WO2021258832A1 - 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法 - Google Patents

基于自适应窗口滤波和小波阈值优化的水声信号去噪方法 Download PDF

Info

Publication number
WO2021258832A1
WO2021258832A1 PCT/CN2021/088635 CN2021088635W WO2021258832A1 WO 2021258832 A1 WO2021258832 A1 WO 2021258832A1 CN 2021088635 W CN2021088635 W CN 2021088635W WO 2021258832 A1 WO2021258832 A1 WO 2021258832A1
Authority
WO
WIPO (PCT)
Prior art keywords
noise
signal
gaussian
new
threshold
Prior art date
Application number
PCT/CN2021/088635
Other languages
English (en)
French (fr)
Inventor
王景景
李嘉恒
董新利
杨星海
施威
徐凌伟
郭瑛
李海涛
Original Assignee
青岛科技大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 青岛科技大学 filed Critical 青岛科技大学
Publication of WO2021258832A1 publication Critical patent/WO2021258832A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208Noise filtering
    • G10L21/0216Noise filtering characterised by the method used for estimating noise
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B11/00Transmission systems employing sonic, ultrasonic or infrasonic waves
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02Preprocessing
    • G06F2218/04Denoising
    • G06F2218/06Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/27Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/45Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of analysis window

Definitions

  • the invention belongs to the technical field of underwater acoustic signal denoising, and specifically relates to an underwater acoustic signal denoising method based on adaptive window filtering (AWFM) and wavelet threshold optimization (GDES) in a Gaussian/non-Gaussian impulse noise environment.
  • AWFM adaptive window filtering
  • GDES wavelet threshold optimization
  • Acoustic waves are widely used in the field of underwater communication.
  • the acoustic signal will be affected by the underwater complex Gaussian/non-Gaussian impulse noise, which leads to the degradation and distortion of the acoustic signal and the degradation of communication quality.
  • Signal denoising technology is a signal processing method used to improve signal quality and reduce the influence of noise. It is widely used in underwater acoustic communications and other fields.
  • SMF Standard Median Filter
  • Gaussian noise can be effectively suppressed by filtering method, wavelet transform method, empirical mode decomposition method and other methods.
  • the denoising method based on wavelet threshold can obtain the asymptotic optimal estimation of the original signal, which has been widely used.
  • the main factors that determine the performance of this method are the accurate estimation of the threshold and the reasonable construction of the threshold function.
  • a unified threshold based on the multi-dimensional independent normal variable decision theory is commonly used under the assumption that the noise model is a Gaussian noise model; however, the unified threshold depends on an accurate estimation of the noise variance, and it is difficult to apply to the actual unknown noise variance.
  • Common threshold functions include hard threshold, soft threshold, and semi-soft threshold. This type of method processes wavelet coefficients according to a fixed structure, which lacks adaptability and reduces the flexibility of signal processing. To overcome the above limitations, a method based on swarm intelligence optimization was introduced to improve the performance of wavelet threshold denoising.
  • the actual ocean background noise environment contains both Gaussian and non-Gaussian impulse noise.
  • the wavelet threshold denoising method based on intelligent optimization is mainly used for Gaussian noise processing. It is difficult to directly apply to the comprehensive processing of marine underwater acoustic noise.
  • the shortcomings are embodied in: First, the lack of a unified general principle for establishing the threshold function, which makes the construction of the threshold function difficult; secondly, the determination of the threshold parameter is an iterative process, usually reaching a sub-optimal value rather than an optimal value; and with the method As the number of iterations increases, the diversity of the population decreases, and the above optimization method may fall into a local minimum.
  • the present invention proposes an underwater acoustic signal denoising method based on adaptive window filtering (AWFM) and wavelet threshold optimization (GDES) in a Gaussian/non-Gaussian impulse noise environment to make up for the deficiencies of the prior art.
  • AWFM adaptive window filtering
  • GDES wavelet threshold optimization
  • the present invention first describes the Gaussian/non-Gaussian impulse noise in the underwater acoustic channel by combining the S ⁇ S distribution and the normal distribution model.
  • This model is a limit distribution that can maintain the generation mechanism and propagation conditions of complex ocean background noise, and can be better described Background noise of marine environment.
  • a median filtering method based on adaptive window is designed. The size of the filter window is corrected according to the number of noise points in the window, non-Gaussian impulse noise is suppressed, the processing of non-noise points is avoided, and the distortion of useful signals is effectively reduced.
  • the noise content adaptively adjusts the size of the filter window, which effectively balances the filtering performance of the method and the computational complexity; then based on an improved artificial bee colony method GDES-ABC, the threshold parameters of the wavelet threshold denoising method are optimized to improve the ability to suppress Gaussian noise .
  • An underwater acoustic signal denoising method based on adaptive window filtering and wavelet threshold optimization including the following steps:
  • S1 Acquire underwater acoustic signal data, and describe the Gaussian/non-Gaussian impulse noise in the underwater acoustic channel through the existing data receiving model;
  • S2 Median filtering method based on adaptive window, suppress non-Gaussian impulse noise, and obtain underwater acoustic signal data without non-Gaussian impulse noise;
  • step S1 is specifically as follows:
  • S1-1 Signal receiving model, in which the signal noise model adopts S ⁇ S distribution and normal distribution model:
  • the time-domain signal y(t) received by the receiving end is expressed in digital form, expressed as a set of discrete samples:
  • s(i) is a noise-free expected signal with random amplitude and phase
  • e(i) is additive marine background noise
  • N is the number of samples
  • the probability density function selected for the Gaussian noise model is as follows: where x is the instantaneous value of the noise pressure;
  • the signal-to-noise ratio SNR is defined as follows:
  • Underwater non-Gaussian noise sources include sound waves from seabed exploration, marine life, sea surface waves, and seabed earthquakes; the probability density function of the sound wave signal from this type of noise source is similar to the normal distribution, but its tailing and strong amplitude The probability is greater, and the duration is shorter, with the characteristic of spikes, which is a kind of burst non-Gaussian pulse signal;
  • the random variable X obeys a stable distribution of ⁇ ; among them, 0 ⁇ 2 indicates the characteristic index, which determines the degree of pulse characteristics. The smaller the ⁇ , the stronger the pulse, the larger the ⁇ , the closer it is to the Gaussian process.
  • -1 ⁇ 1 is the symmetry parameter, used to determine the slope of the distribution
  • ⁇ >0 is the dispersion coefficient, and its meaning is similar to the Gaussian distribution variance
  • MSNR mixed signal-to-noise ratio
  • the underwater acoustic noise model is obtained by superimposing the Gaussian noise and the non-Gaussian impulse noise model, so the underwater acoustic noise e(i) is defined as
  • step S2 The details of the above step S2 are as follows:
  • the new window is recorded as W new , and its length is:
  • the pulse mark N(i) to filter the received signal; among them, the non-noise points in the new window remain unchanged, while the noise points are replaced by the signal median value in the new window; assuming that y ip (i) is the noise point in W new , take out the new Remove all signal samples w new (i) of y ip (i) in the window:
  • the present invention proposes a new adaptive threshold function:
  • the new adaptive threshold function is continuous in (- ⁇ ,- ⁇ j ), (- ⁇ j ,+ ⁇ j ) and (+ ⁇ j ,+ ⁇ ); when When w j,k > ⁇ j , the new adaptive threshold function can be written as:
  • the present invention combines the threshold ⁇ j and exponential factor in the new adaptive threshold function Regarding the unknown threshold parameter, then optimize the estimation of the threshold parameter based on the GDES-ABC method to improve the estimation accuracy and speed, so as to ensure the denoising performance of the proposed method;
  • the population initialization based on the good point set can effectively improve the diversity of the population and prevent the method from falling into the local optimum prematurely;
  • the construction method of the good point is as follows:
  • Ub and Lb are the upper and lower bounds of the solution
  • p max and p min respectively represent the maximum and minimum values of p'; t is the current number of iterations, and t max is the maximum number of iterations; as can be seen from the above formula, in the initial stage of the method, t and p'are very small.
  • the time dynamic elite population contains several optimal solutions. Therefore, based on the elite population, the neighborhood search is more targeted, and the convergence speed can be greatly accelerated; in the later stage of the method, t and p′ are larger, and the dynamic elite The population contains more solutions, which may contain some relatively poor solutions, so the population is more diverse, and the ability of the method to jump out of the local optimum and find the global optimum is strengthened;
  • the improved neighborhood search method of GDES-ABC method based on dynamic elite population is as follows:
  • ⁇ id is a random real number between [-1,1]; Gbest is the global optimal solution, x kd is the random neighborhood of x id , and DXEC d is the dynamic elite population center.
  • the neighborhood search strategy based on the dynamic elite population is described as follows: For each neighborhood search, the hired bees randomly search the neighborhood with the same probability, and a new solution is generated by the improved neighborhood search method; the observation bee randomly selects from the elite population Search the neighborhood, and generate a new solution through an improved neighborhood search method; for the observer, if the new solution is better than the current solution, select the new solution for the next neighborhood search; otherwise, in the next neighborhood search , Re-search the neighborhood randomly from the elite population for the next neighborhood search; this neighborhood search strategy is performed randomly, which not only guarantees the diversity of the population, but also avoids invalid searches;
  • the current temperature is T t
  • the annealing parameter is K
  • the new solution obtained by the improved neighborhood search method is V t
  • its fitness value is fit v ;
  • the simulated annealing selection mechanism is as follows: if fit v > fit i , then accept the new solution, where fit i is the fitness value of the current solution; otherwise, accept the new solution with probability P, and the acceptance probability is defined as
  • the threshold function shown is the parameters ⁇ j and Function; once ⁇ j and Determined, threshold function The new adaptive threshold function is also determined, and the wavelet coefficients can be obtained after the threshold is shrunk, so that the de-noised signal can be reconstructed Therefore, in the GDES-ABC method, the parameters ⁇ j and The composed vector is regarded as the position of the nectar, and the optimal threshold parameter is obtained by minimizing the fitness function.
  • the wavelet coefficients are contracted by constructing a new threshold function and optimal threshold parameters to obtain new wavelet coefficients, and then perform inverse wavelet transform to obtain denoising signals.
  • the invention suppresses the non-Gaussian impulse noise based on the median filter method of the adaptive window, and at the same time, based on the improved artificial bee colony method GDES-ABC, optimizes the threshold parameters of the wavelet threshold denoising method, and improves the suppression ability of the Gaussian noise.
  • the present invention can obtain higher output signal-to-noise ratio (SNR) and noise suppression ratio (NSR), and effectively improves the underwater acoustic communication machine's response to underwater acoustic communication signals such as 2FSK, QPSK, and 16QAM in a Gaussian/non-Gaussian impulse noise environment. Receiving ability.
  • Figure 1 is a brief flow chart of the present invention
  • Figure 2 is a schematic diagram of underwater man-made and natural non-Gaussian noise sources
  • Figure 4 is a comparison diagram of different threshold functions of the present invention.
  • Figure 5 is a comparison chart of output SNR with input SNR after denoising different underwater acoustic communication signals based on different denoising methods (5-1, 5-2, and 5-3 correspond to 2FSK, QPSK, 16QAM, respectively);
  • Figure 6 is a comparison chart of output NSR with input SNR after denoising different underwater acoustic communication signals based on different denoising methods (6-1, 6-2, and 6-3 correspond to 2FSK, QPSK, and 16QAM respectively);
  • Figure 7 is a comparison chart of output SNR with input MSNR after denoising different underwater acoustic communication signals based on different denoising methods (7-1, 7-2, and 7-3 correspond to 2FSK, QPSK, 16QAM, respectively);
  • Figure 8 is a comparison chart of output NSR with input MSNR after denoising different underwater acoustic communication signals based on different denoising methods (8-1, 8-2, 8-3 correspond to 2FSK, QPSK, 16QAM, respectively);
  • Fig. 9 is a comparison diagram of time-domain waveforms after denoising different underwater acoustic communication signals based on the present invention (9-1, 9-2, and 9-3 correspond to 2FSK, QPSK, and 16QAM, respectively).
  • the method for denoising an underwater acoustic signal based on AWFM+GDES in a Gaussian/non-Gaussian impulse noise environment in this embodiment includes the following steps:
  • the time-domain signal y(t) received by the receiving end is expressed in digital form, expressed as a set of discrete samples:
  • s(i) is a noise-free expected signal with random amplitude and phase
  • e(i) is additive marine background noise
  • N is the number of samples
  • the probability density function of the instantaneous value x of the sound pressure of the Gaussian noise source is
  • SNR is defined as follows:
  • underwater man-made and natural non-Gaussian noise sources include sound waves from seabed exploration, marine life, sea surface waves, and seabed earthquakes.
  • the probability density function of the acoustic signal emitted by this type of noise source is similar to the normal distribution, but the probability of its tailing and strong amplitude is greater, and the duration is shorter. It has the characteristics of spikes and belongs to a kind of burst non-Gaussian pulse signal ;
  • the random variable X obeys a stable distribution of ⁇ ; among them, 0 ⁇ 2, which indicates the characteristic index, which determines the degree of pulse characteristics.
  • the stronger the pulse, the larger the ⁇ is, the closer it is to the Gaussian process.
  • -1 ⁇ 1 is the symmetry parameter, used to determine the slope of the distribution
  • ⁇ >0 is the dispersion coefficient, which has a meaning similar to the Gaussian distribution variance
  • the non-Gaussian impulse noise based on the S ⁇ S distribution cannot calculate the variance, so the MSNR is used to describe the noise size, and the MSNR is defined as follows:
  • sort( ⁇ ) is the sorting function.
  • Med median(r(i))
  • median( ⁇ ) represents the median value.
  • differential noise identifier is the differential noise identifier.
  • the sound velocity is c
  • the amplitude is A
  • the sampling frequency is f s
  • the carrier frequency is f c
  • the new window is recorded as W new , and its length is:
  • the new window size And the pulse mark N(i) to filter the received signal.
  • the non-noise points in the new window remain unchanged, and the noise points are replaced by the median value of the signal in the new window.
  • y ip (i) is the noise point in W new
  • the present invention proposes a new adaptive threshold function:
  • ⁇ j is the threshold of the jth layer
  • j 1, 2,..., L
  • L is the number of decomposition layers.
  • the new adaptive threshold function is continuous in (- ⁇ ,- ⁇ j ), (- ⁇ j ,+ ⁇ j ) and (+ ⁇ j ,+ ⁇ ).
  • the new adaptive threshold function can be written as:
  • the new adaptive threshold function can be written as:
  • the threshold function shown in the new adaptive threshold function is a compromise strategy between soft threshold and hard threshold, which has better continuity and smoothness, and retains larger wavelet coefficients. The fidelity of the target signal is stronger.
  • the present invention combines the threshold ⁇ j and exponential factor in the new adaptive threshold function Regarding the unknown threshold parameters, the threshold parameters are optimized and estimated based on the GDES-ABC method to improve the estimation accuracy and speed, so as to ensure the denoising performance of the proposed method.
  • the population initialization based on the good point set can effectively improve the diversity of the population and avoid the method from falling into the local optimum too early.
  • the construction method of the good point is as follows:
  • p is the smallest prime number that satisfies (p-3)/2 ⁇ D
  • D is the dimension of the solution
  • deci ⁇ represents the fractional part
  • r k is a good point. Therefore, the good point set [P SN (1),P SN (2),...,P SN (SN)] T is constructed as follows:
  • Ub and Lb are the upper and lower bounds of the solution
  • the dynamic elite population contains better solutions in the population, and its scale varies with the number of iterations.
  • the ratio is determined according to the following formula:
  • the improved neighborhood search method of GDES-ABC method based on dynamic elite population is as follows:
  • ⁇ id is a random real number between [-1,1]; Gbest is the global optimal solution, x kd is the random neighborhood of x id , and DXEC d is the dynamic elite population center.
  • the neighborhood search strategy based on the dynamic elite population is described as follows: For each neighborhood search, the hired bee randomly searches the neighborhood with the same probability, and a new solution is generated by the improved neighborhood search method. Observing bees randomly search neighborhoods from the elite population, and generate new solutions through improved neighborhood search methods. For the observer bee, if the new solution is better than the current solution, the new solution is selected for the next neighborhood search. Otherwise, in the next neighborhood search, re-search neighborhoods randomly from the elite population for the next neighborhood search.
  • the neighborhood search strategy is carried out randomly, which not only guarantees the diversity of the population, but also avoids invalid search.
  • the current temperature is T t
  • the annealing parameter is K
  • the new solution obtained by the improved neighborhood search method is V t
  • its fitness value is fit v .
  • the simulated annealing selection mechanism is as follows: if fit v > fit i , then accept the new solution, where fit i is the fitness value of the current solution; otherwise, accept the new solution with probability P, and the acceptance probability is defined as
  • the threshold function shown is the parameters ⁇ j and The function. Once ⁇ j and Determined, threshold function The new adaptive threshold function is also determined, and the wavelet coefficients can be obtained after the threshold is shrunk, so that the de-noised signal can be reconstructed Therefore, in the GDES-ABC method, the parameters ⁇ j and The composed vector can be regarded as the position of the nectar, and the optimal threshold parameter is obtained by minimizing the fitness function.
  • common underwater acoustic communication signals such as 2FSK, QPSK, and 16QAM signals are regarded as SOI, and additive white Gaussian noise and non-Gaussian impulse noise are combined into underwater acoustic noise to verify the performance of the present invention.
  • the present invention is denoted as AWFM+GDES; the computer configuration used in the simulation is: Intel i5-4570 processor, Windows 7 operating system, 4G memory, MATLAB R2015b.
  • the output SNR is defined as follows:
  • noise suppression ratio (NSR) is defined as follows:
  • Figure 5 shows the output SNR of 2FSK, QPSK, and 16QAM signals after denoising by AWFM+GDES and other methods as a function of the input SNR.
  • the proposed AWMF method obtains a higher output SNR than the SMF method, and the proposed AWMF+GDES method obtains the highest output SNR than the other five methods.
  • the AWFM+GDES method obtains a higher output SNR than the other three methods.
  • the AWFM+GDES method can obtain a higher output SNR than the other three methods. This means that the proposed AWMF+GDES method is more suitable for 16QAM signal denoising.
  • the output SNR obtained by the AWFM+GDES method is at least 2.3dB, 1.3dB, 1.2dB higher than other methods, respectively.
  • Figure 6 shows the output NSR versus the input SNR after denoising the 2FSK, QPSK, and 16QAM signals with different denoising methods.
  • the parameter setting is shown in Table 1.
  • Input MSNR 20dB. It can be seen from the figure that as the input SNR increases, the output NSR obtained by the six methods first increases and then stabilizes. The trend is basically the same as the result shown in Figure 5, which illustrates the proposed AWMF+GDES from another aspect. Method validity.
  • Figures 7 and 8 respectively show the output SNR and NSR results of 2FSK, QPSK, and 16QAM signals after denoising the 2FSK, QPSK, and 16QAM signals when different denoising methods change with the input MSNR.
  • Figure 9 shows the denoising results of 2FSK, QPSK, and 16QAM signals based on the underwater acoustic signal denoising method based on AWFM+GDES.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)

Abstract

提供了一种基于自适应窗口滤波和小波阈值优化的水声信号去噪方法。该方法首先联合SαS分布和正态分布模型描述水声信道中高斯/非高斯脉冲噪声;并设计了一种基于自适应窗口的中值滤波法,依据窗口内噪声点数量修正滤波窗口大小,抑制非高斯脉冲噪声;然后基于一种改进人工蜂群方法GDES-ABC,优化小波阈值去噪方法的阈值参数,提高对高斯噪声的抑制能力。该方法能有效抑制复杂水声环境中高斯/非高斯脉冲噪声,提高对2FSK、QPSK以及16QAM等水声通信信号的接收能力,获得较高的输出信噪比和噪声抑制比。

Description

基于自适应窗口滤波和小波阈值优化的水声信号去噪方法
本申请要求于2020年06月23日提交中国专利局、申请号为202010509512.6、发明名称为“基于自适应窗口滤波和小波阈值优化的水声信号去噪方法”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本发明属于水声信号去噪技术领域,具体地说,涉及一种高斯/非高斯脉冲噪声环境下基于自适应窗口滤波(AWFM)和小波阈值优化(GDES)的水声信号去噪方法。
背景技术
声波在水下通信领域应用广泛,在水下传输和处理过程中,声波信号会受到水下复杂高斯/非高斯脉冲噪声影响,导致声波信号退化失真,通信质量下降。信号去噪技术是一种用来提高信号质量,降低噪声影响的信号处理方法,被广泛应用在水声通信等领域。
针对水下海底勘探、海洋生物、海面波和海底地震等发出的突发性非高斯脉冲噪声,可通过标准中值滤波法(Standard Median Filter,SMF)进行抑制。然而SMF同时处理了接收信号中有用信号部分,导致有用信号失真。
高斯噪声可被滤波法、小波变换法、经验模式分解法等方法有效抑制。其中基于小波阈值的去噪方法可获得原始信号的渐进最优估计,得到了普遍应用。决定该方法性能的主要因素是对阈值的精确估计和阈值函数的合理构造。目前普遍使用的是在假设噪声模型为高斯噪声模型下,基于多维独立正态变量决策理论的统一阈值;然而统一阈值依赖于噪声方差的精确估计,难以应用于实际未知噪声方差的情况。常见阈值函数包括硬阈值、软阈值以及半软阈值等,该类方法根据固定的结构处理小波系数,缺乏自适应性,降低了信号处理的灵活性。为克服上述限制,基于群智能优化的方法被引入,用来提高小波阈值去噪性能。
然而实际的海洋背景噪声环境中,同时包含高斯和非高斯脉冲噪声,基于智能优化的小波阈值去噪方法主要是用于高斯噪声处理,难以直接适用于海洋水声噪声的全面处理,仍具有大量缺点,具体体现在:首先缺乏统一的建立阈值函数的一般原则,导致阈值函数构造困难;其次阈值参数的确定是一个迭代的过程,通常达到是次优值而不是最优值;并且随着方法迭代次数的增加,种群多样性降低,上述优化方法可能陷入局部极小值。
发明内容
本发明提出一种高斯/非高斯脉冲噪声环境下基于自适应窗口滤波(AWFM)和小波阈值优化(GDES)的水声信号去噪方法,以弥补现有技术的不足。
本发明首先联合SαS分布和正态分布模型描述水声信道中高斯/非高斯脉冲噪声,该模型是一种能够保持和复杂海洋背景噪声的产生机制和传播条件的极限分布,可更好的描述海洋环境背景噪声。并设计了一种基于自适应窗口的中值滤波法,依据窗口内噪声点数量修正滤波窗口大小,抑制非高斯脉冲噪声,避免了对非噪声点的处理,有效降低了有用信号失真,同时基于噪声含量自适应调整滤波窗口大小,有效平衡了方法滤波性能和计算复杂度;然后基于一种改进人工蜂群方法GDES-ABC,优化小波阈值去噪方法的阈值参数,提高对高斯噪声的抑制能力。
为实现上述发明目的,本发明采用下述具体技术方案予以实现:
一种基于自适应窗口滤波和小波阈值优化的水声信号去噪方法,包括以下步骤:
S1:获取水声信号数据,通过现有数据接收模型来描述水声信道中的高斯/非高斯脉冲噪声;
S2:基于自适应窗口的中值滤波法,抑制非高斯脉冲噪声,得到去非高斯脉冲噪声的水声信号数据;
S3:再将S2得到的去非高斯脉冲噪声的水声信号数据进行基于改进人工蜂群小波阈值优化方法(记作GDES-ABC)处理,抑制高斯噪声,最终得到去噪后的水声信号数据。
上述步骤S1具体如下:
S1-1:信号接收模型,其中信号噪声模型采用SαS分布和正态分布模型:
对于单发单收水声通信系统,采用数字形式表示接收端收到的时域信号y(t),用一组离散的样本表示为:
y(i)=s(i)+e(i),i=1,2,...,N
其中s(i)是具有随机幅度和相位的不含噪期望信号;e(i)为加性海洋背景噪声;N是样本数;
S1-2:高斯/非高斯脉冲噪声模型:
高斯噪声模型选用概率密度函数如下:其中,x为噪声压的瞬时值;
Figure PCTCN2021088635-appb-000001
信噪比SNR定义如下:
Figure PCTCN2021088635-appb-000002
其中
Figure PCTCN2021088635-appb-000003
分别为期望信号和高斯噪声的方差;
水下非高斯噪声源,包括海底勘探、海洋生物、海面波和海底地震等发出的声波;这类噪声源发出的声波信号概率密度函数与正态分布相似,但是其拖尾与出现强幅度的概率更大,且持续时间更短,具有尖峰脉冲特性,属于一种突发非高斯脉冲信号;
采用α稳定分布描述水下尖峰脉冲噪声比高斯分布具有更大的优势,若随机变量X的特征函数
Figure PCTCN2021088635-appb-000004
可表示为:
Figure PCTCN2021088635-appb-000005
其中
Figure PCTCN2021088635-appb-000006
-∞<a<∞为实数,表示位置参数,且
Figure PCTCN2021088635-appb-000007
Figure PCTCN2021088635-appb-000008
则随机变量X服从α稳定分布;其中0<α≤2,表示特征指数,决定脉冲特性程度,α越小表示脉冲越强烈,α越大越趋近于高斯过程,当α=2时, 即为高斯分布;-1≤β≤1为对称参数,用于确定分布的斜度;γ>0为分散系数,其含义类似于高斯分布方差;
当β=0时,α稳定分布特征函数表示为
Figure PCTCN2021088635-appb-000009
此时该分布称为对称α稳定分布,记作X~SαS;假设位置参数a=0,此时SαS分布的概率密度函数为:
Figure PCTCN2021088635-appb-000010
基于SαS分布的非高斯脉冲噪声无法计算方差,因此采用混合信噪比MSNR描述噪声大小,MSNR定义如下:
Figure PCTCN2021088635-appb-000011
其中
Figure PCTCN2021088635-appb-000012
和γ分别表示期望信号的方差和非高斯脉冲噪声的分散系数;
为更好的描述实际环境中的噪声,假设水声噪声模型由高斯噪声和非高斯脉冲噪声模型叠加得到,因此定义水声噪声e(i)为
e(i)=e Gauss(i)+e SαS(i)
其中e Gauss(i)、e SαS(i)分别由
Figure PCTCN2021088635-appb-000013
Figure PCTCN2021088635-appb-000014
产生。
上述步骤S2的具体如下:
S2-1:噪声点检测:
假设接收端收到的信号为y=[y(1),y(2),...,y(N)],初始滑动窗口W长度为L W=2n+1,利用初始滑动窗口W取出第i时刻接收信号y中除去中心点y(i)对应的样本w(i):
w(i)=[w 1(i),w 2(i),...,w 2n(i)]
=[y(i-n),...,y(i-1),y(i+1),...,y(i+n)]
对w(i)中信号点从小到大排序,得
r(i)=sort(w(i))
=[r 1(i),r 2(i),...r 2n(i)]
其中sort(·)为排序函数;设Med=median(r(i)),median(·)表示取中值;定义差分噪声识别器为
Figure PCTCN2021088635-appb-000015
对于给定预先设定的脉冲阈值T noise,若d(i)>T noise,则判定y(i)为脉冲噪声点,并令N(i)=1,否则y(i)为期望信号,且N(i)=0,其中N(i)表示脉冲标记;水声接收信号中,设声速为c,振幅为A,采样频率为f s,载波频率为f c,则任意相邻采样点变化率一般不会超过
Figure PCTCN2021088635-appb-000016
且水声接收信号采样长度为
Figure PCTCN2021088635-appb-000017
由此设定脉冲阈值为
Figure PCTCN2021088635-appb-000018
S2-2:自适应窗口大小确定:
对于初始滑动窗口W长度为L W=2n+1,当中心点y(i)为脉冲噪声点时,计算窗口内噪声点数目:
Figure PCTCN2021088635-appb-000019
新窗口记作为W new,长度为:
Figure PCTCN2021088635-appb-000020
S2-3:噪声滤波:
根据新窗口大小
Figure PCTCN2021088635-appb-000021
和脉冲标记N(i)对接收信号进行滤波;其中新窗口内非噪声点不变,而噪声点被新窗口内信号中值替换;假设y ip(i)为W new内噪声点,取出新窗口内除去y ip(i)的所有信号样本w new(i):
Figure PCTCN2021088635-appb-000022
对w new(i)中信号点从小到大排序,得
Figure PCTCN2021088635-appb-000023
则噪声点y ip(i)由下式替代:
y′ ip(i)=median(r new(i))
其中y′ ip(i)为滤波后信号。
上述S3的具体如下:
首先对y′ ip(i)进行小波变换,得到小波系数w j,k,其中,此处j,k表示第j层第k个系数;
再构建新的阈值函数:
S3-1:一种新的阈值函数构造:
本发明提出一种新的自适应阈值函数:
Figure PCTCN2021088635-appb-000024
其中
Figure PCTCN2021088635-appb-000025
为指数因子,取值非负数,λ j为第j层阈值,j=1,2,...,L,L为分解层数;
S3-2:新的阈值函数性质证明:
从连续性的定义来看,很容易证明新的自适应阈值函数在(-∞,-λ j),(-λ j,+λ j)和(+λ j,+∞)是连续的;当w j,k>λ j时,新的自适应阈值函数可写成:
Figure PCTCN2021088635-appb-000026
Figure PCTCN2021088635-appb-000027
当w j,k=λ j
Figure PCTCN2021088635-appb-000028
当|w j,k|<λ j时,新的自适应阈值函数
Figure PCTCN2021088635-appb-000029
Figure PCTCN2021088635-appb-000030
因此
Figure PCTCN2021088635-appb-000031
可得新的自适应阈值函数在点w j,k=λ j连续;当w j,k<-λ j时,新的自适应阈值函数可写成:
Figure PCTCN2021088635-appb-000032
Figure PCTCN2021088635-appb-000033
当|w j,k|≤λ j时,
Figure PCTCN2021088635-appb-000034
Figure PCTCN2021088635-appb-000035
因此,
Figure PCTCN2021088635-appb-000036
可得新的自适应阈值函数在点w j,k=-λ j连续;
当w j,k→+∞时,
Figure PCTCN2021088635-appb-000037
当w j,k→-∞时,
Figure PCTCN2021088635-appb-000038
Figure PCTCN2021088635-appb-000039
因此
Figure PCTCN2021088635-appb-000040
是新的自适应阈值函数的一条渐近线;
S3-3:确定待优化阈值参数:
本发明将新的自适应阈值函数中的阈值λ j和指数因子
Figure PCTCN2021088635-appb-000041
看作未知阈值参数,然后基于GDES-ABC方法对阈值参数进行优化估计,提高估计精度与速度,从而保证所提方法去噪性能;
S3-4:基于佳点集的种群初始化:
基于佳点集的种群初始化可以有效提高种群多样性,避免方法过早陷入局部最优;佳点的构造方法如下:
Figure PCTCN2021088635-appb-000042
其中p是满足(p-3)/2≥D的最小素数,D是解的维度,deci{·}表示取小数部分,r k是佳点;因此,佳点集[P SN(1),P SN(2),...,P SN(SN)] T的构造方法如下:
P SN(i)={deci{r 1*i},...,deci{r D*i}},i=1,2,...,SN
其中[·] T表示转置,SN是种群大小;则初始种群为
X=Lb+(Ub-Lb)*P SN
其中Ub和Lb分别是解的上界和下界,d是当前维解空间,d=1,2,...,D;
S3-5:以动态精英种群指导的领域搜索策略:
动态精英种群包含种群中较好的解,其规模随迭代次数的不同而不同;基于动态精英群的邻域搜索可以有效地加速方法收敛速度,提高搜索效率;首先计算每个X i,i=1,2,...,SN的适应度值,然后取较好的Telite=ceil(p′*SN)个蜜蜂构成动态精英种群DXE i′,i′=1,2,...,Telite,其中ceil(·)表示向上取整,p′是精英种群在所有种群中占的比例,根据下式确定:
Figure PCTCN2021088635-appb-000043
其中p max和p min分别表示p′的最大值和最小值;t是当前迭代次数,t max是最大迭代次数;从上式可以看出,在方法初期阶段,t和p′很小,此时动态精英种群中包含了最优的几个解,因此,基于该精英种群,邻域搜索更具有针对性,收敛速度可大幅加快;在方法后期,t和p′较大,此时动态精英种群中包含了更多的解,其中可能包含一些相对较差的解,因此种群更具有多样性,方法跳出局部最优和找到全局最优的能力被加强;
GDES-ABC方法基于动态精英种群的改进的邻域搜索方法如下:
v id=DXEC did(Gbest d-x kd),d=1,2,...,D
其中,φ id是[-1,1]之间的随机实数;Gbest是全局最优解,x kd是x id的随机邻域,DXEC d是动态精英种群中心,表达式如下:
Figure PCTCN2021088635-appb-000044
基于动态精英种群的邻域搜索策略描述如下:对于每个邻域搜索,雇佣蜂以相同的概率随机地搜索邻域,并且由改进的邻域搜索方法生成新解;观察蜂从精英种群中随机搜索邻域,并通过改进的邻域搜索方法生成新的 解;对于观察蜂,如果新解比当前解更好,则选择该新解进行下一次邻域搜索;否则,在下一次邻域搜索中,重新从精英种群中随机搜索邻域进行下一次邻域搜索;该邻域搜索策略是随机进行的,既保证了种群多样性,又避免了无效搜索;
S3-6:模拟退火选择机制:
假设第t次迭代时,当前温度为T t,退火参数为K,由改进的邻域搜索方法得到新解为V t,其适应度值为fit v;模拟退火选择机制如下:若fit v>fit i,则接受新解,其中fit i为当前解适应度值;否则以概率P接受新解,接受概率定义为
Figure PCTCN2021088635-appb-000045
其中
Figure PCTCN2021088635-appb-000046
随迭代次数改变;其中β≤1是一个常数,取值0.7,σ fit为所有解的适应度的标准差;
由接受概率可知,方法初期,t较小,T t较大,因此P较大,方法接受一定的较差值,蜂群具有较大的开拓能力;方法后期,t变大,T t减小,P减小,方法以较大概率拒绝较差值,保证了方法搜索最优解的能力,避免了无效搜索。
S3-7:将待去噪信号和去噪信号之间的均方误差作为S3中改进的人工蜂群方法的适应度函数,在获取最小均方误差情况下得到最优阈值参数;适应度函数表示为:
Figure PCTCN2021088635-appb-000047
其中s(i)为训练信号,
Figure PCTCN2021088635-appb-000048
为重构训练信号,N为训练信号长度。由新的自适应阈值函数可知,所示阈值函数是参数λ j
Figure PCTCN2021088635-appb-000049
的函数;一旦λ j
Figure PCTCN2021088635-appb-000050
确定,阈值函数新的自适应阈值函数也被确定,阈值收缩后即可得到小波系数,从而可重构出去噪后的信号
Figure PCTCN2021088635-appb-000051
因此,在GDES-ABC方法中,可将由参数λ j
Figure PCTCN2021088635-appb-000052
组成的向量看作是蜜源的位置,并通过最小化适应度函数来获得最优阈值参数。
最后,对小波系数通过构建新的阈值函数和最优阈值参数进行收缩处理得到新的小波系数,再进行逆小波变换,得到去噪信号。
与现有方法相比,本发明优点和技术效果如下:
本发明基于自适应窗口的中值滤波法抑制非高斯脉冲噪声,同时基于改进人工蜂群方法GDES-ABC,优化小波阈值去噪方法的阈值参数,提高对高斯噪声的抑制能力。本发明能获得更高的输出信噪比(SNR)和噪声抑制比(NSR),有效提高了水声通信机在高斯/非高斯脉冲噪声环境下对2FSK、QPSK以及16QAM等水声通信信号的接收能力。
附图说明
图1是本发明的简要流程图;
图2是水下人为和自然非高斯噪声源示意图;
图3 L W=2n+1时初始滑动窗口示意图;
图4是本发明的不同阈值函数对比图;
图5是基于不同去噪方法的对不同水声通信信号去噪后输出SNR随输入SNR变化对比图(5-1、5-2、5-3分别对应2FSK、QPSK、16QAM);
图6是基于不同去噪方法的对不同水声通信信号去噪后输出NSR随输入SNR变化对比图(6-1、6-2、6-3分别对应2FSK、QPSK、16QAM);
图7是基于不同去噪方法的对不同水声通信信号去噪后输出SNR随输入MSNR变化对比图(7-1、7-2、7-3分别对应2FSK、QPSK、16QAM);
图8是基于不同去噪方法的对不同水声通信信号去噪后输出NSR随输入MSNR变化对比图(8-1、8-2、8-3分别对应2FSK、QPSK、16QAM);
图9是基于本发明对不同水声通信信号去噪后的时域波形对比图(9-1、9-2、9-3分别对应2FSK、QPSK、16QAM)。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白,以下将结合附图和实施例,对本发明作进一步详细说明。
实施例1:
参见图1所示,本实施例所述的一种高斯/非高斯脉冲噪声环境下基于 AWFM+GDES的水声信号去噪方法,包括以下步骤:
S1:联合SαS分布和正态分布模型描述水声信道中高斯/非高斯脉冲噪声;具体步骤如下:
S1-1:信号接收模型:
对于单发单收水声通信系统,采用数字形式表示接收端收到的时域信号y(t),用一组离散的样本表示为:
y(i)=s(i)+e(i),i=1,2,...,N
其中s(i)是具有随机幅度和相位的不含噪期望信号;e(i)为加性海洋背景噪声;N是样本数;
S1-2:高斯/非高斯脉冲噪声模型:
高斯噪声源的声压的瞬时值x的概率密度函数为
Figure PCTCN2021088635-appb-000053
SNR定义如下:
Figure PCTCN2021088635-appb-000054
其中
Figure PCTCN2021088635-appb-000055
分别为期望信号和高斯噪声的方差;
参见图2所示,水下人为和自然非高斯噪声源,包括海底勘探、海洋生物、海面波和海底地震等发出的声波。这类噪声源发出的声波信号概率密度函数与正态分布相似,但是其拖尾与出现强幅度的概率更大,且持续时间更短,具有尖峰脉冲特性,属于一种突发非高斯脉冲信号;
采用α稳定分布描述水下尖峰脉冲噪声比高斯分布具有更大的优势,若随机变量X的特征函数可表示为:
Figure PCTCN2021088635-appb-000056
其中
Figure PCTCN2021088635-appb-000057
-∞<a<∞为实数,表示位置参数,且
Figure PCTCN2021088635-appb-000058
Figure PCTCN2021088635-appb-000059
则随机变量X服从α稳定分布;其中,0<α≤2,表示特征指数,决定脉冲特性程度,α越小表示脉冲越强烈,α越大越趋近于高斯过程,当α=2时,即为高斯分布;-1≤β≤1为对称参数,用于确定分布的斜度;γ>0为分散系数,其含义类似于高斯分布方差;
当β=0时,α稳定分布特征函数表示为
Figure PCTCN2021088635-appb-000060
此时该分布称为对称α稳定分布,记作X~SαS;,假设a=0,此时SαS分布的概率密度函数为:
Figure PCTCN2021088635-appb-000061
基于SαS分布的非高斯脉冲噪声无法计算方差,因此采用MSNR描述噪声大小,MSNR定义如下:
Figure PCTCN2021088635-appb-000062
其中
Figure PCTCN2021088635-appb-000063
和γ分别表示期望信号的方差和非高斯脉冲噪声的分散系数;为更好的描述实际环境中的噪声,假设水声噪声模型由高斯噪声和非高斯脉冲噪声模型叠加得到,因此定义水声噪声e(i)为
e(i)=e Gauss(i)+e SαS(i)
其中e Gauss(i)、e SαS(i)分别由
Figure PCTCN2021088635-appb-000064
Figure PCTCN2021088635-appb-000065
产生。
S2:基于自适应窗口的中值滤波法,记作AWFM,抑制非高斯脉冲噪声;具体步骤如下:
S2-1:噪声点检测:
假设接收端收到的信号为y=[y(1),y(2),...,y(N)],初始滑动窗口W长度为L W=2n+1,参见图3所示,利用初始滑动窗口W取出第i时刻接收信号y中除去中心点y(i)对应的样本w(i):
w(i)=[w 1(i),w 2(i),...,w 2n(i)]
=[y(i-n),...,y(i-1),y(i+1),...,y(i+n)]
对w(i)中信号点从小到大排序,得
r(i)=sort(w(i))
=[r 1(i),r 2(i),...r 2n(i)]
其中sort(·)为排序函数。设Med=median(r(i)),median(·)表示取中值。定义差分噪声识别器为
Figure PCTCN2021088635-appb-000066
对于给定预先设定的脉冲阈值T noise,若d(i)>T noise,则判定y(i)为脉冲噪声点,并令N(i)=1,否则y(i)为期望信号,且N(i)=0,其中N(i)表示脉冲标记。水声接收信号中,设声速为c,振幅为A,采样频率为f s,载波频率为f c,则任意相邻采样点变化率一般不会超过
Figure PCTCN2021088635-appb-000067
且水声接收信号采样长度为
Figure PCTCN2021088635-appb-000068
由此设定冲击阈值为
Figure PCTCN2021088635-appb-000069
S2-2:自适应窗口大小确定:
对于初始滑动窗口W长度为L W=2n+1,当中心点y(i)为脉冲噪声点时,计算窗口内噪声点数目:
Figure PCTCN2021088635-appb-000070
新窗口记作为W new,长度为:
Figure PCTCN2021088635-appb-000071
S2-3:噪声滤波:
根据新窗口大小
Figure PCTCN2021088635-appb-000072
和脉冲标记N(i)对接收信号进行滤波。其中新窗口内非噪声点不变,而噪声点被新窗口内信号中值替换。假设y ip(i)为W new内噪声点,取出新窗口内除去y ip(i)的所有信号样本w new(i):
Figure PCTCN2021088635-appb-000073
对w new(i)中信号点从小到大排序,得
Figure PCTCN2021088635-appb-000074
则噪声点y ip(i)由下式替代:
y′ ip(i)=median(r new(i))
其中y′ ip(i)为滤波后信号。
S3:基于GDES-ABC方法,优化小波阈值去噪方法的阈值参数,抑制高斯噪声;具体步骤如下:
S3-1:一种新的阈值函数构造:
本发明提出一种新的自适应阈值函数:
Figure PCTCN2021088635-appb-000075
其中
Figure PCTCN2021088635-appb-000076
为指数因子,取值非负数,λ j为第j层阈值,j=1,2,...,L,L为分解层数。
S3-2:新的阈值函数性质证明:
从连续性的定义来看,很容易证明新的自适应阈值函数在(-∞,-λ j),(-λ j,+λ j)和(+λ j,+∞)是连续的。当w j,k>λ j时,新的自适应阈值函数可写成:
Figure PCTCN2021088635-appb-000077
Figure PCTCN2021088635-appb-000078
当w j,k=λ j
Figure PCTCN2021088635-appb-000079
当|w j,k|<λ j时,新的自适应阈值函数
Figure PCTCN2021088635-appb-000080
Figure PCTCN2021088635-appb-000081
因此
Figure PCTCN2021088635-appb-000082
可得新的自适应阈值函数在点w j,k=λ j连续。当w j,k<-λ j时,新的自适应阈值函数可写成:
Figure PCTCN2021088635-appb-000083
Figure PCTCN2021088635-appb-000084
当|w j,k|≤λ j时,
Figure PCTCN2021088635-appb-000085
Figure PCTCN2021088635-appb-000086
因此,
Figure PCTCN2021088635-appb-000087
可得新的自适应阈值函数在点w j,k=-λ j连续。
当w j,k→+∞时,
Figure PCTCN2021088635-appb-000088
当w j,k→-∞时,
Figure PCTCN2021088635-appb-000089
Figure PCTCN2021088635-appb-000090
因此
Figure PCTCN2021088635-appb-000091
是新的自适应阈值函数的一条渐近线,参见图4所示,其中阈值λ=5,横轴表示小波系数,范围[-10,10],纵轴为阈值收缩后得到的小波系数。从图中可以看出,新的自适应阈值函数所示阈值函数是软阈值和硬阈值之间的一种折衷的策略,具有更好的连续性和平滑性,且保留了较大小波系数,对目标信号保真能力更强。
S3-3:确定待优化阈值参数:
本发明将新的自适应阈值函数中的阈值λ j和指数因子
Figure PCTCN2021088635-appb-000092
看作未知阈值 参数,然后基于GDES-ABC方法对阈值参数进行优化估计,提高估计精度与速度,从而保证所提方法去噪性能。
S3-4:基于佳点集的种群初始化:
基于佳点集的种群初始化可以有效提高种群多样性,避免方法过早陷入局部最优。佳点的构造方法如下:
Figure PCTCN2021088635-appb-000093
其中p是满足(p-3)/2≥D的最小素数,D是解的维度,deci{·}表示取小数部分,r k是佳点。因此,佳点集[P SN(1),P SN(2),...,P SN(SN)] T的构造方法如下:
P SN(i)={deci{r 1*i},...,deci{r D*i}},i=1,2,...,SN
其中[·] T表示转置,SN是种群大小;则初始种群为
X=Lb+(Ub-Lb)*P SN
其中Ub和Lb分别是解的上界和下界,d是当前维解空间,d=1,2,...,D;
S3-5:以动态精英种群指导的领域搜索策略:
动态精英种群包含种群中较好的解,其规模随迭代次数的不同而不同。基于动态精英群的邻域搜索可以有效地加速方法收敛速度,提高搜索效率;首先计算每个X i,i=1,2,...,SN的适应度值,然后取较好的Telite=ceil(p′*SN)个蜜蜂构成动态精英种群DXE i′,i′=1,2,...,Telite,其中ceil(·)表示向上取整,p′是精英种群在所有种群中占的比例,根据下式确定:
Figure PCTCN2021088635-appb-000094
其中p max和p min分别表示p′的最大值和最小值;t是当前迭代次数,t max是最大迭代次数。从上式可以看出,在方法初期阶段,t和p′很小,此时动态精英种群中包含了最优的几个解,因此,基于该精英种群,邻域搜索更具有针对性,收敛速度可大幅加快。在方法后期,t和p′很小较大,此时动态精英种群中包含了更多的解,其中可能包含一些相对较差的解,因此种群更具有多样性,方法跳出局部最优和找到全局最优的能力被加强。
GDES-ABC方法基于动态精英种群的改进的邻域搜索方法如下:
v id=DXEC did(Gbest d-x kd),d=1,2,...,D
其中,φ id是[-1,1]之间的随机实数;Gbest是全局最优解,x kd是x id的随机邻域,DXEC d是动态精英种群中心,表达式如下:
Figure PCTCN2021088635-appb-000095
基于动态精英种群的邻域搜索策略描述如下:对于每个邻域搜索,雇佣蜂以相同的概率随机地搜索邻域,并且由改进的邻域搜索方法生成新解。观察蜂从精英种群中随机搜索邻域,并通过改进的邻域搜索方法生成新的解。对于观察蜂,如果新解比当前解更好,则选择该新解进行下一次邻域搜索。否则,在下一次邻域搜索中,重新从精英种群中随机搜索邻域进行下一次邻域搜索。该邻域搜索策略是随机进行的,既保证了种群多样性,又避免了无效搜索。
S3-6:模拟退火选择机制:
假设第t次迭代时,当前温度为T t,退火参数为K,由改进的邻域搜索方法得到新解为V t,其适应度值为fit v。模拟退火选择机制如下:若fit v>fit i,则接受新解,其中fit i为当前解适应度值;否则以概率P接受新解,接受概率定义为
Figure PCTCN2021088635-appb-000096
其中
Figure PCTCN2021088635-appb-000097
随迭代次数改变。其中β≤1是一个常数,通常取值0.7,σ fit为所有解的适应度的标准差。
由接受概率可知,方法初期,t较小,T t较大,因此P较大,方法接受一定的较差值,蜂群具有较大的开拓能力。方法后期,t变大,T t减小,P减小,方法以较大概率拒绝较差值,保证了方法搜索最优解的能力,避免了无效搜索。
S3-7:将待去噪信号和去噪信号之间的均方误差作为S3中改进的人工蜂群方法的适应度函数,在获取最小均方误差情况下得到最优阈值参数。适应度函数表示为:
Figure PCTCN2021088635-appb-000098
其中s(i)为训练信号,
Figure PCTCN2021088635-appb-000099
为重构训练信号,N为训练信号长度。由新的自适应阈值函数可知,所示阈值函数是参数λ j
Figure PCTCN2021088635-appb-000100
的函数。一旦λ j
Figure PCTCN2021088635-appb-000101
确定,阈值函数新的自适应阈值函数也被确定,阈值收缩后即可得到小波系 数,从而可重构出去噪后的信号
Figure PCTCN2021088635-appb-000102
因此,在GDES-ABC方法中,可将由参数λ j
Figure PCTCN2021088635-appb-000103
组成的向量可以看作是蜜源的位置,并通过最小化适应度函数来获得最优阈值参数。
实施例2:仿真试验结果比较分析
本实施例将2FSK、QPSK和16QAM信号等常见的水声通信信号视为SOI,将加性高斯白噪声和非高斯脉冲噪声合并为水声噪声,验证本发明的性能。其中本发明记作AWFM+GDES;仿真使用的计算机配置是:英特尔i5-4570处理器,Windows 7操作系统,4G内存,MATLAB R2015b。输出SNR定义如下:
Figure PCTCN2021088635-appb-000104
噪声抑制比(Noise suppression ratio,NSR)定义如下:
Figure PCTCN2021088635-appb-000105
其中,s(i)和
Figure PCTCN2021088635-appb-000106
分别为期望信号和估计信号;
Figure PCTCN2021088635-appb-000107
Figure PCTCN2021088635-appb-000108
分别为期望信号和估计信号均值,N为信号长度。
图5分别给出了AWFM+GDES及其他方法对2FSK、QPSK、16QAM信号去噪后的输出SNR随输入SNR变化曲线。其中输入MSNR=20dB,其他参数设置如表1。从图中可以看出,对于每种信号,随着输入SNR的增加,6种方法获得输出SNR先增加,后趋于稳定。此外,所提出的AWMF方法获得比SMF方法更高的输出SNR,并且所提出的AWMF+GDES方法比其他5种方法获得了最高的输出SNR。对于2FSK和QPSK信号,当输入SNR>-5dB时,AWFM+GDES方法获得比其他三种方法更高的输出SNR。对于16QAM信号,当输入SNR>-10dB时,AWFM+GDES方法即可获得比其他三种方法更高的输出SNR。这意味着所提出的AWMF+GDES方法更适合于16QAM信号的去噪。当输入SNR=20dB时,对于2FSK、QPSK、16QAM信号,AWFM+GDES方法获得输出SNR至少分别高出其他方法 2.3dB,1.3dB,1.2dB。
表1 所提方法参数设置
Figure PCTCN2021088635-appb-000109
图6分别给出了不同去噪方法对2FSK、QPSK、16QAM信号去噪后的输出NSR随输入SNR变化曲线。参数设置如表1,输入MSNR=20dB。从图中可以看出,随着输入SNR的增加,6种方法获得输出NSR先增加,后趋于稳定,其趋势基本和图5所示结果一致,从另一个方面说明了所提AWMF+GDES方法有效性。
图7、8分别给出了不同去噪方法随输入MSNR变化时,对2FSK、QPSK、16QAM信号去噪后的输出SNR和NSR结果。其中输入SNR=5dB,其他参数设置如表1。从图中可以看出,随着输入MSNR的增加,6种方法获得输出SNR和NSR逐渐增大。与其他五种方法相比,所提的AWMF+GDES方法获得了最高的输出SNR和NSR,这表明所提出的AWMF+GDES方法在输出SNR和NSR方面都能获得更好的水声信号去噪性能。
图9分别给出了基于AWFM+GDES的水声信号去噪方法对2FSK、QPSK、16QAM信号去噪后的结果。参数设置部分如表1所示,其中输入SNR=5dB,MSNR=20dB。从图中可以看出,所有的期望信号都被水声噪声污染了。然而,基于AWMF+GDES处理后,大部分噪声被消除。且 去噪后的信号保留了更多期望信号的细节信息。由此可见,本发明提出的基于AWFM+GDES的水声信号去噪方法更适合于实际水下环境。
以上所述之实施例子只为本发明之较佳实施例,并非以此限制本发明的实施范围,故凡依本发明之形状、原理所作的变化,均应涵盖在本发明的保护范围内。

Claims (4)

  1. 一种基于自适应窗口滤波和小波阈值优化的水声信号去噪方法,其特征在于,该去噪方法包括以下步骤:
    S1:获取水声信号数据,通过现有数据接收模型来描述水声信道中的高斯/非高斯脉冲噪声;
    S2:基于自适应窗口的中值滤波法,抑制非高斯脉冲噪声,得到去非高斯脉冲噪声的水声信号数据;
    S3:再将S2得到的去非高斯脉冲噪声的水声信号数据进行基于改进人工蜂群小波阈值优化方法处理,抑制高斯噪声,最终得到去噪后的水声信号数据。
  2. 如权利要求1所述的去噪方法,其特征在于,所述步骤S1具体如下:
    S1-1:信号接收模型,其中信号噪声模型采用SαS分布和正态分布模型:
    对于单发单收水声通信系统,采用数字形式表示接收端收到的时域信号y(t),用一组离散的样本表示为:
    y(i)=s(i)+e(i),i=1,2,...,N
    其中s(i)是具有随机幅度和相位的不含噪期望信号;e(i)为加性海洋背景噪声;N是样本数;
    S1-2:高斯/非高斯脉冲噪声模型:
    高斯噪声模型选用概率密度函数如下:其中,x为噪声压的瞬时值;
    Figure PCTCN2021088635-appb-100001
    信噪比SNR定义如下:
    Figure PCTCN2021088635-appb-100002
    其中
    Figure PCTCN2021088635-appb-100003
    分别为期望信号和高斯噪声的方差;
    采用α稳定分布描述水下尖峰脉冲噪声比高斯分布具有更大的优势,若随机变量X的特征函数
    Figure PCTCN2021088635-appb-100004
    可表示为:
    Figure PCTCN2021088635-appb-100005
    其中
    Figure PCTCN2021088635-appb-100006
    -∞<a<∞为实数,表示位置参数,且
    Figure PCTCN2021088635-appb-100007
    Figure PCTCN2021088635-appb-100008
    则随机变量X服从α稳定分布;其中0<α≤2,表示特征指数,决定脉冲特性程度,α越小表示脉冲越强烈,α越大越趋近于高斯过程,当α=2时,即为高斯分布;-1≤β≤1为对称参数,用于确定分布的斜度;γ>0为分散系数,其含义类似于高斯分布方差;
    当β=0时,α稳定分布特征函数表示为
    Figure PCTCN2021088635-appb-100009
    此时该分布称为对称α稳定分布,记作X~SαS;假设位置参数a=0,此时SαS分布的概率密度函数为:
    Figure PCTCN2021088635-appb-100010
    基于SαS分布的非高斯脉冲噪声无法计算方差,因此采用混合信噪比MSNR描述噪声大小,MSNR定义如下:
    Figure PCTCN2021088635-appb-100011
    其中
    Figure PCTCN2021088635-appb-100012
    和γ分别表示期望信号的方差和非高斯脉冲噪声的分散系数;
    假设水声噪声模型由高斯噪声和非高斯脉冲噪声模型叠加得到,因此定义水声噪声e(i)为
    e(i)=e Gauss(i)+e SαS(i)
    其中e Gauss(i)、e SαS(i)分别由
    Figure PCTCN2021088635-appb-100013
    Figure PCTCN2021088635-appb-100014
    产生。
  3. 如权利要求1所述的去噪方法,其特征在于,所述步骤S2的具体如下:
    S2-1:噪声点检测:
    假设接收端收到的信号为y=[y(1),y(2),...,y(N)],初始滑动窗口W长度为 L W=2n+1,利用初始滑动窗口W取出第i时刻接收信号y中除去中心点y(i)对应的样本w(i):
    w(i)=[w 1(i),w 2(i),...,w 2n(i)]
    =[y(i-n),...,y(i-1),y(i+1),...,y(i+n)]
    对w(i)中信号点从小到大排序,得
    r(i)=sort(w(i))
    =[r 1(i),r 2(i),...r 2n(i)]
    其中sort(·)为排序函数;设Med=median(r(i)),median(·)表示取中值;定义差分噪声识别器为
    Figure PCTCN2021088635-appb-100015
    对于给定预先设定的脉冲阈值T noise,若d(i)>T noise,则判定y(i)为脉冲噪声点,并令N(i)=1,否则y(i)为期望信号,且N(i)=0,其中N(i)表示脉冲标记;水声接收信号中,设声速为c,振幅为A,采样频率为f s,载波频率为f c,则任意相邻采样点变化率一般不会超过
    Figure PCTCN2021088635-appb-100016
    且水声接收信号采样长度为
    Figure PCTCN2021088635-appb-100017
    由此设定脉冲阈值为
    Figure PCTCN2021088635-appb-100018
    S2-2:自适应窗口大小确定:
    对于初始滑动窗口W长度为L W=2n+1,当中心点y(i)为脉冲噪声点时,计算窗口内噪声点数目:
    Figure PCTCN2021088635-appb-100019
    新窗口记作为W new,长度为:
    Figure PCTCN2021088635-appb-100020
    S2-3:噪声滤波:
    根据新窗口大小
    Figure PCTCN2021088635-appb-100021
    和脉冲标记N(i)对接收信号进行滤波;其中新窗口内非噪声点不变,而噪声点被新窗口内信号中值替换;假设y ip(i)为 W new内噪声点,取出新窗口内除去y ip(i)的所有信号样本w new(i):
    Figure PCTCN2021088635-appb-100022
    对w new(i)中信号点从小到大排序,得
    Figure PCTCN2021088635-appb-100023
    则噪声点y ip(i)由下式替代:
    y′ ip(i)=median(r new(i))
    其中y′ ip(i)为滤波后信号。
  4. 如权利要求1所述的去噪方法,其特征在于,所述步骤S3具体如下:
    首先对y′ ip(i)进行小波变换,得到小波系数w j,k,其中,此处j,k表示第j层第k个系数;
    再构建新的阈值函数:
    S3-1:一种新的阈值函数构造:
    Figure PCTCN2021088635-appb-100024
    其中
    Figure PCTCN2021088635-appb-100025
    为指数因子,取值非负数,λ j为第j层阈值,j=1,2,...,L,L为分解层数;
    S3-2:确定待优化阈值参数:
    将新的自适应阈值函数中的阈值λ j和指数因子
    Figure PCTCN2021088635-appb-100026
    看作未知阈值参数,然后基于GDES-ABC方法对阈值参数进行优化估计,提高估计精度与速度;
    S3-3:基于佳点集的种群初始化:
    基于佳点集的种群初始化可以有效提高种群多样性,避免方法过早陷入局部最优;佳点的构造方法如下:
    Figure PCTCN2021088635-appb-100027
    其中p是满足(p-3)/2≥D的最小素数,D是解的维度,deci{·}表示取小数部分,r k是佳点;因此,佳点集[P SN(1),P SN(2),...,P SN(SN)] T的构造方法如下:
    P SN(i)={deci{r 1*i},...,deci{r D*i}},i=1,2,...,SN
    其中[·] T表示转置,SN是种群大小;则初始种群为
    X=Lb+(Ub-Lb)*P SN
    其中Ub和Lb分别是解的上界和下界,d是当前维解空间,d=1,2,...,D;
    S3-4:以动态精英种群指导的领域搜索策略:
    动态精英种群包含种群中较好的解,其规模随迭代次数的不同而不同;基于动态精英群的邻域搜索可以有效地加速方法收敛速度,提高搜索效率;首先计算每个X i,i=1,2,...,SN的适应度值,然后取较好的Telite=ceil(p′*SN)个蜜蜂构成动态精英种群DXE i′,i′=1,2,...,Telite,其中ceil(·)表示向上取整,p′是精英种群在所有种群中占的比例,根据下式确定:
    Figure PCTCN2021088635-appb-100028
    其中p max和p min分别表示p′的最大值和最小值;t是当前迭代次数,t max是最大迭代次数;
    GDES-ABC方法基于动态精英种群的改进的邻域搜索方法如下:
    v id=DXEC did(Gbest d-x kd),d=1,2,...,D
    其中,φ id是[-1,1]之间的随机实数;Gbest是全局最优解,x kd是x id的随机邻域,DXEC d是动态精英种群中心,表达式如下:
    Figure PCTCN2021088635-appb-100029
    基于动态精英种群的邻域搜索策略描述如下:对于每个邻域搜索,雇佣蜂以相同的概率随机地搜索邻域,并且由改进的邻域搜索方法生成新解;观察蜂从精英种群中随机搜索邻域,并通过改进的邻域搜索方法生成新的解;对于观察蜂,如果新解比当前解更好,则选择该新解进行下一次邻域搜索;否则,在下一次邻域搜索中,重新从精英种群中随机搜索邻域进行下一次邻域搜索;该邻域搜索策略是随机进行的;
    S3-5:模拟退火选择机制:
    假设第t次迭代时,当前温度为T t,退火参数为K,由改进的邻域搜索方法得到新解为V t,其适应度值为fit v;模拟退火选择机制如下:若fit v>fit i,则接受新解,其中fit i为当前解适应度值;否则以概率P接受新解,接受概率定义为
    Figure PCTCN2021088635-appb-100030
    其中
    Figure PCTCN2021088635-appb-100031
    随迭代次数改变;其中β≤1是一个常数,取值0.7,σ fit为所有解的适应度的标准差;
    由接受概率可知,方法初期,t较小,T t较大,因此P较大,方法接受一定的较差值,蜂群具有较大的开拓能力;方法后期,t变大,T t减小,P减小,方法以较大概率拒绝较差值;
    S3-6:将待去噪信号和去噪信号之间的均方误差作为S3中改进的人工蜂群方法的适应度函数,在获取最小均方误差情况下得到最优阈值参数;适应度函数表示为:
    Figure PCTCN2021088635-appb-100032
    其中s(i)为训练信号,
    Figure PCTCN2021088635-appb-100033
    为重构训练信号,N为训练信号长度。由新的自适应阈值函数可知,所示阈值函数是参数λ j
    Figure PCTCN2021088635-appb-100034
    的函数;一旦λ j
    Figure PCTCN2021088635-appb-100035
    确定,阈值函数新的自适应阈值函数也被确定,阈值收缩后即可得到小波系数,从而可重构出去噪后的信号
    Figure PCTCN2021088635-appb-100036
    最后,对小波系数通过构建新的阈值函数和最优阈值参数进行收缩处理得到新的小波系数,再进行逆小波变换,得到去噪信号。
PCT/CN2021/088635 2020-06-23 2021-04-21 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法 WO2021258832A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010509512.6 2020-06-23
CN202010509512.6A CN111681667B (zh) 2020-06-23 2020-06-23 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法

Publications (1)

Publication Number Publication Date
WO2021258832A1 true WO2021258832A1 (zh) 2021-12-30

Family

ID=72454897

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/088635 WO2021258832A1 (zh) 2020-06-23 2021-04-21 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法

Country Status (2)

Country Link
CN (1) CN111681667B (zh)
WO (1) WO2021258832A1 (zh)

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114734440A (zh) * 2022-04-15 2022-07-12 同济大学 一种基于upf-rbf组合模型的混联双臂搬运机器人运动学参数精准标定方法
CN115100864A (zh) * 2022-06-24 2022-09-23 北京联合大学 一种基于改进麻雀搜索算法的交通信号控制优化方法
CN115816514A (zh) * 2023-02-16 2023-03-21 极限人工智能有限公司 基于测量电变量的机器人关节制动器故障检测方法及系统
CN115953790A (zh) * 2022-09-29 2023-04-11 江苏智联天地科技有限公司 一种标签检测识别方法与系统
CN116304581A (zh) * 2023-05-10 2023-06-23 佛山市钒音科技有限公司 一种空调用智能电控系统
CN116485884A (zh) * 2023-06-28 2023-07-25 四川君安天源精酿啤酒有限公司 基于计算机视觉的精酿啤酒瓶口实时定位方法及系统
CN116559421A (zh) * 2023-04-03 2023-08-08 杭州臻稀生物科技有限公司 一种自动化荧光免疫分析仪及分析方法
CN116597856A (zh) * 2023-07-18 2023-08-15 山东贝宁电子科技开发有限公司 基于蛙人对讲的语音质量增强方法
CN116776749A (zh) * 2023-08-22 2023-09-19 吉林大学 一种考虑不确定性的电动汽车无线充电装置优化方法
CN116859711A (zh) * 2023-06-27 2023-10-10 中科沃业江苏生物有限公司 一种基于iabc优化模糊pid的恒流泵控制系统
CN117370737A (zh) * 2023-12-08 2024-01-09 成都信息工程大学 一种基于自适应高斯滤波器的非稳态非高斯噪声去除方法
CN117395626A (zh) * 2023-12-11 2024-01-12 厦门大学 基于元学习及noma的水声网络水质监测数据搜集方法
CN117633554A (zh) * 2023-12-14 2024-03-01 艾信智慧医疗科技发展(苏州)有限公司 医用箱式物流传输监测及预警系统
CN117668468A (zh) * 2024-01-31 2024-03-08 山东省舜天化工集团有限公司 一种化工制备数据智能分析管理系统
CN117665788A (zh) * 2024-02-01 2024-03-08 湖南科技大学 一种基于微波测量数据的噪声处理方法
CN117786325A (zh) * 2024-02-27 2024-03-29 山东晨晖电子科技有限公司 一种大口径物联网水表的环境温度智慧预警系统
CN117828282A (zh) * 2024-03-06 2024-04-05 山东泰霖信息工程有限公司 一种基于自适应滤波的数据高效处理方法
CN117973429A (zh) * 2024-04-01 2024-05-03 南京信息工程大学 一种应用于非高斯噪声滤波的模型参数比估计方法
CN116559421B (zh) * 2023-04-03 2024-05-31 杭州臻稀生物科技有限公司 一种自动化荧光免疫分析仪及分析方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111681667B (zh) * 2020-06-23 2021-05-04 青岛科技大学 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法
CN112133321B (zh) * 2020-09-23 2021-05-14 青岛科技大学 一种基于盲源分离的水声信号高斯/非高斯噪声抑制方法
CN112530449B (zh) * 2020-10-20 2022-09-23 国网黑龙江省电力有限公司伊春供电公司 基于仿生小波变换的语音增强方法
CN112633225B (zh) * 2020-12-31 2023-07-18 矿冶科技集团有限公司 矿用微震信号滤波方法
CN113822329B (zh) * 2021-08-10 2023-11-03 国网新源控股有限公司 一种水电机组主轴摆度信号处理方法及装置
CN114462458B (zh) * 2022-04-11 2022-07-08 自然资源部第一海洋研究所 一种舰船水下信号降噪及目标增强方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7450779B2 (en) * 2004-05-21 2008-11-11 Imaging Dynamics Company Ltd. De-noising digital radiological images
CN105528768A (zh) * 2015-12-10 2016-04-27 国网四川省电力公司天府新区供电公司 一种图像去噪方法
CN106530254A (zh) * 2016-11-14 2017-03-22 山东理工大学 一种基于小波阈值函数与改进中值滤波融合的抑制图像混合噪声算法
CN107705260A (zh) * 2017-10-03 2018-02-16 陈值英 医学x光图像的去噪系统
CN110515063A (zh) * 2019-08-13 2019-11-29 青岛海洋科学与技术国家实验室发展中心 基于迭代平稳离散小波变换的水声信号处理方法和装置
CN110765834A (zh) * 2019-08-25 2020-02-07 青岛科技大学 一种基于改进人工蜂群算法的参数小波阈值信号去噪方法
CN111681667A (zh) * 2020-06-23 2020-09-18 青岛科技大学 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2427077C2 (ru) * 2005-12-05 2011-08-20 Телефонактиеболагет Лм Эрикссон (Пабл) Обнаружение эхосигнала
JP5641186B2 (ja) * 2010-01-13 2014-12-17 ヤマハ株式会社 雑音抑圧装置およびプログラム
US9053367B2 (en) * 2012-11-09 2015-06-09 Seiko Epson Corporation Detector evolution with multi-order contextual co-occurrence
WO2015170192A2 (en) * 2014-04-25 2015-11-12 King Abdullah University Of Science And Technology System and method for image reconstruction, analysis, and/or de-noising
CN107800659B (zh) * 2017-10-12 2020-09-08 西安电子科技大学 Alpha稳定分布噪声下LFM信号调制参数估计方法
CN109816596B (zh) * 2017-11-21 2020-12-22 中移(杭州)信息技术有限公司 一种图像去噪方法及装置
CN109063676B (zh) * 2018-08-24 2020-12-22 广东石油化工学院 一种用于功率信号的自适应时频峰值滤波方法及系统
CN111144318B (zh) * 2019-12-27 2023-08-29 苏州联视泰电子信息技术有限公司 一种水下声呐系统点云数据降噪方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7450779B2 (en) * 2004-05-21 2008-11-11 Imaging Dynamics Company Ltd. De-noising digital radiological images
CN105528768A (zh) * 2015-12-10 2016-04-27 国网四川省电力公司天府新区供电公司 一种图像去噪方法
CN106530254A (zh) * 2016-11-14 2017-03-22 山东理工大学 一种基于小波阈值函数与改进中值滤波融合的抑制图像混合噪声算法
CN107705260A (zh) * 2017-10-03 2018-02-16 陈值英 医学x光图像的去噪系统
CN110515063A (zh) * 2019-08-13 2019-11-29 青岛海洋科学与技术国家实验室发展中心 基于迭代平稳离散小波变换的水声信号处理方法和装置
CN110765834A (zh) * 2019-08-25 2020-02-07 青岛科技大学 一种基于改进人工蜂群算法的参数小波阈值信号去噪方法
CN111681667A (zh) * 2020-06-23 2020-09-18 青岛科技大学 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
LI JIAHENG: "Research on Complex Ocean Noise Suppression Method Based on Underwater Acoustic Array Beamforming Tehnology", MASTER THESIS, TIANJIN POLYTECHNIC UNIVERSITY, CN, no. 1, 15 January 2021 (2021-01-15), CN , XP055882926, ISSN: 1674-0246 *
PRAMOD N.C., ANAND G.V.: "Nonlinear wavelet denoising for improved bearing estimation in ocean under non gaussian noise conditions.", NONLINEAR SIGNAL AND IMAGE PROCESSING, 2005. NSIP 2005. ABSTRACTS. IEE E-EURASIP SAPPORO, JAPAN MAY 18-20, 2005, PISCATAWAY, NJ, USA,IEEE, 18 May 2005 (2005-05-18) - 20 May 2005 (2005-05-20), pages 40 - 40, XP010833777, ISBN: 978-0-7803-9064-5, DOI: 10.1109/NSIP.2005.1502295 *
WAN XIAO-HONG: "Image Denoising Based on Wavelet Transform and Median Filter", JOURNAL OF BEIHUA UNIVERSITY (NATURAL SCIENCE), 10 June 2012 (2012-06-10), pages 352 - 355, XP055882918, ISSN: 1009-4822 *
WANG JINGJING; LI JIAHENG; YAN SHEFENG; SHI WEI; YANG XINGHAI; GUO YING; GULLIVER T. AARON: "A Novel Underwater Acoustic Signal Denoising Algorithm for Gaussian/Non-Gaussian Impulsive Noise", IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, IEEE, USA, vol. 70, no. 1, 16 December 2020 (2020-12-16), USA, pages 429 - 445, XP011837484, ISSN: 0018-9545, DOI: 10.1109/TVT.2020.3044994 *
WANG XIN: "Image Denoising Based on Adaptive Fuzzy Thresholding Approach", JISUANJI-FANGZHEN = COMPUTER SIMULATION, ZHONGGUO HANGTIAN GONGYE ZONGGONGSI, CN, vol. 28, no. 6, 15 June 2011 (2011-06-15), CN , XP055882920, ISSN: 1006-9348 *
YANG HUI, TANG JIAN-FENG,LU SHI-ZHUAN: "An Effective Method for Removing Mixed Noise in Images", JOURNAL OF HENGYANG NORMAL UNIVERSITY, vol. 31, no. 6, 15 December 2010 (2010-12-15), pages 35 - 39, XP055882915, ISSN: 1673-0313, DOI: 10.13914/j.cnki.cn43-1453/z.2010.06.035 *

Cited By (35)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114734440A (zh) * 2022-04-15 2022-07-12 同济大学 一种基于upf-rbf组合模型的混联双臂搬运机器人运动学参数精准标定方法
CN114734440B (zh) * 2022-04-15 2023-09-05 同济大学 一种混联双臂搬运机器人运动学参数精准标定方法
CN115100864A (zh) * 2022-06-24 2022-09-23 北京联合大学 一种基于改进麻雀搜索算法的交通信号控制优化方法
CN115953790A (zh) * 2022-09-29 2023-04-11 江苏智联天地科技有限公司 一种标签检测识别方法与系统
CN115953790B (zh) * 2022-09-29 2024-04-02 江苏智联天地科技有限公司 一种标签检测识别方法与系统
CN115816514A (zh) * 2023-02-16 2023-03-21 极限人工智能有限公司 基于测量电变量的机器人关节制动器故障检测方法及系统
CN115816514B (zh) * 2023-02-16 2023-06-27 极限人工智能有限公司 基于测量电变量的机器人关节制动器故障检测方法及系统
CN116559421A (zh) * 2023-04-03 2023-08-08 杭州臻稀生物科技有限公司 一种自动化荧光免疫分析仪及分析方法
CN116559421B (zh) * 2023-04-03 2024-05-31 杭州臻稀生物科技有限公司 一种自动化荧光免疫分析仪及分析方法
CN116304581B (zh) * 2023-05-10 2023-07-21 佛山市钒音科技有限公司 一种空调用智能电控系统
CN116304581A (zh) * 2023-05-10 2023-06-23 佛山市钒音科技有限公司 一种空调用智能电控系统
CN116859711A (zh) * 2023-06-27 2023-10-10 中科沃业江苏生物有限公司 一种基于iabc优化模糊pid的恒流泵控制系统
CN116859711B (zh) * 2023-06-27 2024-04-12 中科沃业江苏生物有限公司 一种基于iabc优化模糊pid的恒流泵控制系统
CN116485884B (zh) * 2023-06-28 2023-09-12 四川君安天源精酿啤酒有限公司 基于计算机视觉的精酿啤酒瓶口实时定位方法及系统
CN116485884A (zh) * 2023-06-28 2023-07-25 四川君安天源精酿啤酒有限公司 基于计算机视觉的精酿啤酒瓶口实时定位方法及系统
CN116597856B (zh) * 2023-07-18 2023-09-22 山东贝宁电子科技开发有限公司 基于蛙人对讲的语音质量增强方法
CN116597856A (zh) * 2023-07-18 2023-08-15 山东贝宁电子科技开发有限公司 基于蛙人对讲的语音质量增强方法
CN116776749B (zh) * 2023-08-22 2023-11-17 吉林大学 一种考虑不确定性的电动汽车无线充电装置优化方法
CN116776749A (zh) * 2023-08-22 2023-09-19 吉林大学 一种考虑不确定性的电动汽车无线充电装置优化方法
CN117370737A (zh) * 2023-12-08 2024-01-09 成都信息工程大学 一种基于自适应高斯滤波器的非稳态非高斯噪声去除方法
CN117370737B (zh) * 2023-12-08 2024-02-06 成都信息工程大学 一种基于自适应高斯滤波器的非稳态非高斯噪声去除方法
CN117395626A (zh) * 2023-12-11 2024-01-12 厦门大学 基于元学习及noma的水声网络水质监测数据搜集方法
CN117395626B (zh) * 2023-12-11 2024-02-09 厦门大学 基于元学习及noma的水声网络水质监测数据搜集方法
CN117633554A (zh) * 2023-12-14 2024-03-01 艾信智慧医疗科技发展(苏州)有限公司 医用箱式物流传输监测及预警系统
CN117633554B (zh) * 2023-12-14 2024-05-14 艾信智慧医疗科技发展(苏州)有限公司 医用箱式物流传输监测及预警系统
CN117668468B (zh) * 2024-01-31 2024-04-26 山东省舜天化工集团有限公司 一种化工制备数据智能分析管理系统
CN117668468A (zh) * 2024-01-31 2024-03-08 山东省舜天化工集团有限公司 一种化工制备数据智能分析管理系统
CN117665788B (zh) * 2024-02-01 2024-04-05 湖南科技大学 一种基于微波测量数据的噪声处理方法
CN117665788A (zh) * 2024-02-01 2024-03-08 湖南科技大学 一种基于微波测量数据的噪声处理方法
CN117786325A (zh) * 2024-02-27 2024-03-29 山东晨晖电子科技有限公司 一种大口径物联网水表的环境温度智慧预警系统
CN117786325B (zh) * 2024-02-27 2024-04-30 山东晨晖电子科技有限公司 一种大口径物联网水表的环境温度智慧预警系统
CN117828282A (zh) * 2024-03-06 2024-04-05 山东泰霖信息工程有限公司 一种基于自适应滤波的数据高效处理方法
CN117828282B (zh) * 2024-03-06 2024-06-04 山东泰霖信息工程有限公司 一种基于自适应滤波的数据高效处理方法
CN117973429A (zh) * 2024-04-01 2024-05-03 南京信息工程大学 一种应用于非高斯噪声滤波的模型参数比估计方法
CN117973429B (zh) * 2024-04-01 2024-06-07 南京信息工程大学 一种应用于非高斯噪声滤波的模型参数比估计方法

Also Published As

Publication number Publication date
CN111681667B (zh) 2021-05-04
CN111681667A (zh) 2020-09-18

Similar Documents

Publication Publication Date Title
WO2021258832A1 (zh) 基于自适应窗口滤波和小波阈值优化的水声信号去噪方法
CN110765834B (zh) 一种基于改进人工蜂群算法的参数小波阈值信号去噪方法
Wang et al. ia-PNCC: Noise Processing Method for Underwater Target Recognition Convolutional Neural Network.
CN108109617A (zh) 一种远距离拾音方法
CN112733811B (zh) 基于改进稠密神经网络的水声信号调制方式类间识别方法
CN112331224A (zh) 轻量级时域卷积网络语音增强方法与系统
CN115442191B (zh) 基于相对平均生成对抗网络的通信信号降噪方法及系统
CN113222854B (zh) 一种小波变换的ct图像去噪方法
CN110148419A (zh) 基于深度学习的语音分离方法
CN107360497A (zh) 估算混响分量的计算方法及装置
CN110211598A (zh) 智能语音降噪通信方法及装置
TW201312551A (zh) 語音增強方法
CN111341351B (zh) 基于自注意力机制的语音活动检测方法、装置及存储介质
CN112289337A (zh) 一种滤除机器学习语音增强后的残留噪声的方法及装置
CN115426055B (zh) 一种基于解耦卷积神经网络的含噪水声信号盲源分离方法
CN115065578B (zh) 一种基于改进自适应阈值的dft信道估计方法
Sun et al. Wavelet denoising method based on improved threshold function
CN116520333A (zh) 一种基于复杂干扰下的多基浮标融合检测方法和系统
CN116016055A (zh) 一种基于向量近似消息传递的自适应水声信道均衡方法
CN107393553B (zh) 用于语音活动检测的听觉特征提取方法
CN115932808A (zh) 一种基于多特征融合的被动声纳智能探测方法
CN113611319B (zh) 基于语音成分实现的风噪抑制方法、装置、设备及系统
CN111638501B (zh) 一种自适应匹配随机共振的谱线增强方法
CN111538087B (zh) 基于深度学习频谱分割的动态自适应衰减补偿方法及系统
CN117153178B (zh) 音频信号处理方法、装置、电子设备和存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21829191

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21829191

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 19.05.2023)