Auditory coal gangue identification method based on array multiple beams

Abstract

The invention provides a coal gangue identification method, which comprises the following steps of: firstly, determining the layout of an auditory sensor array; then, coal gangue audio signals are collected, and array element receiving signals are obtained; then, implementing a beam forming algorithm on each array element receiving signal to obtain an audio feature representing coal gangue difference information in a beam forming signal; processing the obtained audio features to obtain a multi-dimensional tensor matrix x containing the coal gangue audio features; and finally, by taking the multi-dimensional tensor matrix chi as input, identifying the coal gangue by using a deep learning network to obtain a coal gangue identification result. According to the auditory coal gangue recognition method provided by the invention, the auditory sensor array is adopted, and the influence of factors such as noise and underground equipment interference on coal gangue audio recognition can be remarkably reduced by performing beam forming on the audio signals of the target area; and the classification identification model is combined to carry out intelligent identification on the coal gangue, so that the coal gangue identification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of coal mine intelligent identification, and in particular to an auditory coal gangue identification method based on array multi-beam. Background Art

[0002] Currently, fully mechanized top-coal caving (LCC) mining is one of the key technologies for the safe and efficient mining of extra-thick coal seams in my country. This technology has been applied to most working faces in extra-thick coal seams in my country, achieving breakthroughs and developments in both theory and key technologies. Gangue, a solid waste generated during coal mining, is complex in composition and diverse in shape. During the caving process, when coal and rock collapse together, gangue is mixed with the mined coal. This results in a lower purity of the mined coal, reducing economic efficiency and increasing the pressure on subsequent gangue sorting. Excessive gangue accumulation after mining also exacerbates land occupation and environmental pollution.

[0003] Currently, the top coal caving process in fully mechanized top coal caving still relies on manual control based on the principle of "closing the window when gangue is seen" or "closing the window when gangue is heard." This approach presents challenges such as lack of safety assurance, subjective judgment of window closing timing, and low efficiency. The number of top coal caving supports in fully mechanized top coal caving faces is large, the working environment during the caving process is poor, and manual control of the caving opening is labor-intensive and inefficient. Therefore, there is an urgent need to make the caving process unmanned, automated, and intelligent, and gangue identification is a key component of this process.

[0004] The complex and ever-changing geological conditions and harsh environments at coal mine working faces pose significant challenges for efficient, accurate, and intelligent gangue identification. Due to low visibility underground, the mine is subject to environmental influences such as dust, water mist, and light, as well as significant interference from mining equipment. Methods based on visual image recognition often struggle to achieve the desired results at the mine site. However, because sound signals are easily acquired and the equipment is relatively inexpensive, they are less susceptible to interference from underground factors such as dust and water mist. Gangue identification methods based on the sound signals emitted by coal and gangue collapsing and impacting the tail beam of hydraulic supports have shown promising application potential. However, sound signals can also be affected by noise or interference from underground sources such as coal cutters, frame shifting, top coal collapse, the movement of adjacent hydraulic supports, and other equipment, resulting in insufficient gangue identification accuracy. Summary of the Invention

[0005] In order to solve the problem of reduced coal gangue recognition rate caused by noise from neighboring hydraulic supports, underground environmental noise, and interference from underground equipment during the process of auditory coal gangue recognition, the present invention provides an auditory coal gangue recognition method, which includes the following steps:

[0006] S1. Determine the layout of an auditory sensor array; the auditory sensor array includes a plurality of array elements;

[0007] S2, using an auditory sensor array to collect coal gangue audio signals and obtain receiving signals of each array element;

[0008] S3. Implementing a beamforming algorithm on the received signal of each array element to obtain audio features representing the difference information of coal gangue in the beamforming signal;

[0009] S4, processing the audio features obtained in step S3 to obtain a multidimensional tensor matrix χ containing the coal gangue audio features;

[0010] S5. Based on the multidimensional tensor matrix χ obtained in S4, the coal gangue is identified using a deep learning network to obtain a coal gangue identification result.

[0011] In some embodiments, step S1 specifically includes:

[0012] S11. Determine the shape of the auditory sensor array according to the shape characteristics of the target monitoring area;

[0013] S12. Determine the element spacing of the auditory sensor array according to the frequency band range of the coal gangue audio signal;

[0014] S13. Determine the number of array elements of the auditory sensor array.

[0015] In some embodiments, the beamforming is multi-directional beamforming directed toward a target area or multi-point focused beamforming covering a target area.

[0016] In some embodiments, the coal gangue audio signal is collected using a parallel multi-channel collection system.

[0017] In some embodiments, the target monitoring area includes several observation points, and step S3 specifically includes:

[0018] S31, determining the relative positions of the centers of each array element and each observation point;

[0019] S32, preprocessing the received signal of each array element to obtain a preprocessed audio signal;

[0020] S33, applying a beamforming algorithm to each observation point based on the relative positions of each array element center and each observation point and the preprocessed audio signal to obtain a beamforming signal for each observation point;

[0021] S34. Based on the beamforming signal of each observation point, obtain the audio features of each frame signal at each observation point that characterize the difference information of coal gangue.

[0022] In some embodiments, step S4 specifically includes:

[0023] S41, recombining the audio features of each frame signal at each observation point to obtain the matrix Il ; The matrix I l Each element in is the audio feature;

[0024] S42, the matrix I of each frame l The multidimensional tensor matrix χ is obtained by reorganizing it in chronological order.

[0025] In some embodiments, the beamforming algorithm is a conventional beamforming algorithm or an advanced adaptive beamforming algorithm.

[0026] In some embodiments, the beamforming algorithm is a minimum variance distortionless response (MVDR) algorithm.

[0027] In some embodiments, the audio features characterizing the difference information of coal gangue are one or more of bandpass power, bandpass energy ratio, short-time Fourier transform spectrum, and Mel-frequency cepstral coefficients.

[0028] The auditory gangue identification method provided by the present invention adopts an auditory sensor array and implements beamforming on the audio signal of the target area, thereby reducing the influence of factors such as noise from neighboring hydraulic supports, underground environmental noise, and interference from underground equipment on the audio recognition of gangue; and combines it with a classification recognition model to perform intelligent recognition of gangue, significantly improving the accuracy of gangue recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] 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.

[0030] Figure 1 This is a schematic diagram of the underground fully mechanized caving mining operation area;

[0031] Figure 2 Flowchart of the auditory gangue identification method provided by the present invention;

[0032] Figure 3 One of the schematic diagrams of the spatial position between an auditory sensor and an observation point in a target monitoring area provided by the present invention;

[0033] Figure 4 The second schematic diagram of the spatial position between an auditory sensor and an observation point in a target monitoring area provided by the present invention;

[0034] Figure 5 Schematic diagram of a multidimensional tensor matrix χ provided by the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings.

[0036] An application scenario of the auditory gangue identification method provided by the present invention is as follows: Figure 1 As shown, Figure 1 The figure shows the working area of ​​underground fully mechanized top-coal caving mining: a working face is set up at the bottom of the coal seam to be mined. During mining, the coal cutter cuts the coal in the lower coal seam section to continuously advance the working face. As the working face advances, the hydraulic support controls the supporting pressure to move the coal body of the pre-cracked coal seam above the working face, forming certain cracks and gaps inside the pre-cracked coal seam, which gradually transforms into a broken coal seam. Finally, the coal in the broken coal seam falls from the top and is gradually released into the rear scraper conveyor from the coal discharge port at the tail beam of the support. This is called "top coal caving."

[0037] Since gangue will inevitably make a sound when falling from the top, in order to prevent gangue from mixing into the mined coal, gangue audio recognition is generally performed during the top coal caving process. Once the sound signal of gangue is identified, the coal caving port is closed to prevent gangue from mixing in.

[0038] The above describes an application scenario of the auditory gangue identification method of the present invention. The following describes the process of the method in detail with reference to the accompanying drawings and specific embodiments. Figure 2 As shown, the following steps are included:

[0039] S1. Determine the layout of the auditory sensor array;

[0040] The auditory sensor is used to collect audio signals from the target monitoring area as a basis for coal gangue identification. The auditory sensor can be a microphone.

[0041] The auditory sensor array refers to a combination of several auditory sensors arranged according to a certain pattern, wherein each auditory sensor is called an array element.

[0042] The layout of the auditory sensor array specifically includes the shape of the array, the number of array elements, and the spacing between array elements.

[0043] This step specifically includes:

[0044] S11. Determine the shape of the auditory sensor array according to the shape characteristics of the target monitoring area.

[0045] Microphone arrays generally include linear arrays, cross arrays, circular arrays, double-L arrays, plane arrays, spiral arrays, and spherical arrays. When the target monitoring area is linear or strip-shaped, a one-dimensional linear array should be selected; when the target monitoring area is area-shaped, a plane array or circular array should be selected; when monitoring multiple different target monitoring areas, and multiple areas are distributed in different spatial positions, a cross array, double-L array, or spherical array should be considered.

[0046] In one embodiment, according to Figure 1 The process of underground top coal caving and the hydraulic support structure are shown. The target monitoring area should be the coal caving port, which is a narrow and long rectangular area. Based on this, the auditory sensor array is arranged near the base of the hydraulic support, and a one-dimensional linear array formation is selected for the auditory sensor array.

[0047] S12. Determine the element spacing d of the auditory sensor array according to the frequency band range of the coal gangue audio signal.

[0048] The upper limit frequency f is set according to the frequency band of the audio signal containing key useful information. max , the upper limit frequency f max The corresponding signal wavelength is λ min =c / f max (c is the speed of sound);

[0049] To avoid array grating lobes, the spacing d between microphone elements should be equal to the minimum wavelength λ of the signal received by the microphone. min Should satisfy d≤λ min / 2, based on which the maximum spacing of microphone array elements that meets the conditions can be determined.

[0050] In one embodiment, the upper limit of the frequency band of the concerned coal gangue audio signal is 5kHz, λ min / 2=0.034m. Then the array element spacing d can be selected to be no greater than 0.034m, and here d=0.034m.

[0051] It should be noted that according to array signal processing theory, in order to make the sound field model meet the far field model, the array source distance needs to be greater than Where D represents the maximum distance between two array elements in the array. In the case of a linear array microphone array, D = (N-1)d. Figure 1 As shown, in this embodiment, the microphone array is installed near the base of the hydraulic support, and there is a distance between it and the tail beam and the coal discharge port. This distance is the actual array source distance, which is estimated to be at least greater than 1.0m. In order to make the sound field model conform to the far-field model, it can be made At this time, N<6.42 is required.

[0052] S13. Determine the specific number N of array elements of the auditory sensor array.

[0053] According to the array formation and element spacing determined in S11 and S12, taking into account the specific hydraulic support structure, the size of the installation space for the array on the support, the number of available channels of the acquisition system, etc., the maximum number of available elements N is determined as the number of elements in the auditory sensor array.

[0054] In one embodiment, Figure 1 As shown, the installation space of the array on the bracket is about 0.2m, and the number of available channels of the parallel multi-channel acquisition system is 5. Based on this, the number of array elements N=5 is selected.

[0055] The above describes how to determine the layout of the auditory sensor array and Figure 1 In the specific application scenario embodiment shown, the sensor array is determined to be a one-dimensional linear array, with an array element spacing d = 0.034m and an array element number N = 5. The following describes the collection of coal gangue audio signals.

[0056] S2. Use the auditory sensor array to collect the coal gangue audio signal and obtain the receiving signal of each array element.

[0057] Using a parallel multi-channel acquisition system, the audio signal is collected through the microphone array in S1 to obtain the receiving signal of each array element.

[0058] In one embodiment, the audio signal received by N array elements is denoted as s n (t), the corresponding discrete sequence is recorded as s n (m), where n is the array element number, 0≤n≤N-1, and the sampling rate of the analog-to-digital converter of each channel is f s .

[0059] The above introduces the acquisition of coal gangue audio signals. The following describes the process of beamforming the signals in detail.

[0060] S3. Implementing a beamforming algorithm on the received signal of each array element to obtain audio features representing the difference information of coal gangue in the beamforming signal;

[0061] Beamforming is a signal processing technique that leverages the phase differences between the signals received by each element in a sensor array to enhance the desired signal from a specific location. Using beamforming in array signal processing can significantly enhance the desired signal and suppress interference and noise, thereby reducing the impact of underground ambient noise on auditory recognition of coal gangue.

[0062] The beamforming algorithm can be a conventional beamforming algorithm or an advanced adaptive beamforming algorithm, such as the Minimum Variance Distortionless Response (MVDR) algorithm. The MVDR algorithm minimizes the variance of the output signal while ensuring that the desired signal passes through without distortion, thereby suppressing interference and noise. This algorithm adaptively adjusts the weighting coefficients of the signals received by the microphones, maintaining the response in the direction of the desired signal while maximally suppressing interference and noise in other directions.

[0063] According to the effect, beamforming can be divided into direction-by-direction beamforming and multi-point focusing beamforming. Direction-by-direction beamforming means that the microphone array forms a beam in a specific direction, thereby enhancing the signal in that direction while suppressing interference in other directions, and obtaining a direction-by-direction beamforming signal. Multi-point focusing beamforming means focusing the acoustic signal on a specific point in space, so that the signal at that point is enhanced, while the signals at other locations are relatively weakened, and a multi-point focusing beamforming signal is obtained. When this step is implemented, one of the two beamforming methods can be selected according to the shape of the sensor array. Generally, when the array is a one-dimensional linear array, either multi-directional beamforming or multi-point focusing can be used; when the array is a two-dimensional surface array or a ring array, multi-point focusing beamforming is applicable.

[0064] The target monitoring area may be one or more areas. When multiple areas are monitored, the beamforming signal directed to each area is calculated separately. Each target monitoring area includes several observation points.

[0065] The audio features characterizing the difference information of coal gangue may be bandpass power, bandpass energy ratio, short-time Fourier transform spectrum, Mel cepstral coefficient, etc. of the audio.

[0066] The concepts involved in step S3 are explained above. Two specific implementations of step S3 are given below. These two specific implementations are respectively implemented using different beamforming algorithms.

[0067] 1. Use the time-domain conventional beamforming algorithm for multi-point focusing beamforming

[0068] S31. Determine the relative position of each array element center and each observation point.

[0069] In this step, a spatial Cartesian coordinate system needs to be established between the auditory sensor and the target monitoring area to determine the position coordinates or corresponding orientations of the centers of each array element in the array and each observation point in the target monitoring area.

[0070] Figure 3A schematic diagram of the Cartesian coordinate system constructed in this embodiment is given in FIG, wherein the yellow circles represent the elements in the two-dimensional array, the center point of the array is the origin of the Cartesian coordinate system, and P(x p ,y p ,z p ) is the observation point, P'(x p ,y p ,0) is the projection of the observation point P on the xoy surface, r cp is the distance between the array center and the observation point P, θ, are the deflection angle and pitch angle, respectively, used to characterize the relative orientation of the observation point, r np is the distance between each array element and the observation point P.

[0071] S32: Preprocess the received signal of each array element to obtain a preprocessed audio signal.

[0072] The preprocessing mainly includes the processing of the received signal s of each array element. n (t) framing, windowing, etc.

[0073] According to formula (1), the received signal s of each array element is n (t) Perform windowing processing:

[0074] s nl (t) = s n (t)w(t) (1)

[0075] Where w(t) is a commonly used window function, such as the rectangular window function shown in formula (2):

[0076]

[0077] Where T is the window width, which is a positive real number.

[0078] S33 , implementing a beamforming algorithm on each observation point according to the relative position of each array element center and each observation point and the preprocessed audio signal to obtain a beamforming signal of each observation point.

[0079] For each sound source monitoring area, according to each array element and each observation point P(x p ,y p ,z p ) np , use formula (3) to estimate the relationship between each array element and the observation point P(x p ,y p ,z p )'s relative delay τ nP The array center point O is used as the reference point for the delay time, r cp is the distance between the center of the array and the observation point P, and c is the speed of sound in air.

[0080]

[0081] According to the relative delay τ nP , using formula (4), perform conventional time domain beamforming on the preprocessed signal to obtain the beamforming signal S at each observation point Pl (t).

[0082]

[0083] S34, beamforming signal S based on each observation point Pl (t), obtain the audio features of each frame signal at each observation point that characterize the difference information of coal gangue.

[0084] Based on the beamforming signal S Pl (t), estimate the passband power P of the lth frame signal at each observation point according to formula (5) Pl .

[0085]

[0086] The above describes how to obtain multi-point focused beamforming signals by implementing the conventional time-domain beamforming algorithm and extract the audio features P representing the difference information of coal gangue from them. Pl The following describes how to use the advanced adaptive beamforming algorithm MVDR to perform multi-directional beamforming and obtain the audio features P that characterize the difference information of coal gangue in each observation direction. Pl .

[0087] 2. Use advanced adaptive beamforming algorithm MVDR for multi-directional beamforming

[0088] S31. Determine the relative position of each array element center and each observation point.

[0089] In this step, a spatial Cartesian coordinate system needs to be established between the auditory sensor and the target monitoring area to determine the relative position vectors of each array element center and each observation point in the array.

[0090] Figure 4 The Cartesian coordinate system diagram constructed in this embodiment is given. The target monitoring area is a narrow rectangular area of ​​the coal caving port. The relative position relationship between the one-dimensional linear array and the axis of the rectangular observation area is shown as follows: Figure 1 As shown, it can be seen that the line where the linear array is located is parallel to the axis of the observation area. Based on this, we get Figure 4 The Cartesian coordinate system shown in the figure, where the sensor array is a one-dimensional linear array, the observation area is a long strip parallel to the linear array, and the observation point P(x p ,y p ,z p ) with respect to the linear array, the direction angle is θ d.

[0091] S32: Preprocess the received signal of each array element to obtain a preprocessed audio signal.

[0092] The specific implementation process of this step is consistent with that in Example 1 and will not be repeated here.

[0093] S33. Based on the relative positions of the centers of the array elements and the observation points and the preprocessed audio signals, beamforming is performed on the observation points to obtain audio features representing the difference information of the coal gangue at each observation point for each frame of the signal.

[0094] In this embodiment, the MVDR algorithm is used to implement beamforming. Since the received coal gangue collapse signal has a certain bandwidth, a broadband MVDR algorithm needs to be used here.

[0095] According to the basic principle of MVDR, under the main lobe constraint condition, the center frequency is f B The output power P of the narrowband signal MVDR for:

[0096]

[0097] Among them, θ d is the observation point P(x p ,y p ,z p ) relative to the direction of the linear array; a(θ d ) is the phase shift vector, also called the direction vector, which is an N-dimensional column vector used to express θ d The phase shift of the signal in the direction on each array element; H represents the conjugate transpose; R l It is the N-order covariance matrix corresponding to the l-th frame signal of the sensor array, and N is the number of array elements.

[0098] Phase shift vector a(θ d ) can be obtained from formula (7):

[0099]

[0100] Where N is the number of array elements, β d Indicates the direction θ d The phase difference between adjacent array elements is where f B represents the center frequency of the narrowband signal, d is the array element spacing, and c is the speed of sound in air.

[0101] Matrix R l It can be obtained from formula (8):

[0102]

[0103] Where X(m) represents the complex envelope vector of the spatial reception signal observed for the mth time, X(m)=[X 0m ,X 1m ,X 2m ,…,X (N-1)m ] T M represents the number of array data samples collected independently in time, also known as the number of snapshots. The element X in the complex envelope vector nm is the complex envelope of the received signal observed by the nth array element for the mth time, which can be obtained by taking the amplitude and phase of the preprocessed audio signal at the center frequency of the narrowband signal after fast Fourier transform, as shown in formula (9):

[0104]

[0105] Among them, FFT refers to fast Fourier transform, s nm (t) is the preprocessed audio signal, which is obtained by preprocessing the audio signal observed by the nth array element for the mth time, f B Indicates the center frequency of a narrowband signal.

[0106] Finally, based on each narrowband (f = f B ) calculated P MVDR (θ d ) is synthesized to obtain the broadband MVDR power P pl (θ d ), as shown in formula (10):

[0107]

[0108] Where abs represents the modulus of a complex variable, f1 and f2 represent the upper and lower frequencies of the FFT frequency domain band, respectively.

[0109] The above describes how to implement a beamforming algorithm to obtain audio features representing the differences in coal gangue from the beamformed signal in the target monitoring area. The following describes how to process the obtained audio features.

[0110] S4. Process the audio features obtained in step S3 to obtain a multidimensional tensor matrix χ containing the coal gangue audio features.

[0111] The characteristic values ​​of the audio features representing the difference information of coal gangue obtained in step S3 are processed and reorganized in a specific way to form a multidimensional tensor matrix χ containing the audio feature information of coal gangue. Where I, J, K, and L are integers representing the dimensions of the tensor, where I, J, and K represent the dimensions of the three-dimensional space of the target monitoring area, and L is the frame number in the time dimension.

[0112] This step specifically includes:

[0113] S41, recombining the audio features of each frame signal at each observation point to obtain a matrix I with coal gangue feature information l (Dimension: I×J×K); the matrix I l Each element in is the audio feature.

[0114] In one embodiment, the audio feature is the bandpass power of the beamforming signal, such as the aforementioned P Pl .

[0115] S42, the matrix I of each frame l The multidimensional tensor matrix χ is obtained by reorganizing it in chronological order.

[0116] Figure 5 The following diagram shows the multidimensional tensor matrix χ for a linear microphone array, a rectangular target observation area, K = 1, and a single feature extraction scenario. In this scenario, the subimages of each frame are two-dimensional, with the matrix width and height corresponding to the width and height of the target observation area. The colors within the subimages represent different eigenvalue magnitudes, with darker colors indicating larger eigenvalues. Arranging the subimages of each frame in chronological order yields the multidimensional tensor matrix χ.

[0117] In some embodiments, there are multiple audio features, and in each sub-image of the multi-dimensional tensor matrix χ, the matrix I l The dimension will exceed three dimensions, I l The dimensions of also include the dimension representing the number of audio features.

[0118] S5. Based on the multidimensional tensor matrix χ obtained in S4, the coal gangue is identified using a deep learning network to obtain a coal gangue identification result.

[0119] The deep learning network can be a Transformer network, a convolutional neural network, etc.

[0120] The gangue identification result may be the presence or absence of gangue, or the type of gangue that exists. The gangue identification result may be presented in a visual form such as a list or an image.

[0121] The above describes the gangue identification method provided by the present invention. This method can be performed in real time during fully mechanized coal caving. Once the gangue identification results are obtained, they can be immediately transmitted to a relevant alarm or control device. The control device controls the opening and closing of the coal caving port based on the real-time gangue identification results and issues an alarm when gangue is detected.

[0122] In the process of constructing the multidimensional tensor matrix χ, the coal gangue audio features are extracted, screened and combined in advance. Using the multidimensional tensor matrix χ as the input of the deep learning network can reduce the burden of the network's self-extraction of features, reduce the complexity of the network training process, and make the network structure simpler and the number of layers better.

[0123] The auditory gangue identification method provided by the present invention adopts a microphone array and implements beamforming on the audio signal of the target area, thereby reducing the influence of factors such as noise from neighboring hydraulic supports, underground environmental noise, and interference from underground equipment on the audio recognition of gangue; and combines the classification recognition model to perform intelligent recognition of gangue, significantly improving the accuracy of gangue recognition.

[0124] In the description of the embodiments of the present application, words such as "exemplary," "for example," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary," "for example," or "for example" in the embodiments of the present application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary," "for example," or "for example" is intended to present the relevant concepts in a concrete manner.

[0125] In the description of the embodiments of this application, the term "and / or" is merely a description of an association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B can represent the following three situations: A exists alone, B exists alone, and A and B exist at the same time. In addition, unless otherwise specified, the term "plurality" means two or more.

[0126] 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 identifying the technical features being referred to. Thus, features specified as "first" or "second" may explicitly or implicitly include one or more of such features. The terms "include," "comprising," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.

[0127] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying coal gangue, characterized in that: The following steps are involved: Using a pre-set auditory sensor array to collect coal gangue audio signals, and obtaining receiving signals of each array element of the auditory sensor array; Implementing a beamforming algorithm on the received signals of each array element to obtain audio features representing the difference information of coal gangue in the beamforming signal; Processing the audio features to obtain a multidimensional tensor matrix χ containing the coal gangue audio features; Based on the multidimensional tensor matrix χ, a deep learning network is used to identify coal gangue to obtain a coal gangue identification result.

2. The coal gangue identification method according to claim 1, characterized in that: The pre-setting of the auditory sensor array specifically includes: Determine the shape of the auditory sensor array based on the shape characteristics of the target monitoring area; Determine the element spacing of the auditory sensor array according to the frequency band range of the coal gangue audio signal; Determine the number of elements in the auditory sensor array.

3. The coal gangue identification method according to claim 1, characterized in that: The beamforming is multi-directional beamforming directed toward a target area or multi-point focusing beamforming covering a target area.

4. The coal gangue identification method according to claim 1, characterized in that: The coal gangue audio signal is collected using a parallel multi-channel collection system.

5. The coal gangue identification method according to claim 1, characterized in that: The target monitoring area includes several observation points; The beamforming algorithm is applied to the received signal of each array element to obtain the audio features representing the difference information of coal gangue in the beamforming signal, which specifically includes: Determine the relative position of each array element center and each observation point; Preprocessing the received signal of each array element to obtain a preprocessed audio signal; Applying a beamforming algorithm to each observation point according to the relative positions of each array element center and each observation point and the preprocessed audio signal to obtain a beamforming signal for each observation point; Based on the beamforming signal of each observation point, the audio features representing the difference information of coal gangue in each frame signal at each observation point are obtained.

6. The coal gangue identification method according to claim 1, characterized in that: The audio features are processed to obtain a multidimensional tensor matrix χ containing the gangue audio features, specifically including: Recombining the audio features of each frame signal at each observation point to obtain the matrix I l ; The matrix I l Each element in is the audio feature; The matrix I of each frame l The multidimensional tensor matrix χ is obtained by reorganizing it in chronological order.

7. The coal gangue identification method according to claim 1, characterized in that: The beamforming algorithm is a conventional beamforming algorithm or an advanced adaptive beamforming algorithm.

8. The coal gangue identification method according to claim 7, characterized in that: The beamforming algorithm is a minimum variance distortionless response (MVDR) algorithm.

9. The coal gangue identification method according to claim 1, characterized in that: The audio features characterizing the difference information of coal gangue are one or more of bandpass power, bandpass energy ratio, short-time Fourier transform spectrum, and Mel-frequency cepstral coefficient.

Images