Underwater leakage sound recognition method and system based on deep learning

Abstract

The invention discloses an underwater leakage sound recognition method and system based on deep learning, and relates to the field of structural water area leakage detection. The problems that according to an existing hydrophone scheme, whether leakage exists or not needs to be judged manually, and according to an electrode method scheme, the measurement process is tedious and consumed time is long are solved. The method comprises the following steps: acquiring underwater sound signals in different environments; effective interval features of the sound signals in the normal environment and the leakage environment are extracted respectively; constructing a deep learning network model; training effective interval features of the extracted signals, and inputting the trained signals into a deep learning network model to obtain a voice recognition model; judging whether leakage exists or not according to the identification model, and if no leakage exists, directly displaying a result; and if the leakage condition exists, determining the position of the leakage point through a sound source positioning technology. According to the invention, the precision and reliability of leakage detection are improved.

Description

Technical Field

[0001] The present invention relates to the field of structural water body leakage detection, and in particular to an underwater leakage sound recognition method based on deep learning. Background Art

[0002] At present, there are two main solutions for pool structural leakage detection, namely the hydrophone solution and the electrode method solution. These two solutions have their own advantages and disadvantages in practical applications and are suitable for different situations. The following is a detailed description of these two solutions:

[0003] 1. Hydrophone solution: The hydrophone solution places a hydrophone in the pool and uses it to monitor the noise changes in the underwater environment, especially the sound characteristics such as water flow sound and bubble sound generated when leaking. The basic principle of this solution is to use the water flow disturbance or bubble generation caused by leakage in the pool to analyze the possibility of leakage. Its advantage is that the hydrophone solution is relatively simple to operate and does not require complex equipment. Workers can judge whether there is a leak by listening around the pool. This solution has no special requirements for the structure of the pool and is suitable for leakage detection of various types of pools. However, due to the reliance on manual listening, there are strong subjective factors. The hearing, experience and emotions of workers may affect the accuracy of judgment. The results of the hydrophone solution may vary among different workers or in different environments, lacking standardization and consistency. At the same time, even if the sound of leakage can be heard, the specific leakage location cannot be determined, and usually only a rough area can be given.

[0004] 2. Electrode method: The electrode method is to set up two transmitting electrodes A and B, which are connected to the upstream and downstream of the pool or different areas of the pool, respectively, and to send special frequency signals with adjustable voltage, and monitor the changes of these signals through the receiver and multiple underwater probes to determine the leakage location. However, the detection process of the electrode method is relatively complicated. It is necessary to arrange multiple electrodes in different areas of the pool and scan and analyze the entire pool one by one. The detection process is relatively cumbersome and time-consuming. Moreover, a set of electrode detection systems usually consists of multiple probes, electrodes, transmitters, receivers, etc., so its equipment investment is large, and professional personnel are required to operate and maintain it, which increases the cost of initial investment and later maintenance. At the same time, although the electrode method can provide more accurate data, the data still needs to be analyzed and processed after measurement to accurately locate the leakage point. This process is relatively cumbersome, and data analysis takes a long time, and the leakage point cannot be found in time. Summary of the invention

[0005] In view of the problems that the existing hydrophone solution requires manual judgment on whether there is leakage and the electrode method solution has a cumbersome and time-consuming measurement process, the present invention proposes an underwater leakage sound recognition method based on deep learning, the method comprising:

[0006] S1: Collect underwater sound signals in different environments;

[0007] S2: Extract the effective interval features of the sound signal in normal environment and leakage environment respectively;

[0008] S3: Build a deep learning network model;

[0009] S4: training the effective interval features of the extracted signal, inputting the trained signal into the deep learning network model, and obtaining a sound recognition model;

[0010] S5: Determine whether there is leakage according to the recognition model, and directly display the result if there is no leakage;

[0011] S6: If leakage exists, determine the location of the leakage point through sound source localization technology.

[0012] Furthermore, a preferred method is proposed, step S1 includes: collecting underwater sound signals in different environments through an underwater acoustic sensor, preprocessing the sound signals through a sensor preamplifier and filter circuit or an external amplifier and filter, and transmitting them to a host computer.

[0013] Furthermore, a preferred method is proposed, in which step S2 comprises:

[0014] S21: Perform short-time Fourier transform on the sound signal to convert the audio signal from the time domain to the frequency domain to obtain the acoustic characteristics of the sound signal;

[0015] S21: subtracting a short-time Fourier transform result of a normal underwater sound signal from a short-time Fourier transform result of an underwater sound signal with leakage to obtain a characteristic value m;

[0016] S23: Set a threshold t, and search for intervals greater than the threshold t in the eigenvalue m;

[0017] S24: Selecting acoustic features in the interval as valid interval features.

[0018] Furthermore, a preferred method is proposed, in step S3, the deep learning network model includes a plurality of convolutional layers and pooling layers stacked, the convolutional layers are used to extract local features, and the pooling layers are used for downsampling;

[0019] The convolution layer includes: using a (3, 3) convolution kernel to capture local features in both frequency and time dimensions; the first convolution layer starts with 1 input channel, and the subsequent convolution layers increase the number of channels, gradually extracting from low-level features to high-level features; using ReLU as an activation function; and performing batch normalization after each convolution layer;

[0020] Each convolutional layer is followed by a pooling layer, and the pooling window size is (2, 2);

[0021] Softmax is used as the activation function to convert the output value into a probability distribution.

[0022] Furthermore, a preferred method is proposed, in which step S4 comprises:

[0023] The sound data samples were divided into three parts: training set, validation set and test set, with a ratio of 8:1:1, and ten-fold cross validation was performed;

[0024] The model is fitted to the sound data samples on the training set, and whether the model meets the required standards is determined by whether the recognition rate reaches the set threshold;

[0025] If the required standard is not met, the learning will be continued. If the required standard is met, the deep learning network model will be verified through the validation set;

[0026] The validation set adjusts the model's hyperparameters and performs a preliminary evaluation of the model's capabilities. It also determines whether the model meets the required standards based on whether the recognition rate reaches the set threshold.

[0027] If the threshold requirement is not met, the program will return to continue learning. If the threshold requirement is met, the program will be tested.

[0028] The test set is used to evaluate the generalization ability of the deep learning network model. If the generalization ability reaches the preset threshold, the training ends, otherwise it returns to retraining.

[0029] Furthermore, a preferred method is proposed, in which step S6 comprises:

[0030] Use multiple hydrophones to form an array to measure underwater signals;

[0031] Compare the arrival times of signals from different hydrophones and calculate the time difference of signal arrival;

[0032] Based on the arrival time difference results, the location of the sound source of the leakage point is derived through trigonometric geometry.

[0033] Furthermore, a preferred embodiment is proposed, in which the hydrophones are arranged in a uniform array.

[0034] Based on the same inventive concept, the present invention also proposes an underwater leakage sound recognition system based on deep learning, the system comprising:

[0035] A signal acquisition unit, used to collect underwater sound signals in different environments;

[0036] A feature extraction unit, used to extract effective interval features of sound signals in a normal environment and a leakage environment respectively;

[0037] A deep learning network model building unit, used to build a deep learning network model;

[0038] A training unit, used for training the effective interval features of the extracted signal, inputting the trained signal into the deep learning network model, and obtaining a sound recognition model;

[0039] A judgment unit, used to judge whether there is leakage according to the recognition model, and directly display the result if there is no leakage;

[0040] The penetration point acquisition unit is used to determine the location of the leakage point through sound source positioning technology if there is a leakage.

[0041] Based on the same inventive concept, the present invention also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes an underwater leakage sound recognition method based on deep learning as described in any one of the above items.

[0042] Based on the same inventive concept, the present invention also proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of an underwater leakage sound recognition method based on deep learning as described above are executed.

[0043] The present invention is beneficial in that:

[0044] The present invention proposes a method for underwater leakage sound recognition based on deep learning, which overcomes the problem that the traditional hydrophone method usually requires manual judgment of the collected underwater sound signals, which is easily affected by human factors and leads to inaccurate judgment or slow response. It proposes a technical concept different from the existing technology, which automatically identifies the leakage characteristics in the underwater sound signals through the deep learning model, reducing the reliance on manual judgment. The deep learning network can efficiently extract potential leakage information from a large amount of sound data, significantly improving the recognition accuracy and response speed.

[0045] The traditional electrode method measurement process is usually cumbersome, requires multiple complex steps, and takes a long time to measure, which is inefficient. In the present invention, deep learning technology is used to process and analyze underwater sound signals in real time, avoiding the long measurement process required by the traditional electrode method. The deep learning model can quickly extract effective information from the collected sound signals, thereby quickly making leakage identification judgments.

[0046] At the same time, in traditional methods, the identification of leakage usually relies on the experience of experts and manual analysis, which has certain errors and delays, and the location of leakage requires additional equipment and technology. In the present invention, through the sound recognition model based on deep learning, it is possible to automatically determine whether leakage exists when it occurs, and accurately determine the location of the leakage point using sound source localization technology. This automated processing method not only improves efficiency, but also significantly reduces manual errors, and improves the accuracy and reliability of leakage detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of an underwater leakage sound recognition method based on deep learning according to implementation mode 1;

[0048] Figure 2 This is a flowchart of deep learning described in Implementation Method 4;

[0049] Figure 3 This is a diagram of the deep learning network structure described in Implementation Method 4.

[0050] Figure 4 A flow chart of the sound source localization technology for determining the leakage point position described in Embodiment 6;

[0051] Figure 5 This is a schematic diagram of the hydrophone arrangement described in Implementation Example 7. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0053] Implementation method 1, see Figure 1 The present embodiment describes a method for underwater leakage sound recognition based on deep learning, the method comprising:

[0054] S1: Collect underwater sound signals in different environments;

[0055] S2: Extract the effective interval features of the sound signal in normal environment and leakage environment respectively;

[0056] S3: Build a deep learning network model;

[0057] S4: training the effective interval features of the extracted signal, inputting the trained signal into the deep learning network model, and obtaining a sound recognition model;

[0058] S5: Determine whether there is leakage according to the recognition model, and directly display the result if there is no leakage;

[0059] S6: If leakage exists, determine the location of the leakage point through sound source localization technology.

[0060] The present embodiment proposes a method for underwater leakage sound recognition based on deep learning, which overcomes the problem that the traditional hydrophone method usually requires manual judgment of the collected underwater sound signals, which is easily affected by human factors and leads to inaccurate judgment or slow response. A different technical concept is proposed, which automatically identifies leakage features in underwater sound signals through a deep learning model, reducing the reliance on manual judgment. The deep learning network can efficiently extract potential leakage information from a large amount of sound data, significantly improving the recognition accuracy and response speed.

[0061] The traditional electrode method measurement process is usually cumbersome, requires multiple complex steps, and takes a long time to measure, which is inefficient. In this embodiment, deep learning technology can be used to process and analyze underwater sound signals in real time, avoiding the long measurement process required by the traditional electrode method. The deep learning model can quickly extract effective information from the collected sound signals, thereby quickly making leakage identification judgments.

[0062] At the same time, in traditional methods, the identification of leakage usually relies on the experience of experts and manual analysis, which has certain errors and delays, and the location of leakage requires additional equipment and technology. In this embodiment, the sound recognition model based on deep learning can automatically determine whether leakage exists when it occurs, and the location of the leakage point can be accurately determined using sound source localization technology. This automated processing method not only improves efficiency, but also significantly reduces manual errors and improves the accuracy and reliability of leakage detection.

[0063] Implementation method 2. This implementation method is a further limitation of the underwater leakage sound recognition method based on deep learning described in implementation method 1. Step S1 includes: collecting underwater sound signals in different environments (such as underwater environments of pools and dams without leakage, underwater environments of pools and dams with leakage, and underwater environments with different shapes and sizes of leakage points) through underwater acoustic sensors, pre-processing the sound signals through sensor pre-amplification and filtering circuits or external amplifiers and filters, and transmitting them to the host computer.

[0064] Embodiment 3: This embodiment further limits the underwater leakage sound recognition method based on deep learning described in embodiment 1. Step S2 includes:

[0065] S21: Perform short-time Fourier transform on the sound signal to convert the audio signal from the time domain to the frequency domain to obtain the acoustic characteristics of the sound signal;

[0066] S21: subtracting a short-time Fourier transform result of a normal underwater sound signal from a short-time Fourier transform result of an underwater sound signal with leakage to obtain a characteristic value m;

[0067] S23: Perform statistical analysis on the feature m of the normal state, and calculate the mean μ and standard deviation σ of its distribution. Set the threshold t and determine the initial threshold. The initial threshold can be set to t=μ+kσ, where k is a multiple used to adjust the sensitivity of the detection. Generally, k=3 is selected. Under normal conditions, the probability of m exceeding this value is extremely low. Then, search for an interval greater than the threshold t in the feature value m;

[0068] S24: Selecting acoustic features in the interval as valid interval features.

[0069] In step S21 of this embodiment, short-time Fourier transform (STFT) is used to convert the sound signal from the time domain to the frequency domain, with the purpose of capturing the frequency components and time changes of the underwater sound signal in detail and extracting valuable acoustic features. The sound features produced by underwater leakage are usually relatively weak, and these slight differences can be revealed by frequency domain analysis in this embodiment, so it has great advantages for the sound recognition of leakage.

[0070] In step S22, the Fourier transform result of the normal underwater sound signal is differentiated from the Fourier transform result of the underwater sound signal with leakage to obtain the characteristic value m. This differentiation process helps to highlight the significant difference between the leakage sound and the normal sound, thereby effectively capturing the characteristics of the leakage signal and improving the sensitivity of recognition. By comparing normal and abnormal signals, the interference of environmental noise can be removed to a certain extent, and the detection accuracy of the model for leakage signals can be improved.

[0071] In step S23, by setting a threshold t and searching for intervals where the characteristic value m is greater than the threshold t, significant features containing leakage information can be effectively identified. This threshold-based method can flexibly adjust the detection criteria according to the sound characteristics in different environments, thereby enhancing the adaptability and robustness of the method.

[0072] In step S24, by selecting the acoustic features in the interval as the effective interval features, the focus of detection is further refined, reducing the impact of irrelevant or unimportant signals on the recognition results, ensuring that the recognition process is more efficient and avoiding interference from excessive irrelevant data.

[0073] The noise in the actual underwater environment is complex and varied, and traditional signal processing methods are difficult to accurately distinguish leakage sounds from other environmental noises. However, through the above methods, especially the combination of short-time Fourier transform and eigenvalue differentiation, the model can still maintain high robustness and stability when facing various environmental changes. The differentiated feature extraction step effectively suppresses the impact of noise on the model, further enhancing its performance in practical applications. Through threshold screening and effective interval extraction, a comprehensive analysis of all data is avoided, allowing the model to focus on the most relevant features and improve recognition speed. This is of great significance for real-time monitoring of underwater leakage. By combining the eigenvalue m with the threshold t for screening, redundant calculation processes are avoided, and the execution efficiency of the algorithm in real-time applications is improved.

[0074] Implementation method 4: See Figure 2 and Figure 3 This embodiment further defines the underwater leakage sound recognition method based on deep learning described in the first embodiment. A deep learning model is designed and optimized for complex underwater environments. The model takes into account the non-stationarity, noise interference and time dependence of underwater signals, and has the advantages of strong anti-noise ability and high robustness. In step S3, the deep learning network model includes a plurality of convolutional layers and pooling layers stacked, wherein the convolutional layer is used to extract local features, and the pooling layer is used for downsampling.

[0075] The convolutional layers include: using a (3, 3) convolution kernel to capture local features in both frequency and time dimensions; the first convolution layer starts with 1 input channel, and the subsequent convolution layers increase the number of channels, gradually extracting from low-level features to high-level features; using ReLU as an activation function to accelerate convergence and alleviate the gradient vanishing problem; and performing batch normalization after each convolution layer to reduce internal covariate shift, stabilize the training process, and accelerate convergence:

[0076] ReLU(x)=max(0,x)

[0077] Each convolutional layer is followed by a pooling layer, using maximum pooling with a pooling window size of (2, 2) to reduce the size of the feature map in time and frequency dimensions while retaining important features;

[0078] Using softmax as the activation function, the output value is converted into a probability distribution:

[0079]

[0080] The convolution layer in the deep learning network model of this embodiment uses a convolution kernel of size (3, 3), which can effectively capture local features in both frequency and time dimensions. This is particularly important for underwater sound signal recognition, because the sound of underwater leakage usually has complex patterns in time and frequency, and the convolution kernel can effectively capture these local features, thereby improving the accuracy of recognition. The convolution layer gradually extracts from low-level features to high-level features, allowing the network to automatically learn features at different levels. From primary low-level features such as edges and textures to subsequent more complex sound patterns and structures, the recognition ability of the model is improved.

[0081] The pooling layer performs downsampling after each convolutional layer, which helps reduce the dimension of the features, thereby reducing computational complexity and effectively preventing overfitting. The pooling layer can also enhance the translation invariance of the model (i.e., reduce sensitivity to position), thereby improving the robustness of the model in different underwater environments. By reducing the spatial size of the feature map, the pooling layer makes the features processed by the subsequent network layers more compact and abstract, thereby better capturing the high-level features of the sound signal.

[0082] Adding batch normalization after each convolutional layer can effectively solve the problem of gradient vanishing or gradient exploding during training, thereby accelerating the convergence process of the model. Batch normalization can maintain the stability of the activation value of each layer by normalizing the input of each layer, thereby improving the stability of training and reducing the risk of overfitting.

[0083] The sound of underwater leakage is often disturbed by various noises, such as water flow, equipment noise, etc. Deep learning networks, especially convolutional neural networks, can extract useful signal features from noise when there is sufficient data. Through hierarchical feature extraction and pooling, the model can effectively ignore the noise and focus on the key signals of the leakage sound.

[0084] Embodiment 5: This embodiment further limits the underwater leakage sound recognition method based on deep learning described in embodiment 1. Step S4 includes:

[0085] The sound data samples are divided into three parts: training set, validation set and test set, with a ratio of 8:1:1, and a ten-fold cross validation is performed; the model is fitted with the sound data samples on the training set, and whether the model meets the required standard is judged by whether the recognition rate reaches the set threshold; if the required standard is not met, the model will return to continue learning; if the required standard is met, the deep learning network model will be verified through the validation set;

[0086] The validation set adjusts the model's hyperparameters and performs a preliminary evaluation of the model's capabilities. It also determines whether the model meets the required standards based on whether the recognition rate reaches the set threshold.

[0087] If the threshold requirement is not met, the program will return to continue learning. If the threshold requirement is met, the program will be tested.

[0088] The test set is used to evaluate the generalization ability of the deep learning network model. If the generalization ability reaches the preset threshold, the training ends, otherwise it returns to retraining.

[0089] In this embodiment, by separating the training, validation, and test sets, the embodiment can carefully optimize the model at different stages. In the training stage, the model is initially trained by fitting on the training set, the validation set is used for hyperparameter adjustment and preliminary evaluation, and the test set is used to evaluate the generalization ability of the model. This step-by-step evaluation process helps the system efficiently discover and correct potential problems in the model. During the training and validation stages, if the recognition rate does not reach the preset threshold, the model will return to continue learning, avoiding premature early stopping and overfitting. This mechanism ensures that the model can end training when the target performance is reached, thereby obtaining the best model effect.

[0090] Implementation Method 6: See Figure 4 This embodiment further defines the underwater leakage sound recognition method based on deep learning described in the first embodiment, and step S6 includes:

[0091] Use multiple hydrophones to form an array to measure underwater signals;

[0092] Compare the arrival times of signals from different hydrophones and calculate the time difference of signal arrival;

[0093] Based on the arrival time difference results, the location of the sound source of the leakage point is derived through trigonometric geometry.

[0094] In this embodiment, by using multiple hydrophones to form an array, underwater sound signals at multiple locations can be captured simultaneously, and then the time difference between the signals reaching different hydrophones (i.e., "arrival time difference") can be compared. This method uses the time difference of sound wave propagation to infer the specific location of the leakage point, and uses trigonometric geometry for calculation, so as to achieve accurate sound source positioning. Compared with the traditional single hydrophone positioning method, multi-point signals provide more information and help reduce positioning errors.

[0095] There are often complex sound wave propagation conditions in underwater environments, such as the influence of water flow, temperature, salinity and other factors on the speed of sound wave propagation. By using multiple hydrophones and comparing the arrival time difference of their signals, this method can adapt to different underwater environments, taking into account the propagation differences between different hydrophones, thereby improving the accuracy of the leak location. This multi-point data fusion method can effectively reduce the impact of the external environment on positioning accuracy.

[0096] The underwater leakage sound recognition method combined with deep learning can use machine learning technology to deeply analyze underwater signals, quickly extract audio signals with leakage characteristics and classify them. This not only improves the accuracy of leakage detection, but also optimizes the sound recognition model by training a large amount of historical data, making the positioning process more intelligent and automated.

[0097] Implementation method 7: See Figure 5 This embodiment is a further limitation of the underwater leakage sound recognition method based on deep learning described in Embodiment 6, wherein the hydrophones are arranged in a uniform array.

[0098] In combination with the underwater environment and actual operation, the hydrophones are arranged in a uniform array. This arrangement can achieve uniform spatial coverage on a two-dimensional plane, avoiding blind spots in spatial sampling, thereby improving the comprehensiveness and consistency of signal acquisition. At the same time, it can effectively improve the accuracy of sound source positioning, especially in three-dimensional space, the estimation of azimuth and elevation angles is more accurate, and it has good beamforming capabilities. By adjusting the beam pointing, the signal in a specific direction can be enhanced, while suppressing interference and noise in other directions. In addition, because the array has a larger spatial dimension, it can provide higher spatial resolution, thereby effectively separating multiple sound sources that are close to each other. Faced with a complex underwater environment, the signal acquisition of the uniform array is more uniform, reducing the impact of environmental noise and the complexity of the sound propagation path on detection.

[0099] Embodiment 8: The underwater leakage sound recognition system based on deep learning described in this embodiment comprises:

[0100] A signal acquisition unit, used to collect underwater sound signals in different environments;

[0101] A feature extraction unit, used to extract effective interval features of sound signals in a normal environment and a leakage environment respectively;

[0102] A deep learning network model building unit, used to build a deep learning network model;

[0103] A training unit, used for training the effective interval features of the extracted signal, inputting the trained signal into the deep learning network model, and obtaining a sound recognition model;

[0104] A judgment unit, used to judge whether there is leakage according to the recognition model, and directly display the result if there is no leakage;

[0105] The penetration point acquisition unit is used to determine the location of the leakage point through sound source positioning technology if there is a leakage.

[0106] Embodiment 9. A computer device described in this embodiment includes a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes an underwater leakage sound recognition method based on deep learning as described in any one of embodiments 1 to 7.

[0107] Embodiment 10. A computer-readable storage medium described in this embodiment stores a computer program, and when the computer program is executed by a processor, the steps of a method for underwater leakage sound recognition based on deep learning as described in any one of embodiments 1 to 7 are executed.

[0108] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure rather than to limit its protection scope. Although the present disclosure has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the disclosed claims to be approved.

Claims

1. A method for underwater leakage sound recognition based on deep learning, characterized in that: The method comprises: S1: Collect underwater sound signals in different environments; S2: Extract the effective interval features of the sound signal in normal environment and leakage environment respectively; S3: Build a deep learning network model; S4: training the effective interval features of the extracted signal, inputting the trained signal into the deep learning network model, and obtaining a sound recognition model; S5: Determine whether there is leakage according to the recognition model, and directly display the result if there is no leakage; S6: If leakage exists, determine the location of the leakage point through sound source localization technology.

2. According to the method for underwater leakage sound recognition based on deep learning in claim 1, it is characterized in that: Step S1 includes: collecting underwater sound signals of different environments through an underwater acoustic sensor, preprocessing the sound signals through a sensor preamplifier and filter circuit or an external amplifier and filter, and transmitting the sound signals to a host computer.

3. The underwater leakage sound recognition method based on deep learning according to claim 1 is characterized in that: Step S2 includes: S21: Perform short-time Fourier transform on the sound signal to convert the audio signal from the time domain to the frequency domain to obtain the acoustic characteristics of the sound signal; S21: subtracting a short-time Fourier transform result of a normal underwater sound signal from a short-time Fourier transform result of an underwater sound signal with leakage to obtain a characteristic value m; S23: Set a threshold t, and search for intervals greater than the threshold t in the eigenvalue m; S24: Selecting acoustic features in the interval as valid interval features.

4. The method for underwater leakage sound recognition based on deep learning according to claim 1 is characterized in that: In step S3, the deep learning network model includes a plurality of convolutional layers and pooling layers stacked together, wherein the convolutional layers are used to extract local features, and the pooling layers are used for downsampling; The convolution layer includes: using a (3, 3) convolution kernel to capture local features in both frequency and time dimensions; the first convolution layer starts with 1 input channel, and the subsequent convolution layers increase the number of channels, gradually extracting from low-level features to high-level features; using ReLU as an activation function; and performing batch normalization after each convolution layer; Each convolutional layer is followed by a pooling layer, and the pooling window size is (2, 2); Softmax is used as the activation function to convert the output value into a probability distribution.

5. The underwater leakage sound recognition method based on deep learning according to claim 1 is characterized in that: Step S4 includes: The sound data samples were divided into three parts: training set, validation set and test set, with a ratio of 8:1:1, and ten-fold cross validation was performed; The model is fitted to the sound data samples on the training set, and whether the model meets the required standards is determined by whether the recognition rate reaches the set threshold; If the required standard is not met, the learning will be continued. If the required standard is met, the deep learning network model will be verified through the validation set; The validation set adjusts the model's hyperparameters and performs a preliminary evaluation of the model's capabilities. It also determines whether the model meets the required standards based on whether the recognition rate reaches the set threshold. If the threshold requirement is not met, the program will return to continue learning. If the threshold requirement is met, the program will be tested. The test set is used to evaluate the generalization ability of the deep learning network model. If the generalization ability reaches the preset threshold, the training ends, otherwise it returns to retraining.

6. The method for underwater leakage sound recognition based on deep learning according to claim 1, characterized in that: Step S6 includes: Use multiple hydrophones to form an array to measure underwater signals; Compare the arrival times of signals from different hydrophones and calculate the time difference of signal arrival; Based on the arrival time difference results, the location of the sound source of the leakage point is derived through trigonometric geometry.

7. The method for underwater leakage sound recognition based on deep learning according to claim 6 is characterized in that: The hydrophones are arranged in a uniform array.

8. An underwater leakage sound recognition system based on deep learning, characterized in that: The system comprises: A signal acquisition unit, used to collect underwater sound signals in different environments; A feature extraction unit, used to extract effective interval features of sound signals in a normal environment and a leakage environment respectively; A deep learning network model building unit, used to build a deep learning network model; A training unit, used for training the effective interval features of the extracted signal, inputting the trained signal into the deep learning network model, and obtaining a sound recognition model; A judgment unit, used to judge whether there is leakage according to the recognition model, and directly display the result if there is no leakage; The penetration point acquisition unit is used to determine the location of the leakage point through sound source positioning technology if there is a leakage.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes an underwater leakage sound recognition method based on deep learning according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of a method for underwater leakage sound recognition based on deep learning as described in any one of claims 1 to 7.

Images