[0018] In a fifth aspect, an embodiment of the present application provides a product comprising a computer program, which, when the computer program product runs on a processor, enables the processor to execute the method as described in any one of the first aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 any creative work.

[0020] Figure 1 This is a flow chart of an auditory fully-mechanized caving coal gangue identification algorithm provided in an embodiment of the present application;

[0021] Figure 2 This is a structural diagram of an auditory fully-mechanized caving coal gangue identification device provided in an embodiment of the present application;

[0022] Figure 3 It is a structural diagram of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0024] The term "and / or" as used herein describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " as used herein indicates that the related objects are in an "or" relationship, for example, A / B means either A or B.

[0025] The terms "first" and "second" in this specification and claims are used to distinguish different objects rather than to describe a specific order of objects. For example, "first response message" and "second response message" are used to distinguish different response messages rather than to describe a specific order of response messages.

[0026] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0027] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more, for example, multiple processing units means two or more processing units, etc.; multiple elements means two or more elements, etc.

[0028] To facilitate understanding of the embodiments of the present application, further explanation will be given below with reference to specific embodiments in conjunction with the accompanying drawings. The embodiments do not constitute a limitation on the embodiments of the present invention.

[0029] This embodiment of the application provides an auditory gangue identification algorithm for fully-mechanized caving coal mines, combining variational mode decomposition (VMD) and Gramian angular difference field (GADF) for gangue identification. This method decomposes raw fully-mechanized caving coal mine data using VMD and extracts gangue feature information using GADF, achieving accurate identification through a classification algorithm.

[0030] For example, Figure 1 The figure shows a flow chart of an auditory coal gangue identification algorithm provided by an embodiment of the present application. Figure 1 As shown, the method may include the following steps:

[0031] S11: collecting vibration signals and sound signals during fully-mechanized coal caving operation and pre-processing the collected original signals;

[0032] In this example, data collection and preprocessing are fundamental steps in the gangue identification process. Data quality directly determines the accuracy of subsequent feature extraction and classification. Vibration or sound signals from fully mechanized caving coal are collected using an acoustic sensor located behind the tail beam of a hydraulic support. The collected data is then preprocessed. This includes noise filtering (eliminating background interference) and standardization (unifying data dimensions), providing clean, uniformly formatted input for subsequent analysis.

[0033] Specifically, the core goal of data acquisition and preprocessing is to obtain valid raw signals from the coal mine operating environment and, through a series of processing steps, eliminate noise and interference to ensure that the data meets analysis standards. Data acquisition typically relies on sound sensors or vibration sensors deployed behind the tail beam of a hydraulic support. These sensors can capture in real time the sound or vibration signals generated by the collision and friction between coal and gangue during fully-mechanized coal caving. Due to the complex environment of coal mine sites, the raw signals collected by sensors often contain a large amount of background noise, such as the sounds of operating machinery and equipment, human operators, and even electromagnetic interference. This noise can mask the valid characteristics of the coal and gangue, necessitating preprocessing of the collected data.

[0034] The first step in preprocessing is noise filtering, which can remove frequency components in the signal that are not related to the target. Common filtering methods include bandpass filtering and wavelet denoising. Bandpass filtering retains the main frequency band of coal and gangue collision by setting a frequency range (for example, 200Hz to 2000Hz), while eliminating low-frequency mechanical noise and high-frequency electromagnetic interference. Wavelet denoising uses multi-scale analysis to decompose the signal into sub-bands of different frequencies, and then suppresses noise through threshold processing. These two methods can be used in combination, for example, first performing coarse denoising through bandpass filtering and then using wavelet transform for fine processing. In addition, if the sensor collects multi-channel data (such as vibration signals at multiple locations), synchronous calibration between channels is also required to ensure time alignment and avoid phase deviation in subsequent analysis.

[0035] After noise filtering, data normalization is required. Because the amplitude of the raw sensor signal output can fluctuate significantly due to environmental or device differences, direct use can make subsequent algorithms sensitive to amplitude, impacting robustness. Normalization typically employs min-max scaling, linearly mapping the signal amplitude to the [0, 1] range. Normalization not only unifies the data dimensions but also prevents the negative impact of extreme values ​​on model training. If the data exhibits significant baseline drift (such as slow fluctuations caused by sensor temperature changes), detrending is also necessary, for example, by removing low-frequency trend terms through high-pass filtering or polynomial fitting.

[0036] During preprocessing, attention must also be paid to signal capture and segmentation. Coal mining operations are continuous processes, but algorithms typically require fixed-length data segments for analysis. A common approach is to segment the continuous signal into several segments using a sliding window. The window length should balance timeliness and feature integrity (e.g., 500 ms to 1 s). Each segment should contain a complete coal or gangue event (e.g., a collision). This can be achieved through energy threshold detection or zero-crossing rate analysis. Furthermore, for non-stationary signals (e.g., dynamic changes in the coal-gangue ratio), dynamic adjustment of the window length or the use of overlapping segmentation strategies may be necessary to improve analysis accuracy. It is important to note that preprocessing parameters (e.g., filter band, wavelet basis function, window length, etc.) need to be optimized based on the actual scenario. For example, if coal mine machinery noise is concentrated below 300 Hz, the lower limit frequency of the bandpass filter can be set to 300 Hz. If the coal-gangue collision signal has significant transient characteristics, a wavelet basis suitable for transient analysis (e.g., Daubechies wavelet) can be selected.

[0037] S12: performing variational mode decomposition on the preprocessed signal to decompose it into k eigenmode components, where k is a preset number of modes and k is greater than 1;

[0038] In this embodiment, after the preprocessing of the collected data is completed, the processed data will enter the variational mode decomposition (VMD) stage. The data at this time should meet the following conditions: the noise level is significantly reduced, the amplitude range is unified, the time is aligned and the segmentation is reasonable. Variational mode decomposition (VMD) is applied to the preprocessed data for time-frequency analysis. VMD effectively extracts representative frequency components by decoupling different modal components in the signal (i.e., the characteristic modes of gangue and coal). This process can overcome the shortcomings of traditional methods in high noise environments and refine the characteristics of gangue. The variational mode decomposition method aims to decompose the preprocessed complex signal into k eigenmode components with clear center frequencies by constructing a variational model. The core idea of ​​VMD is to transform the signal decomposition problem into a variational optimization problem.

[0039] Specifically, VMD uses signal processing to decompose complex coal mining signals into several physically meaningful modal components. This approach fundamentally overcomes the limitations of traditional signal decomposition techniques in processing non-stationary signals, providing a high-quality data foundation for subsequent feature extraction and classification. Because the vibration and sound signals collected by sensors often contain complex waveforms generated by the mixing of coal and gangue, these signals not only contain multiple frequency components but are also often affected by environmental factors such as mechanical equipment noise and electromagnetic interference. VMD is able to separate these mixed signal components, with each separated modal component representing a specific vibration mode or acoustic feature in the signal. This decomposition is not a simple mathematical transformation, but an adaptive process based on the physical properties of the signal itself. In operation, VMD first determines the number of modes to be decomposed, k. This parameter should be neither too small, to prevent inadequate separation of important characteristic components, nor too large, to generate meaningless redundant components. After determining the number of modes, the algorithm initializes a set of center frequencies, which are then used to obtain the signal components that best match them. During VMD's iterative optimization process, the algorithm simultaneously performs three tasks: adjusting the waveform shape of each modal component, fine-tuning their center frequency positions, and ensuring that the sum of all components reconstructs the original signal. This process continues until the decomposition results reach a stable state. It is worth noting that during the optimization process, VMD strives to ensure that each component contains only narrowband frequency components, ensuring that the decomposition results have clear physical meaning.

[0040] The signal processed by VMD exhibits a clear hierarchical structure. Modal components representing coal tend to be concentrated in the lower frequency range because the vibrations generated by the collision of coal blocks are relatively gentle. However, due to its greater hardness, the vibration components generated by gangue often contain more high-frequency components. This difference in frequency distribution provides an important basis for subsequent feature extraction and classification. In gangue identification, key VMD parameters must be selected based on the actual data characteristics. The selection of the number of modes, k, is particularly important. If k is too small, the characteristics of coal and gangue may not be fully separated; if k is too large, redundant components may be introduced. The value of k is usually determined by observing the center frequency distribution or based on prior knowledge. The penalty factor α affects the bandwidth of the modal components. A larger α results in smoother components but may lose detailed information. The convergence threshold ε controls the iteration accuracy. Furthermore, the strategy for initializing the center frequencies (e.g., uniform distribution or based on spectral peaks) also affects the decomposition results. The output of VMD is k modal components and their corresponding center frequencies. In gangue identification, these components can separate the distinct vibration or acoustic characteristics of coal and gangue. For example, the collision of coal may produce low-frequency components, while gangue, due to its high hardness, may appear as high-frequency components. By analyzing the time-frequency characteristics of these components, we can provide a cleaner and more discriminative input for the subsequent Gram angle difference field feature extraction.

[0041] For example, assume that the original signal can be represented as the sum of k modal components, each modal component is a narrowband signal around its center frequency. Each modal component is denoted as u k (t), the component can be expressed as: Where u k (t) represents the kth modal component at time t, A k (t) characterizes the envelope signal, Characterize narrowband signals. The entire decomposition process is to continuously iterate the variational process to minimize the bandwidth of each eigenmode component after decomposition. In order to effectively evaluate the bandwidth of each eigenmode decomposition component after decomposition, the bandwidth evaluation function is determined. For each modal component, its analytical signal is obtained through Hilbert transform and its single-sided spectrum is calculated. The expression is: Where δ(t) represents the Dirac function and j represents the complex number. The calculated spectrum is then multiplied by the exponential tuned to the respective estimated center frequencies to shift the spectrum of the modal components to their respective basebands. The expression is: Where w k Characterize the center frequency. Finally, the Gaussian smoothness of the demodulated signal is used to estimate the bandwidth of each modal component through the square norm of the gradient. The expression is: Where, Characterizes the partial derivative with respect to time t. The iterative process is to minimize the bandwidth expression of each modal component. In addition, in order to ensure that the decomposed intrinsic mode components can reconstruct the original signal, the constraint condition ∑ k u k (t)=f(t), where f(t) represents the original signal.

[0042] The constrained variational model is thus constructed, and its final expression is as follows:

[0043]

[0044] In order to solve the optimal solution of the constructed constrained variational model, the constrained problem is transformed into an unconstrained problem through the quadratic penalty term and Lagrange multiplier, and the final optimization objective function of the variational model is determined, which is expressed as follows:

[0045]

[0046] Where λ represents the Lagrange multiplier, α represents the quadratic penalty term, and f(t) represents the original signal.

[0047] The quadratic penalty term ensures the convergence of the objective function, while the Lagrange multiplier strictly restricts the constraints to ensure accurate reconstruction of the original signal. The optimal value of the objective function is solved using the alternating direction multiplier method. Through continuous iterative updates, the optimal k intrinsic mode components are obtained.

[0048] In some possible embodiments, the iterative update process of the k eigenmode components is as follows:

[0049] S21: Initialize k eigenmode components and the center frequency corresponding to each component, initialize the Lagrange multiplier, and initialize the number of iterations n to 0.

[0050] In this step, the number of eigenmode components k and the center frequency corresponding to each component are initialized. Lagrange multipliers. Each modal component is initialized to zero or a random signal, with the corresponding center frequency uniformly distributed within the signal's effective frequency band. Lagrange multipliers are initialized to a zero vector to enforce subsequent constraints. The iteration counter n is reset to zero, marking the start of the optimization process.

[0051] S22: Update the k components separately. The update expressions are as follows:

[0052]

[0053] In this step, each modal component is independently updated based on the current center frequency estimate and Lagrange multipliers. This update process is essentially an adaptive filtering operation, selectively extracting the signal in the frequency domain. Specifically, the algorithm forms an adaptive bandwidth bandpass filter around the center frequency of each modal component, extracting the component that best matches that mode from the original signal. Simultaneously, the compensation signal provided by the Lagrange multipliers ensures that the sum of all modal components accurately reconstructs the original signal.

[0054] S23: Update the center frequencies corresponding to the k components respectively. The update expressions are as follows:

[0055]

[0056] In this step, the algorithm calculates the weighted centroid of the power spectrum of each modal component and uses the centroid position as the new center frequency estimate. This operation ensures that the modal component can automatically identify the most significant frequency band in the signal without manual intervention. In a physical sense, this is equivalent to allowing each modal component to find the frequency position that best suits the modal component to characterize the specific vibration mode of the signal. For coal gangue signals, the characteristic frequencies of coal and gangue are usually distributed in different frequency bands. Through adaptive frequency adjustment, these characteristic differences can be identified and enhanced.

[0057] S24: Update the Lagrange multiplier. The update expression is as follows:

[0058]

[0059] Where τ represents the Lagrange multiplier update factor.

[0060] In this step, the Lagrange multiplier update is proportional to the difference between the sum of all current modal components and the original signal. This difference reflects the degree to which the constraints are satisfied. As iterations proceed, the multiplier gradually accumulates this difference information and regulates subsequent modal updates. From an optimization theory perspective, this is equivalent to minimizing the bandwidth of each modal component while strictly ensuring the accuracy of signal reconstruction. When processing noisy coal mine signals, the dynamic adjustment of the Lagrange multiplier can effectively suppress noise interference and ensure that the decomposed modal components have clear physical meaning.

[0061] S25: Convergence judgment is performed. The convergence judgment criteria are as follows

[0062]

[0063] Where ε represents the convergence threshold.

[0064] If the convergence criterion is met, the iteration is stopped and the final k eigenmode components are obtained. If the convergence criterion is not met, the iteration is repeated.

[0065] In this step, the relative change in modal components between the current iteration and the previous iteration is calculated. If this change is less than a preset threshold, the decomposition results are stable and the iteration can be stopped; otherwise, the iteration cycle continues. This convergence criterion ensures that the algorithm does not run indefinitely and that the decomposition quality meets the expected standards.

[0066] S13: performing Gram angle difference field conversion on each eigenmode component, mapping the one-dimensional time domain signal of the eigenmode component into a two-dimensional characteristic matrix;

[0067] In this embodiment, the Gram Angular Difference Field (GADF) method is used to further extract features based on the modal signals obtained through VMD decomposition. GADF calculates the angular differences between different modal signals to produce a two-dimensional matrix reflecting the characteristics of the gangue. The angular differences between each modal signal and the other modal signals can effectively reflect its internal structure and frequency characteristics.

[0068] Specifically, the Gram Angle Difference Field is a method for converting one-dimensional time series signals into two-dimensional image representations. The individual modal components obtained after variational mode decomposition (VMD) are still time series signals, but GADF uses a unique conversion mechanism to convert these signals into an image format that is more easily processed by machine learning models while preserving the important characteristic information of the original signal.

[0069] The GADF processing workflow begins with data normalization. Because the modal components after VMD decomposition may have different amplitude ranges, the data for each modal component must first be normalized to the interval [0, 1]. Normalization not only unifies the data scale but also eliminates the impact of amplitude differences on subsequent processing. The normalized data points are then mapped to a polar coordinate system. During the polar coordinate transformation, the value of each data point is converted to a polar angle using the inverse cosine function, while the time information is converted to a polar radius. This transformation leverages the correspondence between trigonometric functions and time series, uniquely mapping each point in the one-dimensional signal to a position on the unit circle. This representation preserves both amplitude and time information of the signal and enhances signal visualization through the geometric properties of the polar coordinate system. After the polar coordinate transformation, GADF constructs a feature matrix by calculating the angular difference between different time points. Specifically, the difference between the polar angles corresponding to any two time points in the signal is measured using the sine function. This processing generates a symmetric two-dimensional matrix, each element of which reflects a specific relationship between two time points in the original signal. This transformation transforms the dynamic characteristics of a time series into spatial distribution features, allowing both the temporal patterns and amplitude variations in the signal to be visually visualized on a two-dimensional plane. For gangue identification, the vibration or sound signals generated by coal and gangue form distinct image patterns after this transformation, providing more discriminative features for subsequent classification and recognition. These generated GADF images can be directly used as input for deep learning models such as convolutional neural networks, or further extracted for statistical features used in traditional machine learning algorithms.

[0070] In the specific scenario of gangue identification, since vibration and sound signals in coal mine environments often contain significant noise and nonlinear components, GADF, by converting time series into images, not only enhances feature representation but also naturally incorporates the signal's time domain and amplitude information. For example, the vibration signal of coal may produce a relatively smooth GADF image, while the harder gangue, due to its signal conversion, may exhibit sharper texture features. This difference can be effectively captured and distinguished in a properly trained machine learning model. Compared to other time-frequency analysis methods, GADF for gangue identification not only avoids the window selection issues of methods like the short-time Fourier transform, but is also more easily integrated with deep learning models than time-frequency analysis methods like the wavelet transform. Features processed by GADF can significantly improve classification accuracy, especially when processing complex signals collected in coal mines, where its robustness and discriminability are particularly outstanding. By combining VMD and GADF, the gangue identification algorithm achieves a complete conversion from raw signals to highly discriminative features, laying a solid foundation for accurate classification.

[0071] In some possible embodiments, the GADF algorithm steps are as follows:

[0072] S31: normalizing the collected vibration signal and sound signal.

[0073] In this step, the collected vibration signal and sound signal are all one-dimensional time domain data, expressed as x = {x1, x2, ... x n-1 ,x n Each element is normalized and scaled to [0,1] using the formula: Where x l Represents the normalized result of the i-th element, x i Represents the i-th element, minX represents the minimum value of signal X, and maxX represents the maximum value of signal X. It is worth noting that X here is a representation, not a limitation of vibration signal and sound signal. The vibration signal can be recorded as X1 and the sound signal as X2, then x1={x11,x12,……x1 n-1 ,x1 n}, x2={x21,x22,……x2 n-1 ,x2 n}.

[0074] S32: Convert the normalized data into polar coordinates.

[0075] In this step, the conversion into polar coordinates can be performed according to the following formula:

[0076]

[0077] Where, t i Representation x i The corresponding timestamp, M represents the normalization factor. The scaled signal value is encoded using the angle cosine so that it falls within the range of [0,π]. i The corresponding timestamp is encoded as the radius, so that each data point is located on the unit circle of polar coordinates. Due to the uniqueness of the timestamp, the corresponding polar coordinate result is also unique.

[0078] S33: Reconstruct the polar coordinate signal using GADF.

[0079] In this step, the reconstruction is performed according to the following formula:

[0080]

[0081] Where θ i Characterizes the polar angle of the i-th point in the time series, θ j Characterizes the polar angle of the j-th point in the time series.

[0082] Converting the single-dimensional vibration and sound signals of the stacker crane into GADF images helps to improve the sparsity of the data, reduce the redundant information of the modal, and reduce the nonlinear characteristics of the data.

[0083] S14: Based on the two-dimensional feature matrix, the coal and gangue in the fully-mechanized coal caving operation are identified and classified through a classification network.

[0084] In this embodiment, the reconstructed data is classified through a classification network, for example, a machine learning classification algorithm (such as support vector machine (SVM), random forest, or deep neural network) is applied to train and classify the features, and finally coal and gangue are identified.

[0085] Specifically, the input to the classification network is a two-dimensional feature matrix generated by GADF transformation. These matrices represent the essential characteristic differences between coal and gangue in the form of images. For this structured data, convolutional neural networks (CNNs) are suitable. Their local connectivity and weight sharing characteristics are well suited for mining spatial hierarchical features in images. A typical CNN architecture may contain 3-5 convolutional layers, each with a ReLU activation function and batch normalization, gradually extracting different levels of features, from edge texture to global patterns. The addition of pooling layers reduces computational effort and improves the spatial invariance of features, while dropout layers effectively prevent overfitting. At the end of the network, fully connected layers map high-level features to class probabilities, outputting a confidence score for coal or gangue via a softmax function. This end-to-end deep learning architecture automatically learns the most discriminative feature combinations, avoiding the tedious manual feature design process required in traditional methods. Given the computational resource constraints at coal mine sites, the design of the classification network must balance accuracy and computational complexity. Lightweight network architectures such as MobileNet and ShuffleNet significantly reduce the number of parameters while maintaining good recognition performance through techniques such as depthwise separable convolution. Another optimization approach is knowledge distillation, which involves first training a large, complex model and then having a smaller network learn its behavior. This significantly improves inference speed while maintaining high accuracy. In actual deployment, network quantization converts floating-point parameters into low-bit integer representations, reducing model size while accelerating computation and enabling the algorithm to run smoothly on embedded devices. In addition to deep learning solutions, traditional machine learning classifiers can also be used for classification. Support vector machines (SVMs) combined with appropriate kernel functions can effectively handle the nonlinear separability of GADF features. Random forests, with their advantages of fast training speed and simple parameter adjustment, perform well with limited data. These traditional methods typically require dimensionality reduction or manual feature extraction of the GADF matrix, such as calculating texture features, statistical moments, or spectral features. While recognition accuracy may be slightly lower than that of deep learning methods, their models offer greater interpretability and computational efficiency, making them a worthy option for coal gangue identification applications.

[0086] The training process of classification networks requires special attention to dataset construction and optimization strategies. High-quality labeled data is the foundation for training reliable models. In the case of gangue identification, this requires collecting a large number of samples covering different coal mines and operating conditions, and accurately annotating them by professionals. Data augmentation techniques, such as rotating, scaling, and adding noise to the GADF matrix, can effectively expand the training set and improve the model's generalization ability. The design of the loss function is also crucial. In the case of an imbalanced gangue ratio, weighted cross entropy or focal loss can be used to alleviate the class imbalance. The choice of optimizer and learning rate scheduling strategy directly impacts training efficiency and final performance. The Adam optimizer combined with a cosine annealing learning rate is a common and effective combination. When deploying the solution, the classification network also needs to consider dynamic adaptability. Coal mine operating conditions may change over time, such as equipment wear and fluctuations in coal quality, which can affect signal characteristics. Online learning mechanisms can enable the model to continuously adapt to new data distributions without requiring complete retraining. Model monitoring systems can track fluctuations in recognition performance in real time, triggering retraining when accuracy drops below a threshold.

[0087] The above is an introduction to the auditory comprehensive coal gangue identification algorithm provided in the embodiment of the present application. The vibration / sound signals in the coal mine operation are collected by sensors. After denoising and standardization preprocessing, the signal is adaptively decomposed into multiple intrinsic mode components using VMD to effectively separate the characteristic frequencies of coal and gangue. Subsequently, the one-dimensional signal is converted into a two-dimensional feature image through GADF to enhance feature sparsity and separability. Finally, a lightweight convolutional neural network or SVM classifier is used to achieve accurate recognition. By integrating the above algorithm into the coal mine monitoring system, the gangue in the comprehensive coal caving can be detected and identified in real time. Through the collection of sensor data, VMD decomposition, GADF feature extraction and classification recognition, the distribution of gangue is automatically determined, and a result report is generated for the operator's reference.

[0088] By combining VMD and GADF methods, this method effectively improves the accuracy of gangue identification and reduces false and missed identifications. Furthermore, it can handle complex gangue data and exhibits strong noise suppression and interference resistance. This method also enables real-time automatic gangue identification, reducing manual intervention and improving coal mine automation and production efficiency. This method is not only applicable to gangue identification in fully mechanized caving coal, but can also be applied to quality monitoring and resource allocation in other types of coal mines.

[0089] It is understandable that the size of the sequence number of each step in the above-mentioned embodiments does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. In addition, in some possible implementations, the steps in the above-mentioned embodiments can be selectively executed according to actual conditions, and can be partially executed or fully executed, which is not limited here. All or part of any features of any embodiment of the present application can be freely and arbitrarily combined without contradiction. The combined technical solution is also within the scope of the present application.

[0090] Based on the method in the above embodiment, the embodiment of the present application also provides an auditory fully-mechanized caving coal gangue identification device. For example, Figure 2 The present invention shows an auditory fully mechanized caving coal gangue identification device, which is deployed on a computing device and used for coal gangue identification. The identification device 200 includes: an acquisition module 201 and a processing module 202.

[0091] The acquisition module 201 is used to collect vibration signals and sound signals in the fully-mechanized coal caving operation and pre-process the collected original signals.

[0092] The processing module is used to perform variational modal decomposition on the preprocessed signal to decompose it into k eigenmodal components, where k is a preset number of modes and k is greater than 1.

[0093] The processing module is further used to perform Gram angle difference field conversion on each eigenmode component, and map the one-dimensional time domain signal of the eigenmode component into a two-dimensional characteristic matrix.

[0094] The processing module is also used to identify and classify coal and gangue in the fully-mechanized coal caving operation through a classification network based on the two-dimensional feature matrix.

[0095] It should be understood that the above-mentioned device is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0096] The present application also provides a computing device 300. Figure 3 As shown, computing device 300 includes a bus 302, a processor 304, a memory 306, and a communication interface 308. Processor 304, memory 306, and communication interface 308 communicate with each other via bus 302. Computing device 300 can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memories in computing device 300.

[0097] The bus 302 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 The fact that only one line is used in the figure does not mean that there is only one bus or only one type of bus. Bus 304 may include a path for transmitting information between various components of computing device 300 (eg, memory 306, processor 304, communication interface 308).

[0098] The processor 304 may include any one or more processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0099] The memory 306 may include volatile memory, such as random access memory (RAM). The processor 304 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0100] Memory 306 stores executable program code, which processor 304 executes to implement the functions of acquisition module 201 and processing module 202, respectively, thereby implementing all or part of the steps of the method in the above-described embodiment. In other words, memory 306 stores instructions for executing all or part of the steps of the method in the above-described embodiment.

[0101] Alternatively, the memory 306 stores executable code, and the processor 304 executes the executable code to implement the functions of the aforementioned auditory fully-mechanized caving coal gangue identification device 200, thereby implementing all or part of the steps in the above-described embodiment method. In other words, the memory 306 stores instructions for executing all or part of the steps in the above-described embodiment method.

[0102] The communication interface 308 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 300 and other devices or a communication network.

[0103] Based on the methods in the above embodiments, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the methods in the above embodiments.

[0104] Based on the methods in the above embodiments, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the methods in the above embodiments.

[0105] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0106] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0107] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state drive (SSD)).

[0108] It will be understood that the various numerical numbers involved in the embodiments of the present application are merely distinctions for the convenience of description and are not intended to limit the scope of the embodiments of the present application.

Claims

1. An auditory fully mechanized caving coal gangue identification algorithm, characterized by: The method comprises: The vibration and sound signals during fully-mechanized coal caving operations are collected through sensors, and the collected raw signals are pre-processed; Perform variational mode decomposition on the preprocessed signal and decompose it into k eigenmode components, where k is the preset number of modes and k is greater than 1; Perform Gram angle difference field transformation on each eigenmode component and map the one-dimensional time domain signal of the eigenmode component into a two-dimensional characteristic matrix; Based on the two-dimensional feature matrix, the coal and gangue in the fully-mechanized coal caving operation are identified and classified through a classification network.

2. The method according to claim 1, characterized in that The variational mode decomposition is performed on the preprocessed signal to decompose it into k eigenmode components, including: Initialize k modal components and the center frequencies corresponding to the modal components; The modal components, center frequencies and Lagrange multipliers are iteratively updated by the alternating direction multiplier method until the convergence conditions are met.

3. The method according to claim 1, characterized in that The performing Gram angle difference field conversion on each eigenmode component includes: Normalize each eigenmode component to the interval [0,1]; The normalized data is mapped to polar angles using the inverse cosine function, and a polar coordinate sequence is generated by combining the timestamps. Calculate the sine of the polar angle difference between any two time points to generate a symmetric two-dimensional feature matrix.

4. The method according to claim 1, wherein The classification network adopts a convolutional neural network, which includes: 3 to 5 convolution layers, each layer is followed by batch normalization and ReLU activation function; a fully connected layer and a Softmax classifier, which outputs the classification probability of coal and gangue.

5. The method according to claim 1, wherein The sensors include a sound sensor and a vibration sensor installed at the rear of the hydraulic support tail beam. The vibration signal and the sound signal are fused after being time-synchronized in a pre-processing stage.

6. The method according to claim 2, characterized in that The variational mode decomposition is solved by optimizing the constrained variational model, and the constrained variational model expression is as follows: Where u k Characterizes the kth modal component, W k Characterize the center frequency, represents the partial derivative with respect to time t, δ(t) represents the Dirac function, j represents the complex number, u k (t) represents the kth modal component at time t, and f(t) represents the original signal.

7. The method according to claim 6, characterized in that In order to solve the constrained variational model, the original constrained optimization problem is transformed into an unconstrained form by introducing a quadratic penalty term and Lagrange multipliers, and the following objective function is constructed for iterative optimization: Where λ represents the Lagrange multiplier, α represents the quadratic penalty term, and f(t) represents the original signal.

8. An auditory fully mechanized caving coal gangue identification device, characterized in that: The device comprises: The acquisition module is used to collect vibration signals and sound signals in fully-mechanized coal caving operations and pre-process the collected original signals; A processing module, configured to perform variational modal decomposition on the preprocessed signal to decompose it into k eigenmodal components, where k is a preset number of modes and k is greater than 1; The processing module is further configured to perform Gram angle difference field conversion on each eigenmode component, and map the one-dimensional time domain signal of the eigenmode component into a two-dimensional characteristic matrix; The processing module is further used to identify and classify coal and gangue in the fully-mechanized coal caving operation through a classification network based on the two-dimensional feature matrix.

9. A computer-readable storage medium comprising computer-readable instructions, which, when a computer reads and executes the computer-readable instructions, causes the computer to execute the method according to any one of claims 1 to 7.

10. A computing device comprising a processor and a memory, wherein: The memory stores computer program instructions, and when the processor executes the computer program instructions, the method according to any one of claims 1 to 7 is executed.

Images