CN115355973A - Truck overload detection method based on acoustic signals - Google Patents

Truck overload detection method based on acoustic signals Download PDF

Info

Publication number
CN115355973A
CN115355973A CN202210916017.6A CN202210916017A CN115355973A CN 115355973 A CN115355973 A CN 115355973A CN 202210916017 A CN202210916017 A CN 202210916017A CN 115355973 A CN115355973 A CN 115355973A
Authority
CN
China
Prior art keywords
truck
model
sound
signal
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210916017.6A
Other languages
Chinese (zh)
Inventor
蒋磊
郭唐仪
薛彪
杜鹏桢
周竹萍
董人龙
呼鑫宇
李卫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202210916017.6A priority Critical patent/CN115355973A/en
Publication of CN115355973A publication Critical patent/CN115355973A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01G—WEIGHING
    • G01G19/00—Weighing apparatus or methods adapted for special purposes not provided for in the preceding groups
    • G01G19/02—Weighing apparatus or methods adapted for special purposes not provided for in the preceding groups for weighing wheeled or rolling bodies, e.g. vehicles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/141—Discrete Fourier transforms
    • G06F17/142—Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Algebra (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • Operations Research (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Probability & Statistics with Applications (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Discrete Mathematics (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses a truck overload detection method based on acoustic signals, which specifically comprises the following steps: collecting original truck running sound of a target model, preprocessing audio signal data of the original truck running sound, and extracting an independent truck running sound signal; carrying out feature extraction on the truck driving sound signal; constructing a truck driving sound model, wherein the truck driving sound model comprises a target model truck driving sound model and an overload detection model, and training a self-encoder network by taking the extracted features as input; and inputting the truck running sound signal to be detected into the trained self-encoder network to detect the truck model and overload. The unsupervised self-encoder has strong feature learning and reconstruction capability, and the network structure is relatively simple, so that the unsupervised self-encoder is very suitable for carrying out a sound detection task.

Description

Truck overload detection method based on acoustic signals
Technical Field
The invention belongs to the technology of sound signal processing, and particularly relates to a truck overload detection method based on sound signals.
Background
At present, roads in China are in an important period from construction to pipe construction and transformation development, but the truck overload phenomenon is forbidden frequently and the damage is huge. In China, overload management is a difficult problem, manual and static weighing detection which is most widely applied has the defects of poor reliability, low efficiency and the like, and the existing dynamic weighing overload control system is high in installation and maintenance cost and cannot be comprehensively popularized and applied in national and provincial trunk lines. The two types of contact weighing overload detection methods are time-consuming and labor-consuming, and are not matched with the current intelligent informatization trend.
In the non-contact overload detection technology, a truck overload detection method based on machine vision is available at present, but the image or video quality is easily limited by the shielding of ambient light and obstacles, and when the weather is dark or a target is shielded, the performance of overload detection through machine vision is greatly damaged.
Disclosure of Invention
In order to solve the technical defects in the prior art, the invention provides a truck overload detection method based on an acoustic signal.
The technical scheme for realizing the purpose of the invention is as follows: a truck overload detection method based on acoustic signals comprises the following specific steps:
collecting original truck running sound of a target model, preprocessing audio signal data of the original truck running sound, and extracting an independent truck running sound signal;
carrying out feature extraction on the truck driving sound signal;
constructing a truck driving sound model, wherein the truck driving sound model comprises a target model truck driving sound model and an overload detection model, and the extracted features are used as input to train a self-encoder network;
and inputting the running sound signal of the truck to be detected into the trained self-encoder network to detect the model and overload of the truck.
Preferably, the obtained audio signal data is preprocessed by using a sparse constraint-based non-negative matrix factorization method, and the specific process is as follows:
step 1.1, constructing an obtained original vehicle running sound signal matrix V = WH, wherein W and H are a truck running sound signal of a target truck and a background noise signal factor matrix respectively;
step 1.2, updating the truck driving sound signal factor matrix W of the target truck through KL divergence deviation, wherein the specific formula is as follows:
Figure BDA0003775672000000021
in the formula, t represents the iteration number, i and j are the number of rows and columns of the matrix, l and k are parameters for iteration through KL divergence deviation, and V ij For values corresponding to the (i, j) subscript observation matrix, W ik Is the value of the source signal matrix W corresponding to the (i, k) index, H kj Is the value of the noise matrix H corresponding to the (k, j) subscript, W il Is the value of the source signal matrix W corresponding to the (i, l) subscript, H lj Is the value of the noise matrix H corresponding to the (l, j) subscript.
Step 1.3, performing column normalization on W according to the following formula, and setting negative elements in W and H to zero:
Figure BDA0003775672000000022
in the formula, W ik t+1 The matrix value before normalization, W, for the t +1 th iteration ik ′ t+1 Is a normalized value.
Step 1.4, updating H:
Figure BDA0003775672000000023
wherein a is a weight.
Step 1.5: and (4) calculating an objective function, if the objective function value is larger than the specified separation index, returning to the step 1.2, and otherwise, performing the step 2.
Preferably, the objective function is specifically:
Figure BDA0003775672000000024
preferably, the specific method for extracting the characteristics of the truck driving sound signal is as follows: performing fast Fourier transform on each frame of signal, converting a time domain signal into a frequency domain, and calculating an energy spectrum; and performing Mel filtering on each frame of energy spectrum in the frequency domain, and converting the linear energy spectrum into a logarithmic energy spectrum to obtain the characteristics.
Preferably, the specific method for extracting the characteristics of the truck driving sound signal comprises the following steps:
step 2.1, performing FFT on each frame of signal, converting the time domain signal into the frequency domain:
Figure BDA0003775672000000031
wherein, x (q) is an input signal, N is the FFT point number, L is the frame length, N represents the signal frame number, and p represents the serial number of the spectral line in the frequency domain;
step 2.2, calculating the energy of the spectral line for the data after each frame of FFT:
E n (p)＝[X n (p)] 2 0≤p＜L
and 2.3, passing the energy spectrum through a Mel filter and summing to obtain output energy:
Figure BDA0003775672000000032
multiplying and adding the energy spectrum of each frame with the frequency domain response of the Mel filter in a frequency domain, wherein M refers to the serial number of the Mel filter, and M is the total number of the Mel filter;
step 2.4, taking logarithm of the energy of the Mel filter, converting the linear energy spectrum into a logarithm energy spectrum, and obtaining the characteristics to be extracted, namely logarithm Mel energy:
X(n，m)＝log[S(n，m)] 0≤m＜M。
preferably, the concrete steps of constructing the target model truck driving sound model are as follows:
step 3.1, inputting the characteristic X of the target model truck driving sound, and obtaining a potential characteristic D through an encoder:
D＝g(GX+b e )
wherein G is a weight matrix of the coding layer and the input layer, b e For coding layer node biasing, g (.) is a node activation function;
step 3.2, obtaining the reconstruction of the input samples through the decoder, wherein b d Biasing for decoding layers:
Figure BDA0003775672000000033
step 3.3, calculating a loss function through a mean square error function of the reconstruction characteristic and the input characteristic:
Figure BDA0003775672000000041
wherein x is r Is an input vector of the r-th dimension,
Figure BDA0003775672000000042
is a reconstructed vector of the r-th dimension, and n is the dimension of the feature matrix;
and 3.4, adopting a gradient descent algorithm as an optimization algorithm, updating the weight and the bias to reduce the loss function, and setting the weight G in the t-th training, wherein the weight G updating formula is as follows:
Figure BDA0003775672000000043
the bias b updates the formula as:
Figure BDA0003775672000000044
wherein eta is the learning rate;
step 3.5, judging whether the reconstruction error is smaller than the expected error
Figure BDA0003775672000000045
If not, repeating the step 3.2 to the step 3.4, otherwise ending the process.
Preferably, the construction process of the overload detection model is the same as that of the target model truck running sound model, and the overload detection model is input as the characteristic of a truck normal load sample.
Preferably, the specific method for inputting the truck driving sound signal to be detected into the trained self-encoder network to perform truck model detection and overload detection comprises the following steps:
step 4.1, inputting the sound features to be detected into a trained sound self-encoder model of the target type truck to obtain a reconstruction error relative to the target type truck;
step 4.2, taking the reconstruction error as an abnormal score and taking the expected error as an expected error
Figure BDA0003775672000000046
As a threshold value of the abnormality score, the abnormality score S is subjected to threshold processing, where 0 indicates that the threshold processing result is that the abnormality score is a normal range, and 1 indicates an abnormality:
Figure BDA0003775672000000047
and 4.3, inputting a detection result into a sample of the target vehicle type, inputting a trained truck overload detection model to carry out overload detection to obtain a reconstruction error, carrying out threshold judgment, judging the sound sample detected as abnormal to be the sound sample of the overloaded truck, and judging the sound sample as not overloaded if not.
Compared with the prior art, the invention has the following remarkable advantages:
firstly, static weighing and dynamic weighing are compared, and overload is detected simply and quickly by using sound, so that the cost is low and the efficiency is high.
And compared with overload detection based on machine vision, the acoustic signal is not influenced by light and obstacles, and the equipment is low in cost and convenient to install.
Finally, the unsupervised self-encoder has strong feature learning and reconstruction capability, and the network structure is relatively simple, so that the unsupervised self-encoder is very suitable for carrying out a voice detection task.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a model training process of the present invention, which takes an auto-encoder as an example.
Detailed Description
It is easily understood that various embodiments of the present invention can be conceived by those skilled in the art according to the technical solution of the present invention without changing the essential spirit of the present invention. Therefore, the following detailed description and the accompanying drawings are merely illustrative of the technical aspects of the present invention, and should not be construed as all of the present invention or as limitations or limitations on the technical aspects of the present invention. Rather, these embodiments are provided so that this disclosure will be thorough and complete. The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form a part hereof, and which together with the embodiments of the invention serve to explain the innovative concepts of the invention.
The invention has the conception that the overload detection is realized by two-stage wagon running sound detection, the first stage is the running sound detection of a target type wagon, and the second stage is the overload detection of the target type wagon. In the first stage, all characteristics of truck driving sound samples belonging to target models are modeled through the encoding and decoding processes of an auto encoder, and therefore vehicle type detection is carried out on the samples to be detected. And in the second stage, a model is built by the characteristics of the normal load running sound sample of the target type truck through an automatic encoder, and overload detection is carried out on the sample to be detected according to the model.
As an embodiment, a truck overload detection method based on acoustic signals comprises the steps of firstly collecting truck running sounds of a target model, carrying out blind source separation on obtained audio signal data, adopting a sparse-constraint non-negative matrix decomposition algorithm aiming at an underdetermined blind source separation problem, adding sparse constraint conditions to source signals and a mixed matrix, and separating and recovering the source signals of the truck running sounds from observation signals.
Then extracting characteristics, performing Fast Fourier Transform (FFT) on each frame of signal, converting a time domain signal into a frequency domain, and calculating an energy spectrum; designing a triangular band-pass filter, carrying out Mel filtering on each frame of energy spectrum in a frequency domain, taking a logarithm after summing, converting a linear energy spectrum into a logarithm energy spectrum, and obtaining characteristics: logarithmic mel-energy;
and then, the extracted features are used as input to train a self-encoder network, and a two-stage truck driving sound detection self-encoder model is constructed, wherein the first stage is driving sound detection of a target type truck, and the second stage is overload detection of the type truck.
The modeling process of the first stage is as follows: inputting the characteristic matrixes of all freight car running sound samples belonging to the target model into a self-encoder, reducing the dimensionality of input characteristics and reconstructing the input characteristics through the encoding and decoding processes, and adjusting network parameters by using a Back Propagation (BP) algorithm and an optimization algorithm to enable reconstruction errors to reach the minimum value, wherein the representation model learns the most effective characteristics of input data, and the running sound model of the freight car of the target model is constructed.
The modeling process of the second stage is as follows: inputting the characteristic matrix of the normal load running sound sample of the truck of the type in the first stage into a self-encoder, and enabling the reconstruction error to reach the minimum value through network calculation of the self-encoder to obtain a truck overload detection model.
And finally, overload detection, namely preprocessing a sample to be detected, extracting characteristics, and sequentially judging whether the model of the target truck is overloaded or not and whether the target truck is overloaded or not through two-stage truck running sound detection.
And in the first stage, carrying out truck model detection, inputting the feature matrix into a trained target model truck driving sound model for reconstruction to obtain a reconstruction error score, judging whether the sample to be detected belongs to a target model truck or not according to a threshold condition of the reconstruction error, if so, entering the second stage, otherwise, failing to complete overload detection, and directly ending the detection process.
And in the second stage, overload detection is carried out, the characteristic matrix of the freight car belonging to the target model is input into the trained freight car overload detection model, whether the sample to be detected belongs to a normal load sample or not is judged according to the threshold condition of the reconstruction error, if yes, overload is not carried out, and if not, the overload sample is judged.
As an embodiment, a method for detecting overload by using two-stage truck running sound includes the following steps:
step 1: in the blind source separation process, a sparse constraint-based non-negative matrix factorization method is used:
step 1.1, in non-negative matrix decomposition, an acquired original vehicle running sound signal matrix V = WH, wherein W and H are factor matrices, that is, a sound signal of a target truck and a background noise signal which need to be separated and recovered.
Step 1.2, updating and iterating W, wherein t represents iteration times, i and j are row number and column number of a matrix, l and k are parameters for iterating through KL divergence deviation, and V ij Observing moments for corresponding (i, j) subscriptsValue of the matrix, W ik Is the value of the source signal matrix W corresponding to the (i, k) subscript, H kj Is the value of the noise matrix H corresponding to the (k, j) subscript, W il Is the value of the source signal matrix W corresponding to the (i, l) index, H lj Is the value of the noise matrix H corresponding to the (l, j) subscript.
Figure BDA0003775672000000071
Step 1.3, performing column normalization on W according to the following formula, and simultaneously setting negative elements in W and H to zero, wherein W ik t+1 The matrix value before normalization, W, for the t +1 th iteration ik ′ t+1 Is a normalized value.
Figure BDA0003775672000000072
Step 1.4, updating H, wherein a is weight:
Figure BDA0003775672000000073
step 1.5, substituting W and H to calculate an objective function,
Figure BDA0003775672000000074
and if the objective function value is larger than the specified separation index, returning to the step 1.2, otherwise, ending the process. At this time, W is the wagon source signal with the background noise removed.
Step 2: the audio signal feature extraction process takes logarithmic mel energy as an example:
step 2.1, performing FFT on each frame of signal, converting the time domain signal into the frequency domain:
Figure BDA0003775672000000075
wherein, x (q) is an input signal, N is the FFT point number, L is the frame length, N represents the signal frame number, and p represents the serial number of the spectral line in the frequency domain;
step 2.2, calculating the energy of the spectral line for the data after each frame of FFT:
E n (p)＝[X n (p)] 2 0≤p＜L
and 2.3, passing the energy spectrum through a Mel filter and summing to obtain output energy:
Figure BDA0003775672000000081
and multiplying and adding the energy spectrum of each frame with the frequency domain response of the Mel filter in the frequency domain, wherein M refers to the serial number of the Mel filter, and M is the total number of the Mel filters.
Step 2.4, taking logarithm of energy of the Mel filter, converting the linear energy spectrum into a logarithm energy spectrum, and obtaining features to be extracted, namely logarithm Mel energy:
X(n，m)＝log[S(n，n)] 0≤m＜M
and step 3: the method comprises the following steps of constructing a two-stage truck driving sound model, including a target model truck driving sound model and an overload detection model, and firstly constructing the former:
step 3.1, inputting the characteristic X of the target model truck driving sound, and obtaining a potential characteristic D through an encoder:
D＝g(GX+b e )
wherein, the weight matrix G of the coding layer and the input layer, the node bias be of the coding layer, and the node activation function G (.) select RELU.
Step 3.2, obtaining the reconstruction of the input sample through a decoder, wherein bd is the decoding layer offset:
Figure BDA0003775672000000082
step 3.3, calculating a loss function through the mean square error function of the reconstruction characteristic and the input characteristic, wherein the loss function is the sample passingReconstruction error obtained by the model. Wherein x is r Is an input vector of the r-th dimension,
Figure BDA0003775672000000083
is a reconstruction vector of the r-th dimension, and n is the dimension of the feature matrix:
Figure BDA0003775672000000084
step 3.4, the optimization algorithm adopts a gradient descent algorithm, weight and bias are updated to reduce a loss function, and the weight G is assumed to be updated according to the following formula in the t-th round of training:
Figure BDA0003775672000000091
the offset b is updated by the formula:
Figure BDA0003775672000000092
here, the learning rate η is set to 0.001.
Step 3.5, judging whether the reconstruction error is smaller than the expected error
Figure BDA0003775672000000093
If not, repeating the step 3.2 to the step 3.4, otherwise, ending the process and completing the construction of the first-stage model.
And 3.6, constructing an overload detection model in the second stage, inputting the characteristics of the normal load sample of the truck of the model, repeating the steps by 3.1-3.5, and finally obtaining the truck overload detection model.
And 4, step 4: the two-stage truck running sound detection comprises the processes of truck model detection and overload detection:
and 4.1, firstly, detecting the vehicle type, inputting the sound characteristics to be detected into the trained sound self-encoder model of the target-type truck, and obtaining the reconstruction error of the sample relative to the target-type truck.
Step 4.2, taking the reconstruction error as an abnormal score and taking the expected error
Figure BDA0003775672000000094
As a threshold value of the abnormality score, threshold processing is performed on the abnormality score S, where 0 indicates that the threshold processing result is that the abnormality score is a normal range, 1 indicates an abnormality:
Figure BDA0003775672000000095
4.3, the sound sample detected as abnormal has larger difference with the sound of the target type truck, and the essence of the sound sample is probably generated by other types of trucks; otherwise, the detection result of the sample belongs to the target vehicle type, the target vehicle type is regarded as the running sound of the target type truck, the second-stage overload detection is continuously carried out, the sample is continuously used as the input of the overload detection model, after the reconstruction error is obtained, the threshold judgment of the step 4.2 is carried out in the same way, the sound sample detected as abnormal is greatly different from the normal load running sound, the sound sample is judged as the sound sample of the overload truck, and otherwise, the overload is not considered.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
It should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes described in a single embodiment or with reference to a single figure, for the purpose of streamlining the disclosure and aiding in the understanding of various aspects of the invention by those skilled in the art. However, the present invention should not be construed such that the features included in the exemplary embodiments are all the essential technical features of the patent claims.
It should be understood that the modules, units, components, and the like included in the apparatus of one embodiment of the present invention may be adaptively changed to be provided in an apparatus different from that of the embodiment. The different modules, units or components comprised by the apparatus of an embodiment may be combined into one module, unit or component or they may be divided into a plurality of sub-modules, sub-units or sub-components.

Claims (8)

1. A truck overload detection method based on acoustic signals is characterized by comprising the following specific steps:
collecting original truck running sound of a target model, preprocessing audio signal data of the original truck running sound, and extracting an independent truck running sound signal;
carrying out feature extraction on the truck driving sound signal;
constructing a truck driving sound model, wherein the truck driving sound model comprises a target model truck driving sound model and an overload detection model, and training a self-encoder network by taking the extracted features as input;
and inputting the running sound signal of the truck to be detected into the trained self-encoder network to detect the model and overload of the truck.
2. The truck overload detection method based on the acoustic signal according to claim 1, wherein the obtained audio signal data is preprocessed by a sparse constraint non-negative matrix factorization-based method, which comprises the following specific processes:
step 1.1, constructing an obtained original vehicle running sound signal matrix V = WH, wherein W and H are a truck running sound signal of a target truck and a background noise signal factor matrix respectively;
step 1.2, updating the truck driving sound signal factor matrix W of the target truck through KL divergence deviation, wherein the specific formula is as follows:
Figure FDA0003775671990000011
where t denotes the number of iterations, i, j are the number of rows and columns of the matrix, l and k are the parameters for the iteration by KL divergence deviations, V ij For values corresponding to the (i, j) subscript observation matrix, W ik Is the value of the source signal matrix W corresponding to the (i, k) subscript, H kj Is the value of the noise matrix H corresponding to the (k, j) subscript, W il Is the value of the source signal matrix W corresponding to the (i, l) index, H lj Is the value of the noise matrix H corresponding to the (l, j) subscript.
Step 1.3, performing column normalization on W according to the following formula, and setting negative elements in W and H to be zero:
Figure FDA0003775671990000012
in the formula, W ik t+1 The matrix value before normalization, W, for the t +1 th iteration ik ′ t+1 Is a normalized value.
Step 1.4, updating H:
Figure FDA0003775671990000021
wherein a is a weight.
Step 1.5: and (4) calculating an objective function, if the objective function value is larger than the specified separation index, returning to the step 1.2, and otherwise, performing the step 2.
3. The truck overload detection method based on the acoustic signal according to claim 2, wherein the objective function is specifically as follows:
Figure FDA0003775671990000022
4. the truck overload detection method based on the acoustic signal according to claim 1, wherein the specific method for extracting the characteristics of the truck running acoustic signal is as follows: performing fast Fourier transform on each frame of signal, converting a time domain signal into a frequency domain, and calculating an energy spectrum; and performing Mel filtering on each frame of energy spectrum in the frequency domain, and converting the linear energy spectrum into a logarithmic energy spectrum to obtain the characteristics.
5. The truck overload detection method based on the acoustic signal according to claim 1 or 4, wherein the specific method for extracting the characteristics of the truck running acoustic signal is as follows:
step 2.1, performing FFT on each frame of signal, converting the time domain signal into the frequency domain:
Figure FDA0003775671990000023
wherein, x (q) is input signal, N is FFT point number, L is frame length, N represents signal frame number, p represents sequence number of spectral line in frequency domain;
step 2.2, calculating the energy of the spectral line for the data after each frame of FFT:
E n (p)＝[X n (p)] 2 0≤p＜L
and 2.3, passing the energy spectrum through a Mel filter and summing to obtain output energy:
Figure FDA0003775671990000024
multiplying and adding the energy spectrum of each frame with the frequency domain response of the Mel filter in a frequency domain, wherein M refers to the serial number of the Mel filter, and M is the total number of the Mel filter;
step 2.4, taking logarithm of energy of the Mel filter, converting the linear energy spectrum into a logarithm energy spectrum, and obtaining features to be extracted, namely logarithm Mel energy:
X(n，m)＝log[S(n，m)] 0≤m＜M。
6. the truck overload detection method based on the acoustic signal as claimed in claim 1, wherein the concrete steps of constructing the truck driving sound model of the target model are as follows:
step 3.1, inputting the characteristic X of the target model truck driving sound, and obtaining a potential characteristic D through an encoder:
D＝g(GX+b e )
wherein G is a weight matrix of the coding layer and the input layer, b e For coding layer node biasing, g (.) is a node activation function;
step 3.2, obtaining a reconstruction of the input samples via a decoder, wherein b d Biasing for decoding layers:
Figure FDA0003775671990000031
step 3.3, calculating a loss function through the mean square error function of the reconstruction characteristic and the input characteristic:
Figure FDA0003775671990000032
wherein x is r Is an input vector of the r-th dimension,
Figure FDA0003775671990000033
is a reconstruction vector of the r-th dimension, and n is the dimension of the feature matrix;
and 3.4, adopting a gradient descent algorithm as an optimization algorithm, updating the weight and the bias to reduce the loss function, and setting the weight G in the t-th training, wherein the weight G updating formula is as follows:
Figure FDA0003775671990000034
the bias b updates the formula as:
Figure FDA0003775671990000035
wherein eta is the learning rate;
step 3.5, judging whether the reconstruction error is smaller than the expected error
Figure FDA0003775671990000041
If not, repeating the step 3.2 to the step 3.4, otherwise ending the process.
7. The method of claim 1, wherein the model for detecting overload is constructed in the same way as the model for constructing the sound model of the target model of truck driving, and the model is inputted as a feature of a sample of normal load of the truck.
8. The truck overload detection method based on the acoustic signal according to claim 1, wherein the truck driving acoustic signal to be detected is input into a trained self-encoder network, and the specific method for truck model detection and overload detection is as follows:
step 4.1, inputting the sound features to be detected into a trained sound self-encoder model of the target type truck to obtain a reconstruction error relative to the target type truck;
step 4.2, taking the reconstruction error as an abnormal score and taking the expected error
Figure FDA0003775671990000042
As a threshold value of the abnormality score, threshold processing is performed on the abnormality score S, where 0 indicates that the threshold processing result is that the abnormality score is a normal range, 1 indicates an abnormality:
Figure FDA0003775671990000043
and 4.3, inputting the detection result into a sample of the target vehicle type, inputting the sample into a trained truck overload detection model to carry out overload detection to obtain a reconstruction error, carrying out threshold judgment, judging the sound sample detected as abnormal to be the sound sample of the overloaded truck, and judging the sound sample as not overloaded otherwise.
CN202210916017.6A 2022-08-01 2022-08-01 Truck overload detection method based on acoustic signals Pending CN115355973A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210916017.6A CN115355973A (en) 2022-08-01 2022-08-01 Truck overload detection method based on acoustic signals

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210916017.6A CN115355973A (en) 2022-08-01 2022-08-01 Truck overload detection method based on acoustic signals

Publications (1)

Publication Number Publication Date
CN115355973A true CN115355973A (en) 2022-11-18

Family

ID=84030943

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210916017.6A Pending CN115355973A (en) 2022-08-01 2022-08-01 Truck overload detection method based on acoustic signals

Country Status (1)

Country Link
CN (1) CN115355973A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117542377A (en) * 2024-01-05 2024-02-09 江西众加利高科技股份有限公司 Information prompting method and related device based on truck weighing

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117542377A (en) * 2024-01-05 2024-02-09 江西众加利高科技股份有限公司 Information prompting method and related device based on truck weighing
CN117542377B (en) * 2024-01-05 2024-04-05 江西众加利高科技股份有限公司 Information prompting method and related device based on truck weighing

Similar Documents

Publication Publication Date Title
CN109117987B (en) Personalized traffic accident risk prediction recommendation method based on deep learning
CN109214575B (en) Ultrashort-term wind power prediction method based on small-wavelength short-term memory network
CN108334948B (en) Mechanical bearing fault diagnosis technology based on wide residual error network learning model
CN110941928B (en) Rolling bearing residual life prediction method based on dropout-SAE and Bi-LSTM
CN110516305B (en) Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model
CN111523509B (en) Equipment fault diagnosis and health monitoring method integrating physical and depth expression characteristics
CN111737911B (en) Bearing recession trend prediction method for deep confidence network and extreme learning machine
CN110672343A (en) Rotary machine fault diagnosis method based on multi-attention convolutional neural network
US11780452B2 (en) Method and system for fault diagnoses of intelligent vehicles
CN115355973A (en) Truck overload detection method based on acoustic signals
CN114548199A (en) Multi-sensor data fusion method based on deep migration network
CN112052551B (en) Fan surge operation fault identification method and system
CN114548154A (en) Intelligent diagnosis method and device for important service water pump
CN112151067B (en) Digital audio tampering passive detection method based on convolutional neural network
CN114331214A (en) Domain-adaptive bearing voiceprint fault diagnosis method and system based on reinforcement learning
CN105823634A (en) Bearing damage identification method based on time frequency relevance vector convolution Boltzmann machine
CN113758709A (en) Rolling bearing fault diagnosis method and system combining edge calculation and deep learning
CN114873425B (en) Escalator driving chain fault diagnosis method based on vibration characteristic enhancement
Li et al. A robust fault diagnosis method for rolling bearings based on deep convolutional neural network
CN113537382A (en) Equipment diagnosis method and system based on residual shrinkage network
CN110599460A (en) Underground pipe network detection and evaluation cloud system based on hybrid convolutional neural network
CN115288994B (en) Improved DCGAN-based compressor abnormal state detection method
CN117349770B (en) Structural health monitoring multi-strain sensor data anomaly detection and repair method
CN116106016A (en) Rolling bearing fault diagnosis method based on joint learning convolutional neural network
CN116659829A (en) Mechanical equipment abnormality detection method, system, storage medium and equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination