Real-time denoising method for disposable sound velocity profiler for ocean underwater sound velocity measurement

Abstract

The application discloses a disposable sound velocity meter real-time denoising method for ocean underwater sound velocity measurement and relates to the field of ocean underwater acoustic detection, and aims at solving the problem that the existing disposable sound velocity meter is easy to lose the sound velocity mutation characteristics during denoising. The application collects underwater original sound velocity data and constructs a sliding window, obtains low-frequency approximate components and high-frequency detail components through three-stage Haar wavelet decomposition; constructs a protection factor according to the local gradient characteristics of the sound velocity, and adaptively corrects the wavelet denoising threshold; combines the MAD algorithm to solve the optimal threshold through layered noise estimation, processes the high-frequency components through an improved exponential decay threshold function, realizes the smooth transition of the soft and hard thresholds, overcomes the distortion defect of the traditional threshold, and reconstructs the denoised sound velocity data through wavelet inverse transformation. The application can adapt to the embedded real-time processing scene of the disposable sound velocity meter, takes into account the noise suppression effect and the sound velocity profile effective feature reservation capability, and effectively improves the precision and reliability of the ocean underwater sound velocity measurement data.

Description

Technical Field

[0001] This invention relates to the field of marine acoustic detection, and more particularly to a real-time noise reduction method for discardable sound velocity meters for underwater sound velocity measurement in the ocean. Background Technology

[0002] Underwater sound velocity profiles are crucial parameters for underwater communication, sonar detection, underwater target localization, multibeam bathymetry, marine mapping, and sound propagation model calculations. Disposable sound velocity meters offer advantages such as convenient deployment, low cost, no need for retrieval, and suitability for rapidly acquiring vertical sound velocity profiles, making them essential equipment in ocean sound velocity measurement. In real-world ocean measurement environments, sound velocity data is easily affected by ocean turbulence, bubble interference, sensor jitter, circuit noise, and ADC sampling errors, leading to random noise, spike anomalies, and local distortions in the sound velocity profile. Traditional fixed threshold filtering, mean filtering, or offline processing methods suffer from insufficient real-time performance, poor environmental adaptability, tendency to over-smooth, and difficulty in preserving thermoclines and abrupt changes in sound velocity gradients. Existing wavelet thresholding denoising methods often suffer from discontinuous hard threshold functions at the threshold, leading to reconstruction oscillations; while soft threshold functions are continuous, they introduce constant deviations for large effective coefficients, weakening the true changes in sound velocity. Therefore, there is a need for a processing method that is suitable for underwater sound velocity measurement scenarios, can run in real time at the acquisition end of a discardable sound velocity meter, and balances noise reduction and sound velocity profile feature fidelity. Summary of the Invention

[0003] This invention provides a real-time denoising method for discardable sound velocity meters for underwater sound velocity measurement in the ocean, in order to solve the problem that existing discardable sound velocity meters are prone to losing sound velocity change characteristics during denoising and are difficult to adapt to underwater embedded real-time measurement.

[0004] A real-time noise reduction method for discardable sound velocity meters used in underwater ocean sound velocity measurement includes the following steps: S1. Collect raw underwater sound velocity measurement data, construct a sliding sound velocity data window, and perform three-level Haar wavelet decomposition on the sound velocity data window to obtain low-frequency approximate components and multi-band high-frequency detail components respectively. S2. Extract smooth gradient features based on local variation features of sound velocity sequence, adaptively calculate gradient protection factor, and complete dynamic protection correction of wavelet threshold. S3. The Median Absolute Deviation (MAD) algorithm is used to estimate the noise level of high-frequency detail components in each layer, and the gradient protection factor is combined to adaptively solve the final denoising threshold of each layer. S4. An improved exponential decay threshold function is used to adaptively denoise the high-frequency detail components. The low-frequency approximation components and the denoised high-frequency detail components are reconstructed using Haar wavelet inverse transform to obtain the denoised sound velocity profile data. Real-time denoised underwater sound velocity profile data is output.

[0005] Further, in step S1, collecting raw underwater sound velocity measurement data and constructing a sliding sound velocity data window includes: The ultrasonic transducer in the jettisonable velocimeter transmits and receives sound wave signals, and the time-to-digital converter chip measures the sound wave flight time. The underwater sound velocity data is then calculated based on the transducer spacing and flight time. : ; In the formula, No. The speed of sound measured this time, The distance between the transducer and the reflector. For the first The measured ultrasonic flight time The fixed delay introduced into the transducer, circuit, and signal transmission link was determined through calibration experiments; The continuously collected sound velocity data is written into the data buffer of the embedded processor to form a fixed-length sound velocity data window: ; In the formula, The total number of sound velocity sampling points contained within the sliding sound velocity data window and Pick ; Each time a new sound velocity data is acquired, the window slides forward one sampling point to complete the real-time update: ; In the formula, This is the iteration number for the data window sliding.

[0006] Further, in S1, the sound velocity data window is subjected to three-level Haar wavelet decomposition to obtain low-frequency approximate components and multi-band high-frequency detail components, including: Perform a three-level wavelet decomposition on the sound velocity data window. First-level decomposition: ; ; In the formula, The sequence number in the sliding sound velocity data window is The original sound speed sample value, The sequence number in the sliding sound velocity data window is The original sound speed sample value, These are the first-order low-frequency approximation coefficients. For the first level of high-frequency detail coefficients, The coefficient index within the wavelet hierarchy and ; Second-level decomposition: ; ; In the formula, These are the second-order low-frequency approximation coefficients. For the second level of high-frequency detail coefficients and ; Third-level decomposition: ; ; In the formula, These are the third-order low-frequency approximation coefficients. For the third level of high-frequency detail coefficients and ; After Haar third-order wavelet decomposition, the original data window is decomposed into... , This represents the low-frequency trend component of the sound velocity profile. For details in the lower frequency band; For detailed changes in the mid-frequency range; This refers to high-frequency, detailed changes.

[0007] Further, in S2, extracting smooth gradient features based on local variation features of the sound speed sequence includes: The reference sound velocity value for each sampling point is obtained by neighborhood median filtering. Reference sound velocity value corresponding to each sampling point Calculated using the median: ; In the formula, For the first The original sound velocity value corresponding to each sampling point. For the first The reference sound velocity value corresponding to each sampling point. To take the median, The radius of the filtering neighborhood; Calculate the local sound velocity variation characteristic of adjacent sampling points based on the change between adjacent sampling points in a continuous sound velocity sampling sequence: ; In the formula, For the first Local sound velocity variation characteristic of the sequence domain of the point and , For the first The and the first The reference sound velocity value corresponding to each sampling point; right Neighborhood averaging is performed, and for sampling points at the data window boundaries that are less than the neighborhood length, endpoint copying is used to pad them. ; In the formula, For the first The smooth local sound velocity variation characteristic near each sampling point The smoothing radius; Within the current sound velocity data window, take the maximum value of the smooth local sound velocity variation characteristic: ; In the formula, This represents the maximum value of all smooth gradient features within the current sliding sound velocity data window.

[0008] Furthermore, in S2, the gradient protection factor is adaptively calculated to complete the dynamic protection correction of the wavelet threshold, including: Set gradient protection trigger threshold Define the gradient protection factor: ; In the formula, This is the gradient protection factor corresponding to the current data window. To protect the intensity adjustment parameters, This is the lower limit of the protective factor. The threshold for gradient protection; Gradient protection trigger threshold The calculation formula is: ; ; ; In the formula, The threshold for gradient protection triggering. For the first The smooth local sound velocity variation characteristic near each sampling point The trigger threshold sensitivity adjustment coefficient, It is a median function. Median absolute deviation function.

[0009] Furthermore, S3 specifically includes: Noise estimation of high-frequency detail coefficients for each layer is performed based on median absolute deviation (MAD). Let the first layer be... High-frequency detail factor set for: ; In the formula, The wavelet decomposition level is denoted as . For the first Number of layer detail coefficients For the first The first in the layer One high-frequency detail coefficient; The total number of sampling points in the sliding sound velocity data window is Then the first The number of high-frequency detail coefficients in the layer meets the following requirements: ; No. Median absolute deviation of layer detail coefficients: ; In the formula, For the first Median of high-frequency detail coefficients for each layer; The first Median of high-frequency detail coefficients of a layer for: ; Estimate the first based on the median absolute deviation Layer noise standard deviation: ; In the formula, For the first Layer-based noise standard deviation estimation; No. Layer-based adaptive threshold: ; In the formula, For the first Layer-based adaptive threshold, For the first Number of layer detail coefficients; Using gradient protection factor for the first Layer-based adaptive threshold correction: ; In the formula, For final use in processing the first The noise reduction threshold for high-frequency detail coefficients of a layer.

[0010] Furthermore, in step S4, an improved exponential decay threshold function is used to adaptively denoise the high-frequency detail components, including: A modified exponential decay threshold function is used to process the high-frequency detail coefficients. The modified exponential decay threshold function is as follows: ; In the formula, The detail coefficients after thresholding. For symbolic functions, For exponential decay parameters; The symbol function is defined as follows: .

[0011] Furthermore, the exponential decay parameter of the improved exponential decay threshold function possesses adaptive soft and hard threshold transition characteristics, specifically including: when As the value approaches 0 in the positive direction, the exponential term approaches 1, and the improved threshold function approaches the traditional soft threshold function, eliminating the problem of discontinuous abrupt changes at the threshold of the hard threshold function; when As the value increases, the rate of exponential decay accelerates, and the amplitude deviation of the high-frequency detail coefficients of the large value decays exponentially and approaches the original value.

[0012] Furthermore, in step S4, the low-frequency approximation component and the denoised high-frequency detail component are used to perform Haar wavelet inverse transform reconstruction to obtain the denoised sound velocity profile data, specifically including: Let the first The low-frequency approximation coefficient of the layer is The high-frequency detail coefficients after thresholding are Then the first The layer approximation coefficient is recovered by the following formula: ; ; In the formula, The first wavelet inverse reconstruction obtained Even-numbered low-frequency approximation coefficients for each layer The first wavelet inverse reconstruction obtained Odd-digit low-frequency approximation coefficients for each layer. ; The initial conditions for reconstruction are: ; Using the third-order low-frequency approximation coefficient and the processed third-level high-frequency detail coefficient The second-order approximation coefficients are obtained through reconstruction: ; ; In the formula, The second-level low-frequency approximation coefficients are the even-numbered coefficients obtained by reconstructing the third-level low-frequency approximation coefficients and the processed third-level high-frequency detail coefficients through inverse wavelet transform. The odd-numbered coefficients of the second-level low-frequency approximation coefficients are obtained by reconstructing the third-level low-frequency approximation coefficients and the processed third-level high-frequency detail coefficients through inverse wavelet transform. ; Using the second-order low-frequency approximation coefficient and the processed second-level high-frequency detail coefficients The first-order approximation coefficients are obtained by reconstruction: ; ; In the formula, The even-numbered coefficients of the first-order low-frequency approximation coefficients obtained by inverse wavelet transform reconstruction. These are the odd-numbered coefficients of the first-order low-frequency approximation coefficients obtained by inverse wavelet transform reconstruction. Using first-order approximation coefficients and the processed first-level high-frequency detail coefficients The reconstructed sound velocity data is then obtained: ; ; In the formula, After reconstruction by inverse wavelet transform, the th The sound velocity sample value corresponding to each sampling point after denoising. After reconstruction by inverse wavelet transform, the th The sound velocity sampled after denoising each sampling point ; After a three-stage Haar wavelet inverse transform, the denoised sound velocity data window is obtained: .

[0013] Furthermore, in step S4, the output of real-time denoised underwater sound velocity profile data includes: After completing the Haar wavelet inverse transform reconstruction to obtain continuous denoised speed data, the embedded main controller frames the denoised speed sample values, sample numbers, and pressure depth information, and outputs real-time denoised underwater sound speed profile data matching the depth position, thus achieving simultaneous completion of sound speed profile noise suppression and effective feature fidelity.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention moves the noise reduction processing of sound velocity data to the acquisition end of the disposable sound velocity meter, enabling real-time noise reduction calculations to be performed locally on the device. This significantly improves the real-time data processing capability at the marine site, reduces reliance on offline post-processing by the host computer, and is suitable for autonomous field exploration conditions of the disposable sound velocity meter. It also employs an improved exponential decay threshold function for wavelet filtering, overcoming the inherent defects of discontinuous hard threshold functions and fixed amplitude deviations in soft threshold functions. The filtered output curve is continuous and smooth, effectively optimizing the accuracy of sound velocity data correction.

[0015] Simultaneously suppressing marine environmental noise, circuit noise, and sampling errors at the acquisition end significantly improves the signal-to-noise ratio and long-term operational stability of underwater sound velocity measurement data, effectively reduces measurement errors caused by various interferences, and comprehensively enhances the real-time measurement performance and field engineering application value of the discardable sound velocity meter. Attached Figure Description

[0016] Figure 1 A comparison of sound velocity profile data processed by different noise reduction methods; Figure 2 This is a magnified comparison of the fine structure of local sound velocity; Figure 3 A comparison of the local sound velocity gradient preservation capabilities of different denoising methods; Figure 4 A comparison chart of sound velocity errors from different methods; Figure 5 This is a flowchart of the technology of the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0018] Example 1 Real-time noise reduction methods for discardable sound velocity meters used in underwater ocean sound velocity measurement, such as Figure 5 As shown, it includes the following steps: The ultrasonic transducer in a discardable sound velocity meter transmits and receives sound wave signals, and a time-to-digital converter chip measures the sound wave flight time. The underwater sound velocity data is then calculated based on the transducer spacing and flight time. The calculation formula is as follows: ; In the formula, No. The speed of sound measured this time, The distance between the transducer and the reflector. For the first The measured ultrasonic flight time The fixed delay introduced into the transducer, circuit, and signal transmission link was determined through calibration experiments; The continuously collected sound velocity data is written into the data buffer of the embedded processor to form a fixed-length sound velocity data window: ; In the formula, The total number of sound velocity sampling points contained within the sliding sound velocity data window and Pick ;For example Wait, for each new sound velocity data point acquired, the window slides forward one sampling point: ; In the formula, This is the iteration number for the data window sliding.

[0019] Multi-level Haar wavelet decomposition is performed on the sound velocity data window to obtain low-frequency approximation coefficients and multi-level high-frequency detail coefficients.

[0020] First-level decomposition: ; ; In the formula, The sequence number in the sliding sound velocity data window is The original sound speed sample value, The sequence number in the sliding sound velocity data window is The original sound speed sample value, These are the first-order low-frequency approximation coefficients. For the first level of high-frequency detail coefficients, The coefficient index within the wavelet hierarchy and ; Second-level decomposition: ; ; In the formula, These are the second-order low-frequency approximation coefficients. For the second level of high-frequency detail coefficients and ; Third-level decomposition: ; ; In the formula, These are the third-order low-frequency approximation coefficients. For the third level of high-frequency detail coefficients and ; After three levels of Haar wavelet decomposition, the original data window is decomposed into: ; In the formula, This represents the low-frequency trend component of the sound velocity profile. For the detailed changes in lower frequency bands, For the detailed changes in the mid-frequency range, For high-frequency detail changes; In underwater sound speed measurement in the ocean It mainly reflects the overall trend of sound speed changing with depth. , , It contains noise components caused by bubble interference, sensor jitter, circuit noise, and ADC sampling errors, as well as effective detailed information corresponding to thermocline, halocline, or abrupt changes in sound velocity gradient.

[0021] A sound velocity gradient protection correction mechanism is further introduced, adjusting the wavelet threshold based on the local variation characteristics in the continuous sound velocity sampling sequence. When a large gradient change is detected in the sound velocity profile, it is determined that the region may correspond to a thermocline, halocline, or sound velocity jump region, and the wavelet threshold of the corresponding data window is reduced to weaken the suppression of effective detail coefficients by threshold processing. This allows the sequence-domain local sound velocity variation characteristics of the ocean sound velocity profile to be preserved while suppressing noise.

[0022] Calculate the local sound velocity variation characteristic in the sequence domain based on the change between adjacent sampling points in a continuous sound velocity sampling sequence: ; In the formula, For the first Local sound velocity variation characteristic near the point and , For the first The and the first The reference sound velocity value corresponding to each sampling point; No. Reference sound velocity value corresponding to each sampling point Calculated using the median: ; In the formula, For the first The original sound velocity value corresponding to each sampling point. For the first The reference sound velocity value corresponding to each sampling point. To take the median, This is the radius of the filtering neighborhood, typically taken as 1 or 2.

[0023] Median filtering can reduce isolated spikes caused by factors such as bubble interference and transient circuit noise, while better preserving the local variation trend of the sound velocity profile.

[0024] At the same time, to avoid small fluctuations affecting the judgment, Neighborhood averaging is performed, and for sampling points at the data window boundaries that are less than the neighborhood length, endpoint copying is used to pad them. ; In the formula, For the first The characteristic quantity of smooth local sound velocity variation near each sampling point; The smoothing radius is typically set to 1. at this time for: ; This processing can enhance the algorithm's ability to identify continuously changing regions and reduce misjudgments caused by single-point noise.

[0025] Within the current sound velocity data window, take the maximum value of the smooth local sound velocity variation characteristic: ; In the formula, This represents the maximum value of all smooth gradient features within the current sliding sound velocity data window; when A larger value indicates that the current window may contain sound velocity profile variation features that need to be protected.

[0026] Set gradient protection trigger threshold Define the gradient protection factor: ; In the formula, This is the gradient protection factor corresponding to the current data window. To protect the intensity adjustment parameters, which are used to control the rate at which the denoising threshold decreases after local changes exceed the trigger threshold, This is the lower limit of the protection factor, used to prevent the threshold from decreasing excessively. The threshold for gradient protection; Gradient protection trigger threshold The calculation formula is: ; ; ; In the formula, The threshold for gradient protection triggering. For the first The smooth local sound velocity variation characteristic near each sampling point The trigger threshold sensitivity adjustment coefficient is used to control the triggering difficulty of the gradient protection mechanism. It was determined experimentally. It is a median function. Median absolute deviation function The total number of sound velocity sampling points contained within the sliding sound velocity data window is a positive integer that satisfies the requirements of Haar wavelet multi-level decomposition, preferably a power of 2. If the number is even, then the average of the two middle numbers after sorting is taken.

[0027] when As the value increases, the gradient protection mechanism becomes more difficult to trigger; when When the gradient is reduced, the gradient protection mechanism is more likely to be triggered.

[0028] After three-level Haar wavelet decomposition, the original sound velocity data window is decomposed into ; Noise estimation of high-frequency detail coefficients for each layer is performed based on median absolute deviation (MAD). Let the first layer be... High-frequency detail factor set for: ; In the formula, The wavelet decomposition level is denoted as . For the first Number of layer detail coefficients For the first The first in the layer One high-frequency detail coefficient; The total number of sampling points in the sliding sound velocity data window is Then the first The number of high-frequency detail coefficients in the layer meets the following requirements: ; No. Median absolute deviation of layer detail coefficients: ; In the formula, For the first Median of high-frequency detail coefficients for each layer; The first Median of high-frequency detail coefficients of a layer for: ; Compared to the mean and standard deviation, MAD is less susceptible to small abnormal spikes, making it more suitable for ocean sound velocity measurement scenarios with bubble interference, sensor jitter, and circuit transient noise.

[0029] Estimate the first based on the median absolute deviation Layer noise standard deviation: ; In the formula, For the first Layer-based noise standard deviation estimation; When the noise in a certain layer is strong and As the noise level increases, both decrease accordingly when the noise level is low.

[0030] The first threshold is calculated using a general threshold method. Layer-based adaptive threshold: ; In the formula, For the first Layer-based adaptive threshold, For the first Number of layer detail coefficients; Subsequently, the gradient protection factor was used to... Layer-based adaptive threshold correction: ; In the formula, For final use in processing the first The noise reduction threshold for high-frequency detail coefficients of a layer.

[0031] An improved exponential decay threshold function is used to process high-frequency detail coefficients. This function provides a continuous transition at the threshold and causes the threshold deviation of large-amplitude wavelet coefficients to decay exponentially with increasing coefficient amplitude.

[0032] No. Layer The high-frequency detail factor is The processed high-frequency detail coefficient is ; The improved exponential decay threshold function is then: ; In the formula, The detail coefficients after thresholding. For symbolic functions, For exponential decay parameters; when At that time, directly ordered , Preset to positive numbers; For exponential decay parameters, and This parameter is used to adjust the speed at which the improved threshold function transitions from soft threshold characteristics to hard threshold characteristics.

[0033] when When the exponent term is close to 1, the improved threshold function approaches the soft threshold function: ; when As the value gradually increases, the rate of decay of the exponential term increases, and the detail coefficients with larger amplitudes are closer to the original values. Therefore: When the size is smaller, the noise reduction result is smoother; When the value is large, the ability to preserve the true characteristics of sound speed changes is stronger; generally Preferred ; The symbolic function is defined as: ; The logic is as follows: when the magnitude of the detail coefficient is less than the threshold: ,make ; This part of the coefficients mainly corresponds to environmental noise, circuit noise, and sampling error, and therefore is suppressed.

[0034] When the detail factor magnitude is greater than or equal to the threshold Instead of directly retaining the original coefficients or using a fixed-amplitude reduction method, it introduces an exponential decay term. The exponential decay term gradually decreases as the magnitude of the detail coefficient increases.

[0035] Compared to traditional hard thresholds: ; Traditional hard thresholding functions exhibit abrupt changes at the threshold position.

[0036] Compared to traditional soft thresholds: ; Although the soft threshold function is continuous, all retained coefficients are reduced by a fixed amount.

[0037] After processing, the result is , and ; Using low-frequency approximation coefficients and processed high-frequency detail coefficients , and Perform Haar wavelet inverse transform to reconstruct the denoised underwater sound velocity data.

[0038] The high-frequency detail coefficients and low-frequency approximation coefficients after thresholding are subjected to inverse wavelet transform to obtain the denoised sound velocity data.

[0039] Let the first The low-frequency approximation coefficient of the layer is The high-frequency detail coefficients after thresholding are Then the first The layer approximation coefficient is recovered by the following formula: ; ; In the formula, The first wavelet inverse reconstruction obtained Even-numbered low-frequency approximation coefficients for each layer The first wavelet inverse reconstruction obtained Odd-digit low-frequency approximation coefficients for each layer. ; The initial conditions for reconstruction are: ; Using the third-order low-frequency approximation coefficient and the processed third-level high-frequency detail coefficient The second-order approximation coefficients are obtained through reconstruction: ; ; In the formula, The second-level low-frequency approximation coefficients are the even-numbered coefficients obtained by reconstructing the third-level low-frequency approximation coefficients and the processed third-level high-frequency detail coefficients through inverse wavelet transform. The odd-numbered coefficients of the second-level low-frequency approximation coefficients are obtained by reconstructing the third-level low-frequency approximation coefficients and the processed third-level high-frequency detail coefficients through inverse wavelet transform. ; Using the second-order low-frequency approximation coefficient and the processed second-level high-frequency detail coefficients The first-order approximation coefficients are obtained by reconstruction: ; ; In the formula, The even-numbered coefficients of the first-order low-frequency approximation coefficients obtained by inverse wavelet transform reconstruction. These are the odd-numbered coefficients of the first-order low-frequency approximation coefficients obtained by inverse wavelet transform reconstruction. Using first-order approximation coefficients and the processed first-level high-frequency detail coefficients The reconstructed sound velocity data is then obtained: ; ; In the formula, After reconstruction by inverse wavelet transform, the th The sound velocity sample value corresponding to each sampling point after denoising. After reconstruction by inverse wavelet transform, the th The sound velocity sampled after denoising each sampling point .

[0040] After a three-stage Haar wavelet inverse transform, the denoised sound velocity data window is obtained: ; by For example, the reconstruction process is as follows: ; After completing the Haar wavelet inverse transform reconstruction to obtain continuous denoised speed data, the embedded main controller frames the denoised speed sample values, sample numbers, and pressure depth information, and outputs real-time denoised underwater sound speed profile data matching the depth position, thus achieving simultaneous completion of sound speed profile noise suppression and effective feature fidelity.

[0041] Example 2 Experimental verification: The algorithm was verified using MATLAB. ArgoGDAC deep-sea temperature and salinity profile data was used as the original data. To better reflect the actual situation, local sound velocity structures were added to the original data: simulating thermocline, halocline, and local water masses. Background random measurement noise, non-Gaussian heavy-tailed noise, bubble pulse interference, bubble cluster disturbance, and low-frequency ocean current disturbance were added. The final original noisy measurement data was used for noise reduction verification. like Figure 1 As shown, Figure 1 The comparison results show the sound velocity profile data processed by different denoising methods. The horizontal axis represents sound velocity, and the vertical axis represents Argo profile pressure. The black curve represents the true sound velocity profile, the gray scatter points represent the noisy measurement data after adding bubbles, ocean current disturbances, and ultrasonic measurement errors, the green dashed line represents the result of traditional hard thresholding, the blue dashed line represents the result of traditional soft thresholding, and the red curve represents the result of the method proposed in this paper.

[0042] from Figure 1 As can be seen, the noisy measurement data exhibits a significant discrete distribution near the true sound velocity profile, with large deviations in local locations. This indicates that bubble disturbances, ocean current disturbances, and random measurement errors cause fluctuations in the sound velocity measurements. After wavelet denoising, all methods can suppress noise to some extent, bringing the processed results closer to the true sound velocity profile.

[0043] Simulation results show that in sound velocity profile data containing bubble disturbances, ocean current disturbances, and random errors in ultrasonic measurements, traditional hard thresholding methods are prone to producing local jumps, while traditional soft thresholding methods tend to weaken the local variation characteristics of the sound velocity profile. In contrast, the proposed method, through pressure domain gradient protection correction and improved exponential decay thresholding, can better preserve the local variation morphology of the sound velocity profile while suppressing noise fluctuations, making the denoising results closer to the true sound velocity profile. This verifies the effectiveness of the proposed method for underwater sound velocity measurement in the ocean.

[0044] To more intuitively compare the actual effects of the various methods, Figure 2The magnified comparison of local sound velocity data reveals that, within pressure ranges exhibiting significant local variations in the sound velocity profile, traditional hard thresholding methods tend to produce step-like jumps, affecting the continuity of the sound velocity profile. While traditional soft thresholding methods offer better smoothness, they can weaken the true local sound velocity structure. In contrast, the proposed method, through pressure domain gradient protection correction and improved exponential decay thresholding, effectively suppresses bubble, current disturbance, and ultrasonic measurement errors while better preserving the changing trend of the local sound velocity structure, making the processed results closer to the true sound velocity profile within local variation ranges. This result demonstrates that the proposed method is more suitable for local structure-preserving noise reduction in underwater sound velocity profile measurements.

[0045] exist Figure 3 In the comparison of local gradient preservation capabilities, traditional hard and soft thresholding methods still tend to generate abnormal gradient spikes after denoising, causing the local rate of change of the sound velocity profile to deviate from the true profile and potentially leading to false local abrupt changes in sound velocity. In contrast, the local rate of change of sound velocity processed by the method in this paper is more consistent with the true sound velocity profile gradient, effectively suppressing false gradients caused by bubbles, ocean current disturbances, and ultrasonic measurement errors, while preserving the local variation characteristics of the true sound velocity profile. This result verifies the effectiveness of the pressure domain gradient protection correction mechanism in sound velocity profile denoising.

[0046] exist Figure 4 The comparison chart of sound velocity errors from different methods shows that the error distribution of noisy measurement data is relatively dispersed, with some pressure position errors exceeding [a certain threshold]. This indicates that bubbles, ocean current disturbances, and random errors in ultrasonic measurements can lead to significant deviations in sound velocity.

[0047] After processing with traditional hard and soft thresholding, the error converges somewhat compared to the noisy data, but significant deviations still occur in some pressure ranges. For example, in shallow layers and local variation ranges, the error curves of traditional methods show obvious jumps, indicating that they may introduce local distortion or over-smoothing during the denoising process. The red curve corresponding to the method in this paper is more concentrated around 0, with a relatively smaller error fluctuation range, indicating that the sound velocity profile processed by the method in this paper is closer to the true sound velocity profile overall.

[0048] Error comparison results show that the sound velocity measurement data containing noise exhibits significant discrete errors at different pressure locations. While traditional hard and soft thresholding methods can reduce some of the noise impact, they still suffer from large local errors and abrupt error changes. The sound velocity error processed by the method presented in this paper is more concentrated near zero, with a smaller error fluctuation range, indicating that this method can more effectively suppress the influence of bubbles, ocean current disturbances, and ultrasonic measurement errors on the sound velocity measurement results.

[0049] Finally, by comparing specific data, we can see that, in terms of overall RMSE and SNR: The original noisy data had an RMSE of 0.7209 and an SNR of 66.33 dB; the traditional hard threshold had an RMSE of 0.6046 and an SNR of 67.86 dB; the traditional soft threshold had an RMSE of 0.6026 and an SNR of 67.89 dB; and the method in this paper had an RMSE of 0.4539 and an SNR of 70.35 dB. The method presented in this paper has the lowest overall error and the highest signal-to-noise ratio.

[0050] In terms of average error and maximum error, the overall MAE is: Hard threshold = 0.4324; Soft threshold = 0.4262; Method in this paper = 0.3481; Maximum absolute error: Hard threshold = 2.2622; Soft threshold = 2.9215; Method in this paper = 1.4997; This demonstrates that the method presented in this paper not only reduces the average error but also significantly reduces the extreme error. In particular, the maximum absolute error is reduced from 2.9215 m / s with the traditional soft threshold to 1.4997 m / s, indicating that the method presented in this paper has a better ability to suppress abnormal measurement points such as bubbles and ocean current disturbances.

[0051] In gradient preservation capability, the RMSE of the gradient in the high gradient region is: Hard threshold = 0.2512; Soft threshold = 0.2890; Method in this paper = 0.1167; The proposed method exhibits the lowest gradient error, representing an improvement of approximately 59.63% compared to traditional soft thresholding. This demonstrates that the proposed method can more accurately recover the local rate of change of the true sound velocity profile, avoiding the generation of spurious gradients or the weakening of the true gradient by traditional methods.

[0052] Simulation results show that, in sound velocity profile data incorporating bubble disturbances, ocean current disturbances, and ultrasonic measurement errors, the proposed method exhibits lower overall error and higher signal-to-noise ratio compared to traditional hard thresholding and soft thresholding methods. Specifically, the overall RMSE of the proposed method is 0.4539 m / s, a reduction of approximately 24.67% compared to the traditional soft thresholding method; the RMSE in the local structure / high gradient region is 0.6282 m / s, a reduction of approximately 28.32% compared to the traditional soft thresholding method; and the gradient RMSE in the high gradient region is 0.1167, a reduction of approximately 59.63% compared to the traditional soft thresholding method. These results demonstrate that the proposed method not only effectively suppresses complex non-Gaussian measurement interference but also better preserves the true local variation characteristics of the sound velocity profile, making it suitable for underwater sound velocity profile measurement scenarios in the ocean.

[0053] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A real-time noise reduction method for discardable sound velocity meters used in underwater sound velocity measurement in the ocean, characterized in that, Includes the following steps: S1. Collect raw underwater sound velocity measurement data, construct a sliding sound velocity data window, and perform three-level Haar wavelet decomposition on the sound velocity data window to obtain low-frequency approximate components and multi-band high-frequency detail components respectively. S2. Extract smooth gradient features based on local variation features of sound velocity sequence, adaptively calculate gradient protection factor, and complete dynamic protection correction of wavelet threshold. The adaptive calculation of the gradient protection factor, which performs dynamic protection correction of the wavelet threshold, includes: Set gradient protection trigger threshold Define the gradient protection factor: ； In the formula, This is the gradient protection factor corresponding to the current data window. To protect the intensity adjustment parameters, This is the lower limit of the protective factor. The threshold for gradient protection triggering. This represents the maximum value of all smooth gradient features within the current sliding sound velocity data window; Gradient protection trigger threshold The calculation formula is: ； ； ； In the formula, The threshold for gradient protection triggering. For the first The smooth local sound velocity variation characteristic near each sampling point The trigger threshold sensitivity adjustment coefficient, It is a median function. Median absolute deviation function; S3. The Median Absolute Deviation (MAD) algorithm is used to estimate the noise level of high-frequency detail components in each layer, and the gradient protection factor is combined to adaptively solve the final denoising threshold of each layer. S3 specifically includes: Noise estimation of high-frequency detail coefficients for each layer is performed based on median absolute deviation (MAD). Let the first layer be... High-frequency detail factor set for: ； In the formula, The wavelet decomposition level is denoted as . For the first Number of layer detail coefficients For the first The first in the layer One high-frequency detail coefficient; The total number of sampling points in the sliding sound velocity data window is Then the first The number of high-frequency detail coefficients in the layer meets the following requirements: ； No. Median absolute deviation of layer detail coefficients: ； In the formula, For the first Median of high-frequency detail coefficients for each layer; The first Median of high-frequency detail coefficients of a layer for: ； Estimate the first based on the median absolute deviation Layer noise standard deviation: ； In the formula, For the first Layer-based noise standard deviation estimation; No. Layer-based adaptive threshold: ； In the formula, For the first Layer-based adaptive threshold, For the first Number of layer detail coefficients; Using gradient protection factor for the first Layer-based adaptive threshold correction: ； In the formula, For final use in processing the first Denoising threshold for high-frequency detail coefficients in a layer; S4. Adaptive denoising of high-frequency detail components is performed using an improved exponential decay threshold function. The low-frequency approximation components and the denoised high-frequency detail components are reconstructed using Haar wavelet inverse transform to obtain the denoised sound velocity profile data. Real-time denoised underwater sound velocity profile data is output. The adaptive denoising process for high-frequency detail components using an improved exponential decay threshold function includes: A modified exponential decay threshold function is used to process the high-frequency detail coefficients. The modified exponential decay threshold function is as follows: ； In the formula, The detail coefficients after thresholding. For symbolic functions, For exponential decay parameters; The symbol function is defined as follows: ； The improved exponential decay threshold function has an adaptive soft and hard threshold transition characteristic in its exponential decay parameter, specifically including: when As the value approaches 0 in the positive direction, the exponential term approaches 1, and the improved threshold function approaches the traditional soft threshold function, eliminating the problem of discontinuous abrupt changes at the threshold of the hard threshold function; when As the value increases, the rate of exponential decay accelerates, and the amplitude deviation of the high-frequency detail coefficients of the large value decays exponentially and approaches the original value.

2. The real-time noise reduction method for a discardable sound velocity meter for underwater sound velocity measurement in the ocean as described in claim 1, characterized in that, In step S1, raw underwater sound velocity measurement data is collected, and a sliding sound velocity data window is constructed, including: The ultrasonic transducer in the jettisonable velocimeter transmits and receives sound wave signals, and the time-to-digital converter chip measures the sound wave flight time. The underwater sound velocity data is then calculated based on the transducer spacing and flight time. : ； In the formula, For the first The speed of sound measured this time, The distance between the transducer and the reflector. For the first The measured ultrasonic flight time The fixed delay introduced into the transducer, circuit, and signal transmission link was determined through calibration experiments; The continuously collected sound velocity data is written into the data buffer of the embedded processor to form a fixed-length sound velocity data window: ； In the formula, The total number of sound velocity sampling points contained within the sliding sound velocity data window and Pick ; Each time a new sound velocity data is acquired, the window slides forward one sampling point to complete the real-time update: ； In the formula, This is the iteration number for the data window sliding.

3. The real-time noise reduction method for a discardable sound velocity meter for underwater sound velocity measurement in the ocean as described in claim 1, characterized in that, In step S1, the sound velocity data window undergoes a three-level Haar wavelet decomposition to obtain low-frequency approximate components and multi-band high-frequency detail components, including: Perform a three-level wavelet decomposition on the sound velocity data window. First-level decomposition: ； ； In the formula, The sequence number in the sliding sound velocity data window is The original sound speed sample value, The sequence number in the sliding sound velocity data window is The original sound speed sample value, These are the first-order low-frequency approximation coefficients. For the first level of high-frequency detail coefficients, The coefficient index within the wavelet hierarchy and ; Second-level decomposition: ； ； In the formula, These are the second-order low-frequency approximation coefficients. For the second level of high-frequency detail coefficients and ; Third-level decomposition: ； ； In the formula, These are the third-order low-frequency approximation coefficients. For the third level of high-frequency detail coefficients and ; After Haar third-order wavelet decomposition, the original data window is decomposed into... , This represents the low-frequency trend component of the sound velocity profile. For details in the lower frequency band; For detailed changes in the mid-frequency range; This refers to high-frequency, detailed changes.

4. The real-time noise reduction method for a discardable sound velocity meter for underwater sound velocity measurement in the ocean as described in claim 1, characterized in that, In step S2, the smooth gradient feature quantity is extracted based on the local variation features of the sound speed sequence, including: The reference sound velocity value for each sampling point is obtained by neighborhood median filtering. Reference sound velocity value corresponding to each sampling point Calculated using the median: ； In the formula, For the first The original sound velocity value corresponding to each sampling point. For the first The reference sound velocity value corresponding to each sampling point. To take the median, The radius of the filtering neighborhood; Calculate the local sound velocity variation characteristic of adjacent sampling points based on the change between adjacent sampling points in a continuous sound velocity sampling sequence: ； In the formula, For the first Local sound velocity variation characteristic of the sequence domain of the point and , For the first The and the first The reference sound velocity value corresponding to each sampling point; right Neighborhood averaging is performed, and for sampling points at the data window boundaries that are less than the neighborhood length, endpoint copying is used to pad them. ； In the formula, For the first The smooth local sound velocity variation characteristic near each sampling point The smoothing radius; Within the current sound velocity data window, take the maximum value of the smooth local sound velocity variation characteristic: ； In the formula, This represents the maximum value of all smooth gradient features within the current sliding sound velocity data window.

5. The real-time noise reduction method for a discardable sound velocity meter for underwater sound velocity measurement in the ocean as described in claim 1, characterized in that, In step S4, the low-frequency approximation component and the denoised high-frequency detail component are used to perform Haar wavelet inverse transform reconstruction to obtain the denoised sound velocity profile data, specifically including: Let the first The low-frequency approximation coefficient of the layer is The high-frequency detail coefficients after thresholding are Then the first The layer approximation coefficient is recovered by the following formula: ； ； In the formula, The first wavelet inverse reconstruction obtained Even-numbered low-frequency approximation coefficients for each layer The first wavelet inverse reconstruction obtained Odd-digit low-frequency approximation coefficients for each layer. ; The initial conditions for reconstruction are: ； Using the third-order low-frequency approximation coefficient and the processed third-level high-frequency detail coefficient The second-order approximation coefficients are obtained through reconstruction: ； ； In the formula, The second-level low-frequency approximation coefficients are the even-numbered coefficients obtained by reconstructing the third-level low-frequency approximation coefficients and the processed third-level high-frequency detail coefficients through inverse wavelet transform. The odd-numbered coefficients of the second-level low-frequency approximation coefficients are obtained by reconstructing the third-level low-frequency approximation coefficients and the processed third-level high-frequency detail coefficients through inverse wavelet transform. ; Using the second-order low-frequency approximation coefficient and the processed second-level high-frequency detail coefficients The first-order approximation coefficients are obtained by reconstruction: ； ； In the formula, The even-numbered coefficients of the first-order low-frequency approximation coefficients obtained by inverse wavelet transform reconstruction. These are the odd-numbered coefficients of the first-order low-frequency approximation coefficients obtained by inverse wavelet transform reconstruction. Using first-order approximation coefficients and the processed first-level high-frequency detail coefficients The reconstructed sound velocity data is then obtained: ； ； In the formula, After reconstruction by inverse wavelet transform, the th The sound velocity sample value corresponding to each sampling point after denoising. After reconstruction by inverse wavelet transform, the th The sound velocity sampled after denoising each sampling point ; After a three-stage Haar wavelet inverse transform, the denoised sound velocity data window is obtained: 。 6. The real-time noise reduction method for a discardable sound velocity meter for underwater sound velocity measurement in the ocean as described in claim 1, characterized in that, In step S4, the real-time denoised underwater sound velocity profile data is output, including: After completing the Haar wavelet inverse transform reconstruction to obtain continuous denoised speed data, the embedded main controller frames the denoised speed sample values, sample numbers, and pressure depth information, and outputs real-time denoised underwater sound speed profile data matching the depth position, thus achieving simultaneous completion of sound speed profile noise suppression and effective feature fidelity.

Images