Audio-based pipeline environment event identification method and system

Abstract

According to the pipeline environment event recognition method and system based on the audio, when a pipeline environment event occurs, pipeline environment sound is collected, and an event audio set is generated; generating a classification label of a surrounding event type for the pipeline environment sound in the event audio set, and performing feature extraction on the pipeline environment sound by using a Mel function and a Welch spectrum analysis algorithm to obtain feature data; according to the feature data and the classification labels corresponding to the feature data, training an ECAPA-TDNN model, establishing a model sample library, and performing identification and early warning of pipeline environment events by using the model sample library; the method is higher in maintainability and stability, and is better in anti-noise capability and higher in recognition precision.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline environment event recognition, and in particular to an audio-based pipeline environment event recognition method and system. Background Art

[0002] Power facilities are one of the important driving forces and guarantees for economic and social development. With the rapid development of smart grids, urban power grid transformation work continues to deepen, and the area of ​​power pipeline laying is becoming more and more extensive. With the acceleration of urbanization, urban pipelines are facing more and more complex and diverse external threats. Pipeline damage caused by municipal construction, pipeline fires, external intrusions and other events has brought immeasurable consequences to the city's public safety and the development of the local national economy.

[0003] CN112488371B is a pipeline intelligent early warning method and system based on big data, which uses optical fiber sensing technology combined with mixed Gaussian models to identify pipeline events. However, optical fiber sensing technology is difficult to deploy or maintain in complex terrains such as swamps, shallow waters or streams, and may be interfered with in some environments with electromagnetic interference.

[0004] Therefore, a technical solution is needed that can identify and warn pipeline environmental event models while having higher maintainability and stability. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an audio-based pipeline environmental event recognition method and system, which can realize pipeline environmental event model recognition and early warning while having higher maintainability and stability.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: A pipeline environment event recognition method based on audio comprises the following steps: S1. When a pipeline environment event occurs, collect pipeline environment sounds and generate an event audio set; S2, generating a classification label of a surrounding event type for the pipeline environment sound in the event audio set, and extracting features of the pipeline environment sound using a Mel function and a Welch spectrum analysis algorithm to obtain feature data; S3. According to the feature data and the corresponding classification labels, the ECAPA-TDNN model is trained to establish a model sample library, and the model sample library is used to identify and warn pipeline environmental events.

[0007] In order to solve the above technical problems, another technical solution adopted by the present invention is: An audio-based pipeline environment event recognition system, comprising: Data collection module: when a pipeline environment event occurs, it collects pipeline environment sounds and generates an event audio set; Data processing module: generating classification labels of surrounding event types for the pipeline environment sound in the event audio set, and extracting features of the pipeline environment sound using Mel function and Welch spectrum analysis algorithm to obtain feature data; Model training module: training the ECAPA-TDNN model according to the feature data and the corresponding classification labels, and establishing a model sample library; Event recognition module: uses the model sample library to identify and warn pipeline environmental events.

[0008] The beneficial effects of the present invention are as follows: an audio-based pipeline environmental event recognition method and system of the present invention utilizes audio recognition technology to recognize pipeline environmental events, and has higher maintainability and stability compared to optical fiber sensing technology. At the same time, the present invention utilizes Mel function and Welch spectrum analysis algorithm to extract features of pipeline environmental sound, and has better anti-noise capability; and the ECAPA-TDNN model commonly used in speech recognition is used for the recognition of pipeline environmental events. Compared with complex language recognition, it is applied to the recognition of environmental sounds around power pipelines of relatively simple types, and can achieve higher recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a flowchart of an audio-based pipeline environment event recognition method according to an embodiment of the present invention; Figure 2 This is a structural example diagram of an audio-based pipeline environment event recognition system according to an embodiment of the present invention; Description of labels: 1. An audio-based pipeline environment event recognition system; 2. Data acquisition module; 3. Data processing module; 4. Model training module; 5. Event recognition module. DETAILED DESCRIPTION

[0010] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0011] Please refer to Figure 1 , a pipeline environment event recognition method based on audio, comprising the steps of: S1. When a pipeline environment event occurs, collect pipeline environment sounds and generate an event audio set; S2, generating a classification label of a surrounding event type for the pipeline environment sound in the event audio set, and extracting features of the pipeline environment sound using a Mel function and a Welch spectrum analysis algorithm to obtain feature data; S3. According to the feature data and the corresponding classification labels, the ECAPA-TDNN model is trained to establish a model sample library, and the model sample library is used to identify and warn pipeline environmental events.

[0012] From the above description, it can be seen that the beneficial effects of the present invention are: an audio-based pipeline environmental event recognition method of the present invention uses audio recognition technology to identify pipeline environmental events, which has higher maintainability and stability than optical fiber sensing technology. At the same time, the present invention uses Mel function and Welch spectral analysis algorithm to extract features of pipeline environmental sounds, and has better anti-noise ability; and the ECAPA-TDNN model commonly used in speech recognition is used for the recognition of pipeline environmental events. Compared with complex language recognition, it is applied to the recognition of environmental sounds around power pipelines of relatively simple types, which can achieve higher recognition accuracy.

[0013] Furthermore, the characteristic data is an audio spectrum waveform generated using a Mel function and a Welch spectrum analysis algorithm.

[0014] From the above description, it can be seen that the use of Mel function and Welch spectrum analysis algorithm can extract the audio spectrum waveform of pipeline environmental sound as feature data, which can more effectively capture the key information in the audio signal, improve the efficiency and performance of feature representation, and enhance the robustness, recognition ability and generalization ability of the model.

[0015] Furthermore, when a pipeline environmental event occurs, the pipeline environmental sound is collected as follows: Each time a pipeline event is sensed, audio is recorded at a preset frequency and duration to obtain audio data of the pipeline event.

[0016] From the above description, it can be seen that recording audio according to the preset frequency and duration can better capture the audio data of pipeline environmental event characteristics.

[0017] Furthermore, the steps between step S1 and step S2 include: Noise reduction processing is performed on the audio data in the event audio set.

[0018] From the above description, it can be seen that before using the Mel function and Welch spectral analysis algorithm for feature extraction, denoising the audio data can further reduce the impact of noise and further improve the effectiveness of model training.

[0019] Furthermore, in step S3, the identification and early warning of pipeline environmental events using the model sample library are specifically as follows: Real-time audio is collected at different positions of the pipeline, and the model sample library is used to identify and warn pipeline environmental events based on the real-time collected audio data.

[0020] From the above description, it can be seen that the real-time collected audio data is used to identify and warn pipeline environmental events through the model sample library.

[0021] Please refer to Figure 2 , an audio-based pipeline environment event recognition system, comprising: Data collection module: when a pipeline environment event occurs, it collects pipeline environment sounds and generates an event audio set; Data processing module: generating classification labels of surrounding event types for the pipeline environment sound in the event audio set, and extracting features of the pipeline environment sound using Mel function and Welch spectrum analysis algorithm to obtain feature data; Model training module: training the ECAPA-TDNN model according to the feature data and the corresponding classification labels, and establishing a model sample library; Event recognition module: uses the model sample library to identify and warn pipeline environmental events.

[0022] The audio-based pipeline environment event recognition system of the present invention utilizes audio recognition technology to recognize pipeline environment events, and has higher maintainability and stability compared with optical fiber sensing technology. At the same time, the present invention adopts Mel function and Welch spectrum analysis algorithm to extract features of pipeline environment sound, and has better anti-noise ability; and the ECAPA-TDNN model commonly used in speech recognition is used for the recognition of pipeline environment events. Compared with complex language recognition, it is applied to the recognition of environmental sound around power pipelines of relatively simple types, and can achieve higher recognition accuracy.

[0023] Furthermore, the characteristic data is an audio spectrum waveform generated using a Mel function and a Welch spectrum analysis algorithm.

[0024] From the above description, it can be seen that the use of Mel function and Welch spectrum analysis algorithm can extract the audio spectrum waveform of pipeline environmental sound as feature data, which can more effectively capture the key information in the audio signal, improve the efficiency and performance of feature representation, and enhance the robustness, recognition ability and generalization ability of the model.

[0025] Furthermore, when a pipeline environmental event occurs, the pipeline environmental sound is collected as follows: Each time a pipeline event is sensed, audio is recorded at a preset frequency and duration to obtain audio data of the pipeline event.

[0026] From the above description, it can be seen that recording audio according to the preset frequency and duration can better capture the audio data of pipeline environmental event characteristics.

[0027] Furthermore, before the data processing module generates a classification label of the surrounding event type for the pipeline environment sound in the event audio set, the data processing module further includes the steps of: Noise reduction processing is performed on the audio data in the event audio set.

[0028] From the above description, it can be seen that before using the Mel function and Welch spectral analysis algorithm for feature extraction, denoising the audio data can further reduce the impact of noise and further improve the effectiveness of model training.

[0029] Furthermore, the identification and early warning of pipeline environmental events using the model sample library are specifically as follows: Real-time audio is collected at different positions of the pipeline, and the model sample library is used to identify and warn pipeline environmental events based on the real-time collected audio data.

[0030] From the above description, it can be seen that the real-time collected audio data is used to identify and warn pipeline environmental events through the model sample library.

[0031] The present invention discloses an audio-based pipeline environment event recognition method and system, which are suitable for the recognition and early warning of pipeline environment events, and are particularly suitable for the recognition and early warning of pipeline environment events of power pipelines.

[0032] Please refer to Figure 1 , Embodiment 1 of the present invention is: A pipeline environment event recognition method based on audio comprises the following steps: S1. When a pipeline environment event occurs, collect pipeline environment sounds and generate an event audio set; When a pipeline environmental event occurs, the specific steps for collecting pipeline environmental sound are as follows: Each time a pipeline event is sensed, audio is recorded at a preset frequency and duration to obtain audio data of the pipeline event.

[0033] In this embodiment, the existing intelligent early warning system based on light sensing technology is used, based on optical fiber sensing technology and the principle of elastic light effect, to collect pipeline environmental sound and generate event audio sets each time a pipeline event is sensed. Specifically, optical fiber monitors the power pipeline line in real time. When multiple alarms are triggered on the line or a special sound triggers an alarm, the system selects a period of time when the alarm occurs for recording, rather than just the 10 seconds before and after the alarm moment, and the recording duration and frequency are one audio every 10 seconds.

[0034] The steps between step S1 and step S2 also include: Noise reduction processing is performed on the audio data in the event audio set.

[0035] In this embodiment, all recorded audio is subjected to noise reduction processing and saved. The noise reduction method uses Python noisereduce library to perform noise reduction processing (spectral gating technology noise reduction technology).

[0036] S2. Generate a classification label of the surrounding event type for the pipeline environment sound in the event audio set, and extract features of the pipeline environment sound using a Mel function and a Welch spectrum analysis algorithm to obtain feature data.

[0037] The characteristic data is an audio spectrum waveform generated using a Mel function and a Welch spectrum analysis algorithm.

[0038] In this embodiment, the audio after noise reduction is manually classified, for example: large jackhammer, drilling rig, handheld jackhammer, machinery-other, normal environmental sound, water flow sound, and vehicle pressure manhole cover, etc.

[0039] Then use the Mel function and Welch spectrum analysis algorithm to extract feature values ​​of various types of audio data. The steps can be referred to in the following example: 1. Normalization: Normalize the audio volume and adjust the effective value (in decibels) of the audio signal to the specified target decibel value, thereby optimizing the performance and quality of the audio signal.

[0040] In this embodiment, the audio data is mainly a 0-2KHz audio signal.

[0041] 2. Window function application: A window function is applied to each frame. In this embodiment, a Hamming window is used to reduce edge effects and spectrum leakage.

[0042] 3. Short-time Fourier transform: Perform a fast Fourier transform (FFT) on each frame to convert it from the time domain to the frequency domain to obtain a frequency domain representation.

[0043] 4. Mel filter bank: The frequency domain representation is processed through a Mel filter bank. The Mel filter bank simulates the characteristics of the human auditory system, maps the frequency range to the Mel scale, and calculates the energy of each filter output.

[0044] 5. Logarithmic compression: Logarithmic compression is performed on the output results of the Mel filter bank to better simulate the perceptual characteristics of the human ear.

[0045] 6. Discrete Cosine Transform: Use discrete cosine transform to convert the logarithmic Mel energy into Mel frequency cepstral coefficients (MFCC). DCT can remove the high correlation between filter coefficients and thus extract useful features.

[0046] 7. Welch spectrum analysis: Welch spectrum analysis is a power spectrum density estimation method used to calculate the spectral characteristics of audio signals. Its main steps include: Split the signal into multiple overlapping frames.

[0047] Perform window function processing on each frame.

[0048] Compute the FFT of each frame.

[0049] The spectral estimation results of multiple frames are averaged to reduce the influence of noise.

[0050] 8. Feature statistical processing: Perform statistical processing on the extracted MFCC features, such as calculating the mean, variance, and delta, so that they can be further used for classification or recognition tasks.

[0051] S3. According to the feature data and the corresponding classification labels, the ECAPA-TDNN model is trained to establish a model sample library, and the model sample library is used to identify and warn pipeline environmental events.

[0052] In this embodiment, the structure of the ECAPA-TDNN model includes an input layer, a convolutional layer, a SE-Res2Block module, an attention statistics pooling layer, a multi-layer feature aggregation layer, a fully connected layer, and an AAM-Softmax loss function.

[0053] Input layer: In this embodiment, the input of the input layer is an 80-dimensional Mel-frequency cepstral coefficient (MFCC) feature with a shape of 80×T, where T is the number of time frames.

[0054] Convolutional layer: The first layer has a convolution kernel size of k=5 and a step size of d=1, which is used to extract local features. It is then processed using batch normalization (BN) and ReLU activation function.

[0055] SE-Res2Block module: SE-Res2Block combines the Squeeze-Excitation (SE) module and the Res2Net structure, and enhances the feature representation capability through multi-scale feature fusion and channel attention mechanism. Each SE-Res2Block contains multiple convolutional layers with a convolution kernel size of k=3 and a step size of d=2,3,4. Residual connections are used to enhance the network's expressive power.

[0056] Attention Statistics Pooling Layer: The attention statistics pooling layer calculates the mean and standard deviation of frame-level features and weights these statistics through the attention mechanism to capture richer temporal information.

[0057] Multi-layer feature aggregation layer: The multi-layer feature aggregation layer fuses features at different levels to further improve the feature representation capability of the model.

[0058] Fully connected layer: Finally, two fully connected layers are used to extract the low-dimensional speaker embedding vector. The first fully connected layer acts as a bottleneck layer with an output dimension of 192.

[0059] Cross-entropy loss function: Use the cross-entropy loss function for optimization, which improves classification performance by measuring the difference between the probability distribution predicted by the model and the probability distribution of the true label, making the model more discriminative between different categories and more similar within the same category.

[0060] In addition, in this embodiment, SpecAugment data enhancement technology is used to improve the robustness of the model to noise and voice changes, and jump connections and multi-scale feature fusion are introduced to enhance the network's expressive power and feature diversity.

[0061] In step S3, the identification and early warning of pipeline environmental events using the model sample library are specifically as follows: Real-time audio is collected at different positions of the pipeline, and the model sample library is used to identify and warn pipeline environmental events based on the real-time collected audio data.

[0062] In this embodiment, the audio data from which the features are classified and extracted is used to establish a model for machine learning and training to form a model sample library, and the type of sound of events around the power pipeline is identified based on the established model sample library.

[0063] In this embodiment, a model is established based on the audio data that is classified and has its feature values ​​extracted to perform machine learning training, and a model sample library containing multiple event recognition models is obtained for different types of pipeline environmental events.

[0064] Please refer to Figure 2 , Embodiment 2 of the present invention is: An audio-based pipeline environment event recognition system, comprising: Data collection module: when a pipeline environment event occurs, it collects pipeline environment sounds and generates an event audio set; Data processing module: generating classification labels of surrounding event types for the pipeline environment sound in the event audio set, and extracting features of the pipeline environment sound using Mel function and Welch spectrum analysis algorithm to obtain feature data; Model training module: training the ECAPA-TDNN model according to the feature data and the corresponding classification labels, and establishing a model sample library; Event recognition module: uses the model sample library to identify and warn pipeline environmental events.

[0065] In this embodiment, based on the data acquisition module, the data processing module, the model training module and the event recognition module, the steps in the audio-based pipeline environment event recognition method in the above embodiment 1 are implemented.

[0066] In summary, the audio-based pipeline environmental event recognition method and system provided by the present invention utilize audio recognition technology to identify pipeline environmental events, which has higher maintainability and stability compared with optical fiber sensing technology. At the same time, the present invention uses Mel function and Welch spectrum analysis algorithm to extract features of pipeline environmental sounds, and has better anti-noise ability; and the ECAPA-TDNN model commonly used in speech recognition is used for the recognition of pipeline environmental events. Compared with complex language recognition, it is applied to the recognition of environmental sounds around power pipelines of relatively simple types, which can achieve higher recognition accuracy.

[0067] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A pipeline environment event recognition method based on audio, characterized in that: Includes steps: S1. When a pipeline environment event occurs, collect pipeline environment sounds and generate an event audio set; S2, generating a classification label of a surrounding event type for the pipeline environment sound in the event audio set, and extracting features of the pipeline environment sound using a Mel function and a Welch spectrum analysis algorithm to obtain feature data; S3. According to the feature data and the corresponding classification labels, the ECAPA-TDNN model is trained to establish a model sample library, and the model sample library is used to identify and warn pipeline environmental events.

2. The method for identifying pipeline environment events based on audio according to claim 1, characterized in that: The characteristic data is an audio spectrum waveform generated using a Mel function and a Welch spectrum analysis algorithm.

3. The method for identifying pipeline environment events based on audio according to claim 1, characterized in that: When a pipeline environmental event occurs, the specific steps for collecting pipeline environmental sound are as follows: Each time a pipeline event is sensed, audio is recorded at a preset frequency and duration to obtain audio data of the pipeline event.

4. The method for identifying pipeline environment events based on audio according to claim 1, characterized in that: The steps between step S1 and step S2 also include: Noise reduction processing is performed on the audio data in the event audio set.

5. The method for identifying pipeline environment events based on audio according to claim 1, characterized in that: In step S3, the identification and early warning of pipeline environmental events using the model sample library are specifically as follows: Real-time audio is collected at different positions of the pipeline, and the model sample library is used to identify and warn pipeline environmental events based on the real-time collected audio data.

6. An audio-based pipeline environment event recognition system, comprising a processor, a memory, and a computer program stored in the memory, characterized in that: include: Data collection module: when a pipeline environment event occurs, it collects pipeline environment sounds and generates an event audio set; Data processing module: generating classification labels of surrounding event types for the pipeline environment sound in the event audio set, and extracting features of the pipeline environment sound using Mel function and Welch spectrum analysis algorithm to obtain feature data; Model training module: training the ECAPA-TDNN model according to the feature data and the corresponding classification labels, and establishing a model sample library; Event recognition module: uses the model sample library to identify and warn pipeline environmental events.

7. The audio-based pipeline environment event recognition system according to claim 6, characterized in that: The characteristic data is an audio spectrum waveform generated using a Mel function and a Welch spectrum analysis algorithm.

8. The audio-based pipeline environment event recognition system according to claim 6, characterized in that: When a pipeline environmental event occurs, the specific steps for collecting pipeline environmental sound are as follows: Each time a pipeline event is sensed, audio is recorded at a preset frequency and duration to obtain audio data of the pipeline event.

9. The audio-based pipeline environment event recognition system according to claim 6, characterized in that: Before the data processing module generates a classification label of the surrounding event type for the pipeline environment sound in the event audio set, the data processing module further includes the steps of: Noise reduction processing is performed on the audio data in the event audio set.

10. The audio-based pipeline environment event recognition system according to claim 6, characterized in that: The identification and early warning of pipeline environmental events using the model sample library are specifically as follows: Real-time audio is collected at different positions of the pipeline, and the model sample library is used to identify and warn pipeline environmental events based on the real-time collected audio data.

Images