Speckle noise detection and removal method based on acoustic-laser cross-medium communication

Abstract

The application discloses a speckle noise detection and removal method based on acoustic wave-laser cross-medium communication, which comprises the following steps: based on a dual-feature joint detection mechanism of kurtosis ratio and Teager energy operator, noise positions are identified and recognized through sliding window analysis; adaptive boundary expansion is performed on each detected noise position, and adjacent noise positions are merged to form continuous noise segments; two methods of bidirectional linear prediction and empirical mode decomposition are used to process the noise segments in parallel, and prediction signals of the noise segments are respectively generated; according to the local signal-to-noise ratio of the prediction signals and the boundary continuity, the fusion weights of the two kinds of prediction signals are respectively adaptively calculated, the two-way prediction signals are weighted and fused, and a fusion signal is obtained; through linear transformation, the end-point continuity of the fusion signal is matched, the smooth connection between the repaired segment and the original signal is ensured, and the speckle noise removal is completed. The application can detect and effectively suppress the speckle noise caused by dynamic water surface disturbance, and significantly improve the integrity and clarity of signal transmission.

Description

Technical Field

[0001] This invention relates to the fields of cross-water and air medium communication and laser detection, and particularly to a method for detecting and removing speckle noise based on acoustic-laser cross-medium communication. Background Technology

[0002] Cross-water-air medium communication is a key technology for enabling direct communication between underwater platforms (submarines, underwater sensor networks, deep-sea probes, etc.) and airborne platforms (aircraft, drones, satellites, etc.), and has significant application value in fields such as marine monitoring, underwater rescue, and military communications. Its basic principle is as follows: an underwater sound source emits sound waves carrying information; these sound waves propagate to the water surface and cause minute water surface vibrations (typically at the nanometer to micrometer level); an airborne laser Doppler vibrometer detects these vibrations, achieving sound-to-light signal conversion; and the laser signal is demodulated to obtain the original sound wave information.

[0003] Laser Doppler Vibrometry (LDV) is a non-contact vibration measurement instrument widely used in precision vibration measurement, remote voice detection, and structural health monitoring. However, when a laser beam illuminates a rough surface, speckle noise is generated, severely affecting measurement accuracy and signal quality. Natural water surfaces, influenced by wind, current, and underwater sound pressure, exhibit complex random waves, forming an optically rough reflective interface. When a laser beam illuminates the dynamically undulating water surface, multiple scattering and interference occur, forming a speckle pattern that changes rapidly and randomly over time and space. The movement of the water surface itself also causes scattering changes, resulting in a speckle field exhibiting strong non-stationary dynamic characteristics, with its rate of change directly related to the water surface velocity. This noise is non-stationary and highly abrupt, making it difficult to effectively remove using traditional filtering methods.

[0004] In existing technologies, the main methods for removing speckle noise are:

[0005] (1) Frequency domain filtering method: By designing a filter with a specific frequency response to remove noise frequency components, this method is very likely to filter out useful signals that overlap with the noise spectrum while removing noise, resulting in information loss.

[0006] (2) Wavelet transform method: Wavelet multi-resolution analysis is used for signal decomposition and threshold denoising. However, this method faces the inherent problem of difficulty in selecting the threshold function and the number of decomposition layers. Furthermore, it is easy to introduce false ripples or artifacts during the processing, which affects the signal fidelity.

[0007] (3) Statistical methods are used to model and suppress noise based on its statistical characteristics, but they are not very adaptable to non-stationary and sudden speckle noise, and their processing effect is limited.

[0008] The common core defects of these methods are: they cannot accurately locate the sudden location of speckle noise in the time domain, resulting in a lack of specificity in the denoising process; while suppressing noise, it is difficult to avoid the loss of useful signals, especially the destruction of weak and effective signals; and for sudden speckle noise with dynamic intensity changes and short correlation time, the processing effect of existing methods is generally not ideal.

[0009] Therefore, developing a noise reduction method that can accurately locate the sudden noise source, adaptively repair it, and retain weak useful signals to address dynamic speckle noise interference at the water-air interface is of great significance for improving the reliability of acoustic-laser cross-medium communication and promoting the development of marine observation and communication technologies. Summary of the Invention

[0010] The purpose of this invention is to provide a method for detecting and removing speckle noise in an acoustic-laser cross-medium communication system, which solves the technical problems of severe impact of water surface speckle noise on communication quality and unsatisfactory noise removal effect of traditional methods in the prior art, and realizes high-quality real-time communication across water and air media.

[0011] The objective of this invention is achieved through the following technical solution:

[0012] A method for speckle noise detection and removal based on acoustic-laser cross-medium communication includes:

[0013] S1: A dual-feature joint detection mechanism based on kurtosis ratio and Teager energy operator is adopted, and the location of speckle noise is identified through sliding window analysis;

[0014] S2: Perform adaptive boundary expansion on each detected speckle noise location, merge adjacent speckle noise locations to form a continuous noise segment, and record the start position, end position and segment length of each noise segment;

[0015] S3: Two methods, bidirectional linear prediction and empirical mode decomposition, are used in parallel to process the noise segment and generate prediction signals for the noise segment respectively;

[0016] S4: Based on the local signal-to-noise ratio and boundary continuity of the predicted signal, adaptively calculate the fusion weights of the two predicted signals respectively, and weight and fuse the two predicted signals to obtain the fused signal;

[0017] S5: By performing endpoint continuity matching on the fused signal through linear transformation, a smooth connection between the repaired segment and the original signal is ensured, thus completing the removal of speckle noise.

[0018] Furthermore, step S1 specifically includes the following sub-steps:

[0019] S1.1: Divide the input signal into sliding windows according to the set window size and the overlap length of adjacent windows, and calculate the original signal kurtosis of each window;

[0020] S1.2: Calculate the lower and upper thresholds of the window signal using the inverse cumulative distribution function, and retain the sampling points located between the upper and lower thresholds to obtain the trimmed signal;

[0021] S1.3: Calculate the kurtosis, kurtosis ratio, and Teager energy operator of the pruned signal:

[0022] S1.4: Calculate the kurtosis ratio of the trimmed signal and the weighted sum of the Teager energy operator to obtain a noise decision based on dual features. If the noise decision is greater than the noise decision threshold, it is determined that there is speckle noise in the window and the noise location is recorded.

[0023] S1.5: Merge and deduplicate the detected noise locations to obtain a set of speckle noise locations.

[0024] Furthermore, step S2 specifically includes the following sub-steps:

[0025] S2.1: Adaptive boundary expansion is performed on each detected speckle noise location, expanding multiple sampling points to the left and right;

[0026] S2.2: Merge the noise locations that are less than 10 adjacent sampling points after expansion to form M continuous noise segments;

[0027] S2.3: Record the start position, end position, and length of each noise segment.

[0028] Furthermore, in step S3, a bidirectional linear prediction method is used to process the noise segment and generate a prediction signal for the noise segment, specifically including the following sub-steps:

[0029] S3.1: Search forward and backward from the starting position of the noise segment to extract continuous clean data as the forward context and backward context respectively;

[0030] S3.2: Use the Levinson-Durbin algorithm to calculate the forward linear prediction coefficients and the backward linear prediction coefficients respectively;

[0031] S3.3: Perform forward and backward recursive predictions respectively to generate forward prediction signals and backward prediction signals;

[0032] S3.4: The forward prediction signal and the backward prediction signal are fused using a cross-fade-in / fade-out method to obtain the prediction signal generated by the bidirectional linear prediction method;

[0033] S3.5: Extract the prediction signal corresponding to the noise segment from the prediction signal generated by the bidirectional linear prediction method.

[0034] Furthermore, in step S3, the empirical mode decomposition method is used to process the noise segment and generate a prediction signal for the noise segment, specifically including the following sub-steps:

[0035] S3.6: Centered on the noise segment, add a clean signal of fixed length before its start position and after its end position as context, which together with the noise segment form an extended signal;

[0036] S3.7: Perform empirical mode decomposition on the extended signal and obtain a set of intrinsic mode functions through an iterative sieving process;

[0037] S3.8: Perform feature analysis on each intrinsic mode function to identify noise-dominated intrinsic mode functions;

[0038] S3.9: Remove noise-dominated intrinsic mode functions to obtain the reconstructed signal corresponding to the empirical mode decomposition method;

[0039] S3.10: Extract the predicted signal corresponding to the noise segment from the reconstructed signal corresponding to the empirical mode decomposition method.

[0040] Furthermore, empirical mode decomposition is performed on the extended signal, and a set of intrinsic mode functions is obtained through an iterative sieving process, specifically including:

[0041] S3.7.1: Identify all local maxima and local minima of the extended signal corresponding to the noise segment;

[0042] S3.7.2: Use cubic spline interpolation to connect local maxima and local minima respectively, and construct upper and lower envelopes;

[0043] S3.7.3: Calculate the mean of the upper and lower envelopes;

[0044] S3.7.4: Subtract the mean from the extended signal corresponding to the noise segment to obtain the candidate intrinsic mode functions;

[0045] S3.7.5: Check whether the candidate intrinsic mode function satisfies the intrinsic mode function condition. If not, set the extended signal equal to the candidate intrinsic mode function and return to step S3.7.1; until the intrinsic mode function condition is satisfied, the intrinsic mode function is obtained.

[0046] S3.7.6: Subtract the intrinsic mode function from the extended signal to obtain the residual, and return to step S3.7.1 until the residual is a monotonic function or meets the stopping criterion.

[0047] Furthermore, feature analysis is performed on each intrinsic mode function to identify noise-dominated intrinsic mode functions, specifically including:

[0048] S3.8.1: Calculate the instantaneous amplitude of each intrinsic mode function using the Hilbert transform;

[0049] S3.8.2: Based on the instantaneous amplitude, calculate the energy of each sampling point, sort all sampling points by energy from smallest to largest, and calculate the energy concentration; at the same time, calculate the zero crossover rate and the Pearson correlation coefficient between each intrinsic mode function and the noise segment respectively;

[0050] S3.8.3: When the energy concentration is less than the energy concentration threshold, the zero crossover rate is greater than the zero crossover rate threshold, and the Pearson correlation coefficient is greater than the correlation coefficient threshold, the intrinsic mode function is determined to be a noise-dominated intrinsic mode function.

[0051] Furthermore, step S5 includes the following sub-steps:

[0052] S5.1: Extract the clean sample values ​​of the left and right neighbors of the noisy segment;

[0053] S5.2: Construct a linear transformation matrix to obtain the calculation formulas for the scaling factor and offset, and solve for the values ​​of the scaling factor and offset using the least squares method;

[0054] S5.3: Based on the values ​​of scaling factor and offset, perform a linear transformation on the fused signal to obtain the repaired signal segment;

[0055] S5.4: Replace the original noise segment with the repaired signal segment.

[0056] A speckle noise detection and removal system based on acoustic-laser cross-medium communication, the system including a noise detection module and a signal repair module;

[0057] The noise detection module employs a dual-feature joint detection mechanism based on kurtosis ratio and Teager energy operator. It identifies speckle noise locations through sliding window analysis, adaptively expands the boundaries of the detected speckle noise locations, merges adjacent speckle noise locations to form continuous noise segments, and records the start and end positions and segment length of each noise segment.

[0058] The signal restoration module employs two parallel methods—bidirectional linear prediction and empirical mode decomposition—to process the noise segment, generating predicted signals for each segment. Based on the local signal-to-noise ratio and boundary continuity of the predicted signals, it adaptively calculates the fusion weights of the two predicted signals, weights the two predicted signals, and fuses them to obtain a fused signal. Finally, it performs endpoint continuity matching on the fused signal through linear transformation to ensure a smooth connection between the restored segment and the original signal, thus completing speckle noise removal.

[0059] An electronic device, comprising:

[0060] One or more processors;

[0061] A storage device for storing one or more programs, which, when executed by the electronic device, enable the electronic device to implement a speckle noise detection and removal method based on acoustic-laser cross-medium communication.

[0062] The beneficial effects of this invention are as follows:

[0063] This invention employs a dual-feature detection mechanism that integrates kurtosis ratio and the Teager energy operator, enabling accurate identification of noise bursts in strong noise environments and overcoming the limitations of traditional methods in temporal localization. Combining adaptive boundary processing and a dual-path parallel repair strategy, it not only comprehensively covers the noise-affected area but also achieves high-quality signal reconstruction in both temporal correlation and frequency-domain adaptive decomposition through the complementary effects of bidirectional linear prediction and empirical mode decomposition. Furthermore, the introduction of quality-assessment-based adaptive fusion and endpoint smoothing techniques ensures a natural connection between the repaired segment and the original signal, effectively avoiding distortion or artifacts that may occur during the repair process. This invention significantly improves the reliability and stability of communication between underwater acoustic sources and airborne laser detectors, ensuring that the system can accurately detect and effectively suppress speckle noise caused by dynamic surface disturbances in complex marine environments, especially under conditions of significant surface disturbance. It enables continuous high-quality cross-medium information transmission, significantly improving signal integrity and clarity. This invention is particularly suitable for all demanding application scenarios that utilize laser vibrometric technology for underwater-air link communication. Attached Figure Description

[0064] Figure 1 This is a flowchart of a speckle noise detection and removal method based on acoustic-laser cross-medium communication, according to one embodiment of the present invention.

[0065] Figure 2 This is a flowchart illustrating speckle noise location identification in one embodiment.

[0066] Figure 3 This is a flowchart of noise location processing and grouping for one embodiment.

[0067] Figure 4 This is a flowchart of a dual-path parallel prediction and repair algorithm for one embodiment.

[0068] Figure 5 This is a schematic diagram of a speckle noise detection and removal system based on acoustic-laser cross-medium communication, according to another embodiment of the present invention. Detailed Implementation

[0069] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0070] like Figure 1 As shown, the speckle noise detection and removal method based on acoustic-laser cross-medium communication in this embodiment includes the following steps one through five:

[0071] Step 1: Speckle noise detection: A dual-feature joint detection mechanism based on kurtosis ratio (KR) and Teager energy operator (TEO) is adopted to identify the location of speckle noise through sliding window analysis.

[0072] like Figure 2 As shown, step one specifically includes the following sub-steps:

[0073] (1.1) Set the sliding window size W, select the analysis window length, and set the overlap length O between adjacent windows to ensure the continuity of detection; set the noise decision threshold D. th .

[0074] The recommended sliding window size is 128-512 sampling points, adjusted according to the sampling frequency and signal characteristics. For a typical laser Doppler vibrometer system (sampling frequency 20-100kHz), a window size of 256 sampling points is preferred, ensuring sufficient statistical characteristics while maintaining high temporal resolution. The recommended overlap length O is 80%-90% of the window size, meaning a high degree of overlap between adjacent windows ensures the continuity and accuracy of speckle noise detection. That is, when the window size is 256 sampling points, the recommended overlap length is 224 sampling points.

[0075] (1.2) Divide the input signal x(n) into sliding windows according to the set window size W and overlap length O, and calculate the window step size S and the total number of windows M; calculate the signal x of the u-th window. u The original signal kurtosis K of (n) orig :

[0076] S=WO

[0077]

[0078]

[0079] Where N is the length of the input signal; μ is the mean of the window signal; and σ is the standard deviation. This indicates the expected operation.

[0080] (1.3) Remove outliers from the window signal using a thresholding method, specifically by calculating the lower threshold thr using the inverse cumulative distribution function ICDF. low and upper threshold thr high These correspond to the k% quantile and (100-k)% quantile of the signal, respectively; retain those that satisfy thr low <x u (n) < thr high The conditional sampling points are used to obtain the trimming signal x. trim (n);

[0081] thr low =x k / 100

[0082] thr high = x 1-k / 100

[0083] (1.4) Calculate the pruning signal x trim Kurtosis K of (n) trim Kurtosis ratio (KR) and Teager energy operator (TEO):

[0084]

[0085]

[0086]

[0087] Where, μ trim and σ trim The pruning signal x is respectively trim The mean and standard deviation of (n).

[0088] (1.5) Calculate noise decision based on dual features When noise decision Greater than If the window contains speckle noise, its location is recorded.

[0089]

[0090] in , These are the weighting coefficients.

[0091] (1.6) The detected noise locations are merged and deduplicated to obtain a set of speckle noise locations.

[0092] Step 2: Noise location processing and grouping: Adaptive boundary expansion is performed on the detected noise locations, and adjacent noise locations are merged into continuous segments.

[0093] like Figure 3As shown, step two includes the following sub-steps:

[0094] (2.1) For each detected noise location, perform adaptive boundary expansion, extending E sampling points to the left and right:

[0095]

[0096] in This represents the local standard deviation near the location of the noise.

[0097] (2.2) Merge the noise locations where the adjacent sampling points are less than 10 after expansion to form M continuous noise segments;

[0098] (2.3) Record the starting position s of each noise segment. m End position e m and segment length L m :

[0099] L m =e m - s m +1

[0100] Step 3: Dual-path parallel prediction repair: Two techniques, linear predictive coding (LPC) and empirical mode decomposition (EMD), are used in parallel to process the noise segment and generate prediction signals for the noise segment respectively.

[0101] like Figure 4 As shown, step three uses a bidirectional linear prediction method to process the noise segment and generate a prediction signal for the noise segment, including the following sub-steps:

[0102] (3.1) From the starting position s of the noise segment m Search forward, skipping other noisy regions, and extract continuous clean data of length L as the forward context; starting from the end position e m The backward search extracts continuous clean data of length L as the backward context, where L ≥ 2p, and p is the linear prediction order. ;

[0103] (3.2) Calculate the LPC coefficients a from the forward context data using the Levinson-Durbin algorithm. f Reverse the backward context data in time and calculate the LPC coefficient a. b , used for backward prediction.

[0104] (3.3) Using a f The forward context recursion generates a length of L. m Predicted signal :

[0105]

[0106] Where i is the lag index of the linear prediction model, which represents the time lag of the historical sample used for prediction relative to the current prediction point n; x(ni) is a known clean signal sample located i sampling points before the noise segment and before the current prediction point n.

[0107] Use a b The backward recursion generates a length of L. m Predicted signal :

[0108]

[0109] (3.4) The forward and backward prediction results are fused using a cross-fade-in / fade-out method to obtain the prediction signal generated by the bidirectional linear prediction method. :

[0110]

[0111]

[0112] Where n is the discrete-time index of the signal within the noise segment, n=1,2,…,L m α(n) is a weight function that decreases linearly from 1 to 0.

[0113] (3.5) From the predicted signal Extract the corresponding position of the noise segment. .

[0114] Step three employs two methods, empirical mode decomposition (EMD), to process the noise segment and generate a predicted signal for the noise segment. This includes the following sub-steps:

[0115] (3.6) Taking the noise segment as the center, at its starting position s m Before and end positions e m Then, take a segment of length L from each. c The continuous clean signal serves as the context, and together with the noise segment, they form an extended signal x. ext L c The context length for EMD analysis should be long enough to ensure that the complete oscillation of the signal near the noise segment is captured.

[0116] (3.7) For the extended signal x ext EMD decomposition is performed, and a set of intrinsic mode functions (IMFs) is obtained through an iterative sieving process. The specific iterative process is as follows:

[0117] (3.7.1) Identify x ext All local maxima and local minima;

[0118] (3.7.2) Use cubic spline interpolation to connect the local maxima and local minima respectively, and construct the upper envelope e. upper and lower envelope e lower ;

[0119] (3.7.3) Calculate the envelope mean m:

[0120]

[0121] (3.7.4) From the extended signal x ext Subtracting the mean m from the mean yields the IMF candidate h:

[0122]

[0123] (3.7.5) Check whether h satisfies the IMF condition, that is, the difference between the number of zeros and the number of extreme points does not exceed 1, and the upper and lower envelopes are locally symmetrical; if the IMF condition is satisfied, then let the final obtained intrinsic mode function IMF. j =h;

[0124] Otherwise, let x ext =h, return to step (3.6.1);

[0125] (3.7.6) From the extended signal x ext Subtract IMF j The residual r is obtained. Return to step (3.6.1) until the residual is a monotonic function or the stopping criterion is met;

[0126] (3.8) Perform feature analysis on each IMF to identify noise-dominated IMFs;

[0127] Step (3.8) includes the following sub-steps:

[0128] (3.8.1) Calculate the instantaneous amplitude of each eigenmode function using the Hilbert transform. ;

[0129] (3.8.2) Calculate the energy E at each sampling point. z The energy of all sampling points is sorted from smallest to largest, and the energy concentration EC is calculated according to the following formula. At the same time, the zero crossover rate ZCR and the energy concentration of each IMF are calculated separately. j PEARSON correlation coefficient with the noise segment:

[0130]

[0131]

[0132]

[0133]

[0134] Among them, M j For IMF j The total number of sampling points; ρ is the proportionality coefficient, usually taken as a value between 0.05 and 0.2; K is the number of the first K largest energy samples.

[0135] (3.8.3) When EC < T EC ZCR>T ZCR Furthermore, when the PEARSON correlation coefficient is greater than the correlation coefficient threshold, it is determined to be a noise-dominated IMF; where T EC T is the energy concentration threshold. ZCR The zero crossover rate threshold;

[0136] (3.9) Remove the noise-dominated IMF to obtain the reconstructed signal corresponding to the empirical mode decomposition method. :

[0137] ;

[0138] Wherein, IMFs represents the set of intrinsic mode functions;

[0139] (3.10) From the reconstructed signal Extract the prediction signal corresponding to the noise segment. .

[0140] Step 4: Adaptive fusion based on quality assessment: Based on the local signal-to-noise ratio and boundary continuity of the predicted signal, the fusion weights of the two predicted signals are adaptively calculated respectively, and the two prediction results are weighted and fused to obtain the fused signal.

[0141] Step four includes the following sub-steps:

[0142] (4.1) Calculate the local signal-to-noise ratio (SNR) of the bidirectional linear prediction signal respectively. LPC and boundary continuity BC LPC And the local signal-to-noise ratio (SNR) of the empirical mode decomposition reconstructed signal. EMD and boundary continuity BC EMD ;

[0143] (4.2) Calculate the quality weighting coefficient w of the bidirectional linear prediction signal respectively. LPC The quality weighting coefficient w of the Empirical Mode Decomposition (EMD) reconstructed signal EMD :

[0144]

[0145]

[0146] (4.3) The bidirectional linear prediction signal and the empirical mode decomposition reconstructed signal are weighted and fused to obtain the fused signal x. repair (n):

[0147]

[0148] Step 5: Endpoint continuity matching and signal reconstruction: Endpoint continuity matching is performed through linear transformation to ensure a smooth connection between the repaired segment and the original signal, thus completing speckle noise removal.

[0149] Step five includes the following sub-steps:

[0150] (5.1) Extract the clean sample values ​​x from the left and right neighbors of the noisy segment. left and x right :

[0151]

[0152]

[0153] (5.2) Construct a linear transformation matrix with parameters (α,β) such that:

[0154]

[0155] Where α is the scaling factor and β is the offset;

[0156] Based on the linear transformation matrix, the following formulas for calculating α and β are obtained, and the values ​​of α and β are solved using the least squares method:

[0157]

[0158]

[0159] (5.3) Perform a linear transformation on the fused signal to obtain the repaired signal segment x. final (n);

[0160]

[0161] Among them, L noise The length of the noise segment;

[0162] (5.4) Using the repaired signal segment x final (n) Replace the original noise segment.

[0163] Another embodiment of the present invention also provides a speckle noise detection and removal system based on acoustic-laser cross-medium communication, such as... Figure 5 As shown, the system includes a noise detection module and a signal repair module, wherein:

[0164] The noise detection module adopts a dual-feature joint detection mechanism based on kurtosis ratio and Teager energy operator. It identifies speckle noise locations through sliding window analysis, performs adaptive boundary expansion on the detected speckle noise locations, merges adjacent speckle noise locations to form continuous noise segments, and records the start position, end position, and segment length of each noise segment.

[0165] The signal restoration module uses two methods, bidirectional linear prediction and empirical mode decomposition, to process the noise segment in parallel, generating prediction signals for the noise segment respectively. Based on the local signal-to-noise ratio and boundary continuity of the prediction signals, the fusion weights of the two prediction signals are adaptively calculated, and the two prediction signals are weighted and fused to obtain a fused signal. Finally, the endpoint continuity of the fused signal is matched by linear transformation to ensure a smooth connection between the restored segment and the original signal, thus completing the speckle noise removal.

[0166] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for detecting and removing speckle noise based on acoustic-laser cross-medium communication, characterized in that, include: S1: A dual-feature joint detection mechanism based on kurtosis ratio and Teager energy operator is adopted to identify the location of speckle noise through sliding window analysis; S1 specifically includes the following sub-steps: S1.1: Divide the input signal into sliding windows according to the set window size and the overlap length of adjacent windows, and calculate the original signal kurtosis of each window; S1.2: Calculate the lower and upper thresholds of the window signal using the inverse cumulative distribution function, and retain the sampling points located between the upper and lower thresholds to obtain the trimmed signal; S1.3: Calculate the kurtosis, kurtosis ratio, and Teager energy operator of the pruned signal: S1.4: Calculate the kurtosis ratio of the trimmed signal and the weighted sum of the Teager energy operator to obtain a noise decision based on dual features. If the noise decision is greater than the noise decision threshold, it is determined that there is speckle noise in the window and the noise location is recorded. S1.5: Merge and deduplicate the detected noise locations to obtain a set of speckle noise locations; S2: Perform adaptive boundary expansion on each detected speckle noise location, merge adjacent speckle noise locations to form a continuous noise segment, and record the start position, end position and segment length of each noise segment; S3: Two methods, bidirectional linear prediction and empirical mode decomposition, are used in parallel to process the noise segment and generate prediction signals for the noise segment respectively; S4: Based on the local signal-to-noise ratio and boundary continuity of the predicted signal, adaptively calculate the fusion weights of the two predicted signals respectively, and weight and fuse the two predicted signals to obtain the fused signal; S5: Perform endpoint continuity matching on the fused signal through linear transformation to ensure a smooth connection between the repaired segment and the original signal, thus completing speckle noise removal; S5 includes the following sub-steps: S5.1: Extract the clean sample values ​​of the left and right neighbors of the noisy segment; S5.2: Construct a linear transformation matrix to obtain the calculation formulas for the scaling factor and offset, and solve for the values ​​of the scaling factor and offset using the least squares method; S5.3: Based on the values ​​of scaling factor and offset, perform a linear transformation on the fused signal to obtain the repaired signal segment; S5.4: Replace the original noise segment with the repaired signal segment.

2. The method for speckle noise detection and removal based on acoustic-laser cross-medium communication according to claim 1, characterized in that, S2 specifically includes the following sub-steps: S2.1: Adaptive boundary expansion is performed on each detected speckle noise location, expanding multiple sampling points to the left and right; S2.2: Merge the noise locations that are less than 10 adjacent sampling points after expansion to form M continuous noise segments; S2.3: Record the start position, end position, and length of each noise segment.

3. The method for speckle noise detection and removal based on acoustic-laser cross-medium communication according to claim 1, characterized in that, In step S3, a bidirectional linear prediction method is used to process the noise segment and generate a prediction signal for the noise segment. This specifically includes the following sub-steps: S3.1: Search forward and backward from the starting position of the noise segment to extract continuous clean data as the forward context and backward context respectively; S3.2: Use the Levinson-Durbin algorithm to calculate the forward linear prediction coefficients and the backward linear prediction coefficients respectively; S3.3: Perform forward and backward recursive predictions respectively to generate forward prediction signals and backward prediction signals; S3.4: The forward prediction signal and the backward prediction signal are fused using a cross-fade-in / fade-out method to obtain the prediction signal generated by the bidirectional linear prediction method; S3.5: Extract the prediction signal corresponding to the noise segment from the prediction signal generated by the bidirectional linear prediction method.

4. The method for speckle noise detection and removal based on acoustic-laser cross-medium communication according to claim 1, characterized in that, In step S3, the empirical mode decomposition method is used to process the noise segment and generate the prediction signal for the noise segment. Specifically, this includes the following sub-steps: S3.6: Centered on the noise segment, add a clean signal of fixed length before its start position and after its end position as context, which together with the noise segment form an extended signal; S3.7: Perform empirical mode decomposition on the extended signal and obtain a set of intrinsic mode functions through an iterative sieving process; S3.8: Perform feature analysis on each intrinsic mode function to identify noise-dominated intrinsic mode functions; S3.9: Remove noise-dominated intrinsic mode functions to obtain the reconstructed signal corresponding to the empirical mode decomposition method; S3.10: Extract the predicted signal corresponding to the noise segment from the reconstructed signal corresponding to the empirical mode decomposition method.

5. The method for speckle noise detection and removal based on acoustic-laser cross-medium communication according to claim 4, characterized in that, Empirical mode decomposition (EMD) is performed on the extended signal, and a set of intrinsic mode functions (EMFs) is obtained through an iterative sieving process, specifically including: S3.7.1: Identify all local maxima and local minima of the extended signal corresponding to the noise segment; S3.7.2: Use cubic spline interpolation to connect local maxima and local minima respectively, and construct upper and lower envelopes; S3.7.3: Calculate the mean of the upper and lower envelopes; S3.7.4: Subtract the mean from the extended signal corresponding to the noise segment to obtain the candidate intrinsic mode functions; S3.7.5: Check whether the candidate intrinsic mode function satisfies the intrinsic mode function condition. If not, set the extended signal equal to the candidate intrinsic mode function and return to step S3.7.1; until the intrinsic mode function condition is satisfied, the intrinsic mode function is obtained. S3.7.6: Subtract the intrinsic mode function from the extended signal to obtain the residual, and return to step S3.7.1 until the residual is a monotonic function or meets the stopping criterion.

6. The method for speckle noise detection and removal based on acoustic-laser cross-medium communication according to claim 4, characterized in that, Feature analysis is performed on each intrinsic mode function to identify noise-dominated intrinsic mode functions, specifically including: S3.8.1: Calculate the instantaneous amplitude of each intrinsic mode function using the Hilbert transform; S3.8.2: Based on the instantaneous amplitude, calculate the energy of each sampling point, sort all sampling points by energy from smallest to largest, and calculate the energy concentration; at the same time, calculate the zero crossover rate and the Pearson correlation coefficient between each intrinsic mode function and the noise segment respectively; S3.8.3: When the energy concentration is less than the energy concentration threshold, the zero crossover rate is greater than the zero crossover rate threshold, and the Pearson correlation coefficient is greater than the correlation coefficient threshold, the intrinsic mode function is determined to be a noise-dominated intrinsic mode function.

7. A speckle noise detection and removal system based on acoustic-laser cross-medium communication, characterized in that, This system is used to implement the speckle noise detection and removal method based on acoustic-laser cross-medium communication as described in any one of claims 1 to 6. The system includes a noise detection module and a signal repair module. The noise detection module employs a dual-feature joint detection mechanism based on kurtosis ratio and Teager energy operator. It identifies speckle noise locations through sliding window analysis, adaptively expands the boundaries of the detected speckle noise locations, merges adjacent speckle noise locations to form continuous noise segments, and records the start and end positions and segment length of each noise segment. The signal restoration module employs two parallel methods—bidirectional linear prediction and empirical mode decomposition—to process the noise segment, generating predicted signals for each segment. Based on the local signal-to-noise ratio and boundary continuity of the predicted signals, it adaptively calculates the fusion weights of the two predicted signals, weights the two predicted signals, and fuses them to obtain a fused signal. Finally, it performs endpoint continuity matching on the fused signal through linear transformation to ensure a smooth connection between the restored segment and the original signal, thus completing speckle noise removal.

8. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, cause the electronic device to implement the speckle noise detection and removal method based on acoustic-laser cross-medium communication as described in any one of claims 1 to 6.

Images