Tactile sensing system based on vibration signals and sensing signal acquisition method

Abstract

The invention discloses a tactile sensing system based on a vibration signal and a sensing signal acquisition method, and belongs to the field of robot tactile sensing. The invention aims to solve the problems of large volume, complex structure and wiring and high cost of the conventional touch sensor capable of covering the hand of the robot. The diameter of the front end of the sensing support is smaller than that of the rear end of the sensing support, the top face of the front end of the sensing support is a protruding cambered surface, four round through holes are formed in the protruding cambered surface at equal intervals, and the round through holes penetrate through a sensing support body in the axial direction; four flexible connecting plates are fixed on the side surface of the sensing bracket body at equal intervals, and an accelerometer is fixed in the center of each flexible connecting plate; the sensing end of each accelerometer is connected with one end of a metal wire, and the other end of each metal wire penetrates through the sensing support body through a round through hole in the front end of the sensing support and is fixed to the rear end of the sensing support body. The accelerometer is used for collecting vibration signals of the metal wire. The method is mainly used for acquiring and processing tactile signals.

Description

Technical Field

[0001] The invention belongs to the field of robot tactile sensing. Background Art

[0002] In recent years, with the continuous development of artificial intelligence and robotics technology, the perception capabilities of robots have been continuously improved. Among them, the vision and hearing of robots have developed rapidly and have reached a high level, but tactile perception has been developing slowly. Therefore, tactile perception technology has become a key technology that urgently needs to be developed.

[0003] Tactile perception is a crucial component of a robot's environmental awareness, providing rich and reliable information for its manipulation tasks, such as contact location, force magnitude, and surface texture. This rich information enables robots to perform complex and precise operations such as grasping, hand manipulation, motor coordination, and tool use, enabling deeper interaction with their surroundings.

[0004] To achieve these complex operations, robotic hands often require high-precision and versatile tactile sensing capabilities. The traditional approach involves deploying a large tactile sensor array within the robot hand. This approach not only makes circuit routing difficult but also incurs significant data transmission and computational overhead. Furthermore, a large number of tactile sensors is expensive and presents challenges in adapting software algorithms. These limitations significantly limit the development of robotic tactile sensing technology. Summary of the Invention

[0005] The present invention aims to solve the problems of large size, complex structure and wiring, and high cost of existing tactile sensors that can cover the hands of robots, and now provides a tactile sensing system based on vibration signals.

[0006] The tactile sensing system based on vibration signals of the present invention includes a sensing bracket, a flexible connecting plate and an accelerometer;

[0007] The front end diameter of the sensor bracket is smaller than the rear end diameter. The top surface of the front end of the sensor bracket is a convex arc surface. The convex arc surface is evenly spaced with four circular through holes. The circular through holes penetrate the sensor bracket body in the axial direction.

[0008] Four flexible connection plates are fixed at equal intervals on the side of the sensor bracket body, and an accelerometer is fixed at the center of each flexible connection plate;

[0009] The sensing end of each accelerometer is connected to one end of a metal wire, and the other end of each metal wire passes through the sensing bracket body through the circular through hole at the front end of the sensing bracket and is fixed to the rear end of the sensing bracket body;

[0010] The accelerometer is used to collect the vibration signal of the metal wire.

[0011] Furthermore, in the present invention, the sensor bracket includes a cylindrical member, a truncated cone member and a rectangular parallelepiped member;

[0012] The cylindrical part, the truncated cone part and the rectangular parallelepiped part are coaxially connected in sequence from front to back to form an integral part;

[0013] Four through holes are opened in the axial direction inside the integral part, and the through holes correspond to the four through holes on the convex arc surface;

[0014] A flange is fixed on the bottom end of the rectangular parallelepiped member; and the top end of the cylindrical member is a convex arc surface.

[0015] The flange is used to connect the sensor bracket to the robotic arm;

[0016] A flexible connecting plate is fixed on each side of the rectangular parallelepiped member.

[0017] Furthermore, in the present invention, a rectangular groove is provided on each of the four sides of the cuboid, and a second-order groove is provided in each groove; the frame of the flexible connecting plate is fixed on the rectangular groove, and the accelerometer corresponds to the center of the second-order groove.

[0018] Furthermore, in the present invention, an FPC flexible cable socket is fixed in the rectangular groove of the cuboid; the FPC flexible cable socket is used to connect the FPC flexible cable, and the signal output by the accelerometer is transmitted through the FPC flexible cable.

[0019] Furthermore, in the present invention, the other ends of the four metal wires are connected to the four side surfaces of the rectangular parallelepiped member through four bolts.

[0020] Furthermore, in the present invention, the flexible connection plate includes a frame, a flexible connection member is fixed in the frame, and the accelerometer is fixed at the center of the flexible connection member; the flexible connection member includes a cross member and a flexible connection strip, and the flexible connection strip is vertically connected to the outer end of the cross member.

[0021] Furthermore, in the present invention, the metal wire is made of nickel-titanium material and has a diameter of 0.1 mm.

[0022] A method for identifying sensing signals of a tactile sensing system based on vibration signals is provided. The method is implemented using vibration signals collected by the tactile sensing system based on vibration signals, and specifically comprises:

[0023] Step 1: Segment the vibration signal collected by the tactile sensing system based on the vibration signal using a sliding window method, and perform short-time Fourier transform on the segmented data segments to obtain the frequency components of the vibration signal at different time points;

[0024] Step 2: For the frequency components of the vibration signal at different time points, square the modulus to obtain a power spectrum, input the power spectrum into a Mel filter bank to obtain Mel-band energy of the power spectrum, and concatenate the energy of the Mel-bands to obtain a Mel-spectrogram;

[0025] Step 3: Input the Mel-spectrogram into a pre-trained ResNet neural network to identify the position and force of the tactile sensor contacting the object.

[0026] Furthermore, in the present invention, in step 1, the method of performing short-time Fourier transform on the segmented data segments is:

[0027]

[0028] Where x[n] is a discrete time signal, w[n] is a window function of length N, X(k,t) represents the short-time Fourier transform result of x[n] at time n=th, t is the time index, indicating the tth window, h is the step size, k is the frequency index, indicating the kth frequency component, N is the length of FFT, j is the imaginary unit, and j 2 =-1.

[0029] Furthermore, in the present invention, in step 2, the process of obtaining the mel-spectrogram is as follows:

[0030] The power spectrum is obtained by taking the square of the modulus of the short-time Fourier transform result:

[0031] P(k,t)=|X(k,t)| 2

[0032] Apply a Mel filter bank to the power spectrum to convert the linear frequency f into Mel frequency. The Mel frequency M(f) is:

[0033]

[0034] Each filter H m (k) is a triangular filter, H m The frequency response of (k) is at f m to f m+1 Increase linearly between f m+1 to f m+2 Linearly decrease between:

[0035]

[0036] Where k is the frequency index, corresponding to the kth frequency component in the spectrum of the short-time Fourier transform result;

[0037] Pass the power spectrum P(k,t) through the Mel filter bank to obtain the energy of each Mel frequency band:

[0038]

[0039] The energy of each Mel frequency band is spliced ​​together, with the horizontal axis being the time window and the vertical axis being the frequency band, to obtain the Mel spectrum.

[0040] Furthermore, in the present invention, the pre-trained ResNet neural network includes 4 convolutional layers and 1 fully connected layer.

[0041] The present invention utilizes a combination design of an accelerometer, a flexible connecting plate, a nickel-titanium wire, and a sensor bracket to realize the detection of subtle vibrations when the robot interacts with external objects, and effectively reduces the number of longitudinal sensors, thereby reducing the amount of data transmission and the amount of calculation. In addition, the raw materials are cheap and the shape is easy to manufacture. Using a vibration sensing algorithm based on Mel spectrum and ResNet neural network, the contact position, force magnitude, and surface material of the object can be decoded from the vibration signal with extremely high accuracy. This innovative design not only improves the robot's tactile perception ability, but also provides strong technical support for the robot when performing complex manipulation tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the overall structure of the tactile sensing system based on vibration signals;

[0043] Figure 2 Schematic diagram of the structure of the sensor bracket in the tactile sensing system based on vibration signals;

[0044] Figure 3 for Figure 2 Front view of

[0045] Figure 4 for Figure 2 A top view of

[0046] Figure 5 Schematic diagram of the assembled flexible circuit board and accelerometer in a tactile sensing system based on vibration signals. DETAILED DESCRIPTION

[0047] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described 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 work are within the scope of protection of the present invention. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other in the absence of conflict.

[0048] Specific implementation method 1: refer to Figures 1 to 4 Specifically describing this embodiment, a tactile sensing system based on vibration signals described in this embodiment includes a sensing bracket 1, a flexible connecting plate 2 and an accelerometer 3;

[0049] The front diameter of the sensor bracket 1 is smaller than the rear diameter. The top surface of the front end of the sensor bracket is a convex arc surface 101. The convex arc surface 101 is evenly spaced with four circular through holes. The circular through holes penetrate the sensor bracket body in the axial direction.

[0050] Four flexible connecting plates 2 are fixed at equal intervals on the side of the sensor bracket body, and an accelerometer 3 is fixed at the center of each flexible connecting plate 2;

[0051] The sensing end of each accelerometer 3 is connected to one end of a metal wire, and the other end of each metal wire passes through the sensing bracket body through the circular through hole at the front end of the sensing bracket and is fixed to the rear end of the sensing bracket body;

[0052] The accelerometer 3 is used to collect the vibration signal of the metal wire.

[0053] Furthermore, in this embodiment, the sensor bracket 1 includes a cylindrical member 102, a truncated cone member 103 and a rectangular parallelepiped member 104;

[0054] The cylindrical member 102, the truncated table member 103 and the rectangular parallelepiped member 104 are coaxially connected in sequence from front to back to form an integral member;

[0055] The integral part has four through holes opened in the axial direction, and the through holes correspond to the four through holes of the convex arc surface 101;

[0056] A flange 105 is fixed to the bottom end of the rectangular parallelepiped member 104 ; and a convex arc surface 101 is formed on the top end of the cylindrical member 102 .

[0057] The flange 105 is used to connect the sensor bracket 1 to the robotic arm;

[0058] A flexible connecting plate 2 is fixed on each side of the rectangular parallelepiped member 104 .

[0059] Furthermore, in this embodiment, a rectangular groove is opened on each of the four sides of the rectangular parallelepiped 104, and a second-order groove is opened in each groove; the frame of the flexible connecting plate 2 is fixed on the rectangular groove, and the accelerometer 3 corresponds to the center of the second-order groove.

[0060] Furthermore, in this embodiment, an FPC flexible cable socket is fixed in the rectangular groove of the cuboid 104; the FPC flexible cable socket is used to connect the FPC flexible cable, and the signal output by the accelerometer 3 is transmitted through the FPC flexible cable.

[0061] Furthermore, in this embodiment, the other ends of the four metal wires are connected to the four side surfaces of the rectangular parallelepiped member 104 through four bolts.

[0062] Furthermore, in this embodiment, the flexible connection plate 2 includes a frame, a flexible connection member is fixed inside the frame, and the accelerometer 3 is fixed at the center of the flexible connection member; the flexible connection member includes a cross member and a flexible connection strip, and the flexible connection strip is vertically connected to the outer end of the cross member.

[0063] Furthermore, in this embodiment, the metal wire is made of nickel-titanium material and has a diameter of 0.1 mm.

[0064] A method for identifying sensing signals of a tactile sensing system based on vibration signals is provided. The method is implemented using vibration signals collected by the tactile sensing system based on vibration signals, and specifically comprises:

[0065] Step 1: Segment the vibration signal collected by the tactile sensing system based on the vibration signal using a sliding window method, and perform short-time Fourier transform on the segmented data segments to obtain the frequency components of the vibration signal at different time points;

[0066] Step 2: For the frequency components of the vibration signal at different time points, square the modulus to obtain a power spectrum, input the power spectrum into a Mel filter bank to obtain Mel-band energy of the power spectrum, and concatenate the energy of the Mel-bands to obtain a Mel-spectrogram;

[0067] Step 3: Input the Mel-spectrogram into a pre-trained ResNet neural network to identify the position and force of the tactile sensor contacting the object.

[0068] Furthermore, in this embodiment, in step 1, the method of performing short-time Fourier transform on the segmented data segments is:

[0069]

[0070] Where x[n] is a discrete time signal, w[n] is a window function of length N, X(k,t) represents the short-time Fourier transform result of x[n] at time n=th, t is the time index, indicating the tth window, h is the step size, k is the frequency index, indicating the kth frequency component, N is the length of FFT, j is the imaginary unit, and j 2 =-1.

[0071] Furthermore, in this embodiment, in step 2, the process of obtaining the mel-spectrogram is as follows:

[0072] The power spectrum is obtained by taking the square of the modulus of the short-time Fourier transform result:

[0073] P(k,t)=|X(k,t)| 2

[0074] Apply a Mel filter bank to the power spectrum to convert the linear frequency f into Mel frequency. The Mel frequency M(f) is:

[0075]

[0076] Each filter H m (k) is a triangular filter, H m The frequency response of (k) is at f m to f m+1 Increase linearly between f m+1 to f m+2 Linearly decrease between:

[0077]

[0078] Where k is the frequency index, corresponding to the kth frequency component in the spectrum of the short-time Fourier transform result;

[0079] Pass the power spectrum P(k,t) through the Mel filter bank to obtain the energy of each Mel frequency band:

[0080]

[0081] The energy of each Mel frequency band is spliced ​​together, with the horizontal axis being the time window and the vertical axis being the frequency band, to obtain the Mel spectrum.

[0082] Furthermore, in this embodiment, the pre-trained ResNet neural network includes 4 convolutional layers and 1 fully connected layer.

[0083] The training process of the ResNet neural network described in the present invention is to collect vibration signals, add labels to the vibration signals, and mark the magnitude of force, contact position and material category respectively; after the vibration signal is divided by a sliding window, the mel spectrum of the vibration signal is calculated. The mel spectrum is input into the ResNet neural network in batches, and the amount of mel spectrum is 64. The ResNet neural network learns the spectral characteristics of the vibration signal through convolution operations and automatically obtains the relationship between the spectrum and the label. The training rounds of the ResNet neural network are 100 rounds, the learning rate is 0.0005, and the output is a one-dimensional vector or a scalar value. The one-dimensional vector represents the probability of belonging to different material categories, and the material category corresponding to the maximum probability is the predicted category. The scalar value represents the contact position or the magnitude of the force.

[0084] In the present invention, when the tactile sensor makes instantaneous partial contact with an external object, the contact position on the nickel-titanium wire exposed outside the tactile sensor is s, and the contact force is F. The corresponding vibration signal of the nickel-titanium wire can be captured by the accelerometer as a time series signal that changes with time. (Three-dimensional vector, corresponding to the x, y, z axis components of the accelerometer). Through segmentation, spectrum conversion and neural network learning, the contact position s and contact force F that generate the signal can be predicted.

[0085] The contact point position s is the position on the robot finger. The vibration signal x has the following mathematical relationship with the contact point position s and the magnitude of the force.

[0086]

[0087] Where x(t) is the signal value of the vibration signal at time t, and f(t) is the magnitude of the force at time t. r is the position of the accelerometer. i (t) is the modal shape function, which describes the vibration shape of the structure in the i-th mode. i (t) is the impulse response function, which represents the response under the action of unit impulse force.

[0088] Ultimately, the vibration signal can be represented as the product of a matrix A(s) and a force F. A(s) is a matrix related to the contact position s. The purpose of this invention is to solve for s and F by analyzing the vibration signal. From a mathematical perspective, s and F have multiple solutions. However, by leveraging the powerful learning capabilities of neural networks, the contact position s and contact force F can be derived from the vibration signal.

[0089] In an embodiment, the sensing system will be installed at the end of the robot hand. The end of the robot touches an external object to generate a vibration signal. The vibration signal is transmitted to the accelerometer through a metal wire. The accelerometer communicates with the external host using a multi-channel I2C communication protocol. After acquiring the data, the vibration signal data is first segmented using a sliding window method, and the segmented data segments are then subjected to short-time Fourier transform.

[0090] The sliding window method is to divide a long string of original vibration data into small segments of the same length according to a certain interval. Assuming that the window size is W and the step size is S1, the subscript of the nth window can be expressed as an interval:

[0091] [n*S1,n*S1+W-1]

[0092] The Short-Time Fourier Transform (STFT) is a commonly used method for analyzing non-stationary signals. It divides the signal into several short time periods (windows) and then applies the Fourier transform to each time period to obtain the frequency components of the signal at different time points. The mathematical expression of the STFT is as follows:

[0093] Assuming x[n] is a discrete-time signal and w[n] is a window function of length N (such as a Hanning window, a rectangular window, etc.), the STFT of x[n] at time n=th can be expressed as:

[0094]

[0095] Where t is the time index, indicating the tth window. h is the step size, i.e. the number of samples between adjacent windows. k is the frequency index, indicating the kth frequency component. N is the length of the FFT, which is usually the same as the window length. j is the imaginary unit, satisfying j 2 =-1. The Hanning window function can be expressed as:

[0096]

[0097] Where n is the sample index, ranging from 0 to N-1. N is the length of the window.

[0098] Next, we calculate the Mel spectrum. Calculating the Mel spectrum is a two-step process. The first step is to calculate the power spectrum. This is done by square the modulus of the STFT result.

[0099] P(k,t)=|X(k,t)| 2

[0100] Then, a Mel filter bank is applied to the power spectrum. The Mel filter bank is used to convert the linear frequency into the Mel frequency. The Mel frequency M(f) can be converted from the linear frequency f by the following formula:

[0101]

[0102] Each filter H m (k) is a triangular filter with a frequency response of f m to f m+1 Increase linearly between f m+1 to f m+2 The specific formula is as follows:

[0103]

[0104] where k is the frequency index corresponding to the frequency component in the STFT spectrum.

[0105] Pass the power spectrum P(k,t) through the Mel filter bank to obtain the energy of each Mel frequency band:

[0106]

[0107] The energy of each Mel frequency band is spliced ​​together, with the horizontal axis being the time window and the vertical axis being the frequency band, to obtain the Mel spectrum.

[0108] The Mel spectrum map is then input into a ResNet neural network with 4 convolutional layers, each with a kernel size of 3, 4 input channels and output channels of the convolutional layers, and 1 fully connected layer.

[0109] Suppose there is a spectrum map I of size H*W, and a convolution kernel of size K*K. The convolution kernel slides on the image, sliding a step size S each time, and for each output position (p,q), the value of the output feature map O p,q It can be calculated by the following formula:

[0110]

[0111] Among them, i = p*S, j = q*S, i and j are the starting positions on the input image, and m and n are the relative positions in the convolution kernel.

[0112] After four convolutional layers, the spectrogram is expanded into a one-dimensional vector. This one-dimensional vector is then multiplied by a matrix through a fully connected layer, transforming it into a vector or scalar data that identifies a category. A vector represents the probability of belonging to a certain category, while a scalar represents the magnitude of force or contact location.

[0113] The ResNet neural network outputs tactile information such as surface material, force magnitude, and contact position.

[0114] This tactile information will help robots perceive the external environment and perform complex and precise operations.

[0115] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely illustrative of the principles and applications of the invention. It should be understood that many modifications may be made to the illustrative embodiments, and that other arrangements may be devised, without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in ways other than those described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in conjunction with other described embodiments.

Claims

1. A tactile sensing system based on vibration signals, characterized in that: It comprises a sensing bracket (1), a flexible connecting plate (2) and an accelerometer (3); The front end diameter of the sensor bracket (1) is smaller than the rear end diameter; the front end top surface of the sensor bracket is a convex arc surface (101); the convex arc surface (101) is evenly spaced with four circular through holes; the circular through holes penetrate the sensor bracket body in the axial direction; Four flexible connection plates (2) are fixed at equal intervals on the side of the sensor bracket body, and an accelerometer (3) is fixed at the center of each flexible connection plate (2); The sensing end of each accelerometer (3) is connected to one end of a metal wire, and the other end of each metal wire passes through the sensing bracket body through the circular through hole at the front end of the sensing bracket and is fixed to the rear end of the sensing bracket body; The accelerometer (3) is used to collect vibration signals of the metal wire.

2. A tactile sensing system based on vibration signals according to claim 1, characterized in that: The sensing bracket (1) comprises a cylindrical part (102), a truncated cone part (103) and a rectangular parallelepiped part (104); The cylindrical member (102), the truncated table member (103) and the rectangular parallelepiped member (104) are coaxially connected in sequence from front to back to form an integral member; Four through holes are opened in the axial direction inside the integral part, and the through holes correspond to the four through holes of the convex arc surface (101); A flange 105 is fixed to the bottom end of the rectangular parallelepiped member (104); the top end of the cylindrical member (102) is a convex arc surface (101); The flange 105 is used to connect the sensor bracket (1) to the robotic arm; A flexible connecting plate (2) is fixed on each side of the rectangular parallelepiped member (104).

3. The tactile sensing system based on vibration signals according to claim 2, characterized in that: Each of the four sides of the rectangular parallelepiped (104) is provided with a rectangular groove, and each groove is further provided with a second-order groove; the frame of the flexible connecting plate (2) is fixed on the rectangular groove, and the accelerometer (3) corresponds to the center of the second-order groove.

4. The tactile sensing system based on vibration signals according to claim 3, characterized in that: An FPC flexible cable socket is also fixed in the rectangular groove of the cuboid (104); the FPC flexible cable socket is used to connect an FPC flexible cable, and the signal output by the accelerometer (3) is transmitted through the FPC flexible cable.

5. The tactile sensing system based on vibration signals according to claim 4, characterized in that: The other ends of the four metal wires are connected to the four sides of the rectangular parallelepiped member (104) through four bolts.

6. The tactile sensing system based on vibration signals according to claim 5, characterized in that: The flexible connection plate (2) comprises a frame, a flexible connection piece is fixed in the frame, and the accelerometer (3) is fixed at the center of the flexible connection piece; the flexible connection piece comprises a cross piece and a flexible connection strip, and the flexible connection strip is vertically connected to the outer end of the cross piece.

7. The tactile sensing system based on vibration signals according to claim 5, characterized in that: The metal wire is made of nickel titanium and has a diameter of 0.1 mm.

8. A method for identifying sensing signals of a tactile sensing system based on a vibration signal, wherein the method is implemented using a vibration signal collected by the tactile sensing system based on a vibration signal, and is characterized in that: Specifically: Step 1: Segment the vibration signal collected by the tactile sensing system based on the vibration signal using a sliding window method, and perform short-time Fourier transform on the segmented data segments to obtain the frequency components of the vibration signal at different time points; Step 2: For the frequency components of the vibration signal at different time points, square the modulus to obtain a power spectrum, input the power spectrum into a Mel filter bank to obtain Mel-band energy of the power spectrum, and concatenate the energy of the Mel-bands to obtain a Mel-spectrogram; Step 3: Input the Mel-spectrogram into a pre-trained ResNet neural network to identify the position and force of the tactile sensor contacting the object.

9. The method for identifying sensing signals of a tactile sensing system based on vibration signals according to claim 8, wherein: In step 1, the method of performing short-time Fourier transform on the segmented data segments is: Where x[n] is a discrete time signal, w[n] is a window function of length N, X(k,t) represents the short-time Fourier transform result of x[n] at time n=th, t is the time index, indicating the tth window, h is the step size, k is the frequency index, indicating the kth frequency component, N is the length of FFT, j is the imaginary unit, and j 2 =-1.

10. The method for identifying sensing signals of a tactile sensing system based on vibration signals according to claim 9, wherein: In step 2, the process of obtaining the Mel spectrum is as follows: The power spectrum is obtained by taking the square of the modulus of the short-time Fourier transform result: P(k,t)=|X(k,t)| 2 Apply a Mel filter bank to the power spectrum to convert the linear frequency f into Mel frequency. The Mel frequency M(f) is: Each filter H m (k) is a triangular filter, H m The frequency response of (k) is at f m to f m+1 Increase linearly between f m+1 to f m+2 Linearly decrease between: Where k is the frequency index, corresponding to the kth frequency component in the spectrum of the short-time Fourier transform result; Pass the power spectrum P(k,t) through the Mel filter bank to obtain the energy of each Mel frequency band: The energy of each Mel frequency band is spliced ​​together, with the horizontal axis being the time window and the vertical axis being the frequency band, to obtain the Mel spectrum.

Images