Weak underwater acoustic signal adaptive detection method based on sound field complexity

Abstract

The present application relates to the technical field of underwater acoustic signal detection, and particularly relates to a weak underwater acoustic signal adaptive detection method based on sound field complexity. The method comprises the following steps: underwater acoustic signal collection and preprocessing; multi-dimensional feature analysis is performed on the preprocessed underwater acoustic signal, and multi-dimensional features reflecting the current sound field environment complexity are extracted; the multi-dimensional feature parameters extracted in step S2 are fused to generate a comprehensive complexity score, the complexity level of the current underwater acoustic environment is evaluated, and adaptive detection algorithm decision is made according to the sound field complexity level obtained after evaluation; a specific underwater acoustic signal detection task is executed, weak underwater acoustic signals are detected and processed, and corresponding detection algorithms are executed according to the sound field complexity level; and the detection result is output. The method can dynamically select a suitable detection algorithm according to the change of the underwater acoustic environment, thereby improving the detection stability and reliability of weak underwater acoustic signals in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic signal detection technology, and in particular to an adaptive detection method for weak underwater acoustic signals based on acoustic field complexity. Background Technology

[0002] Underwater acoustic detection is a key technology in underwater information acquisition and target perception, primarily used to identify and determine the presence of target acoustic signals from background noise in complex underwater acoustic environments. Underwater acoustic detection technology is widely used in underwater target detection, underwater engineering status monitoring, and the front-end signal processing of underwater measurement and survey systems, and is one of the fundamental technologies in the underwater information processing chain. Due to the limited propagation of optical and electromagnetic signals in the underwater environment, underwater acoustic signals become the primary information carrier for underwater target detection and perception.

[0003] Existing methods for detecting weak underwater acoustic signals typically employ fixed detection procedures based on energy characteristics, statistical features, or threshold decisions, and complete target identification through preset parameters. These methods generally rely on a predetermined signal processing flow, performing framing, filtering, and feature extraction on the acquired signal, and then using preset thresholds or decision rules to complete target detection. In engineering implementation, these methods are relatively simple in structure and have a clear implementation path, making them easy to deploy and run in practical systems. However, their core processing logic is usually determined during the system design phase, and the detection algorithm and its parameters are difficult to dynamically adjust according to the actual operating environment, resulting in an overall technical characteristic dominated by fixed algorithms and fixed procedures.

[0004] In the aforementioned underwater acoustic detection methods based on fixed algorithms, the detection process and its parameters are usually predetermined during the system design phase. However, during actual operation, there is a lack of effective characterization and feedback mechanisms for the current underwater acoustic environment, making it difficult to quantitatively represent the complexity of the sound field. Due to the time-varying and uncertain nature of the underwater acoustic environment, the spectral structure, statistical characteristics, and interference composition of background noise change over time. Fixed detection strategies have limited adaptability to environmental changes, easily leading to unstable detection performance under different environmental conditions. Furthermore, in highly complex underwater acoustic environments, weak underwater acoustic signals may exhibit different forms such as low-frequency weak line spectra, broadband weak energy, or short-term transients. If a single fixed detection algorithm is still used, it is difficult to simultaneously adapt to the detection needs of different types of weak signals. Therefore, there is an urgent need for a weak underwater acoustic signal detection method that can quantitatively assess the complexity of the current underwater acoustic environment and adaptively select the detection algorithm based on the morphological characteristics of weak signals in highly complex environments. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned defects in the existing technology and propose an adaptive detection method for weak underwater acoustic signals based on sound field complexity. By introducing a sound field complexity assessment and adaptive switching mechanism for detection algorithms into the underwater acoustic detection and processing link, it can dynamically select a suitable detection algorithm according to changes in the underwater acoustic environment, thereby improving the detection stability and reliability of weak underwater acoustic signals in complex environments.

[0006] The technical solution of this invention is: an adaptive detection method for weak underwater acoustic signals based on acoustic field complexity, comprising the following steps: S1. Acoustic signal acquisition and preprocessing; S2. Perform multidimensional feature analysis on the preprocessed underwater acoustic signal to extract multidimensional features that can reflect the complexity of the current acoustic field environment. S3. The multidimensional feature parameters extracted in step S2 are fused to generate a comprehensive complexity score, which assesses the complexity level of the current underwater acoustic environment. Based on the sound field complexity level obtained after the assessment, an adaptive detection algorithm decision is made. S4. Perform specific underwater acoustic signal detection tasks, detect and process weak underwater acoustic signals, and execute corresponding detection algorithms according to the sound field complexity level. S5. Output the detection results.

[0007] In this invention, discrete underwater acoustic signals are divided into continuous processing frames, with a frame length of [missing information]. Sampling rate Then the number of sampling points per frame is ; The specific implementation process of step S1 is as follows: S1.1. Collect acoustic signals in the underwater environment and complete the acoustic-to-electric conversion: The underwater acoustic sensor array receives underwater acoustic signals and performs acoustic-to-electric conversion, outputting analog signals. The signal is converted to a digital signal at a sampling rate of Discretize: , in, This is the global sampling point index; For continuous signals With sampling rate The discrete sequence obtained by sampling; S1.2, Preliminary processing of the raw underwater acoustic signal: The discrete underwater acoustic signal is divided into continuous processing frames, the first... The intra-frame signal of a frame is represented as follows: , in, For frame numbering, For intra-frame sampling point index, The number of sampling points per frame; For the first The frame signal undergoes frame-by-frame mean removal processing, the first... The average value of the frames is: , Signal after mean removal Represented as: , Subsequently, the mean-removed signal is bandpass filtered to obtain the preprocessed signal. : , in, These are the coefficients of the bandpass filter; This is the filter coefficient index, used to suppress low-frequency environmental noise and high-frequency electronic noise; This is the filter length; The filtered signal is normalized frame by frame: Define the first... Root mean square magnitude of the frame for: , The normalized signal for: .

[0008] in, The extremely small positive number introduced to prevent the denominator from being zero.

[0009] In step S2, the multidimensional statistical features reflecting the complexity of the current sound field environment include spectral distribution features, time-domain fluctuation features, noise stability features, and signal separability features. For the normalized first Frame signal Performing a short-time Fourier transform yields the frequency domain representation: , in, Indicates the first Frame signal in the Frequency domain representation at each frequency point; Frequency index; For window functions; Calculate the normalized power spectral density based on the short-time spectrum: , in, Indicates the first Frame signal in the The normalized power spectral density at each frequency index satisfies the probability distribution characteristics: , Spectral complexity Represented by spectral entropy, it is defined as: , in, Represents the logarithm to the base 2; Based on the filtered signal Constructing short-time energy sequences, the first The short-time energy of a frame signal is defined as: , In length Within the time statistics window, define the average short-time energy. for: , Define time fluctuation complexity for: , in, The number of frames used for statistics within a time window, reflecting the degree of fluctuation of short-term energy within adjacent time windows; To characterize the stationarity of background noise, the first... The autocorrelation function of the frame-normalized signal: , in, It is a delay quantity, and satisfies ; For the first Frame signal in delay The autocorrelation function under the following conditions; Define stationarity index for: , in, This is the maximum delay length; To characterize the separability of the target signal from the background noise, local signal-to-noise ratio estimation is introduced. : , in, For the target frequency band; This is the noise frequency band, and ; For the first Frame signal in target frequency band Frequency domain representation at the corresponding frequency point within the range; For the first Frame signal in noisy reference band Frequency domain representation at the corresponding frequency point within the range.

[0010] In step S3, the multidimensional feature parameters obtained in step S2 are used to construct a complexity vector: ， definition: ， ， ， in, , , All of these are feature quantities after nonlinear compression mapping; Used to reflect the degree of fluctuation of short-term energy in the time dimension; It is used to reflect the stability of the statistical structure of background noise and to suppress the influence of outliers through exponential mapping; Used to reflect the degree of inseparability between the target signal and the background noise; Overall complexity score The weighted nonlinear mapping method is used to represent it as follows: ， in, , , All are weighting coefficients. And satisfy ; Introducing exponential smoothing, we obtain the smoothed overall complexity score: , in, This is the exponential smoothing coefficient, used to adjust the weight between the current frame's overall complexity score and historical smoothing scores. And the initial conditions are: , Score based on smoothed overall complexity The current sound field complexity is divided into different levels: , in, This is the first complexity threshold; It is the second complexity threshold, and satisfies ; When the smoothed overall complexity score is lower than the first complexity threshold When choosing a low-complexity detection algorithm When the smoothed overall complexity score is greater than or equal to And less than the second complexity threshold When choosing a detection algorithm with medium complexity. When the smoothed overall complexity score is greater than or equal to the second complexity threshold At this time, it enters a high-complexity detection mode; The set of high-complexity detection algorithms is as follows: ， in, An algorithm for accumulating and detecting low-frequency weak line spectrum trajectories. This is a sub-band peak energy selective accumulation detection algorithm. This is a matched filter detection algorithm; Calculate the morphological fit score for weak signals under high-complexity detection modes: , in, For low-frequency line spectrum continuity adaptation fraction; For the significant fit score of broadband subband energy, Match the appropriate score to the template; Based on the above suitability score, the corresponding detection algorithm is selected from the set of high-complexity detection algorithms.

[0011] In step S4, the current state of the algorithm execution is: At that time, a single-frame energy detection algorithm was used; Definition of the first The detection energy of the frame signal is: , Based on energy detection Make a binary hypothesis decision: , in, This indicates the presence of a target signal; This indicates that the target signal does not exist; The threshold for single-frame energy detection; The single-frame energy detection algorithm is in the first... The decision output of the frame.

[0012] In step S4, the current state of the algorithm execution is: At that time, an adaptive normalized energy detection algorithm is adopted; Define the detection statistic as: , in, The background reference power estimate is used to characterize the noise level in the current received signal, and its calculation process is as follows: For the first Spectral analysis is performed on the preprocessed frame signal to obtain its frequency domain representation. , and by Represent the spectral amplitude and calculate the spectral power: , in, Indicates the first Frame number The unnormalized spectral power at each frequency point is used for background reference power estimation; In analyzing frequency bands Within this frame, the median of the spectral power is used as a temporary background baseline estimate for the current frame. , The result obtained using smooth update is: , in, This represents the smoothing coefficient for the background baseline power estimate. ; Based on statistics Judgment: , in, This indicates the presence of a target signal; This indicates that the target signal does not exist; This is the normalized energy detection threshold; The adaptive normalized energy detection algorithm is represented in the first... The decision output of the frame.

[0013] In step S4, when the smoothed overall complexity score indicates that the current environment is a high-complexity environment, the morphology of the weak signals in the high-complexity environment is judged, and the corresponding detection algorithm is selected from the set of high-complexity detection algorithms: For continuous Analyze the short-time spectrum of the frame, and assume... For the first Frame number Normalized spectral energy at each frequency point For low-frequency analysis bands, It is a candidate line spectrum trajectory that satisfies the frequency continuity constraint between adjacent frames: , in, The maximum allowable frequency drift range; The low-frequency line spectrum continuity fit fraction is defined as: , The target operating frequency band is divided into Individual belt: , No. The number of belts in the first The energy saliency of a frame is defined as: , in, For the first Individual belt; This represents the number of frequency points within the sub-band. For the first Estimators of the local background noise variance for each sub-band; For the first Frame signal in the Normalized power spectral energy at each frequency point; Select the one with the highest energy significance Each sub-band constitutes a set The broadband subband energy saliency fit score is defined as: , Let the pre-stored target weak signal template be , No. The signal after frame preprocessing is The template length is Template delayed search range is The template matching adaptation score is defined as follows: , Let the maximum of the three fit scores above be: , Let the second largest of the three fit scores above be: , The algorithm selection rule for high-complexity environments is then expressed as: ， in, For signal morphology discrimination margin; This represents a low-frequency weak line spectrum trajectory accumulation detection algorithm; This represents the sub-band peak energy selective accumulation detection algorithm; This represents the matched filter detection algorithm.

[0014] In step S4, when selecting under high complexity environment At that time, a low-frequency weak line spectrum trajectory accumulation detection algorithm was used to detect weak underwater acoustic signals; For continuous Short-time spectrum analysis of the frame signal yields the first... Frame number Frequency domain representation at each frequency point To reduce the influence of local background fluctuations, the normalized time-frequency energy is defined as: , in, For the first Frame number Local background energy estimation near each frequency point; Preserve the set of candidate spectral points in each frame: ,in, Set a threshold for selecting candidate line spectrum points; Assume continuous One candidate line trajectory within the frame is: And it satisfies the frequency continuity constraint between adjacent frames: ,in, This indicates that the candidate line trajectory is at the th... The frequency point index corresponding to the frame; This indicates that the candidate line trajectory is in the current position. The frequency point index corresponding to the frame; Define trajectory accumulation score for: , in, This is the trajectory smoothing penalty coefficient; Corresponding detection statistics for: , Based on statistics Judgment: , in, Accumulate detection thresholds for low-frequency weak line spectrum trajectories; The low-frequency weak line spectrum trajectory accumulation detection algorithm is represented in the first... The decision output of the frame.

[0015] In step S4, when selecting under high complexity environment At that time, the sub-band peak energy selective accumulation detection algorithm was used to detect weak underwater acoustic signals; The target operating frequency band is divided into Individual belt: , No. The number of belts in the first The significance of the normalized energy of a frame is defined as follows: , in, For the first Local background noise variance estimate corresponding to each sub-band For the first Number of frequency points within each sub-band; from Select subbands with higher energy significance. Each strip is denoted as: , in, Indicates from The index with the largest value among the subband energy significance indicators is selected. The set of sub-band indexes corresponding to each indicator; For the first The set of selected candidate subbands in a frame; In continuous Within a frame, the energy of candidate sub-bands is selectively accumulated, and the detection statistic is defined as follows: , in, Indicates the first The frame with the highest energy significance A set of candidate subbands; For the first Frame number The weight coefficients of candidate subbands, satisfying: .

[0016] Based on statistics Judgment: , in, The threshold for selective accumulation of subband peak energy; The subband peak energy selective accumulation detection algorithm is shown in the first... The decision output of the frame.

[0017] In step S4, when selecting under high complexity environment At that time, a matched filter detection algorithm was used to detect weak underwater acoustic signals; Let the pre-stored target weak signal template be , No. The filtered signal after frame preprocessing is The template length is The matched filter detection statistic is defined as follows: , in, This is the template delay amount; Based on statistics Judgment: , in, The detection threshold for matched filtering; This indicates that the matched filter detection algorithm is in the first... The decision output of the frame.

[0018] In step S5, the first The final detection result of the frame is represented as follows: .

[0019] The beneficial effects of this invention are: (1) This application addresses the problem of insufficient adaptability of fixed algorithms in the detection of weak underwater acoustic signals in complex underwater acoustic environments by introducing a sound field complexity assessment mechanism. This mechanism comprehensively analyzes the characteristics of the current underwater acoustic signal, such as its spectral distribution, time fluctuation, stability, and local separability, to form a sound field complexity score. Based on this score, low-complexity, medium-complexity, and high-complexity environments are distinguished, and corresponding detection algorithms are selected. This prevents the detection process from being fixed in a single processing method for a long time, thereby improving the stability of the detection results under different environmental conditions. (2) In low-complexity environments, this application adopts a single-frame energy detection algorithm to avoid unnecessary computational overhead caused by using a high-complexity algorithm when the background is relatively stable and the interference is weak; in medium-complexity environments, an adaptive normalized energy detection algorithm is adopted to reduce the impact of slow noise fluctuations on the detection statistics through background noise estimation, thus balancing detection performance and computational cost. (3) In a highly complex environment, this application further introduces a weak signal morphology judgment mechanism: for low-frequency weak line spectrum signals, a low-frequency weak line spectrum trajectory accumulation detection algorithm is selected; for broadband weak energy signals, a sub-band peak energy selective accumulation detection algorithm is selected; for template-correlated weak signals, a matched filtering detection algorithm is selected; compared with the existing fixed multi-frame joint detection, this method does not use the same accumulation strategy for all signals, but selects the corresponding detection path according to the actual manifestation of the weak signal, thereby reducing the problem of noise, interference and indiscriminate accumulation of the target signal.

[0020] (4) In the process of algorithm switching, the present invention combines constraints such as complexity score smoothing, hysteresis threshold, minimum dwell time and signal morphology adaptability margin, so that the algorithm selection does not completely depend on the instantaneous result of a single frame, reducing the frequent switching caused by small fluctuations in complexity score or morphology judgment result. Therefore, the present application can maintain a relatively stable detection state during long-term continuous operation and provide more reliable operating information for the real-time processing and status maintenance of underwater acoustic monitoring equipment. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method described in this application. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Specific details are set forth in the following description to provide a full understanding of the invention. However, the invention can be practiced in many ways other than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] The flowchart of the weak underwater acoustic signal adaptive detection method based on sound field complexity described in this application is as follows: Figure 1 As shown. This embodiment addresses an engineering scenario where a shipborne or fixed observation platform passively detects underwater vehicles, towed equipment, or other acoustic source targets. By adaptively evaluating the complexity of the sound field and dynamically selecting the detection algorithm based on the evaluation results, detection stability is maintained in complex environments. The following explanation uses the underwater acoustic signal output from any array element or any receiving channel as an example.

[0025] To facilitate real-time processing, the discrete underwater acoustic signal is divided into continuously processed frames. Let the frame length be... Sampling rate Then the number of sampling points per frame is .

[0026] in Typically, a value between 20ms and 100ms is used; when substituting this value into the formula, it should be converted to seconds. It is preferable to take the sampling period as an integer multiple of the sampling period, so that It is an integer.

[0027] The method includes the following steps.

[0028] The first step is underwater acoustic signal acquisition and preprocessing.

[0029] First, the underwater acoustic sensor array is used to collect acoustic signals in the underwater environment and complete the acoustic-to-electric conversion. Then, the raw underwater acoustic signal is preliminarily processed, including signal framing, DC component removal, amplitude normalization, and filtering, to suppress the influence of environmental noise and system noise on signal quality, so that the signal meets the requirements of subsequent feature analysis and detection processing in terms of amplitude and frequency range, thereby obtaining a stable and reliable preprocessed underwater acoustic signal.

[0030] The underwater acoustic sensor array receives underwater acoustic signals and performs acoustic-to-electric conversion, outputting analog signals. The signal is converted to a digital signal at a sampling rate of... Discretize: , in, This is the global sampling point index; For continuous signals With sampling rate The discrete sequence obtained by sampling.

[0031] To facilitate real-time processing, the discrete underwater acoustic signal is divided into continuously processed frames. The intra-frame signal of a frame is represented as follows: , in, For frame numbering, For intra-frame sampling point index, This represents the number of sampling points per frame.

[0032] To eliminate sensor bias and low-frequency gradual drift, the first The frame signal undergoes frame-by-frame mean reduction processing to suppress DC components and gradually varying biases. The average value of the frames is: , Signal after mean removal Represented as: , Subsequently, the mean-removed signal is bandpass filtered to obtain the preprocessed signal. : , in, These are the coefficients of the bandpass filter; For filter coefficient index, its passband frequency range is preset according to the frequency characteristics of the target underwater acoustic signal and the background noise distribution characteristics or determined by prior experience, and is used to suppress low-frequency environmental noise and high-frequency electronic noise. This is the filter length. For In such cases, zero-fill boundary treatment is preferred. The filter is preferably implemented using a finite impulse response (FIR) structure.

[0033] To unify the amplitude scale across different frames, the filtered signal is normalized frame by frame. Define the... Root mean square magnitude of the frame for: , The normalized signal for: .

[0034] in, The extremely small positive number introduced to prevent the denominator from being zero.

[0035] After the above preprocessing, the normalized signal The signal is fed into the sound field complexity assessment module for subsequent complexity feature extraction; simultaneously, based on the input requirements of the selected detection algorithm, the filtered signal... or normalized signal It is fed into the weak underwater acoustic signal detection algorithm module group for subsequent detection and processing.

[0036] The second step is to extract the sound field complexity features.

[0037] Multidimensional feature analysis is performed on the preprocessed underwater acoustic signal to extract multidimensional statistical features that reflect the complexity of the current acoustic field environment, including spectral distribution features, time-domain fluctuation features, noise stationarity features, and signal separability features. These features are used to describe the changes in the underwater acoustic signal in the frequency domain, time domain, and statistical characteristics, thereby characterizing the complexity of the current underwater acoustic environment.

[0038] First, the normalized first... Frame signal Performing a short-time Fourier transform yields the frequency domain representation: , in, Indicates the first Frame signal in the Frequency domain representation at each frequency point; Frequency index; The window function is used to weight the signal, reducing the discontinuity of frame boundaries, thereby reducing spectral leakage and improving the reliability of subsequent spectral feature extraction.

[0039] Calculate the normalized power spectral density based on the short-time spectrum: , in, Indicates the first Frame signal in the The normalized power spectral density at each frequency index satisfies the probability distribution characteristics: .

[0040] Spectral complexity Represented by spectral entropy, it is defined as: , in, It represents the logarithm to the base 2; spectral entropy is used to characterize the dispersion of the spectral distribution. A larger value indicates a more uniform spectrum and higher environmental frequency domain complexity.

[0041] To characterize the time fluctuation characteristics of a signal, based on the filtered signal Construct short-time energy sequences. The short-time energy of a frame signal is defined as: , Among them, the filtered signal Short-time energy is constructed to preserve the true energy fluctuation information between different frames, avoiding the weakening of energy fluctuation characteristics by frame-by-frame normalization.

[0042] In length Within the time statistics window, define the average short-time energy. for: , Therefore, time fluctuation complexity is defined. for: , in, The number of frames used for statistics within a time window is used to reflect the degree of fluctuation of short-term energy within adjacent time windows. The larger the value, the more pronounced the time fluctuation. The definitions of average short-time energy and time fluctuation complexity mentioned above apply to... The situation where the initial amount is insufficient; In the case of frames, it is preferable to use existing frame statistics or zero-padding method for initialization.

[0043] To characterize the stationarity of background noise, the first... The autocorrelation function of the frame-normalized signal: , in, It is a delay quantity, and satisfies ; For the first Frame signal in delay The autocorrelation function under the given conditions.

[0044] Further define stationarity indicators for: , in, This represents the maximum delay length. This stationarity index reflects the overall stability of the signal structure by accumulating and normalizing the autocorrelation values ​​under different delays.

[0045] In addition, to characterize the separability of the target signal from the background noise, a local signal-to-noise ratio estimation is introduced. : , in, For the target frequency band; This is the noise frequency band, and ; Used to characterize the The relative magnitude of the target frequency band energy relative to the background noise reference frequency band energy in the frame signal; For the first Frame signal in target frequency band Frequency domain representation at the corresponding frequency point within the range; For the first Frame signal in noisy reference band Frequency domain representation at the corresponding frequency point within the range.

[0046] The above spectral complexity Time fluctuation complexity Correlation indicators and local signal-to-noise ratio Together, they constitute the multidimensional feature parameters output by the sound field complexity assessment module, which are used for subsequent complexity fusion and scoring.

[0047] The third step involves complexity fusion and environmental condition assessment, followed by adaptive detection algorithm decision-making based on the sound field complexity level obtained after assessment.

[0048] In this step, the multidimensional feature parameters extracted in the previous step are fused, and a comprehensive complexity score is generated through weighted fusion, rule-based judgment, or statistical analysis. This score can be represented in continuous numerical form or divided into multiple discrete complexity levels, such as low-complexity environment, medium-complexity environment, and high-complexity environment, to quantify the overall complexity of the current underwater acoustic environment and provide a unified evaluation basis for the selection of subsequent detection algorithms.

[0049] The multidimensional feature parameters obtained in the second step are used to construct a complexity vector: ， in, Indicates the first The spectral complexity of a frame signal. This represents the complexity of energy fluctuations based on time window statistics. Indicates the first Correlation index of frame signals Indicates the first The local signal-to-noise ratio of a frame.

[0050] To improve the fusionability of different features and reduce the impact of differences in the dimensions and numerical ranges of different features on the comprehensive score, a bounded nonlinear mapping is first applied to some features. Definition: ， ， ， in, , , These are all feature quantities after nonlinear compression mapping, used to improve the stability of the multidimensional feature fusion process. Derived from time fluctuation complexity It is used to reflect the degree of fluctuation of short-term energy in the time dimension; Derived from stationarity related indicators It is used to reflect the stability of the statistical structure of background noise and to suppress the influence of outliers through exponential mapping. Derived from local signal-to-noise ratio A reverse mapping method is used to reflect the degree of inseparability between the target signal and background noise; that is, the lower the local signal-to-noise ratio, the larger the mapping value, and the more significant its contribution to the complexity score. The above three features participate in the comprehensive complexity fusion from the perspectives of time fluctuation, stationary change, and signal separability, respectively, thereby avoiding the problem that a single feature is insufficient to describe the sound field complexity.

[0051] Overall complexity score The weighted nonlinear mapping method is used to represent it as follows: ， in, , , All are weighting coefficients. And satisfy .

[0052] The aforementioned comprehensive complexity score is achieved through the nonlinear coupling of multidimensional features. and Multiplicative coupling is employed to characterize the combined effects of spectral complexity and temporal volatility; an exponential function is used to assess the stationarity index. A nonlinear mapping is performed to make it saturate in the high-value region, thereby enhancing the sensitivity to changes with low stationarity; by... After the fractional normalization mapping, we get This reflects the degree to which the target signal and background noise can be separated, thus enabling a comprehensive quantitative assessment of the sound field complexity.

[0053] To suppress the impact of short-term fluctuations on the scoring results, exponential smoothing is introduced to obtain the smoothed comprehensive complexity score: , in, The exponential smoothing coefficient is used to adjust the weight between the current frame's overall complexity score and historical smoothing scores, and satisfies the following conditions: In a preferred embodiment, The value range is 0.6 to 0.8, with the preferred value being... .when When the value is large, the smoothing result responds faster to changes in the complexity of the current frame; when When the value is small, the smoothing result is more strongly influenced by historical scores, and the suppression effect on short-term disturbances is more obvious. The initial conditions are set as follows: , Score based on smoothed overall complexity The current sound field complexity is divided into different levels: , in, This is the first complexity threshold; It is the second complexity threshold, and satisfies .

[0054] By coupling and mapping multidimensional features, continuous quantitative evaluation of sound field complexity is achieved, providing a basis for adaptive switching of subsequent detection algorithms.

[0055] When the overall complexity score is lower than the first complexity threshold, the current environment is determined to be a low-complexity environment, and a low-computational-load detection algorithm is selected; when the overall complexity score is between the first and second complexity thresholds, the current environment is determined to be a medium-complexity environment, and an adaptive normalized energy detection algorithm is selected; when the overall complexity score is higher than the second complexity threshold, the current environment is determined to be a high-complexity environment, and the high-complexity detection mode is entered.

[0056] The pre-defined detection algorithms include low-complexity, medium-complexity, and high-complexity algorithms. Let the low-complexity detection algorithm be... The medium-complexity detection algorithm is The set of high-complexity detection algorithms is as follows: , in, An algorithm for accumulating and detecting low-frequency weak line spectrum trajectories. This is a sub-band peak energy selective accumulation detection algorithm. This is a matched filter detection algorithm.

[0057] Based on the above sound field complexity evaluation results, a first-level algorithm selection function is constructed. When the smoothed overall complexity score is lower than the first complexity threshold... When choosing a low-complexity detection algorithm When the smoothed overall complexity score is greater than or equal to And less than the second complexity threshold When choosing a detection algorithm with medium complexity. When the smoothed overall complexity score is greater than or equal to the second complexity threshold At that time, it enters the high-complexity detection mode.

[0058] In high-complexity detection modes, the weak signal morphology fit score is further calculated: , in, For low-frequency line spectrum continuity adaptation fraction; For the significant fit score of broadband subband energy, Match the appropriate score to the template.

[0059] Based on the above suitability score, the corresponding detection algorithm is selected from the set of high-complexity detection algorithms.

[0060] To prevent frequent algorithm switching caused by fluctuations in the overall complexity score around a threshold, a hysteresis threshold and a minimum dwell time constraint are introduced. This constraint applies to both the first-level complexity level switching and the second-level algorithm selection process in high-complexity mode. Algorithm switching is only allowed when the complexity level change or signal morphology adaptability change continuously meets the preset frame count, and the time since the last algorithm switch is greater than the minimum dwell time.

[0061] The fourth step is to detect and process weak underwater acoustic signals, and execute the corresponding detection algorithm according to the complexity level of the sound field.

[0062] This step is used to perform specific underwater acoustic signal detection tasks. The algorithms are configured hierarchically according to sound field complexity and weak signal morphology, including low-complexity detection algorithms, medium-complexity detection algorithms, and a set of high-complexity detection algorithms. The low-complexity detection algorithm is preferably a single-frame energy detection algorithm, the medium-complexity detection algorithm is preferably an adaptive normalized energy detection algorithm, and the high-complexity detection algorithm set includes a low-frequency weak line spectrum trajectory accumulation detection algorithm, a sub-band peak energy selective accumulation detection algorithm, and a matched filter detection algorithm.

[0063] Based on the output of the previous step, the detection mode is selected according to the sound field complexity level. When entering the high complexity detection mode, a specific high complexity detection algorithm is further selected based on the weak signal morphology characteristics, thereby achieving adaptive detection of different types of weak underwater acoustic signals.

[0064] After selecting a detection algorithm based on the current acoustic field complexity and the characteristics of the weak signal, corresponding detection processing is performed on the input underwater acoustic signal. The final set of selectable detection algorithms is represented as follows: , Current number The algorithm state corresponding to the frame is denoted as: .

[0065] (I) Low-complexity detection algorithm .

[0066] The low-complexity detection algorithm in this embodiment is preferably a single-frame energy detection algorithm. The current algorithm execution state is... At that time, a single-frame energy detection algorithm is used. Define the first... The detection energy of the frame signal is: , Based on energy detection Make a binary hypothesis decision: , in, This indicates the presence of a target signal; This indicates that the target signal does not exist; The threshold for single-frame energy detection; The single-frame energy detection algorithm is in the first... The decision output of the frame.

[0067] (ii) Medium complexity detection algorithm.

[0068] In this embodiment, the medium-complexity detection algorithm is preferably an adaptive normalized energy detection algorithm. When the current execution algorithm state is... At that time, an adaptive normalized energy detection algorithm is used. The detection statistic is defined as: , in, It serves as a background reference power estimate, used to characterize the noise level in the currently received signal.

[0069] The background reference power estimate is obtained by estimating the received signal itself. Specifically, for the first... Spectral analysis is performed on the preprocessed frame signal to obtain its frequency domain representation. And calculate the spectral power: , in, Indicates the first Frame number The unnormalized spectral power at each frequency point is used for background reference power estimation.

[0070] In analyzing frequency bands Within this frame, the median of the spectral power is used as a temporary background baseline estimate for the current frame. , Furthermore, a smooth update can be used to obtain: , in, This represents the smoothing coefficient for the background baseline power estimate. This is used to adjust the weights between the background baseline estimate of the previous frame and the temporary background baseline estimate of the current frame.

[0071] By using median and smoothing update methods, the impact of local spectral peaks, weak target components, or transient disturbances on background baseline estimation can be reduced.

[0072] Based on statistics Judgment: , in, This is the normalized energy detection threshold; The adaptive normalized energy detection algorithm is represented in the first... The decision output of the frame.

[0073] (III) Determining the form of weak signals in highly complex environments and deciding on the corresponding detection algorithm for different weak signal forms.

[0074] First, determine the morphology of weak signals under highly complex environments.

[0075] When the smoothed sound field comprehensive complexity score indicates that the current environment is a high-complexity environment, instead of directly using single multi-frame joint cumulative detection, we further judge the weak signal morphology and select the corresponding detection algorithm from the set of high-complexity detection algorithms.

[0076] In high-complexity detection mode, the adaptability indexes for weak signal morphology are further calculated, including low-frequency line spectrum continuity index, broadband subband energy significance index, and template matching significance index. Based on the relative magnitudes and confidence margins of these indices, a corresponding detection algorithm is selected from the set of high-complexity detection algorithms. Therefore, not only can the algorithm be switched according to the sound field complexity, but a more suitable detection strategy can also be selected based on the weak signal morphology in complex environments.

[0077] Definition of the first The weak signal morphology adaptation vector of the frame is: , in, For low-frequency line spectrum continuity adaptation fraction; The significant fit score for broadband subband energy; Match the appropriate score to the template.

[0078] In a preferred embodiment, for continuous Analyze the short-time spectrum of the frame. Let... For the first Frame number Normalized spectral energy at each frequency point For low-frequency analysis bands, It is a candidate line spectrum trajectory that satisfies the frequency continuity constraint between adjacent frames: , in, This represents the maximum allowable frequency drift range.

[0079] The low-frequency line spectrum continuity fit fraction is defined as: , The adaptation score is used to characterize whether there is a stable or slowly drifting low-frequency weak line spectrum structure within consecutive frames.

[0080] The target operating frequency band is divided into Individual belt: , No. The number of belts in the first The energy saliency of a frame is defined as: , in, For the first Individual belt; Select the frequency with the highest energy significance for this sub-band. Each sub-band constitutes a set Then the number of broadband subband power points; For the first Estimators of the local background noise variance for each sub-band; Indicates the first Frame signal in the Normalized power spectral density at each frequency index.

[0081] Select the one with the highest energy significance Each sub-band constitutes a set The broadband subband energy saliency fit score is defined as: , The fit score is used to characterize whether there is a weak energy uplift relative to the local background within multiple target subbands.

[0082] Let the pre-stored target weak signal template be , No. The signal after frame preprocessing is The template length is Template delayed search range is This is used to limit the set of positions where the template is allowed to slide and match within the current frame. The template matching adaptation score is then defined as: , The matching score is used to characterize the degree of matching between the current input signal and the preset target weak signal template.

[0083] Let the maximum of the three fit scores above be: , Let the second largest of the three fit scores above be: , The algorithm selection rule for high-complexity environments is then expressed as: ， in, For signal morphology discrimination margin; This represents a low-frequency weak line spectrum trajectory accumulation detection algorithm; This represents the sub-band peak energy selective accumulation detection algorithm; This represents the matched filter detection algorithm.

[0084] If the difference between the largest fit score and the second largest fit score is greater than If the difference is not greater than 1, then select the high-complexity detection algorithm corresponding to the maximum fit score; if the difference is not greater than 1, then select the high-complexity detection algorithm corresponding to the maximum fit score. If the state of the previous high-complexity detection algorithm is maintained, it will avoid erroneous switching caused by instantaneous feature fluctuations.

[0085] when When the maximum margin condition is met, A31, i.e., the low-frequency weak line spectrum trajectory accumulation detection algorithm, is selected; when When the maximum margin condition is met, A32, i.e., the sub-band peak energy selective accumulation detection algorithm, is selected; when When the maximum margin condition is met, choose A33, which is the matched filter detection algorithm.

[0086] Second, for different weak signal forms, corresponding detection algorithms are used to detect weak underwater acoustic signals.

[0087] A. Low-frequency weak line spectrum trajectory accumulation detection algorithm.

[0088] When choosing in a highly complex environment At that time, a low-frequency weak line spectrum trajectory accumulation detection algorithm was used to detect weak underwater acoustic signals.

[0089] For continuous Short-time spectrum analysis of the frame signal yields the first... Frame number Frequency domain representation at each frequency point To mitigate the impact of local background fluctuations, the normalized time-frequency energy is defined as: , in, For the first Frame number Local background energy estimation near a frequency point.

[0090] Preserve the set of candidate spectral points in each frame: .in, The threshold for selecting candidate spectral points.

[0091] Assume continuous One candidate line trajectory within the frame is: And it satisfies the frequency continuity constraint between adjacent frames: .in, For continuous Frame number within the frame time window ; This indicates that the candidate line trajectory is at the th... The frequency point index corresponding to the frame; This indicates that the candidate line trajectory is in the current position. The frequency point index corresponding to the frame; This represents the maximum allowable frequency drift range.

[0092] Define trajectory accumulation score for: , in, This is the trajectory smoothing penalty coefficient.

[0093] Corresponding detection statistics for: , Based on statistics Judgment: , in, Accumulate detection thresholds for low-frequency weak line spectrum trajectories; The low-frequency weak line spectrum trajectory accumulation detection algorithm is represented in the first... The decision output of the frame.

[0094] B. Subband peak energy selective accumulation detection algorithm.

[0095] When choosing in a highly complex environment At that time, the sub-band peak energy selective accumulation detection algorithm was used to detect weak underwater acoustic signals.

[0096] The target operating frequency band is divided into Individual belt: , No. The number of belts in the first The significance of the normalized energy of a frame is defined as follows: , in, For the first Local background noise variance estimate corresponding to each sub-band For the first The number of frequency points within each subband.

[0097] from Select subbands with higher energy significance. Each strip is denoted as: , in, Indicates from The index with the largest value among the subband energy significance indicators is selected. The set of sub-band indexes corresponding to each indicator; Indicates the first The selected candidate subband set in the frame. This step allows for the selection of candidate subbands with the most significant energy increase from all subbands, reducing the impact of irrelevant frequency band noise on subsequent detection statistics.

[0098] To obtain more stable test results, in continuous Within a frame, the energy of candidate sub-bands is selectively accumulated, and the detection statistic is defined as follows: , in, Indicates the first The frame with the highest energy significance A set of candidate subbands; For the first Frame number The weight coefficients of candidate subbands, satisfying: .

[0099] Based on statistics Judgment: , in, The threshold for selective accumulation of subband peak energy; The subband peak energy selective accumulation detection algorithm is shown in the first... The decision output of the frame.

[0100] C. Matched filter detection algorithm.

[0101] When choosing in a highly complex environment At that time, a matched filtering detection algorithm was used to detect weak underwater acoustic signals.

[0102] Let the pre-stored target weak signal template be , No. The filtered signal after frame preprocessing is The template length is To measure the correlation between the input signal and the preset target template, the matched filter detection statistic is defined as: , in, This is the template delay amount; A very small positive number is introduced to prevent the denominator from being zero. This detection statistic represents the maximum normalized correlation between the current input signal and the preset template at different delay positions; when the input signal contains a weak signal component consistent with the template shape... It will reach the maximum value.

[0103] Based on statistics Judgment: , in, The detection threshold for matched filtering; This indicates that the matched filter detection algorithm is in the first... The decision output of the frame.

[0104] The fifth step is to output the test results and monitor the operation.

[0105] Based on the above detection algorithms, the first The final detection result of the frame is represented as follows: , in, This represents the decision output of the single-frame energy detection algorithm; This represents the decision output of the adaptive normalized energy detection algorithm; This represents the decision output of the low-frequency weak line spectrum trajectory accumulation detection algorithm; This represents the decision output of the sub-band peak energy selective accumulation detection algorithm; This represents the decision output of the matched filter detection algorithm.

[0106] After detection, the system outputs the detection result indicating the presence of the target signal, along with relevant detection parameters. Simultaneously, it records and outputs key status information during operation, including the current sound field complexity level, the detection algorithm's running status, and the number of algorithm switching times. This provides necessary information support for the operation and status diagnosis of the entire method, improving the long-term reliability and maintainability of the method described in this application.

[0107] To verify the effectiveness of the adaptive detection method for weak underwater acoustic signals based on acoustic field complexity proposed in this invention, an underwater acoustic signal detection simulation platform was built using MATLAB to compare and verify the fixed multi-frame joint detection scheme with the method described in this application.

[0108] In the simulation, the sampling rate was set to 12 kHz, the single frame duration was set to 50 ms, and 1000 frames of signal were collected under each working condition. The target signal was set as a conventional weak energy signal, a low-frequency weak line spectrum signal, a broadband weak energy signal, and a template-related weak signal according to different working conditions, and background noise, multipath, narrowband interference, broadband interference, and pulse interference were superimposed to simulate underwater acoustic environments with different levels of complexity.

[0109] To demonstrate the adaptability of this invention under different sound field complexities and weak signal morphologies, five typical operating conditions S1 to S5 are set. S1 represents a low-complexity environment with relatively stable background noise and weak interference. S2 represents a medium-complexity environment with slowly fluctuating background noise and superimposed with mild colored noise. S3 represents a high-complexity low-frequency line spectrum environment with strong colored noise and multi-source narrowband interference in the background; the target signal exhibits a weak low-frequency line spectrum or a slowly drifting line spectrum. S4 represents a high-complexity broadband environment with strong colored noise and broadband interference in the background; the target signal exhibits weak energy rise within multiple subbands. S5 represents a high-complexity template-dependent environment with pulse interference in the background; the target signal exhibits a weak signal structure related to a preset template.

[0110] Comparison method: A fixed multi-frame joint detection algorithm is always used, without judging the sound field complexity or the weak signal morphology.

[0111] This application's solution: First, the current environmental state is determined based on the sound field complexity score. When the environment is low complexity (condition S1), a single-frame energy detection algorithm is selected. When the environment is medium complexity (condition S2), an adaptive normalized energy detection algorithm is selected. When the environment is high complexity, further analysis is performed based on signal morphology characteristics. In condition S3, a low-frequency weak line spectrum trajectory accumulation detection algorithm is selected. In condition S4, a sub-band peak energy selective accumulation detection algorithm is selected. In condition S5, a matched filter detection algorithm is selected.

[0112] The detection performance was evaluated using detection probability, false alarm rate, false negative rate, and average computational cost. The average computational cost was represented by a normalized relative value to reflect the computational overhead of different detection algorithms. Table 1 shows the detection results of the fixed multi-frame joint detection scheme and the scheme of this invention under five operating conditions.

[0113] Table 1. Comparison of the performance of fixed multi-frame joint detection and this application in five types of environments. As shown in Table 1, in the low-complexity environment S1, this application selects the single-frame energy detection algorithm. Compared with fixed multi-frame joint detection, the detection probability and false negative rate are similar, but the computational cost of this application is reduced from 1.50 to 0.72. This indicates that in relatively simple environments, using multi-frame joint detection does not bring significant detection benefits, but instead increases computational overhead.

[0114] In a moderately complex environment S2, this application selects an adaptive normalized energy detection algorithm. This algorithm normalizes the detection statistics by estimating background noise, increasing the detection probability from 92.8% to 94.3% and reducing the false alarm rate from 5.2% to 4.5%, while also reducing the computational cost compared to fixed multi-frame joint detection. This demonstrates that under conditions of slowly fluctuating noise, adaptive normalized detection is more suitable for the current environment than fixed multi-frame joint detection.

[0115] In the high-complexity low-frequency line spectrum environment S3, this application selects a low-frequency weak line spectrum trajectory accumulation detection algorithm. Compared with fixed multi-frame joint detection, this method can utilize the cross-frame continuity of the low-frequency line spectrum for trajectory accumulation, increasing the detection probability from 79.6% to 89.4% and reducing the false negative rate from 20.4% to 10.6%. This result shows that under the condition of low-frequency weak line spectrum targets, ordinary multi-frame energy accumulation alone is insufficient to fully utilize the line spectrum structure features.

[0116] In the highly complex broadband environment S4, this application selects a subband peak energy selective accumulation detection algorithm. This method does not indiscriminately accumulate energy across the entire frequency band, but instead prioritizes retaining candidate subbands where the target is dominant. Experimental results show that the detection probability increases from 77.1% to 86.1%, and the false alarm rate decreases from 10.5% to 6.2%, indicating that selective subband accumulation can reduce the impact of broadband noise and interference on the detection statistics.

[0117] In the highly complex template-correlated environment S5, this application selects a matched filtering detection algorithm. Since the target signal and the preset template have a correlated structure, matched filtering can enhance weak signal components consistent with the template. Compared with fixed multi-frame joint detection, the detection probability increases from 72.8% to 88.7%, and the false negative rate decreases from 27.2% to 11.3%, indicating that this method is more suitable for detecting template-correlated weak underwater acoustic signals.

[0118] In summary, while fixed multi-frame joint detection possesses a certain cross-frame accumulation capability, its fixed detection process makes it difficult to simultaneously adapt to weak underwater acoustic signals of low, medium, and varying morphologies. This application adaptively selects the detection algorithm based on the acoustic field complexity and the weak signal morphology, reducing computational overhead in low-complexity environments, maintaining stable detection performance in medium-complexity environments, and improving the detection capabilities for low-frequency weak line spectra, broadband weak energy, and template-correlated weak signals in high-complexity environments. Experimental results verify that this application has better detection adaptability and engineering application value in complex underwater acoustic environments.

[0119] The above provides a detailed description of the adaptive detection method for weak underwater acoustic signals based on acoustic field complexity provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of the claims of this invention. The above description of the disclosed embodiments enables those skilled in the art to implement or use this invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of this invention. Therefore, this invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A weak underwater acoustic signal adaptive detection method based on sound field complexity, characterized in that, Includes the following steps: S1. Acoustic signal acquisition and preprocessing; S2. Perform multidimensional feature analysis on the preprocessed underwater acoustic signal to extract multidimensional features that can reflect the complexity of the current acoustic field environment. S3. The multidimensional feature parameters extracted in step S2 are fused to generate a comprehensive complexity score, which assesses the complexity level of the current underwater acoustic environment. Based on the sound field complexity level obtained after the assessment, an adaptive detection algorithm decision is made. S4. Perform specific underwater acoustic signal detection tasks, detect and process weak underwater acoustic signals, and execute corresponding detection algorithms according to the sound field complexity level. S5. Output the detection results; In step S3, the multidimensional feature parameters obtained in step S2 are used to construct a complexity vector: ， wherein, denotes the spectral complexity; denotes the temporal fluctuation complexity; denotes the stationarity indicator; denotes the local SNR estimate; definition: ， ， ， in, , , All of these are feature quantities after nonlinear compression mapping; Used to reflect the degree of fluctuation of short-term energy in the time dimension; It is used to reflect the stability of the statistical structure of background noise and to suppress the influence of outliers through exponential mapping; Used to reflect the degree of inseparability between the target signal and the background noise; Overall complexity score The weighted nonlinear mapping method is used to represent it as follows: ， in, , , All are weighting coefficients. And satisfy ; Introducing exponential smoothing, we obtain the smoothed overall complexity score: ， in, This is the exponential smoothing coefficient, used to adjust the weight between the current frame's overall complexity score and historical smoothing scores. And the initial conditions are: ， Score based on smoothed overall complexity The current sound field complexity is divided into different levels: ， in, This is the first complexity threshold; It is the second complexity threshold, and satisfies ; When the smoothed overall complexity score is lower than the first complexity threshold When choosing a low-complexity detection algorithm When the smoothed overall complexity score is greater than or equal to And less than the second complexity threshold When choosing a detection algorithm with medium complexity. When the smoothed overall complexity score is greater than or equal to the second complexity threshold At this time, it enters a high-complexity detection mode; The set of high-complexity detection algorithms is as follows: ， in, An algorithm for accumulating and detecting low-frequency weak line spectrum trajectories. This is a sub-band peak energy selective accumulation detection algorithm. This is a matched filter detection algorithm; Calculate the morphological fit score for weak signals under high-complexity detection modes: ， in, For low-frequency line spectrum continuity adaptation fraction; For the significant fit score of broadband subband energy, Match the appropriate score to the template; Based on the above suitability score, the corresponding detection algorithm is selected from the set of high-complexity detection algorithms.

2. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 1, characterized in that, The discrete underwater acoustic signal is divided into continuous processing frames, with a frame length of 1. Sampling rate Then the number of sampling points per frame is ; The specific implementation process of step S1 is as follows: S1.

1. Collect acoustic signals in the underwater environment and complete the acoustic-to-electric conversion: The underwater acoustic sensor array receives underwater acoustic signals and performs acoustic-to-electric conversion, outputting analog signals. The signal is converted to a digital signal at a sampling rate of Discretize: ， in, This is the global sampling point index; For continuous signals With sampling rate The discrete sequence obtained by sampling; S1.2, Preliminary processing of the raw underwater acoustic signal: The discrete underwater acoustic signal is divided into continuous processing frames, the first... The intra-frame signal of a frame is represented as follows: ， in, For frame numbering, For intra-frame sampling point index, The number of sampling points per frame; For the first The frame signal undergoes frame-by-frame mean removal processing, the first... The average value of the frames is: , Signal after mean removal Represented as: ， Subsequently, the mean-removed signal is bandpass filtered to obtain the preprocessed signal. : ， in, These are the coefficients of the bandpass filter; This is the filter coefficient index, used to suppress low-frequency environmental noise and high-frequency electronic noise; This is the filter length; The filtered signal is normalized frame by frame: Define the first... Root mean square magnitude of the frame for: ， The normalized signal for: ， in, The extremely small positive numbers introduced to prevent the denominator from being zero; In step S2, the multidimensional statistical features reflecting the complexity of the current sound field environment include spectral distribution features, time-domain fluctuation features, noise stability features, and signal separability features. For the normalized first Frame signal Performing a short-time Fourier transform yields the frequency domain representation: ， in, Indicates the first Frame signal in the Frequency domain representation at each frequency point; Frequency index; For window functions; Calculate the normalized power spectral density based on the short-time spectrum: ， in, Indicates the first Frame signal in the The normalized power spectral density at each frequency index satisfies the probability distribution characteristics: ， Spectral complexity Represented by spectral entropy, it is defined as: ， in, Represents the logarithm to the base 2; Based on the filtered signal Constructing short-time energy sequences, the first The short-time energy of a frame signal is defined as: ， In length Within the time statistics window, define the average short-time energy. for: ， Define time fluctuation complexity for: ， in, The number of frames used for statistics within a time window, reflecting the degree of fluctuation of short-term energy within adjacent time windows; To characterize the stationarity of background noise, the first... The autocorrelation function of the frame-normalized signal: ， in, It is a delay quantity, and satisfies ; For the first Frame signal in delay The autocorrelation function under the following conditions; Define stationarity index for: ， in, This is the maximum delay length; To characterize the separability of the target signal from the background noise, local signal-to-noise ratio estimation is introduced. : ， in, For the target frequency band; This is the noise frequency band, and ; For the first Frame signal in target frequency band Frequency domain representation at the corresponding frequency point within the range; For the first Frame signal in noisy reference band Frequency domain representation at the corresponding frequency point within the range.

3. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 1, characterized in that, In step S4, the current state of the algorithm execution is: At that time, a single-frame energy detection algorithm was used; Definition of the first The detection energy of the frame signal is: ， Based on energy detection Make a binary hypothesis decision: ， in, This indicates the presence of a target signal; This indicates that the target signal does not exist; The threshold for single-frame energy detection; The single-frame energy detection algorithm is in the first... The decision output of the frame.

4. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 1, characterized in that, In step S4, the current state of the algorithm execution is: At that time, an adaptive normalized energy detection algorithm is adopted; Define the detection statistic as: ， in, The background reference power estimate is used to characterize the noise level in the current received signal, and its calculation process is as follows: For the first Spectral analysis is performed on the preprocessed frame signal to obtain its frequency domain representation. And calculate the spectral power: ， in, Indicates the first Frame number The unnormalized spectral power at each frequency point is used for background reference power estimation; In analyzing frequency bands Within this frame, the median of the spectral power is used as a temporary background baseline estimate for the current frame. ， The result obtained using smooth update is: ， in, This represents the smoothing coefficient for the background baseline power estimate. ; Based on statistics Judgment: ， in, This indicates the presence of a target signal; This indicates that the target signal does not exist; This is the normalized energy detection threshold; The adaptive normalized energy detection algorithm is represented in the first... The decision output of the frame.

5. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 1, characterized in that, In step S4, when the smoothed overall complexity score indicates that the current environment is a high-complexity environment, the morphology of the weak signals under the high-complexity environment is judged, and the corresponding detection algorithm is selected from the set of high-complexity detection algorithms: For continuous Analyze the short-time spectrum of the frame, and assume... For the first Frame number Normalized spectral energy at each frequency point For low-frequency analysis bands, It is a candidate line spectrum trajectory that satisfies the frequency continuity constraint between adjacent frames: ， in, The maximum allowable frequency drift range; The low-frequency line spectrum continuity fit fraction is defined as: , The target operating frequency band is divided into Individual belt: ， No. The number of belts in the first The energy saliency of a frame is defined as: ， in, For the first Individual belt; This represents the number of frequency points within the sub-band. For the first Estimators of the local background noise variance for each sub-band; For the first Frame signal in the Normalized power spectral density at each frequency index; Select the one with the highest energy significance Each sub-band constitutes a set The broadband subband energy saliency fit score is defined as: ， Let the pre-stored target weak signal template be , No. The signal after frame preprocessing is The template length is Template delayed search range is The template matching adaptation score is defined as follows: ， Let the maximum of the three fit scores above be: ， Let the second largest of the three fit scores above be: ， The algorithm selection rule for high-complexity environments is then expressed as: ， in, For signal morphology discrimination margin; This represents a low-frequency weak line spectrum trajectory accumulation detection algorithm; This represents the sub-band peak energy selective accumulation detection algorithm; This represents the matched filter detection algorithm.

6. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 5, characterized in that, In step S4, when selecting under high complexity environment At that time, a low-frequency weak line spectrum trajectory accumulation detection algorithm was used to detect weak underwater acoustic signals; For continuous Short-time spectrum analysis of the frame signal yields the first... Frame number Frequency domain amplitude at each frequency point To reduce the influence of local background fluctuations, the normalized time-frequency energy is defined as: ， in, For the first Frame number Local background energy estimation near each frequency point; Preserve the set of candidate spectral points in each frame: ,in, Set a threshold for selecting candidate line spectrum points; Assume continuous One candidate line trajectory within the frame is: And it satisfies the frequency continuity constraint between adjacent frames: ,in, This indicates that the candidate line trajectory is at the th... The frequency point index corresponding to the frame; This indicates that the candidate line trajectory is in the current position. The frequency point index corresponding to the frame; Define trajectory accumulation score for: ， in, This is the trajectory smoothing penalty coefficient; Corresponding detection statistics for: ， Based on statistics Judgment: ， in, Accumulate detection thresholds for low-frequency weak line spectrum trajectories; The low-frequency weak line spectrum trajectory accumulation detection algorithm is represented in the first... The decision output of the frame.

7. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 5, characterized in that, In step S4, when selecting under high complexity environment At that time, the sub-band peak energy selective accumulation detection algorithm was used to detect weak underwater acoustic signals; The target operating frequency band is divided into Individual belt: , No. The number of belts in the first The significance of the normalized energy of a frame is defined as follows: ， in, For the first Local background noise variance estimate corresponding to each sub-band For the first Number of frequency points within each sub-band; from Select subbands with higher energy significance. Each strip is denoted as: ， in, Indicates from The index with the largest value among the subband energy significance indicators is selected. The set of sub-band indexes corresponding to each indicator; For the first The set of selected candidate subbands in a frame; In continuous Within a frame, the energy of candidate sub-bands is selectively accumulated, and the detection statistic is defined as follows: ， in, Indicates the first The frame with the highest energy significance A set of candidate subbands; For the first Frame number The weight coefficients of candidate subbands, satisfying: , Based on statistics Judgment: ， in, The threshold for selective accumulation of subband peak energy; The subband peak energy selective accumulation detection algorithm is shown in the first... The decision output of the frame.

8. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 5, characterized in that, In step S4, when selecting under high complexity environment At that time, a matched filter detection algorithm was used to detect weak underwater acoustic signals; Let the pre-stored target weak signal template be , No. The filtered signal after frame preprocessing is The template length is The matched filter detection statistic is defined as follows: ， in, This is the template delay amount; Based on statistics Judgment: ， in, The detection threshold for matched filtering; This indicates that the matched filter detection algorithm is in the first... The decision output of the frame.

9. The adaptive detection method for weak underwater acoustic signals based on sound field complexity according to claim 5, characterized in that, In step S5, the first The final detection result of the frame is represented as follows: 。

Images