Leak detector data acquisition and transmission system and method

Abstract

The invention discloses a leak detector data acquisition and transmission system and method, and the method comprises the steps: actively collecting on-site original sound waves through an ultrasonic microphone, and carrying out the processing of the original sound waves, so as to generate a specific noise fingerprint of a current environment, thereby depicting the spectrum distribution and intensity characteristics of background noise; targeted noise reduction is carried out on subsequently collected sound waves on the basis, so that a real leakage signal is accurately stripped from strong interference. On this basis, the terminal device performs local real-time reasoning by using the extracted high-quality feature vector so as to quickly identify potential leakage. Finally, in response to a leakage event which is locally judged to be high in confidence, the original sound wave data stream without information loss is transmitted to the cloud. Through accurate noise fingerprint noise reduction, the detection sensitivity and accuracy of the leak detector on early weak leakage signals in a complex industrial environment are greatly improved, and false alarm and missing alarm caused by environmental noise and transient interference are effectively avoided.

Description

Technical Field

[0001] The present application relates to the field of data processing, and more specifically, to a leak detector data acquisition and transmission system and method. Background Art

[0002] The stable operation of pipelines is crucial in critical infrastructure networks such as urban water supply, petrochemicals, and natural gas transmission. However, pipeline leaks are common due to factors such as pipeline aging, external damage, and geological changes. These leaks not only result in a significant waste of precious resources but can also cause serious environmental pollution and public safety incidents. Therefore, timely and accurate detection of pipeline leaks is crucial for ensuring the safe and efficient operation of pipeline networks. Acoustic detection, particularly the use of ultrasonic technology to detect high-frequency sound waves, has become a mainstream and effective leak detection method. To achieve real-time monitoring and intelligent analysis of extensive pipeline networks, it is crucial to develop an efficient leak detector data acquisition and transmission system.

[0003] However, in practical applications, especially in complex industrial environments such as large chemical plants, the use of acoustic signals for leak detection faces severe technical challenges. The inventors found in their research that the core technical problem of existing leak detection schemes is how to effectively extract weak leakage signals under high noise interference. Specifically, stable and high-intensity background noise is prevalent in on-site working conditions, such as the high-decibel, wide-spectrum noise generated by the operation of large compressors, whose energy may be several orders of magnitude higher than the signal of early tiny leaks, directly resulting in an extremely low signal-to-noise ratio. Even more challenging is that the environment is also filled with a large number of transient interference sounds that are extremely similar to the acoustic characteristics of real leaks, such as the sharp "hissing" sound generated when pneumatic valves are opened and closed. This interference can easily cause system misjudgment. Since the leakage signal itself has the characteristics of evolution from weak to significant, traditional general noise reduction algorithms are difficult to effectively remove these complex and changeable background noises and transient interferences without damaging the target signal, resulting in a significant reduction in the accuracy and reliability of detection.

[0004] Therefore, an optimized leak detector data acquisition and transmission solution is expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a leak detector data acquisition and transmission system and method, which actively collects the original sound waves at the scene through an ultrasonic microphone, and processes them to generate a noise fingerprint unique to the current environment, so as to characterize the spectral distribution and intensity characteristics of the background noise. Then, based on this, the subsequent collected sound waves are targeted for noise reduction, so as to accurately separate the real leakage signal from the strong interference. On this basis, the terminal device uses the extracted high-quality feature vectors for local real-time reasoning to quickly identify potential leaks. Finally, in response to a local judgment of a high-confidence leakage event, the original sound wave data stream without information loss is transmitted to the cloud. Through precise noise fingerprint denoising, the detection sensitivity and accuracy of the leak detector for early weak leakage signals in complex industrial environments are greatly improved, effectively avoiding false alarms and missed alarms caused by environmental noise and transient interference.

[0006] According to one aspect of the present application, a leak detector data acquisition and transmission method is provided, comprising: In response to an environmental noise learning instruction, collecting original analog sound waves through an ultrasonic microphone, and performing data processing on the original analog sound waves to obtain a noise fingerprint; Acquire the leak detection acoustic wave digital stream collected by the measuring instrument; Based on the noise fingerprint, the digital stream of the leak detection sound wave is subjected to the noise fingerprint denoising to obtain the noise-reduced leak detection sound wave spectrum; Performing feature extraction on the noise-reduced leak detection sound wave spectrum to obtain a noise-reduced leak detection sound wave feature vector; Generate inference results based on the noise-reduced leak detection acoustic wave feature vector; In response to the inference result that a leak exists and the confidence level is greater than a preset threshold, the leak detection acoustic wave digital stream is transmitted to the cloud.

[0007] According to another aspect of the present application, a leak detector data acquisition and transmission system is provided, comprising: a noise fingerprint acquisition module, configured to respond to an environmental noise learning instruction, collect original analog sound waves through an ultrasonic microphone, and perform data processing on the original analog sound waves to obtain a noise fingerprint; A leak detection acoustic wave digital stream acquisition module is used to acquire the leak detection acoustic wave digital stream collected by the measuring instrument; A voiceprint fingerprint noise reduction module is used to perform voiceprint fingerprint noise reduction on the leak detection sound wave digital stream based on the noise fingerprint to obtain the leak detection sound wave spectrum after noise reduction; A leak detection acoustic wave feature extraction module is used to extract features from the noise-reduced leak detection acoustic wave spectrum to obtain a noise-reduced leak detection acoustic wave feature vector; An inference result generation module, used to generate an inference result based on the noise-reduced leak detection acoustic wave feature vector; The leakage detection acoustic wave digital stream transmission module is used to transmit the leakage detection acoustic wave digital stream to the cloud in response to the inference result that there is a leak and the confidence level is greater than a preset threshold.

[0008] Compared with the prior art, the present application provides a leak detector data acquisition and transmission system and method, which actively collects the original sound waves on site through an ultrasonic microphone, and processes them to generate a noise fingerprint unique to the current environment to characterize the spectral distribution and intensity characteristics of the background noise. Then, based on this, the subsequent collected sound waves are targeted for noise reduction, thereby accurately separating the real leakage signal from the strong interference. On this basis, the terminal device uses the extracted high-quality feature vectors for local real-time reasoning to quickly identify potential leaks. Finally, in response to a local judgment of a high-confidence leakage event, the original sound wave data stream without information loss is transmitted to the cloud. Through precise noise fingerprint denoising, the leak detector's detection sensitivity and accuracy for early weak leakage signals in complex industrial environments are greatly improved, effectively avoiding false alarms and missed alarms caused by environmental noise and transient interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 Flowchart of a method for collecting and transmitting data from a leak detector according to an embodiment of the present application; Figure 2 Schematic diagram of data flow of the leak detector data collection and transmission method according to an embodiment of the present application; Figure 3 4 is a block diagram of a leak detector data acquisition and transmission system according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0012] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0016] In the technical solution of the present application, a method for collecting and transmitting leak detector data is proposed. Figure 1 Flowchart of a method for collecting and transmitting leak detector data according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the leak detector data collection and transmission method according to the embodiment of the present application. Figure 1 and Figure 2 As shown, the leak detector data collection and transmission method according to the embodiment of the present application includes the following steps: S1, in response to the environmental noise learning instruction, collecting the original analog sound wave through the ultrasonic microphone, and performing data processing on the original analog sound wave to obtain a noise fingerprint; S2, obtaining the leak detection sound wave digital stream collected by the measuring instrument; S3, based on the noise fingerprint, performing voiceprint fingerprint noise reduction on the leak detection sound wave digital stream to obtain a leak detection sound wave spectrum after noise reduction; S4, performing feature extraction on the leak detection sound wave spectrum after noise reduction to obtain a leak detection sound wave feature vector after noise reduction; S5, generating an inference result based on the leak detection sound wave feature vector after noise reduction; S6, in response to the inference result that there is a leak and the confidence level is greater than a preset threshold, transmitting the leak detection sound wave digital stream to the cloud.

[0017] Specifically, S1, in response to an environmental noise learning instruction, collects raw analog sound waves via an ultrasonic microphone and processes the raw analog sound waves to obtain a noise fingerprint. It should be understood that industrial sites are often filled with background noise generated by the operation of various equipment. If this noise is mixed with the leakage sound, it will seriously interfere with the subsequent leakage signal identification. Therefore, in the technical solution of this application, by pre-collecting and analyzing pure background noise, a stable and representative noise fingerprint is generated, which serves as a benchmark for subsequent noise reduction processing.

[0018] During the specific implementation, the operator first triggers a specific instruction to start the noise learning process after confirming that there is no leak at the scene; then, after receiving the instruction, the system will collect environmental sound waves through its built-in ultrasonic microphone, and process the collected analog signals through a series of digital processing to finally form a noise fingerprint.

[0019] The environmental noise learning command is a user-initiated signal that initiates the background noise collection and analysis process. A raw analog sound wave refers to the continuously changing sound wave signal captured directly by an ultrasonic microphone without any digital processing. A noise fingerprint is a processed data structure that stably represents the spectral characteristics of background noise in a specific environment, typically expressed as an average energy spectrum.

[0020] Specifically, the specific process of digitally processing the collected analog signal includes: first, performing ADC conversion on the original analog sound wave to obtain a digital stream of the original sound wave. It should be understood that computers and digital signal processors can only process discrete digital signals, while the original sound waves collected by the ultrasonic microphone are continuous analog signals. Therefore, in the technical solution of this application, the original analog sound wave is first converted into digital form to facilitate the system's subsequent calculations and analysis. Specifically, this step is performed using an analog-to-digital converter (ADC). During this process, the ADC periodically samples the continuous analog sound wave, measuring the amplitude of the sound wave once at each extremely small time interval and quantizing the amplitude value into a binary digit. By continuously sampling at a sufficiently high frequency (sampling rate), the continuous analog waveform can be approximately represented as a digital stream of the original sound wave consisting of a digital sequence.

[0021] Next, the original sound wave digital stream is framed to obtain a sequence of original sound wave digital frames. It should be understood that sound signals are usually time-varying, and their frequency characteristics may be different in different time periods. Directly performing Fourier transform on the entire long-term digital stream will lose the local time characteristics of the signal. Therefore, in the technical solution of the present application, the original sound wave digital stream is framed to cut the long string of digital streams into multiple short-term frames. In this way, the system analyzes each frame separately, thereby capturing the changes in the signal in the time dimension.

[0022] Subsequently, a fast Fourier transform is performed on each original sound wave digital frame in the sequence of original sound wave digital frames to obtain a sequence of power spectrum data distribution. It should be understood that the signal frame obtained after framing is still a time domain signal (the amplitude changes with time), and it is difficult to intuitively see its frequency composition. Therefore, in the technical solution of the present application, a fast Fourier transform is performed on each original sound wave digital frame in the sequence of original sound wave digital frames to convert the signal from the time domain to the frequency domain. In this way, the system can analyze the frequency components in the signal and the intensity of each component.

[0023] Among them, the fast Fourier transform (FFT) is a fast algorithm that converts time domain signals into frequency domain representation. Its output result is a complex number sequence, the square of its modulus is the power spectrum of the frame signal, which shows the distribution of signal energy at different frequencies.

[0024] Then, the power spectrum density of each power spectrum data distribution in the sequence of power spectrum data distribution is calculated to obtain a sequence of power spectrum density. It should be understood that the numerical value of the power spectrum obtained directly by FFT is related to the number of FFT points (i.e., frame length), which is not convenient for comparison under different parameter settings. Therefore, in the technical solution of the present application, by calculating the power spectrum density of each power spectrum data distribution, the energy density of the signal at each frequency is more accurately and physically represented, making the result more universal and comparable. Among them, the power spectrum density (PSD) is a function that describes the distribution of signal power in the frequency domain and represents the signal power within a unit frequency interval.

[0025] Furthermore, the sequence of power spectral density is averaged and smoothed to obtain an average energy spectrum as a noise fingerprint. It should be understood that the power spectral density of a single frame may fluctuate greatly due to instantaneous disturbances and is not representative. Therefore, in the technical solution of the present application, the sequence of power spectral density is averaged and smoothed to stably reflect the final fingerprint of the steady-state background noise characteristics of the current environment, thereby eliminating the influence of random noise and instantaneous burst signals. In this process, the system first averages the power spectral density calculated for all frames in the time dimension, that is, for each frequency point, its PSD value in all frames is added and divided by the total number of frames; then, a smoothing filter (such as a moving average filter) is used to process the obtained average energy spectrum to make its curve smoother, further eliminate burrs and noise, and obtain an average energy spectrum as a noise fingerprint.

[0026] Specifically, step S2 acquires the digital stream of leak detection sound waves collected by the measuring instrument. By acquiring the raw, uncompressed or preprocessed digital stream, it is possible to preserve all details of the on-site sound waves, providing a high-quality raw data foundation for subsequent refined noise reduction, feature extraction, and ultimately transmission to the cloud for in-depth analysis.

[0027] In practice, the leak detector is aimed at the target (such as a pipe or valve) and continuously collects acoustic data. The leak detection acoustic wave digital stream refers to the raw, unprocessed digital time series representing the acoustic field information at the test point, collected by the leak detector's ultrasonic microphone and ADC during the actual detection task. This leak detection acoustic wave digital stream is a time-domain signal that fully records the amplitude information of all acoustic waves received by the leak detector at the moment of detection (including both stable background noise and high-frequency ultrasonic signals that may be generated by gas leaks), varying over time.

[0028] In particular, the S3 performs voiceprint fingerprint denoising on the leak detection sound wave digital stream based on the noise fingerprint to obtain the leak detection sound wave spectrum after denoising. It should be understood that there are background noise and target leakage signals in the leak detection sound wave digital stream collected by the leak detector. If it is not subjected to denoising, the strong background noise will cover up or even drown out the key leakage sound characteristics, resulting in misjudgment or missed judgment in subsequent feature extraction and model reasoning. Therefore, in the technical solution of the present application, by performing voiceprint fingerprint denoising on the leak detection sound wave digital stream, the signal-to-noise ratio of the signal can be significantly improved, so that subsequent processing can focus on a purer signal basis that can better highlight the leakage characteristics.

[0029] In a specific example of the present application, the following steps can be used to perform voiceprint fingerprint noise reduction on the leak detection sound wave digital stream: first, a fast Fourier transform is performed on the leak detection sound wave digital stream to obtain the current spectrum of the side leakage detection sound wave signal. It should be understood that the sound signal in the time domain (amplitude changes with time) often appears as a complex waveform, and the background noise and the target leakage signal are superimposed on each other, making it difficult to separate them directly. However, in the frequency domain (energy is distributed with frequency), sounds from different sources usually have their own unique frequency characteristics. Therefore, in the technical solution of the present application, first, a fast Fourier transform is performed on the leak detection sound wave digital stream to transform the signal into the frequency domain.

[0030] Among them, the current spectrum of the side leakage detection sound wave signal is a data set (usually a vector or array), which describes the energy or amplitude distribution of the mixed sound waves (including background noise and potential leakage sound) collected at different frequencies within a very short time window (one frame).

[0031] Then, the current spectrum of the side leakage detection sound wave signal and the noise fingerprint are spectrally encoded to obtain the current spectrum feature vector of the side leakage detection sound wave signal and the noise fingerprint coding vector. It should be understood that if the amplitude or energy of the two spectra are directly subtracted, distortions such as musical noise will be introduced, and the situation where the signal and noise are highly overlapped or have a nonlinear relationship in the frequency band cannot be effectively handled. In order to achieve more effective noise suppression, in the technical solution of the present application, the current spectrum of the side leakage detection sound wave signal and the noise fingerprint are spectrally encoded to map the spectrum data into a high-dimensional feature space. The model can learn and capture more abstract and essential patterns in the spectrum, such as the overall shape of the spectrum, harmonic structure, and the distribution law of energy in different frequency bands. In this feature space, the characteristics of the noise and the characteristics of the signal can be more clearly distinguished.

[0032] In practice, a pre-trained deep learning model can be used to convert one-dimensional spectral data (typically a sequence of frequencies and energy / amplitudes) into a high-dimensional, dense feature vector. In one example, this spectral encoder can be implemented using a Transformer-based encoding layer. Due to its self-attention mechanism, the Transformer model can capture long-range dependencies between frequency points in the spectral sequence, thereby generating feature vectors that are rich in information and highly resolvable.

[0033] Then, the positional difference between the current spectrum feature vector of the side leak detection acoustic signal and the noise fingerprint encoding vector is calculated to obtain the noise-reduced leak detection acoustic spectrum encoding vector. Because the current spectrum of the leak detection acoustic signal contains mixed information of both leakage sound and background noise, directly judging leaks based on the original spectrum will result in a high false detection rate. Therefore, in the technical solution of the present application, by subtracting the noise fingerprint encoding vector (representing the characteristics of pure background noise) from the current spectrum feature vector, preliminary noise elimination can be achieved within the feature space, providing cleaner input for subsequent feature enhancement and decoding, and ultimately improving the accuracy of leak detection.

[0034] Furthermore, feature distribution enhancement is performed on the noise-reduced leak detection acoustic wave spectrum code vector to obtain an enhanced noise-reduced leak detection acoustic wave spectrum code vector. It should be understood that the feature distribution in the noise-reduced leak detection acoustic wave spectrum code vector may contain noise or weakly correlated patterns unrelated to the leak, which can lead to an increased false detection rate in the subsequent inference stage. This feature distribution enhancement allows the system to focus on high-frequency, persistent leak acoustic signatures (such as the specific soundprint pattern of a pipeline leak), suppressing background interference and thereby improving detection accuracy and confidence.

[0035] Specifically, first, the spectrum feature segmentation of the sound wave spectrum coding vector after noise reduction for leak detection is performed based on a preset window to obtain a set of the spectrum feature segmentation coding vectors of the sound wave spectrum coding vector after noise reduction for leak detection. It should be understood that although the spectrum feature segmentation coding vector of the sound wave spectrum coding vector after noise reduction has suppressed some interference through preliminary processing, it still remains as a macroscopic one-dimensional feature vector, which may miss weak leakage features or local abnormal patterns. Therefore, the spectrum feature segmentation of the sound wave spectrum coding vector after noise reduction is performed based on a preset window to decompose the entire spectrum into multiple short-time, continuous local segments, each segment corresponding to a specific frequency band and time window, thereby adapting to the non-stationary characteristics of the sound wave signal, solving the feature ambiguity problem caused by residual noise or overall homogenization of the spectrum, and providing an operational basis for subsequent fine processing.

[0036] In a specific example of the present application, the spectrum feature segmentation of the noise-reduced leakage detection sound wave spectrum encoding vector is performed based on a preset window using the following formula to obtain a set of noise-reduced leakage detection sound wave spectrum feature segment encoding vectors; wherein, the formula is:

[0037] in, is a set of encoding vectors of characteristic segments of the leak detection sound wave spectrum after noise reduction, are respectively the first, second, and third in the set of encoding vectors of characteristic segments of the noise-reduced leakage detection sound wave spectrum. and A coding vector of characteristic segments of the leak detection sound wave spectrum after noise reduction.

[0038] Next, the collective feature shape of the noise-reduced leak detection sound wave spectrum feature segment encoding vector is reshaped into a noise-reduced leak detection sound wave spectrum feature segment encoding feature map, and then content perception based on convolutional coding is performed to obtain a noise-reduced leak detection sound wave spectrum local semantic perception encoding feature map. It should be understood that although the one-dimensional feature vector that has undergone preliminary noise reduction has suppressed some noise, it is still difficult to effectively capture the local frequency domain patterns unique to the leak sound (such as continuous pulses or harmonic structures in a specific frequency band). These patterns have spatial correlation in time-frequency distribution, and traditional one-dimensional feature representations cannot fully exploit their structured information. Therefore, in the technical solution of this application, the one-dimensional segment vector is converted into a two-dimensional feature map structure through feature shape reshaping. This allows the inherent local connection and weight sharing characteristics of the convolutional layer to scan and analyze the energy distribution, waveform continuity, and mutation patterns between adjacent frequency bands. This structured representation enables the system to identify acoustic fingerprints with spatial locality, such as the high-frequency hissing sound unique to gas leaks and the low-frequency surging sound of liquid leaks.

[0039] That is, in the technical solution of the present application, first, the collective characteristic shape of the coding vector of the characteristic segment of the noise-reduced leakage detection sound wave spectrum is reshaped into a coding characteristic map of the characteristic segment of the noise-reduced leakage detection sound wave spectrum. The process is expressed by the formula:

[0040] in, represents the feature reshaping operation, is the convolutional coding function, Encoding a characteristic graph for characteristic segments of the noise-reduced leak detection sound wave spectrum; Next, the coding feature map of the characteristic segment of the sound wave spectrum of the leak detection after noise reduction is subjected to dynamic convolution coding based on phase disorder correction to obtain the content-aware coding feature map of the trigger signal pattern characteristic segment. In particular, it should be understood here that traditional static convolution kernels are difficult to adapt to the dynamic distortion of sound wave signals in complex physical environments. For example, when a wide-band soundprint generated by a high-pressure gas leak encounters structures such as pipe elbows and valves during propagation, its phase distribution will show local fractures or non-uniform offsets, resulting in the destruction of the spatial continuity of the feature map. If this phase disorder phenomenon is not corrected, it will seriously hinder the effective capture of leakage features by the convolution layer. Therefore, in the technical solution of the present application, real-time compensation for phase disorder is achieved by dynamically generating convolution kernel parameters that are adaptive to the acoustic scene.

[0041] That is, in the technical solution of the present application, first, the coding feature map of the noise-reduced leakage detection sound wave spectrum feature segment is globally mean pooled along the channel dimension to obtain the global pooled coding vector of the noise-reduced leakage detection sound wave spectrum feature. The process is expressed by the formula:

[0042] in, is the global mean pooling operation, A coding vector for a characteristic segment of the noise-reduced leak detection sound wave spectrum; Next, the global pooling coding vector of the noise-reduced leak detection sound wave spectrum feature is subjected to dynamic content perception based on fully connected coding to obtain a dynamic convolution kernel. The process is expressed as follows:

[0043] in, and are the first and second weight matrices, and denote the first and second bias vectors, represents the dynamic convolution kernel fully connected perception vector, represents the dynamic convolution kernel, represents the characteristic shape reshaping function; Then, based on the dynamic convolution kernel, the noise-reduced leakage detection sound wave spectrum feature segment encoding feature map is convolutionally encoded to obtain the trigger signal pattern feature segment content perception encoding feature map, and the process is expressed by the formula:

[0044] in, is the convolution operation, A content-aware coding feature map of the trigger signal pattern feature segment.

[0045] In particular, in the technical solution of the present application, the process of performing dynamic content perception based on fully connected coding on the coding vector of the characteristic segment of the noise-reduced leak detection sound wave spectrum to obtain a dynamic convolution kernel includes: First, extract the first fully connected weight matrix, the second fully connected weight matrix, the first initial bias vector, and the second initial bias vector; Then, based on the first fully connected weight matrix and the second fully connected weight matrix, the first initial bias vector and the second initial bias vector are iteratively corrected for phase disorder to obtain a first bias vector and a second bias vector. This process can be expressed as follows:

[0046] in, and are the first and second weight matrices, and are the first and second initial bias vectors, represents the dynamic convolution kernel fully connected perception vector, represents the dynamic convolution kernel, Indicates point multiplication by position, and are the first and second surface plane state vectors, is the transition operator vector, represents vector multiplication, represents the two-norm of the matrix, and denote the first and second bias vectors; Furthermore, based on the first fully connected weight matrix, the second fully connected weight matrix, the first bias vector and the second bias vector, the global pooling coding vector of the noise reduction leakage detection sound wave spectrum feature is fully connected encoded to obtain the dynamic convolution kernel fully connected perception vector.

[0047] Finally, the dynamic convolution kernel fully connected perception vector is reshaped to obtain the dynamic convolution kernel.

[0048] Furthermore, the local semantic perception coding feature map of the sound wave spectrum after noise reduction for leak detection is input into the spatial attention layer to obtain the spatial significant perception coding feature map of the feature segment content of the sound wave spectrum after noise reduction for leak detection. It should be understood that although the previous convolutional coding has extracted preliminary features from the segmented spectrum segments, the local semantic perception coding feature map of the sound wave spectrum after noise reduction for leak detection still contains a large amount of interference information or low-value features that are not related to leakage. If such features are used directly for leak judgment, it is easy to cause high-frequency false alarms or missed detections in low signal-to-noise ratio scenarios. Therefore, in the technical solution of the present application, the feature map is input into the spatial attention layer to establish a mapping relationship between the spatial position of the feature map and the leakage probability, and the weights of leakage-sensitive areas (such as high-frequency energy mutation areas) are adaptively enhanced while suppressing the response of background noise-dominated areas.

[0049] In a specific example of the present application, the local semantic perception coding feature map of the noise reduction leakage detection sound wave spectrum is input into the spatial attention layer to obtain the spatial significant perception coding feature map of the content of the noise reduction leakage detection sound wave spectrum feature segment; wherein, the formula is:

[0050] Among them, it means function, is the spatial attention weight of the feature segment content of the noise reduction leakage detection sound wave spectrum, It is a content space significant perceptual coding feature map of the characteristic segment of the leakage detection sound wave spectrum after noise reduction.

[0051] Subsequently, the spatially significant perceptual coding feature map of the characteristic fragment content of the sound wave spectrum of the leak detection after noise reduction is reshaped to obtain an enhanced sound wave spectrum coding vector for leak detection after noise reduction. It should be understood that in the wide-area pipe network monitoring scenario, the significant areas extracted by the spatial attention mechanism (such as continuous high-energy pulses in the 40-60kHz frequency band) accurately mark the suspected leakage areas, but these areas are often discontinuous and scattered (such as multi-band harmonics generated by pipeline cracks). Therefore, in the technical solution of the present application, the spatially significant perceptual coding feature map of the characteristic fragment content of the sound wave spectrum of the leak detection after noise reduction is reshaped to rearrange the two-dimensional feature map into a one-dimensional vector according to spatial correlation. In this way, the core information of the spatial attention weighting is retained, and the redundant dimensions are eliminated to adapt to the linear structure of the terminal device.

[0052] In a specific example of the present application, the following formula is used to reshape the feature shape of the content space significant perceptual coding feature map of the noise-reduced leakage detection sound wave spectrum feature segment to obtain the enhanced noise-reduced leakage detection sound wave spectrum coding vector; wherein, the formula is:

[0053] in, It is the encoding vector of the leak detection sound wave spectrum after the enhancement and noise reduction.

[0054] Subsequently, the enhanced noise-reduced leak detection acoustic wave spectrum encoding vector is feature decoded to obtain the noise-reduced leak detection acoustic wave spectrum. Although the enhanced noise-reduced leak detection acoustic wave spectrum encoding vector has optimized the discriminability of leakage features, as an abstract high-dimensional vector, it cannot directly reflect the energy distribution characteristics of the acoustic wave. In the technical solution of the present application, through feature decoding, the system can reversely map the enhanced feature representation back to the frequency domain space, generating a noise-reduced leak detection acoustic wave spectrum that retains the key leakage signal and suppresses noise.

[0055] In specific implementation, the enhanced noise-reduced leak detection sound wave spectrum encoding vector can be input into the feature decoder, and the decoder gradually restores the spectrum structure through multi-layer inverse transformation operations. It is worth mentioning that if the encoder in the previous step uses multi-layer convolution and pooling to compress the data dimension and extract features, the decoder will use the corresponding deconvolution (or transposed convolution) and upsampling layers to gradually restore the original dimension of the data. In this process, the decoder uses its internal trained weight parameters to perform a series of nonlinear transformations on the input feature vector, and finally reconstructs a one-dimensional sequence. Each point in the sequence represents the energy or amplitude value at a specific frequency, that is, the noise-reduced leak detection sound wave spectrum.

[0056] In particular, S4 performs feature extraction on the noise-reduced leak detection acoustic wave spectrum to obtain a noise-reduced leak detection acoustic wave feature vector. That is, in the technical solution of the present application, the noise-reduced leak detection acoustic wave spectrum is passed through a leak detection acoustic wave feature extractor based on a convolutional neural network to obtain a noise-reduced leak detection acoustic wave feature vector. It should be understood that although the noise-reduced leak detection acoustic wave spectrum has been largely removed, it is still essentially a high-dimensional data sequence or image containing a large amount of raw information. Making judgments directly based on such high-dimensional data is not only computationally intensive but also difficult to capture the critical, subtle patterns that distinguish leaks from non-leakages. Therefore, in the technical solution of the present application, the noise-reduced leak detection acoustic wave spectrum is passed through a leak detection acoustic wave feature extractor based on a convolutional neural network to learn and extract from the clean spectrum the fingerprint information most valuable for leak judgment, such as specific frequency peak combinations, energy distribution patterns across frequency bands, harmonic structures, etc., and condense these complex patterns into a low-dimensional vector, namely, the noise-reduced leak detection acoustic wave feature vector. This greatly simplifies the subsequent classification task and improves its accuracy.

[0057] Convolutional neural networks are typically constructed by alternating multiple convolutional layers, activation function layers (such as ReLU), and pooling layers. Convolutional layers perform convolution operations on the input spectrogram by sliding small filters (or convolution kernels) over it to capture local features, such as energy spikes within a certain frequency range. Activation function layers introduce nonlinearity, enabling the network to learn more complex patterns. Pooling layers downsample the feature map, reducing the amount of data while retaining the most significant features and providing a certain degree of translation invariance.

[0058] In particular, the S5 generates an inference result based on the characteristic vector of the sound wave detected after noise reduction. It should be understood that although the characteristic vector of the sound wave detected after noise reduction contains key information on whether there is a leak or not, it does not have direct business significance. In order to enable the equipment to independently judge the on-site situation, a classification model is introduced in the technical solution of the present application to interpret the characteristic vector of the sound wave detected after noise reduction, so as to map the complex characteristic vector to a predefined category. The categories include "leakage exists" or "no leakage". In this process, the classifier finally outputs a clear classification label, that is, whether the currently detected sound wave belongs to a leak; and a confidence score corresponding to the judgment. This score is usually a value between 0 and 1, which indicates the degree of confidence of the model in the current judgment.

[0059] In particular, in S6, in response to the inference result that there is a leak and the confidence level is greater than a preset threshold, the leak detection sound wave digital stream is transmitted to the cloud. It should be understood that if all collected data are transmitted indiscriminately, it will place a huge burden on the network module and power consumption of the device, and will also generate high cloud storage and computing costs. On the contrary, if the final judgment result is completely dependent on the local, some leakage details that are difficult to judge or require more complex algorithms to confirm may be lost due to the limitations of the terminal device's computing power. Therefore, in the technical solution of the present application, in response to the inference result that there is a leak and the confidence level is greater than a preset threshold, the leak detection sound wave digital stream is transmitted to the cloud. This can not only filter out the vast majority of non-leakage or background noise data, but also ensure that the original sound wave data of all high-value and high-probability leakage events can be retained for the cloud platform to conduct more in-depth analysis, archiving and traceability.

[0060] In summary, the data acquisition and transmission method of the leak detector according to the embodiment of the present application is explained, which actively collects the original sound waves at the scene through an ultrasonic microphone, and processes them to generate a noise fingerprint unique to the current environment to characterize the spectral distribution and intensity characteristics of the background noise. Then, based on this, the subsequent collected sound waves are targeted for noise reduction, thereby accurately separating the real leakage signal from the strong interference. On this basis, the terminal device uses the extracted high-quality feature vectors for local real-time reasoning to quickly identify potential leaks. Finally, in response to a local judgment of a high-confidence leakage event, the original sound wave data stream without information loss is transmitted to the cloud. Through precise noise fingerprint denoising, the detection sensitivity and accuracy of the leak detector for early weak leakage signals in complex industrial environments are greatly improved, effectively avoiding false alarms and missed alarms caused by environmental noise and transient interference.

[0061] Furthermore, a leak detector data acquisition and transmission system is also provided.

[0062] Figure 3 FIG is a block diagram of a leak detector data acquisition and transmission system according to an embodiment of the present application. Figure 3As shown, the leak detector data acquisition and transmission system 300 according to the embodiment of the present application includes: a noise fingerprint acquisition module 310, which is used to collect original analog sound waves through an ultrasonic microphone in response to an environmental noise learning instruction, and perform data processing on the original analog sound waves to obtain a noise fingerprint; a leak detection sound wave digital stream acquisition module 320, which is used to obtain the leak detection sound wave digital stream collected by the measuring instrument; a voiceprint fingerprint noise reduction module 330, which is used to perform voiceprint fingerprint noise reduction on the leak detection sound wave digital stream based on the noise fingerprint to obtain a leak detection sound wave spectrum after noise reduction; a leak detection sound wave feature extraction module 340, which is used to perform feature extraction on the leak detection sound wave spectrum after noise reduction to obtain a leak detection sound wave feature vector after noise reduction; an inference result generation module 350, which is used to generate an inference result based on the leak detection sound wave feature vector after noise reduction; and a leak detection sound wave digital stream transmission module 360, which is used to transmit the leak detection sound wave digital stream to the cloud in response to the inference result that there is a leak and the confidence level is greater than a preset threshold.

[0063] As described above, the leak detector data acquisition and transmission system 300 according to the embodiments of the present application can be implemented in various wireless terminals, such as a server equipped with a leak detector data acquisition and transmission algorithm. In one possible implementation, the leak detector data acquisition and transmission system 300 according to the embodiments of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the leak detector data acquisition and transmission system 300 can be a software module within the wireless terminal's operating system, or an application developed specifically for the wireless terminal. Of course, the leak detector data acquisition and transmission system 300 can also be one of the wireless terminal's many hardware modules.

[0064] Alternatively, in another example, the leak detector data acquisition and transmission system 300 and the wireless terminal may be separate devices, and the leak detector data acquisition and transmission system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0065] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A leak detector data collection and transmission method, characterized in that: include: In response to an environmental noise learning instruction, collecting original analog sound waves through an ultrasonic microphone, and performing data processing on the original analog sound waves to obtain a noise fingerprint; Acquire the leak detection acoustic wave digital stream collected by the measuring instrument; Based on the noise fingerprint, the digital stream of the leak detection sound wave is subjected to the noise fingerprint denoising to obtain the noise-reduced leak detection sound wave spectrum; Performing feature extraction on the noise-reduced leak detection sound wave spectrum to obtain a noise-reduced leak detection sound wave feature vector; Generate inference results based on the noise-reduced leak detection acoustic wave feature vector; In response to the inference result that a leak exists and the confidence level is greater than a preset threshold, the leak detection acoustic wave digital stream is transmitted to the cloud.

2. The leak detector data acquisition and transmission method according to claim 1, characterized in that: Perform data processing on the original simulated sound wave to obtain the noise fingerprint, including: Performing ADC conversion on the original analog sound wave to obtain the original sound wave digital stream; Performing frame processing on the original sound wave digital stream to obtain a sequence of original sound wave digital frames; Performing a fast Fourier transform on each original sound wave digital frame in the sequence of original sound wave digital frames to obtain a sequence of power spectrum data distribution; Performing power spectrum density calculation on each power spectrum data distribution in the power spectrum data distribution sequence to obtain a power spectrum density sequence; The power spectral density sequence is averaged and smoothed to obtain the average energy spectrum as the noise fingerprint.

3. The leak detector data acquisition and transmission method according to claim 1, characterized in that: Based on the noise fingerprint, the digital stream of the leak detection sound wave is subjected to voiceprint fingerprint noise reduction to obtain the noise-reduced leak detection sound wave spectrum, including: Performing a fast Fourier transform on the leakage detection acoustic wave digital stream to obtain the current spectrum of the side leakage detection acoustic wave signal; Calculate the differential characteristics of the current spectrum of the side leakage detection acoustic wave signal and the noise fingerprint to obtain the spectrum characteristics of the leakage detection acoustic wave after noise reduction; Performing feature distribution enhancement on the noise-reduced leakage detection sound wave spectrum coding vector to obtain an enhanced noise-reduced leakage detection sound wave spectrum coding vector; Based on the enhanced noise-reduced leakage detection acoustic wave spectrum coding vector, the noise-reduced leakage detection acoustic wave spectrum is obtained.

4. The leak detector data acquisition and transmission method according to claim 3, characterized in that: Calculate the differential features of the current spectrum of the side leakage detection acoustic wave signal and the noise fingerprint to obtain the spectrum features of the leakage detection acoustic wave after noise reduction, including: Performing spectrum encoding on the current spectrum and noise fingerprint of the side leakage detection acoustic wave signal to obtain a current spectrum feature vector of the side leakage detection acoustic wave signal and a noise fingerprint encoding vector; The position difference between the current spectrum feature vector of the side leakage detection sound wave signal and the noise fingerprint coding vector is calculated to obtain the spectrum coding vector of the leakage detection sound wave after noise reduction as the spectrum feature of the leakage detection sound wave after noise reduction.

5. The leak detector data acquisition and transmission method according to claim 3, characterized in that: The feature distribution enhancement is performed on the noise-reduced leakage detection sound wave spectrum coding vector to obtain an enhanced noise-reduced leakage detection sound wave spectrum coding vector, including: Performing local sound spectrum semantic perception on the noise-reduced leakage detection sound wave spectrum coding vector to obtain a local semantic perception coding feature map of the noise-reduced leakage detection sound wave spectrum; The spatial dimension of the local semantic perception coding feature map of the noise-reduced leak detection sound wave spectrum is marked and the feature shape is flattened to obtain the enhanced noise-reduced leak detection sound wave spectrum coding vector.

6. The leak detector data acquisition and transmission method according to claim 3, characterized in that: Based on the enhanced noise-reduced leakage detection acoustic wave spectrum coding vector, the noise-reduced leakage detection acoustic wave spectrum is obtained, including: The enhanced noise-reduced leakage detection acoustic wave spectrum encoding vector is feature decoded to obtain the noise-reduced leakage detection acoustic wave spectrum.

7. The leak detector data collection and transmission method according to claim 1, characterized in that: Feature extraction is performed on the noise-reduced leak detection sound wave spectrum to obtain a feature vector of the noise-reduced leak detection sound wave, including: The noise-reduced leak detection acoustic wave spectrum is passed through a leak detection acoustic wave feature extractor based on a convolutional neural network to obtain a noise-reduced leak detection acoustic wave feature vector.

8. A leak detector data acquisition and transmission system, characterized in that: include: a noise fingerprint acquisition module, configured to respond to an environmental noise learning instruction, collect original analog sound waves through an ultrasonic microphone, and perform data processing on the original analog sound waves to obtain a noise fingerprint; A leak detection acoustic wave digital stream acquisition module is used to acquire the leak detection acoustic wave digital stream collected by the measuring instrument; A voiceprint fingerprint noise reduction module is used to perform voiceprint fingerprint noise reduction on the leak detection sound wave digital stream based on the noise fingerprint to obtain the leak detection sound wave spectrum after noise reduction; A leak detection acoustic wave feature extraction module is used to extract features from the noise-reduced leak detection acoustic wave spectrum to obtain a noise-reduced leak detection acoustic wave feature vector; An inference result generation module, used to generate an inference result based on the noise-reduced leak detection acoustic wave feature vector; The leakage detection acoustic wave digital stream transmission module is used to transmit the leakage detection acoustic wave digital stream to the cloud in response to the inference result that there is a leak and the confidence level is greater than a preset threshold.

Images