CN110440148B - A method, device and system for classifying and identifying leaked acoustic signals - Google Patents
A method, device and system for classifying and identifying leaked acoustic signals Download PDFInfo
- Publication number
- CN110440148B CN110440148B CN201910635177.1A CN201910635177A CN110440148B CN 110440148 B CN110440148 B CN 110440148B CN 201910635177 A CN201910635177 A CN 201910635177A CN 110440148 B CN110440148 B CN 110440148B
- Authority
- CN
- China
- Prior art keywords
- leakage
- signal
- acoustic signal
- classification
- voiceprint
- 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.)
- Active
Links
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F17—STORING OR DISTRIBUTING GASES OR LIQUIDS
- F17D—PIPE-LINE SYSTEMS; PIPE-LINES
- F17D5/00—Protection or supervision of installations
- F17D5/02—Preventing, monitoring, or locating loss
- F17D5/06—Preventing, monitoring, or locating loss using electric or acoustic means
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Examining Or Testing Airtightness (AREA)
- Pipeline Systems (AREA)
Abstract
本申请公开了一种漏损声信号分类识别方法，用于管道的漏损检测，该方法包括：获取管道的漏损声信号，所述漏损声信号为时间‑振幅信号；预处理所述漏损声信号，得到所述漏损声信号的声纹图；其中，所述声纹图中横坐标为时间，纵坐标为频率，灰度值为振幅；输入所述声纹图至卷积神经网络分类模型；分类识别处理所述声纹图，获得所述漏损声信号的分类识别结果；该方法将管道漏损声信号的一维离散信号变换得到漏损声信号的声纹图，然后用卷积神经网络分类模型对该声纹图进行分类识别处理，根据分类识别处理结构判断是否存在漏损及漏损类型，解决了需要人工分析与判断而主观性较大的技术问题，实现了提高判断的准确性的技术效果。
The present application discloses a method for classification and identification of acoustic leakage signals, which is used for leakage detection of pipelines. The method includes: acquiring acoustic leakage signals of pipelines, where the acoustic leakage signals are time-amplitude signals; Leakage sound signal, and obtain a voiceprint image of the leaked acoustic signal; wherein, the abscissa in the voiceprint image is time, the ordinate is frequency, and the gray value is amplitude; input the voiceprint image to convolution A neural network classification model; the voiceprint image is processed by classification and identification to obtain the classification and identification result of the leaked acoustic signal; the method transforms the one-dimensional discrete signal of the pipeline leaked acoustic signal to obtain the voiceprint of the leaked acoustic signal, Then use the convolutional neural network classification model to classify and identify the voiceprint, and judge whether there is leakage and the type of leakage according to the classification and identification processing structure, which solves the technical problem that requires manual analysis and judgment and is highly subjective. The technical effect of improving the accuracy of judgment is achieved.
Description
Technical Field
The invention relates to the technical field of pipeline leakage detection, in particular to a method, a device and a system for detecting a pipeline leakage acoustic signal.
Background
Due to various complex reasons such as corrosion, the water supply network has a leakage phenomenon, which not only causes waste of water resources, but also causes pollution to water in the pipe network, and in addition, causes damages such as ground settlement. Therefore, it is necessary to detect and locate leaks in the water supply network. At present, sensors are often widely deployed in water supply networks to monitor leak points in the water supply network.
However, in the process of implementing the technical solution in the embodiment of the present application, the inventors of the present application find that the above-mentioned technology has at least the following technical problems:
in the above scheme, the sensor generally sends the acquired information to the handheld monitoring controller, and then the operator performs subjective judgment according to the information on the handheld monitoring controller, which results in low judgment accuracy.
Disclosure of Invention
The embodiment of the application provides a method, a device and a system for classifying and identifying a leakage sound signal, wherein a one-dimensional discrete signal of the pipeline leakage sound signal is converted to obtain a voiceprint image of the leakage sound signal, then a convolutional neural network classification model is used for classifying, identifying and processing the voiceprint image, whether the leakage and the leakage type exist is judged according to a classification and identification processing structure, the technical problem that manual analysis and judgment are needed and the subjectivity is high is solved, and the technical effect of improving the judgment accuracy is achieved.
The embodiment of the application provides a method for classifying and identifying leakage acoustic signals, which is used for detecting the leakage of a pipeline and comprises the following steps:
acquiring a leakage acoustic signal of a pipeline, wherein the leakage acoustic signal is a time-amplitude signal;
preprocessing the leakage acoustic signal to obtain a voiceprint image of the leakage acoustic signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
inputting the voiceprint image into a convolutional neural network classification model;
and carrying out classification recognition processing on the voiceprint image to obtain a classification recognition result of the leakage acoustic signal.
The embodiment of the present application further provides a leakage acoustic signal classification and identification device for the leakage detection of the pipeline, include:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a leakage acoustic signal of a pipeline, and the leakage acoustic signal is a time-amplitude signal;
the preprocessing module is used for preprocessing the leakage sound signal to obtain a voiceprint image of the leakage sound signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
the input module is used for inputting the voiceprint image to a convolutional neural network classification model;
and the classification identification module is used for classifying, identifying and processing the voiceprint image to obtain a classification identification result of the leakage acoustic signal.
The embodiment of the present application further provides a leakage acoustic signal classification and identification device for the leakage detection of the pipeline, including:
a memory for storing a computer program;
a processor for implementing the steps of the method for classifying and identifying a missing acoustic signal as described above when executing the computer program.
The embodiment of the present application further provides a leakage acoustic signal collection device for the leakage acoustic signal of collection pipeline, include:
the signal receiving unit is arranged on the pipeline and used for converting the received leakage acoustic signal into an electric signal;
the signal processing unit is connected with the signal receiving unit and used for receiving and processing the electric signal of the signal receiving unit to obtain a discrete electric signal;
the GPS positioning unit is used for collecting the geographical position information of the acoustic signal acquisition device;
the data transmission unit is simultaneously connected with the signal processing unit and the GPS positioning unit and is used for sending the discrete electric signals and the geographical position information to the leaked acoustic signal classification and identification device;
and the power supply unit provides electric power support for the acoustic signal acquisition device.
The embodiment of the present application further provides a leakage acoustic signal classification and identification system for the leakage detection of the pipeline, including:
the leakage sound signal acquisition devices are used for acquiring leakage sound signals and transmitting the leakage sound signals to the classification and identification devices;
the leakage sound signal classification and identification device is used for classifying the leakage sound signals;
wherein, the leakage acoustic signal collection system includes:
the signal receiving unit is arranged on the pipeline and used for converting the received acoustic signals into electric signals;
the signal processing unit is connected with the signal receiving unit and used for receiving and processing the electric signal of the signal receiving unit to obtain a discrete electric signal;
the GPS positioning unit is used for collecting the geographical position information of the acoustic signal acquisition device;
the data transmission unit is simultaneously connected with the signal processing unit and the GPS positioning unit and is used for sending the discrete electric signals and the geographical position information to the leaked acoustic signal classification and identification device;
the power supply unit provides electric support for the acoustic signal acquisition device;
the leakage acoustic signal classification and identification device comprises:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a leakage acoustic signal of a pipeline, and the leakage acoustic signal is a time-amplitude signal;
the preprocessing module is used for preprocessing the leakage sound signal to obtain a voiceprint image of the leakage sound signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
the input module is used for inputting the voiceprint image to a convolutional neural network classification model;
and the classification identification module is used for classifying, identifying and processing the voiceprint image to obtain a classification identification result of the leakage acoustic signal.
The embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the method for classifying and identifying a missing acoustic signal as described above.
One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:
in the embodiment of the application, the leakage acoustic signal is acquired through the acquisition device, the leakage acoustic signal is a one-dimensional discrete electric signal based on time-amplitude, the time-amplitude signal is utilized to obtain the voiceprint image of the leakage acoustic signal, and then the voiceprint image is classified, identified and processed by utilizing the deep convolutional neural network classification model, so that whether the pipeline is leaked or not and the shape, size and other conditions of a leakage point can be automatically identified according to the voiceprint image information of the pipeline, manual analysis and judgment are not needed, the accuracy of leakage identification and detection is improved, and the method has strong practicability;
meanwhile, the plurality of collecting devices can be arranged in a net shape, and the distribution situation of the leakage points in the visual space can be generated by combining the geographical position information of the collecting devices, so that the relation among different kinds of leakage points and the future development trend of the leakage points can be researched and prejudged, and the overall control of the pipe network is realized.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments of the present invention will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without inventive labor.
Fig. 1 is a flowchart of a method for classifying and identifying a leakage acoustic signal according to an embodiment of the present application.
Fig. 2 is a time-amplitude signal of the leakage acoustic signal described in the embodiment of the present application.
Fig. 3 is a voiceprint diagram as described in the embodiments of the present application.
Fig. 4 is a schematic structural diagram of an unsupervised model based on a convolutional auto-encoder in the embodiment of the present application.
Fig. 5 is a schematic diagram of the classification result in the embodiment of the present application.
Fig. 6 is a schematic structural diagram of the leaked acoustic signal classification and identification device in the embodiment of the present application.
Fig. 7 is a schematic structural diagram of the leaked acoustic signal classification and identification device in the embodiment of the present application.
Fig. 8 is a schematic structural diagram of the leakage acoustic signal acquisition device in the embodiment of the present application.
Fig. 9 is a schematic structural diagram of a leakage acoustic signal classification and identification system in an embodiment of the present application.
Detailed Description
In order to better understand the technical solution, the technical solution will be described in detail with reference to the drawings and the specific embodiments.
The embodiment of the application provides a method for classifying and identifying a leakage acoustic signal, which is used for detecting the leakage of a pipeline and comprises the following steps in combination with fig. 1:
and 4, carrying out classification recognition processing on the voiceprint image to obtain a classification recognition result of the leakage acoustic signal.
The classification recognition result may include information on whether the pipe is leaking or not, and the size and shape of the leaking point of the pipe.
In the method of the embodiment of the invention, a convolutional neural network classification algorithm is utilized, and whether the pipeline is leaked or not and the conditions of the shape, the size and the like of the leakage are analyzed based on the characteristics of the sound pattern of the leakage sound signal; firstly, acquiring a leakage acoustic signal of a pipeline, converting a time-amplitude signal of the leakage acoustic signal into a voiceprint image, and performing classification and identification processing on the voiceprint image by adopting a convolutional neural network classification model to obtain a result; the convolutional neural network classification model can automatically identify and classify the missing points based on a deep neural network classification algorithm, manual analysis and judgment are not needed, and the accuracy of identification and classification is greatly improved.
In step 1 of this embodiment, a leakage acoustic signal of a pipeline is acquired, where the acquired leakage acoustic signal is a time-amplitude signal, and is represented as a one-dimensional discrete electrical signal X ═ { X1, X2, x3... xn }, where xn represents the amplitude of the leakage acoustic signal when time is n; for example, an experimental platform can be established with reference to table 1 below, and the leakage sound signals of different leakage points are collected.
TABLE 1 acquisition of different leakage acoustic signals
In step 2 of this embodiment, the preprocessing the leakage acoustic signal to obtain a voiceprint image of the leakage acoustic signal specifically includes:
step 21, performing discrete Fourier transform on the time-amplitude signal to obtain a time-frequency signal of the leakage acoustic signal;
and step 22, squaring the absolute value of the time-frequency signal to obtain a voiceprint image of the leakage acoustic signal.
More specifically, in step 21, as shown in fig. 2, the time-amplitude signal of the leakage acoustic signal is a one-dimensional acoustic waveform, and then discrete fourier transform is performed by using equation 1 under a small time window,
further obtaining a time-frequency signal of the leakage acoustic signal, namely a frequency distribution diagram of the leakage acoustic signal;
more specifically, in step 22, the time-frequency signal of the leakage acoustic signal is used to obtain the voiceprint map of the leakage acoustic signal by using the following formula 2:
SP(t，f)＝|X(t，f)|2x (t, f) formula 2
In equations 1 and 2, t is time, f is frequency, X (τ) is a wave function of the acoustic signal in the time domain, w (t- τ) is a window function, X (t, f) is a time-frequency signal of the leakage acoustic signal, and SP (t, f) is a voiceprint of the leakage acoustic signal.
In the voiceprint image, as shown in fig. 3, the abscissa represents time, the ordinate represents frequency, and the gray-scale value represents amplitude.
In step 3 of this embodiment, inputting the voiceprint image to the convolutional neural network classification model specifically includes:
converting the voiceprint image into a two-dimensional matrix, wherein rows and columns of the two-dimensional matrix respectively correspond to horizontal coordinates and vertical coordinates of the voiceprint image, and element values of the two-dimensional matrix correspond to gray values of the voiceprint image;
and inputting the two-dimensional matrix to an input layer of a convolutional neural network classification model.
In this embodiment, on one hand, the acquired one-dimensional leakage acoustic signals can only respectively reflect changes of amplitude or frequency along with time, and cannot globally reflect changes of amplitude, frequency and time of the leakage acoustic signals, and the reflected characteristics are few; on the other hand, since the leakage acoustic signal is extremely unstable and the variation range of the amplitude or frequency is large with time, it is necessary to know the variation of the amplitude and frequency of the acoustic signal with time; therefore, in the present embodiment, the collected leakage acoustic signal is converted into a voiceprint map, which reflects the variation of the energy intensity of the leakage acoustic signal at each frequency with time. The method is to cut the original sound signal by a small time window and then to perform discrete Fourier transform, and the voiceprint image shows the change of the energy intensity of the signal in time and frequency in one image at the same time, so that the characteristics of the sound signal are more preserved. In addition, in this embodiment, the voiceprint image is converted into a two-dimensional matrix (
Xnm, representing the amplitude of the acoustic signal at different time and frequency), and then inputting the two-dimensional matrix into the convolutional neural network classification model, so as to match the input of the convolutional neural network classification model and facilitate the extraction of features; and the identification and judgment of the classification model are facilitated.
Illustratively, the pretreatment in step 2, for example, may also employ a cepstrum analysis method: in order to analyze the main frequency components of the leaky sound signal, it is necessary to extract the envelope of the leaky sound signal spectrum to analyze the position, peak or even derivative, etc. of the peak in the frequency of the signal. In order to achieve the above purpose, it is necessary to first perform Fourier transform on the leakage sound signal X t to obtain a frequency spectrum X k, where a logarithmic spectrum (in decibels), i.e., log (X k), is used. The frequency spectrum log (X [ k ]) can be decomposed into log (X [ k ]) log (H [ k ]) + log (E [ k ]), where log (H [ k ]) represents the low frequency envelope and log (E [ k ]) represents the high frequency details. And performing inverse Fourier transform on the two sides to obtain x [ t ] ═ h [ t ] + e [ t ], and filtering out high-frequency e [ t ] through a low-pass filter to obtain low-frequency h [ t ]. This h t is the extracted feature, i.e. the input to the model.
In the embodiment of the present application, the preprocessing manner in step 2 is exemplified, and it can be understood that the method is not limited to the two preprocessing manners, and other transformation manners may also be adopted to convert the one-dimensional leakage acoustic signal into the voiceprint image.
The convolutional neural network classification model of the present embodiment includes an unsupervised model based on a convolutional auto-encoder and a classifier based on a support vector machine.
Specifically, the convolutional neural network classification model is based on a deep neural network algorithm, and in terms of structure, the convolutional neural network classification model comprises an unsupervised model based on a convolutional self-encoder and a classifier based on a support vector machine; unsupervised model based convolution autoencoder capable of matching input leakage acoustic signal
And performing cyclic training to extract the acoustic signal characteristics, and then performing classification and identification processing on the acoustic signal characteristics by the classifier based on the acoustic signal characteristics to obtain a classification and identification result of the leakage acoustic signal.
It can be understood that the unsupervised model based on the convolution self-encoder can project the high-dimensional input into the low-dimensional subspace, so as to obtain a better representation of the signal, which is embodied in that after the high-dimensional input is compressed into the low-dimensional (two-dimensional or three-dimensional) output, the trained effect of the model can be visually seen through a visualization method. Then, only a shallow classifier, namely a classifier based on a support vector machine, is needed to be connected to the rear end of the model, and classification of the leakage acoustic signals can be completed by using a supervised learning method.
It can be understood that the unsupervised learning method can be used for not only finishing the simple task of classifying the leakage point signals and the interference signals, but also visually generating the distribution situation of the leakage points in the subspace, so that a researcher can research and prejudge the relation among different types of leakage points and the future development trend of the leakage points, and the overall control of the pipe network is realized.
Fig. 4 is a schematic structural diagram of the unsupervised model based on the convolutional auto-encoder, which is combined with fig. 4, the unsupervised model based on the convolutional auto-encoder is composed of an encoder 41 and a decoder 42, the encoder 41 is composed of a convolutional layer, a batch normalization layer, an activation function ReLU, a max-pooling layer and a first fully-connected layer; the decoder 42 is composed of a convolutional layer, a batch normalization layer, an activation function ReLU, an upsampled layer, and a second fully-connected layer, and the encoder 41 and the decoder 42 are connected end-to-end by a number of first fully-connected layers and second fully-connected layers.
It should be noted that fig. 4 is only a schematic diagram of the model, and the actual structure of the model is more complicated, including more convolutional layers 44.
Specifically, the encoder 41 can extract image features of the input leakage acoustic signal, that is, the leakage acoustic signal features 43, through convolution operation, as the depth of the convolution layer 44 increases, the dimensions of the length and width of the input image gradually decrease, the thickness gradually increases, and the extracted features also abstract evolve from a simple direction; the decoder 42 is able to recover the input image from the image features, with the goal of training being to minimize the difference between the real input image and the image recovered by the decoder, which is also called the loss function, and is represented by the mean square error; for example, when the judgment accuracy of the model on the training set can reach more than 0.95, the training of the model can be considered to be completed.
In the encoder, the batch of normalization layers are used for adjusting the parameter specification of each layer in the encoder network, and model convergence is facilitated; the activation function ReLU gives the model the ability to fit complex nonlinear functions; the maximum pooling layer can compress the length and width of an input image, and reduce the burden of deep network calculation; the first fully-connected layer is capable of converting the output of a series of convolution operations into a feature vector.
In decoder 42, the second fully-connected layer can restore the feature vectors to acceptable inputs for the higher-dimensional convolutional layer.
In the embodiment of the application, the classifier based on the support vector machine obtains the support vector by using a Lagrange method and a KKT condition as an optimization method so as to obtain the hyperplane which is divided into the best features and achieve the final classification purpose.
Wherein, as shown in fig. 5, A, B, C, D, E shows the classification and identification results in step 4, for example, when the leakage acoustic signals which are acquired from different pipeline acquisition points for multiple times and are unknown whether to leak or not are input into a convolutional neural network classification model for processing, A, B, C, D, E is classified into two types, one type (A, B, E) is represented by a triangular symbol 52 and is a point which is not leaked, and the other type (C, D) is represented by a circle 53 and is a point which is leaked; it can be seen that the convolutional neural network classification model can distinguish two different types of signals based on the leaky acoustic signal features 51; in other examples, the shape, size, leakage amount, etc. of the leakage points may also be classified.
In addition, in the embodiment of the present invention, the process of obtaining the convolutional neural network classification model includes:
step a, collecting a leakage acoustic signal sample and establishing a leakage voiceprint pattern database;
b, inputting a leakage voiceprint image sample to an unsupervised model based on a convolution self-encoder for training to obtain the characteristics of the acoustic signal;
and c, classifying the acoustic signal features by using a classifier based on a support vector machine.
In the step a, collecting a sample of the leakage acoustic signal and establishing a database of a sample of the leakage acoustic pattern specifically comprise:
collecting a leakage sound signal sample;
preprocessing the leakage acoustic signal sample to obtain a leakage voiceprint image sample;
and establishing a leakage voiceprint pattern database.
Specifically, different kinds of analog leaky point acoustic signals with fixed lengths need to be collected to establish a leaky voiceprint pattern database. The database needs to contain sound signals of at least ten leak points of different shapes and sizes. Because of the training involved on deep neural networks, a significant sample size is required. For each leak signal, five seconds as a length of one sample, at least 2000 samples need to be collected. Since the leak points can be classified into 10 or more classes depending on the size of the shape, a total of 20000 or more samples of the leaked acoustic signal are required. Meanwhile, an acoustic signal in a leak-point-free state is collected to serve as a blank control group, noise which is frequently generated in a real environment paved by an urban pipe network is added to serve as an interference item, and the blank control group also needs twenty thousand samples which are equivalent to the leak-point group data in scale. After the database is built, the samples in the database need to be sorted disorderly, 4/5 of the samples are taken as a training set, and the rest 1/5 is taken as a test set.
In the step b, inputting the missing voiceprint image sample to an unsupervised model based on a convolution self-encoder to train the voiceprint image sample comprises three steps: forward propagation, backward propagation, gradient descent; the above three steps will be called circularly until the model converges. The forward propagation takes a voiceprint image of an acoustic signal sample as an input, and the parameters of a hidden layer in the whole model are updated. The back propagation is to derive each parameter in the entire model hidden layer with a loss function and update the old derivative with the newly found derivative. The gradient descent procedure is to use the derivative multiplied by a small step size to update the parameters in the hidden layer. And circularly calling the three steps until the model converges.
In conclusion, in the embodiment of the application, a leakage acoustic signal sample is collected, a leakage voiceprint pattern database is established, then an unsupervised model based on a convolution self-encoder and a classifier based on a support vector machine are adopted to train the voiceprint pattern sample to obtain a convolution neural network classification model, the classification model can be used for carrying out classification and identification on the leakage acoustic signal, manual analysis and judgment are not needed, and the accuracy of identification and classification is greatly improved; and the leakage acoustic signals can be collected at different positions of the pipeline, and compared with a method for detecting the leakage point of the pipeline by moving a control device in the pipeline, the method is more suitable for the pipe network condition of a complex topological structure and the pipeline with a smaller caliber.
In the embodiment of the application, the database for establishing the missing voiceprint pattern is provided for the first time. Through the establishment of the missing voiceprint pattern database, newly collected signals can be compared with tens of thousands of samples in the signal database, and the judgment of missing points can be obtained. Compared with a method for directly analyzing a newly collected signal, the method is more accurate and stable. Moreover, a database is established by a two-dimensional leakage voiceprint image instead of a one-dimensional discrete acoustic signal, so that more information of the leakage point acoustic signal in two dimensions of a time domain and a frequency domain can be contained, and the accuracy of model prediction is higher.
In one possible implementation manner, in step 1, the missing acoustic signal includes a location tag of the missing acoustic signal, and then the classification and identification result includes a location tag corresponding to the missing acoustic signal.
It can be understood that when the leakage acoustic signals of multiple positions of the pipeline are obtained, the distribution situation of different leakage points in the visual space can be generated by combining the geographical position information of the leakage acoustic signals, and then the relation among different types of leakage points and the future development trend of the leakage points can be researched and prejudged according to the distribution situation, so that the global control of the pipe network is realized.
With reference to fig. 6, an embodiment of the present application further provides a leakage acoustic signal classification and identification apparatus, which is used for detecting leakage of a pipeline, and includes:
an obtaining module 61, configured to obtain a leakage acoustic signal of the pipeline, where the leakage acoustic signal is a time-amplitude signal;
a preprocessing module 62, configured to preprocess the leakage acoustic signal to obtain a voiceprint image of the leakage acoustic signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
an input module 63, configured to input the voiceprint image to a convolutional neural network classification model;
a classification recognition module 64, configured to perform classification recognition on the voiceprint image to obtain a classification recognition result of the leakage acoustic signal;
in this embodiment, the preprocessing module includes:
the first transformation submodule is used for performing discrete Fourier transformation on the time-amplitude signal to obtain a time-frequency signal of the leakage sound signal;
and the second transformation submodule is used for squaring the absolute value of the time-frequency signal to obtain a voiceprint image of the leakage sound signal.
In this embodiment, the second transform submodule obtains a voiceprint image of the leakage acoustic signal by using the following formula:
SP(t，f)＝|X(t，f)2＝X(t，f)X*(t，f)
where t is time, f is frequency, X (t, f) is a time-frequency signal of the leakage acoustic signal, and SP (t, f) is a voiceprint of the leakage acoustic signal.
In this embodiment, the input module includes:
the first input submodule is used for converting the voiceprint image into a two-dimensional matrix, wherein the row and the column of the two-dimensional matrix respectively correspond to the abscissa and the ordinate of the voiceprint image, and the element value of the two-dimensional matrix corresponds to the gray value of the voiceprint image;
and the second input submodule is used for inputting the two-dimensional matrix to the convolutional neural network classification model.
With reference to fig. 7, an embodiment of the present application further provides a leakage acoustic signal classification and identification device, which is used for detecting leakage of a pipeline, and includes: a memory 71 for storing a computer program; the processor 72 is configured to implement the steps of the above-mentioned method for classifying and identifying a missing acoustic signal when executing a computer program.
With reference to fig. 8, an embodiment of the present application further provides a leakage acoustic signal collecting device, configured to collect a leakage acoustic signal of a pipeline, and then transmit the leakage acoustic signal to a classification and identification device in a wired or wireless manner, so as to perform classification and identification on the leakage acoustic signal.
This collection system includes:
a signal receiving unit 81 installed at the pipe 80 for converting the received acoustic signal into an electrical signal;
a signal processing unit 82 connected to the signal receiving unit 81, for receiving and processing the electrical signal of the signal receiving unit 81 to obtain a discrete electrical signal;
a GPS positioning unit 83 that collects geographical position information of the leaked sound signal acquisition device;
the data transmission unit 84 is connected with the signal processing unit 82 and the GPS positioning unit 83, and is used for sending the discrete electric signals and the geographical position information to the leaked acoustic signal classification and identification device;
and the power supply unit 85 is used for providing power support for the leakage sound signal acquisition device.
In practice, the signal receiving unit 81 is an underwater acoustic transducer, which belongs to an underwater acoustic transducer of piezoelectric ceramic material, the front end of the underwater acoustic transducer is provided with a black sphere for collecting underwater leak point sound signals in an omnibearing manner, and the underwater acoustic transducer can convert the sound signals generated by the leak points in the water supply pipeline 80 into electric signals; the underwater acoustic transducer is mounted to the pipe 80 by a mounting base 86, wherein a black sphere, which is used by the underwater acoustic transducer to collect signals, is provided at the axial center of the pipe.
In practice, the signal processing unit 82 includes an amplifying module 821, a filtering module 822 and an analog-to-digital conversion module 823 which are connected in sequence; the amplifying module 821 is connected with the signal receiving unit 81 through a lead, and the analog-to-digital conversion module 823 is connected with the data transmission unit 84; the electrical signal collected by the underwater acoustic transducer is first transmitted to the amplifying module 821 for amplification, then subjected to noise reduction processing by the filtering module 822, and finally converted into a discrete electrical signal which can be recognized and processed by the computer by the analog-to-digital conversion module 823.
The data transmission unit 84 may transmit the geographical location information and the discrete electrical signal acquired by the GPS positioning unit 83 to the classification and identification device in a wired or wireless manner, so as to perform a step of classifying and identifying the missing acoustic signal.
In practice, the power supply unit 85 is a water flow generator, which provides power to the entire collection device and does not require external charging.
In this embodiment, the leakage acoustic signal acquisition device can collect the real-time monitored leakage acoustic signals and the GPS geographical location information to the classification and identification device. The plurality of leakage sound signal acquisition devices can be arranged in a net shape, for example, the leakage sound signal acquisition devices can be arranged on a complex urban pipe network by adopting a certain optimization algorithm, and urban fire hydrants can be generally selected as installation sites, so that the whole area to be monitored is ensured to be covered under the monitoring of the system. And then, the classification and identification device is used for classifying and identifying the collected leakage sound signals, so that the comprehensive automation and the intellectualization of the monitoring of the leakage of the pipe network can be realized, and the labor cost is greatly saved.
With reference to fig. 9, an embodiment of the present application further provides a leakage acoustic signal classification and identification system for leakage detection of a pipeline, where the classification and identification system includes a plurality of leakage acoustic signal acquisition devices 93 and a leakage acoustic signal classification and identification device 92;
the leakage sound signal acquisition device 93 is arranged on a pipe network between the water works 90 and the users 91, and the leakage sound signal acquisition device 93 is used for acquiring leakage sound signals and transmitting the leakage sound signals to the classification and identification device; wherein, this leakage acoustic signal collection device 93 includes:
the signal receiving unit is arranged on the pipeline and used for converting the received acoustic signals into electric signals;
the signal processing unit is connected with the signal receiving unit and used for receiving and processing the electric signal of the signal receiving unit to obtain a discrete electric signal;
the GPS positioning unit is used for collecting the geographical position information of the acoustic signal acquisition device;
the data transmission unit is simultaneously connected with the signal processing unit and the GPS positioning unit and is used for sending the discrete electric signals and the geographical position information to the leaked acoustic signal classification and identification device;
the power supply unit provides electric support for the acoustic signal acquisition device;
the leaked sound signal classification and identification device 92 is used for classifying the leaked sound signal; the classification recognition apparatus includes:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a leakage acoustic signal of a pipeline, and the leakage acoustic signal is a time-amplitude signal;
the preprocessing module is used for preprocessing the leakage sound signal to obtain a voiceprint image of the leakage sound signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
the input module is used for inputting the voiceprint image to a convolutional neural network classification model;
and the classification identification module is used for classifying, identifying and processing the voiceprint image to obtain a classification identification result of the leakage acoustic signal.
The embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the method for classifying and identifying a missing acoustic signal as described above.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (5)
1. A method for classifying and identifying a leakage acoustic signal is used for detecting leakage of a pipeline and is characterized by comprising the following steps:
acquiring a leakage acoustic signal of a pipeline, wherein the leakage acoustic signal is a time-amplitude signal;
preprocessing the leakage acoustic signal to obtain a voiceprint image of the leakage acoustic signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
inputting the voiceprint image into a convolutional neural network classification model;
classifying, identifying and processing the voiceprint image to obtain a classification identification result of the leakage acoustic signal;
preprocessing the leakage acoustic signal, wherein the voiceprint image of the leakage acoustic signal is obtained by:
performing discrete Fourier transform on the time-amplitude signal to obtain a time-frequency signal of the leakage sound signal;
squaring the absolute value of the time-frequency signal by the following formula 2 to obtain a voiceprint image of the leakage acoustic signal;
SP(t，f)＝|X(t，f)|2＝X(t，f)X*(t, f) equation 2
In the formula 2, t is time, f is frequency, X (t, f) is a time-frequency signal of the leakage acoustic signal, and SP (t, f) is a voiceprint image of the leakage acoustic signal;
the input of the voiceprint image to the convolutional neural network classification model is as follows:
converting the voiceprint image into a two-dimensional matrix, wherein rows and columns of the two-dimensional matrix respectively correspond to horizontal coordinates and vertical coordinates of the voiceprint image, and element values of the two-dimensional matrix correspond to gray values of the voiceprint image;
inputting the two-dimensional matrix to an input layer of a convolutional neural network classification model;
the convolutional neural network classification model comprises an unsupervised model based on a convolutional self-encoder and a classifier based on a support vector machine; the unsupervised model based on the convolution self-encoder consists of an encoder and a decoder, wherein the encoder consists of a convolution layer, a batch normalization layer, an activation function ReLU, a maximum pooling layer and a first full-link layer; the decoder consists of a convolutional layer, a batch normalization layer, an activation function ReLU, an upsampling layer and a second fully-connected layer, and the encoder and the decoder are connected end to end through a plurality of first fully-connected layers and second fully-connected layers; the classifier based on the support vector machine obtains the support vector by using a Lagrange method and a KKT condition as an optimization method so as to obtain a hyperplane with the best division of the characteristics and achieve the final classification purpose;
the process of obtaining the convolutional neural network classification model comprises the following steps:
step a, collecting a leakage acoustic signal sample and establishing a leakage voiceprint pattern database;
b, inputting the leakage voiceprint image sample to the unsupervised model based on the convolution self-encoder for training to obtain the characteristics of the acoustic signal;
c, classifying the acoustic signal features by using the classifier based on the support vector machine;
in step b, the process of inputting the missing voiceprint image samples to the unsupervised model based on the convolutional auto-encoder for training includes three steps: forward propagation, backward propagation, gradient descent;
and the leakage sound signal comprises a position label of the leakage sound signal, and the classification and identification result comprises the position label corresponding to the leakage sound signal.
2. A kind of leakage acoustic signal classification recognition device, the leakage used for pipeline measures, characterized by that, including:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a leakage acoustic signal of a pipeline, and the leakage acoustic signal is a time-amplitude signal;
the preprocessing module is used for preprocessing the leakage sound signal to obtain a voiceprint image of the leakage sound signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
the input module is used for inputting the voiceprint image to a convolutional neural network classification model;
the classification identification module is used for classifying, identifying and processing the voiceprint image to obtain a classification identification result of the leakage acoustic signal;
wherein the preprocessing module comprises:
the first transformation submodule is used for performing discrete Fourier transformation on the time-amplitude signal to obtain a time-frequency signal of the leakage sound signal;
a second transformation submodule for squaring the absolute value of the time-frequency signal by the following formula 2 to obtain a voiceprint image of the leakage acoustic signal;
SP(t，f)＝|X(t，f)|2＝X(t，f)X*(t, f) equation 2
In the formula 2, t is time, f is frequency, X (t, f) is a time-frequency signal of the leakage acoustic signal, and SP (t, f) is a voiceprint image of the leakage acoustic signal;
the input module includes:
the first input submodule is used for converting the voiceprint image into a two-dimensional matrix, wherein rows and columns of the two-dimensional matrix respectively correspond to horizontal coordinates and vertical coordinates of the voiceprint image, and element values of the two-dimensional matrix correspond to gray values of the voiceprint image;
the second input submodule is used for inputting the two-dimensional matrix to an input layer of a convolutional neural network classification model;
wherein the convolutional neural network classification model comprises an unsupervised model based on a convolutional auto-encoder and a classifier based on a support vector machine; the unsupervised model based on the convolution self-encoder consists of an encoder and a decoder, wherein the encoder consists of a convolution layer, a batch normalization layer, an activation function ReLU, a maximum pooling layer and a first full-link layer; the decoder consists of a convolutional layer, a batch normalization layer, an activation function ReLU, an upsampling layer and a second fully-connected layer, and the encoder and the decoder are connected end to end through a plurality of first fully-connected layers and second fully-connected layers; the classifier based on the support vector machine obtains the support vector by using a Lagrange method and a KKT condition as an optimization method so as to obtain a hyperplane with the best division of the characteristics and achieve the final classification purpose;
the process of obtaining the convolutional neural network classification model comprises the following steps:
step a, collecting a leakage acoustic signal sample and establishing a leakage voiceprint pattern database;
b, inputting the leakage voiceprint image sample to the unsupervised model based on the convolution self-encoder for training to obtain the characteristics of the acoustic signal;
c, classifying the acoustic signal features by using the classifier based on the support vector machine;
in step b, the process of inputting the missing voiceprint image samples to the unsupervised model based on the convolutional auto-encoder for training includes three steps: forward propagation, backward propagation, gradient descent;
and the leakage sound signal comprises a position label of the leakage sound signal, and the classification and identification result comprises the position label corresponding to the leakage sound signal.
3. A kind of leakage acoustic signal classification identification equipment, the leakage used for pipeline is detected, characterized by that, including:
a memory for storing a computer program;
a processor for implementing the steps of the method for classification and identification of a missing acoustic signal as claimed in claim 1 when executing the computer program.
4. The utility model provides a leakage acoustic signal collection system for the leakage acoustic signal of gathering pipeline, its characterized in that includes:
the signal receiving unit is arranged on the pipeline and used for converting the received leakage acoustic signal into an electric signal;
the signal processing unit is connected with the signal receiving unit and used for receiving and processing the electric signal of the signal receiving unit to obtain a discrete electric signal;
the GPS positioning unit is used for collecting the geographical position information of the acoustic signal acquisition device;
a data transmission unit, connected to both the signal processing unit and the GPS positioning unit, for sending the discrete electrical signal and the geographical location information to a leaked acoustic signal classification and identification device, where the leaked acoustic signal classification and identification device is the leaked acoustic signal classification and identification device according to claim 2;
and the power supply unit provides electric power support for the acoustic signal acquisition device.
5. A leakage acoustic signal classification and identification system for leakage detection of a pipeline, comprising:
the leakage sound signal acquisition devices are used for acquiring leakage sound signals and transmitting the leakage sound signals to the classification and identification devices;
the leakage sound signal classification and identification device is used for classifying the leakage sound signals;
wherein, the leakage acoustic signal collection device is the leakage acoustic signal collection device of claim 4, the leakage acoustic signal classification and identification device is the leakage acoustic signal classification and identification device of claim 2, the leakage acoustic signal collection device includes:
the signal receiving unit is arranged on the pipeline and used for converting the received acoustic signals into electric signals;
the signal processing unit is connected with the signal receiving unit and used for receiving and processing the electric signal of the signal receiving unit to obtain a discrete electric signal;
the GPS positioning unit is used for collecting the geographical position information of the acoustic signal acquisition device;
the data transmission unit is simultaneously connected with the signal processing unit and the GPS positioning unit and is used for sending the discrete electric signals and the geographical position information to the leaked acoustic signal classification and identification device;
the power supply unit provides electric support for the acoustic signal acquisition device;
the leakage acoustic signal classification and identification device comprises:
the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a leakage acoustic signal of a pipeline, and the leakage acoustic signal is a time-amplitude signal;
the preprocessing module is used for preprocessing the leakage sound signal to obtain a voiceprint image of the leakage sound signal; wherein, the horizontal coordinate in the voiceprint image is time, the vertical coordinate is frequency, and the gray value is amplitude;
the input module is used for inputting the voiceprint image to a convolutional neural network classification model;
and the classification identification module is used for classifying, identifying and processing the voiceprint image to obtain a classification identification result of the leakage acoustic signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910635177.1A CN110440148B (en) | 2019-07-15 | 2019-07-15 | A method, device and system for classifying and identifying leaked acoustic signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910635177.1A CN110440148B (en) | 2019-07-15 | 2019-07-15 | A method, device and system for classifying and identifying leaked acoustic signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110440148A CN110440148A (en) | 2019-11-12 |
| CN110440148B true CN110440148B (en) | 2021-07-13 |
Family
ID=68430405
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910635177.1A Active CN110440148B (en) | 2019-07-15 | 2019-07-15 | A method, device and system for classifying and identifying leaked acoustic signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110440148B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111899390B (en) * | 2020-08-04 | 2021-10-22 | 福建爱巴士网络科技有限公司 | Intelligent access control method and terminal integrating sound codes and two-dimensional codes |
| CN111881999A (en) * | 2020-08-04 | 2020-11-03 | 武汉易维环境工程有限公司 | Water service pipeline leakage detection method and system based on deep convolutional neural network |
| CN113154268B (en) * | 2021-03-18 | 2023-07-14 | 西安交通大学 | A wireless passive damage leakage monitoring system and its control method |
| CN113217826B (en) * | 2021-06-03 | 2022-06-14 | 邯郸市水利水电勘测设计研究院 | Pipeline water supply pipe network leakage alarm control method, device and medium |
| CN114417940A (en) * | 2022-03-25 | 2022-04-29 | 阿里巴巴（中国）有限公司 | Equipment for detecting data center, method and device for obtaining equipment detection model |
| CN115146684A (en) * | 2022-07-29 | 2022-10-04 | 上海威派格智慧水务股份有限公司 | Water supply pipe network leakage detection method and device based on pipeline vibration signals |
| CN115655594B (en) * | 2022-12-12 | 2023-03-10 | 杭州兆华电子股份有限公司 | Method and device for distinguishing and identifying throttling and leakage of pipeline based on acoustic cloud chart |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105546361A (en) * | 2016-03-08 | 2016-05-04 | 钱昊铖 | Acoustic-wave-method gas pipeline leakage monitoring method based on ANN (Artificial Neural Network) |
| CN106369288B (en) * | 2016-08-31 | 2018-10-02 | 瀚沃环境技术(上海)有限公司 | Water supply network leakage loss monitoring system |
| US20180347763A1 (en) * | 2017-05-31 | 2018-12-06 | Schneider Electric Software, Llc | Machine learning detection of pipeline ruptures |
| CN109654383A (en) * | 2018-11-06 | 2019-04-19 | 三川智慧科技股份有限公司 | A kind of leak degree of water supply line determines method and system |
| CN109556797B (en) * | 2018-11-19 | 2020-10-09 | 浙江工业大学 | Pipeline leak detection and localization method based on spline local mean decomposition and convolutional neural network |
| CN109945075B (en) * | 2019-03-07 | 2021-03-02 | 三川智慧科技股份有限公司 | Method and device for detecting leakage degree of water supply pipeline |
| CN109931506A (en) * | 2019-03-14 | 2019-06-25 | 三川智慧科技股份有限公司 | Pipeline leakage detection method and device |
-
2019
- 2019-07-15 CN CN201910635177.1A patent/CN110440148B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| 城市燃气管道泄漏诊断方法;王新颖等;《消防科学与技术》;20180630;第37卷(第6期);第834-838页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110440148A (en) | 2019-11-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110440148B (en) | A method, device and system for classifying and identifying leaked acoustic signals | |
| Oliveira et al. | Ultrasound-based identification of damage in wind turbine blades using novelty detection | |
| CN109858408B (en) | Ultrasonic signal processing method based on self-encoder | |
| CN110427878B (en) | Method and system for identifying rapid radio storm signals | |
| CN115356397B (en) | Method and device for diagnosing hollow defects of concrete-filled steel tube structures based on sound signals | |
| CN107435817A (en) | A kind of 2 leak detection accurate positioning methods of pressure pipeline | |
| CN115034271B (en) | An acoustic identification method for gas leakage in pressure vessels with automatic feature extraction | |
| CN112985574B (en) | High-precision classification identification method for optical fiber distributed acoustic sensing signals based on model fusion | |
| CN110007355A (en) | A convolutional autoencoder and method and device for detecting abnormality inside an object | |
| CN111695465B (en) | Pipe network fault diagnosis and positioning method and system based on pressure wave mode identification | |
| CN112762362A (en) | Underwater pipeline leakage acoustic emission detection method based on convolutional neural network | |
| CN105488520A (en) | Multi-resolution singular-spectrum entropy and SVM based leakage acoustic emission signal identification method | |
| CN111275108A (en) | A method for sample expansion of partial discharge data based on generative adversarial network | |
| CN112990082B (en) | Detection and identification method of underwater sound pulse signal | |
| CN118424369B (en) | Urban environment detection and comprehensive evaluation method based on remote sensing data | |
| CN118731172B (en) | System and method for detecting baseline of gathering and conveying pipeline based on multi-technology fusion | |
| CN117368651A (en) | Comprehensive analysis system and method for faults of power distribution network | |
| CN112711873A (en) | Method, system and medium for carrying out damage positioning based on lamb wave and neural network | |
| US20220291175A1 (en) | Acoustic resonance diagnostic method for detecting structural degradation and system applying the same | |
| CN118817859B (en) | Nondestructive ultrasonic flaw detection method and system for seamless oil casing | |
| CN119150020A (en) | Method, system, equipment and medium for constructing sliding bearing fault diagnosis model | |
| CN108834043A (en) | Priori knowledge-based compressed sensing multi-target passive positioning method | |
| CN113536894A (en) | Defect detection method, defect detection apparatus and computer readable storage medium | |
| CN117287643A (en) | Water leakage monitoring method, device, equipment and storage medium for heating and ventilation system | |
| CN117689967A (en) | Phased array nonlinear imaging crack length prediction method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |