A water supply network sound leakage detection method based on an enhanced semi-supervised model

Abstract

The application discloses a kind of water supply network sound leakage detection methods based on enhanced semi-supervised model, comprising: step 1, obtain data set, and weakly enhance operation and strong enhancement operation are carried out to data;Step 2, based on semi-supervised network framework, build prediction network, the prediction network includes the marked learning branch and the unmarked learning branch of weight sharing, with the image sample of label in weakly enhanced sample set is input into the marked learning branch to output corresponding prediction distribution;Step 3, using weakly enhanced sample set and enhanced sample set, the prediction network is trained to obtain pipe network sound leakage monitoring model.The method provided in the application can guarantee the recognition accuracy and robustness of sound leakage detection model under the condition of limited labeled sample.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of water supply network leakage detection, and particularly relates to a water supply network sound leakage detection method based on an enhanced semi-supervised model. BACKGROUND

[0002] Water supply pipeline systems are an important part of urban infrastructure, and ensuring their efficient and reliable operation is crucial for public health and economic development. However, as the urbanization process continues to advance, the scale and complexity of water supply pipeline networks are increasing, and pipeline leakage problems have become increasingly serious. Pipeline leakage not only wastes valuable water resources, but also can cause serious environmental pollution and economic losses. Therefore, developing efficient, reliable, and field-adapted leakage detection methods is of great significance to ensure the stable operation of water supply systems.

[0003] In recent years, leakage detection technology based on vibration sound signals has received widespread attention in academia and practical applications due to its high sensitivity and non-invasive characteristics. Vibration sound signals can reflect subtle changes in the state of water flow in the pipeline, thus producing obvious characteristics when leakage occurs. However, relying solely on manually labeled vibration sound signal data for fully supervised leakage detection model development is often very costly, and most single water companies are unable to obtain sufficient labeled data. In addition, the collection of a large amount of unlabeled sound signal data in the field is relatively easy, but it has not been effectively utilized.

[0004] Academic literature Novel Leakage Detection and Localization Method Based on LineSpectrum Pair and Cubic Interpolation Search[J]. WATER RESOURCES MANAGEMENT, 2020, 34(12): 3895-911. proposes a new leakage detection and localization method based on line spectrum pair and random forest, which has high precision and strong noise resistance, but may have high computational complexity in complex pipeline networks.

[0005] Patent document CN118757697A discloses a water supply network leakage intelligent detection system and method, wherein the system part includes an ultrasonic module, a preprocessing module, an extraction module, and an identification module. The ultrasonic module is used to emit ultrasonic waves to the water supply network and collect waveform data of the water supply network. The preprocessing module is used to preprocess the waveform data to obtain processed data. The extraction module is used to extract features from the processed data to obtain sound wave features. The identification module is used to perform cluster analysis on the sound wave features to identify the leakage of the water supply network. SUMMARY

[0006] The application aims to provide a water supply network sound leakage detection method based on an enhanced semi-supervised model, which can ensure the recognition accuracy and robustness of the sound leakage detection model under the condition of limited labeled samples.

[0007] In order to achieve the purpose of the application, the following technical scheme is provided: a water supply network sound leakage detection method based on an enhanced semi-supervised model, comprising the following steps:

[0008] Step 1: obtaining the field sound signal of the water supply network and converting it into the corresponding log mel spectrum graph through a filter, selecting part of the log mel spectrum graph to label whether there is a pipe network leakage, so as to obtain image samples with labels and image samples without labels;

[0009] Weak enhancement operation is performed on all image samples to obtain a corresponding weak enhancement sample set;

[0010] Strong enhancement operation is performed on the image samples without labels to obtain a corresponding enhanced sample set;

[0011] Step 2: constructing a prediction network based on a semi-supervised network framework, the prediction network comprising a labeled learning branch and an unlabeled learning branch sharing weights, inputting the image samples with labels in the weak enhancement sample set into the labeled learning branch to output the corresponding prediction distribution;

[0012] The unlabeled image samples in the weak enhancement sample set and the enhanced sample set are respectively input into the unlabeled learning branch to output the first prediction result corresponding to the unlabeled image samples in the weak enhancement sample set and the second prediction result corresponding to the enhanced sample set, the unlabeled image samples corresponding to the prediction results with confidence higher than the adaptive threshold in the first prediction result are given pseudo labels, and the unlabeled image samples with pseudo labels are added to the labeled learning branch to update the prediction distribution, and the adaptive threshold is gradually reduced by a fixed value according to the number of updates;

[0013] Step 3: training the prediction network using the weak enhancement sample set and the enhanced sample set to obtain a pipe network sound leakage monitoring model.

[0014] The application fully utilizes unlabeled data and acoustic data enhancement, and is based on a semi-supervised network framework to reduce the dependence on labeled data, so that the accuracy and reliability of sound leakage detection can still be ensured under the condition of limited labeled data.

[0015] Specifically, the weak enhancement operation comprises one or more of brightness adjustment, horizontal flip or contrast adjustment on the image samples.

[0016] Specifically, the strong enhancement operation adjusts the image sample in time domain and frequency domain by SpecAugment.

[0017] Specifically, the adjustment includes one or more of time warping based on time domain, time masking based on time domain or frequency masking based on frequency domain.

[0018] Specifically, the prediction network adopts a three-layer residual convolutional neural network, and a residual block expression in the three-layer residual convolutional neural network is as follows:

[0019]

[0020] Y n =X n +F(X n ,W n )

[0021]

[0022] Wherein, X n and X n+1 are the input and output of the residual block respectively, F is a residual function, f is a ReLU activation function, X N represents the learned features in the residual block, W i is a weight coefficient vector of the residual block.

[0023] Specifically, the calculation formula of the pseudo label is as follows:

[0024]

[0025] Wherein, I is an index function; q=p(y|x) is a probability vector predicted by the model; H is a cross-entropy function; pse=argmax(q) is a pseudo label.

[0026] Specifically, in the training process, a consistency regularization method is used for the unlabeled learning branch to ensure that the first prediction result and the second prediction result are consistent in form, and the consistency regularization method is only executed when the confidence of the prediction result is greater than an adaptive threshold.

[0027] Specifically, the comprehensive loss function is used to jointly guide and train the labeled learning branch and the unlabeled learning branch.

[0028] Specifically, the expression of the comprehensive loss function is as follows:

[0029] L=L S +λ u ×L u

[0030]

[0031] wherein, L S is the loss function of the labeled learning branch; L u is the loss function of the unlabeled learning branch; lambda u is the weight coefficient, m is the data amount of a batch, y i is the true label, f(x i ) is the predicted value, k is the number of categories; pse ij is the pseudo label after weak enhancement; is the probability vector after strong enhancement.

[0032] Compared with the prior art, the beneficial effects of the present application are:

[0033] Through data weak enhancement and strong enhancement operations on the data set containing a small amount of labels, and model training based on the semi-supervised model framework, the ESSM can utilize the unlabeled acoustic data, significantly improve the accuracy and reliability of acoustic leakage detection in the case of limited labeled data, and better support the optimal operation and reliability management of the water supply network. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The flowchart of the water supply network acoustic leakage detection method based on the enhanced semi-supervised model provided in the present embodiment;

[0035] Figure 2 The log mel spectrum after weak enhancement provided in the present embodiment;

[0036] Figure 3 The log mel spectrum before and after strong enhancement provided in the present embodiment;

[0037] Figure 4 The structural schematic diagram of the semi-supervised network framework provided in the present embodiment;

[0038] Figure 5 The schematic diagram of the residual convolutional neural network provided in the present embodiment;

[0039] Figure 6 The spectrum after data enhancement of different sizes of Cutout provided in the present embodiment;

[0040] Figure 7 The influence of different data enhancement methods on the semi-supervised network framework provided in the present embodiment. DETAILED DESCRIPTION

[0041] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0042] The embodiment provides a water supply network acoustic leakage detection method based on an enhanced semi-supervised model, which simultaneously considers reducing label data dependence, fully utilizing unlabelled data and acoustic data enhancement. By effectively utilizing unlabelled acoustic data, the ESSM can significantly improve the accuracy and reliability of acoustic leakage detection even under the condition of limited labelled data, and can even exceed the fully supervised model, and can better support the optimized operation and reliability management of the water supply network.

[0043] As shown in Figure 1 , the specific steps of the water supply network acoustic leakage detection method based on the enhanced semi-supervised model provided by the embodiment are as follows:

[0044] Step 1, acquiring the field acoustic signals of the water supply network and converting them into corresponding log mel spectrograms through a filter, selecting part of the log mel spectrograms to label whether there is a pipe network leakage, so as to obtain image samples with labels and image samples without labels;

[0045] Performing weak enhancement operation on all image samples to obtain a corresponding weak enhancement sample set;

[0046] Performing strong enhancement operation on the image samples without labels to obtain a corresponding enhanced sample set.

[0047] More specifically, the acoustic signal data set in the embodiment comes from the water supply networks of multiple southwestern cities in China, and includes 17602 audio signals, including 8801 leakage signals and 8801 non-leakage signals, and each signal has a duration of 1s. Since each audio data is manually labelled, there will be some unavoidable errors. Each acoustic signal sample is converted into a log mel spectrogram through a filter bank to facilitate the basic model to identify its frequency domain features, and the parameters are shown in Table 1. In order to meet the training needs of the semi-supervised model, 90% (including labelled and unlabelled) audio signals are selected for training of the semi-supervised model, and 10% (labelled) data are used as a test set to verify the performance of the trained semi-supervised model.

[0048] Table 1

[0049]

[0050] In view of the fact that labeled acoustic signals are usually scarce and costly in water supply networks, while unlabeled acoustic signal data is relatively abundant but not fully utilized, data augmentation helps to bridge the gap between the two. Weak augmentation usually makes subtle perturbations to the input spectrogram, but does not destroy its basic time-frequency features.

[0051] As shown in FIG. 1, a weak augmentation operation demonstration diagram provided by the embodiment is shown, wherein Figure 2 a in FIG. 1 is a schematic diagram of brightness adjustment, Figure 2 b in FIG. 1 is a schematic diagram of horizontal flip, Figure 2 c in FIG. 1 is a schematic diagram of contrast adjustment. Figure 2

[0052] The above three data augmentation methods will not bring too much interference to the time-frequency features of the log-mel spectrogram.

[0053] As shown in FIG. 2, a strong augmentation operation demonstration diagram provided by the embodiment is shown, wherein Figure 3 a in FIG. 2 is an original log-mel spectrogram, Figure 3 b in FIG. 2 is a schematic diagram of time warping, Figure 3 c in FIG. 2 is a schematic diagram of time masking, Figure 3 d in FIG. 2 is a schematic diagram of frequency masking, Figure 3 e in FIG. 2 is a schematic diagram of superimposed use of the above three methods. Figure 3

[0054] As shown in FIG. 3, a strong augmentation operation demonstration diagram provided by the embodiment is shown, wherein Figure 4 ​​As shown in the semi-supervised network framework provided by the embodiment, the embodiment builds a prediction network using a three-layer residual convolutional neural network, which includes a labeled learning branch and an unlabeled learning branch. This dual-path strategy ensures effective use of the rich information contained in the labeled and unlabeled acoustic signals on site. First, a large number of field acoustic signals are collected, and after data processing, part of the signals are manually annotated by experts. Second, in the labeled learning branch, we apply weak augmentation technology to generate weak augmented samples from the labeled acoustic signal dataset. Then it is input into the base model to generate the prediction distribution. A key step in this process involves calculating and optimizing the supervised loss between the model prediction and the true label to ensure its effectiveness on labeled data. Third, the unlabeled learning branch is used to process unlabeled acoustic signal samples. Since there is no real label, each spectrum is simultaneously weakly augmented and strongly augmented. Subsequently, it is input into the base model to produce two different predictions for computing consistency regularization loss (unsupervised loss). Fourth, in the unlabeled branch, to improve the quality and reliability of the generated pseudo-labels, an adaptive threshold strategy for dynamically assigning pseudo-labels is introduced, which mainly judges the prediction results after weak augmentation, and can more accurately utilize a large amount of unlabeled acoustic signal data. Fifth, the two training branches are jointly guided by the total loss function for training, reducing the overfitting of the prediction network to a small amount of labeled data, thereby improving the generalization ability and overall performance of the final model.

[0055] As Figure 5 shown, the residual convolutional neural network provided by the embodiment can be represented as follows:

[0056] Y n = X n + F(X n , W n ) (1)

[0057] X n+1 = f(Y n ) (2)

[0058] where X n and X n+1 are the input and output of the residual block, F is the residual function, and f is the ReLU activation function. From (1) and (2), equation (3) can be obtained to obtain the learned features in the residual block.

[0059]

[0060] In the labeled learning branch, the labeled log-mel spectrogram is also input to the base model after weak enhancement for training, mainly to reduce overfitting in the supervised learning process. In addition, for the leakage detection problem, the loss function of the supervised part is obtained by calculating the cross-entropy loss during training, which can be expressed as equation (4).

[0061]

[0062] where L s is the supervised loss, m is the data amount of a batch, y i is the true label, and f(x i ) is the predicted value of the base model.

[0063] In the unlabeled learning branch, a small amount of labeled acoustic signals in the field can provide some effective information for the base model during training, which enables the model to have a preliminary cognitive pattern of acoustic signals. When using the trained model to predict unlabeled data in the unsupervised branch, high-confidence prediction results can be used as pseudo-labels, thereby converting unlabeled data into part of labeled data. This can increase additional information during training and improve the performance and generalization ability of the model. The pseudo-label calculation formula is shown in equation (5).

[0064]

[0065] where I is the indicator function; q = p(y|x) is the probability vector predicted by the model; H is the cross-entropy function; and pse = argmax(q) is the pseudo-label.

[0066] Traditional semi-supervised models set a fixed threshold when generating pseudo-labels, which cannot be adjusted according to changes in training data and may not perform well in different data sets or different training stages. In this work, the threshold is dynamically adjusted based on the training progress. First, a high initial threshold is set to ensure the accuracy of the base model in judging pseudo-labels in the early stage of training. Then, as the base model improves, the threshold gradually decreases, allowing more unlabeled acoustic signal data to be used for leakage detection.

[0067] The consistency regularization strategy adopted in this embodiment aims to ensure that the base model can maintain consistent prediction results after introducing different perturbations (weak enhancement and strong enhancement) to the unlabeled acoustic signals. Through consistency regularization, a large amount of unlabeled acoustic signal data can be effectively utilized to improve the generalization ability and stability of the semi-supervised acoustic leakage detection model. The consistency regularization loss calculation formula (6) in the unsupervised learning branch is as follows:

[0068]

[0069] where L uis the consistency regularization loss (i.e., unsupervised loss) for the unsupervised branch; m is the total number of samples; k is the number of classes; pse ij is the pseudo label after weak augmentation; is the probability vector predicted by the base model after strong augmentation.

[0070] It is worth noting that not every sample after weak and strong augmentation will calculate the unsupervised loss. Only when the confidence of the sample after weak augmentation is predicted by the base model is greater than the set adaptive threshold, the consistency regularization loss (i.e., unsupervised loss) will be calculated.

[0071] By jointly training the labeled learning branch and the unlabeled branch, the potential information in the limited labeled acoustic signals and unlabeled acoustic signals is more effectively utilized. As can be seen from equations (4) and (6), the total optimization objective function of the ESSM is shown in equation (7).

[0072] L = L S + λ u × L u (7)

[0073] where L S is the loss function of the labeled learning branch; L u is the loss function of the unlabeled learning branch; λ u is the weight coefficient.

[0074] To prove the superiority of the proposed pipe network acoustic leakage monitoring model (ESSM model) trained in this embodiment, the commonly used semi-supervised model and the fully supervised model are compared under the condition of the same annotation rate. As can be seen from Table 2, the fully supervised model shows extremely high accuracy and Specificity when using 100% labeled data, reaching 98.30% and 98.61%, respectively. This shows that with the support of sufficient labeled data, fully supervised learning can maximize the performance of the model. However, when the proportion of labeled data is reduced to 20% or less, the performance of the fully supervised model decreases significantly. Especially when there is only 5% labeled data, the accuracy and Specificity decrease to 88.80% and 87.83%, respectively. In contrast, the semi-supervised models such as ESSM and Fixmatch perform more excellently in the case of less labeled data, and the model deviation in multiple cross-validation is smaller, proving the application potential and advantage of semi-supervised models compared with fully supervised models in acoustic leakage detection. Especially the ESSM, its performance is close to or even exceeds the performance of the fully supervised model under the condition of high labeled data.

[0075] ESSM shows excellent performance at 5%, 10%, and 20% labeled data ratio, outperforming other semi-supervised models. Especially at 20% label rate, its accuracy, precision, F1-score, sensitivity, and specificity all reach the highest, respectively 98.45%, 98.46%, 98.45%, 98.32%, and 98.59%. In addition, when the label rate is 20%, the average deviation of each evaluation index of ESSM is the smallest, which further illustrates its strong stability in leakage detection.

[0076] Fixmatch also performs well at 10% and 20% labeled data ratio, especially in precision and F1-score. This is due to its use of strong and weak data augmentation and consistency regularization, effectively utilizing unlabeled data to improve model performance. Mean Teacher performs close to Fixmatch, especially at 20% label rate. SimCLR performs slightly worse at low labeled data ratio, especially at 5% labeled data, with an accuracy and sensitivity of only 92.16% and 92.01%. This may be due to SimCLR's reliance on contrastive learning, which limits the learning efficiency of feature representation when there is less labeled data.

[0077] Table 2 Comparison of identification results of each model

[0078]

[0079] To demonstrate the superiority of the data augmentation method SpecAugment in the ESSM model, the Cutout method and the weak augmentation method (i.e., replacing the strong augmentation of ESSM with weak augmentation) are compared and analyzed.

[0080] As shown in Figure 6 , the results of log Mel spectrogram after different size Cutout augmentation, where Figure 6 a is Cutout-10, Figure 6 b is Cutout-15, Figure 6 c is Cutout-20. As the size of Cutout increases, the area of the log Mel spectrogram that is masked also increases, and the time-frequency features lost after data augmentation also increase.

[0081] As shown in Figure 7The influence of different data enhancement methods on the ESSM recognition pipeline leakage under the three labeling rates is as follows: in the case of 5% labeled data, the accuracy (94.24%), precision (94.26%), F1-score (94.12%), sensitivity (94.80%), and specificity (93.68%) of SpecAugment are all higher than those of Cutout and weak enhancement. This shows that when the labeled data is less, SpecAugment has the most significant effect and can better improve the various indicators of the model. SpecAugment enhances the diversity of data through time domain masking, frequency domain masking, and time warping techniques, which not only enables the model to learn features at different times and frequencies, but also enhances the model's robustness to time series and frequency changes, and more comprehensively improves the model's generalization ability.

[0082] In contrast, the masking technique in the Cutout method mainly randomly masks certain areas of the input image or feature map, with a masking size of (10, 15, 20). Although Cutout increases data diversity by masking part of the image area, due to the fixed masking size, it cannot flexibly mask different areas in the time and frequency domains like SpecAugment, nor can it achieve time warping.

[0083] When the amount of labeled data increases to 10%, ESSM still performs best, especially in accuracy and sensitivity, leading Cutout and weak enhancement by about 1 percentage point. This shows that SpecAugment not only has a significant advantage when there is less data, but even if the labeled data increases, it can still maintain a leading position. When the amount of labeled data reaches 20%, the performance of ESSM is further improved. In contrast, although Cutout and weak enhancement have improved at this labeling ratio, they still lag behind SpecAugment in F1-score, sensitivity, and specificity.

[0084] In water supply networks, compared to labeled acoustic signal data, the cost of collecting unlabeled data is relatively small. Therefore, it is meaningful to explore the influence of the amount of unlabeled data on improving the accuracy of ESSM leakage detection while ensuring the amount of labeled data remains unchanged (10%).

[0085] Table 3

[0086]

[0087] As shown in Table 3, the performance of ESSM is constantly improved while keeping the proportion of labeled data unchanged and gradually increasing the number of unlabeled data. When the number of unlabeled data is from (0-80%), the recognition accuracy of ESSM is improved by about 5%, which shows that the unlabeled data has a significant contribution to the accuracy improvement of the semi-supervised model. In addition, when the unlabeled data is 40%, the accuracy of ESSM has improved by about 4% compared with the fully supervised model. However, when the number of unlabeled data is further increased from 40% to 80%, the indicators of ESSM are only improved by about 1%. The results can show that the number of unlabeled data is not unlimited for the accuracy improvement of ESSM. It may be that as the number of unlabeled data increases, ESSM gradually masters the distribution and time-frequency characteristics of the sound signal data, and the marginal contribution of the newly added unlabeled data to the model also gradually decreases.

[0088] In summary, the acoustic leakage detection based on deep learning plays an irreplaceable role in the optimized operation and reliability management of the water supply network. However, the limited scale of labeled sound signals in the field restricts the recognition accuracy and robustness of the acoustic leakage detection model, and a large amount of unlabeled sound signals is not effectively utilized. The present application only needs a small amount of labeled data to carry out acoustic leakage detection of the water supply network, has high generalization, accuracy and robustness, and can greatly reduce the high sound signal labeling cost of water companies and improve the efficiency of field data utilization. It has important significance for enhancing the stable and efficient operation of the water supply network and reducing resource consumption. Even when the same number of labeled sound signals is collected in the field, the semi-supervised model modeling using the present application can obtain more robust and accurate recognition results than the fully supervised model.

[0089] In addition, the terms "upper", "lower", "inner", "outer", "front", "back" are only for the purpose of description, and cannot be understood as indicating or implying relative importance. Unless otherwise specified, the relative steps, numerical expressions and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0090] Of course, the above only describes specific embodiments of the present application, and does not limit the scope of the present application. Equivalent changes or modifications made according to the structure, features and principles described in the patent application scope of the present application shall be included in the patent application scope of the present application.

[0091] Finally, it should be noted that the above-described embodiments are merely specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, but not to limit the same. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that any person skilled in the art can still modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some of the technical features, within the technical scope disclosed by the present application. The modifications, changes or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A water supply network acoustic leakage detection method based on enhanced semi-supervised model, characterized in that: The following steps are involved: Step 1: Obtain the on-site acoustic signal of the water supply network and convert it into a corresponding logarithmic Mel spectrum through a filter. Select some logarithmic Mel spectrums to label whether there is leakage in the network to obtain labeled image samples and unlabeled image samples; Perform weak enhancement operations on all image samples to obtain corresponding weak enhancement sample sets; Perform strong enhancement operations on unlabeled image samples to obtain corresponding enhanced sample sets; Step 2: Build a prediction network based on a semi-supervised network framework. The prediction network includes a weight-shared labeled learning branch and an unlabeled learning branch. Input the labeled image samples in the weakly enhanced sample set into the labeled learning branch to output the corresponding predicted distribution. The unlabeled image samples and the enhanced sample set in the weakly enhanced sample set are respectively input into the unlabeled learning branch to output a first prediction result corresponding to the unlabeled image samples in the weakly enhanced sample set and a second prediction result corresponding to the enhanced sample set, a pseudo-label is assigned to the unlabeled image samples corresponding to the prediction results in the first prediction result whose confidence is higher than the adaptive threshold, and the unlabeled image samples with the pseudo-label are added to the labeled learning branch to update the prediction distribution, and the adaptive threshold is gradually reduced by a fixed value according to the number of updates; The calculation formula of the pseudo label is as follows: ;in, is the indicator function; is the probability vector predicted by the model; is the cross entropy function; is a pseudo label; step 3, using the weak enhancement sample set and the enhancement sample set to train the prediction network to obtain a pipe network acoustic leakage monitoring model, and using a comprehensive loss function to jointly guide the training of the labeled learning branch and the unlabeled learning branch. The expression of the comprehensive loss function is as follows: ; ; ;in, is the loss function of the label learning branch; is the loss function of the unlabeled learning branch; is the weight coefficient, is the amount of data in a batch, For the real label, is the predicted value, is the number of categories; is the pseudo label after weak enhancement; is the probability vector after strong enhancement.

2. The water supply network acoustic leakage detection method based on enhanced semi-supervised model according to claim 1 is characterized in that: The weak enhancement operation includes one or more of brightness adjustment, horizontal flipping, or contrast adjustment on the image sample.

3. The water supply network acoustic leakage detection method based on enhanced semi-supervised model according to claim 1 is characterized in that: The strong enhancement operation uses SpecAugment to adjust image samples in both the time domain and the frequency domain.

4. The water supply network acoustic leakage detection method based on enhanced semi-supervised model according to claim 3 is characterized in that: The adjustment includes one or more of time warping based on the time domain, time masking based on the time domain, or frequency masking based on the frequency domain.

5. The water supply network acoustic leakage detection method based on enhanced semi-supervised model according to claim 1 is characterized in that: The prediction network adopts a three-layer residual convolutional neural network. The residual block expression in the three-layer residual convolutional neural network is as follows: ; ;in, and are the input and output of the residual block, is the residual function, represents the features learned in the residual block, is the ReLU activation function.

6. The water supply network acoustic leakage detection method based on enhanced semi-supervised model according to claim 1 is characterized in that: During the training process, a consistency regularization method is used for the unlabeled learning branch to ensure that the first prediction result and the second prediction result are consistent in form. The consistency regularization method is only executed when the confidence of the prediction result is greater than the adaptive threshold.

Images